Datasets:

Hendrik-a

/

arxiv

like 0

we experimentally demonstrate phase - insensitive linear optical amplification which preserves the idler at the output . since our amplification operation is unitary up to small excess noise , it is reversible beyond the classical limit . the entanglement between the two output modes is the resource for the reversibility . the amplification gain of is demonstrated . in addition , combining this amplifier with a beamsplitter , we also demonstrate approximate cloning of coherent states where an anticlone is present . we investigate the reversibility by reconstructing the initial state from the output correlations , and the results are slightly beyond the cloning limit . furthermore , full characterization of the amplifier and cloner is given by using coherent states with several different mean values as inputs . our amplifier is based on linear optics , offline preparation of nonclassical ancillas , and homodyne measurements followed by feedforward . squeezed states are used as the ancillas , and nonlinear optical effects are exploited only for their generation . the ancillas introduce nonclassicality into the amplifying operation , making entanglement at the output .

1101.1139

quantum optics is governed by rules imposed by commutation relations which have to be kept during time evolution . optical amplification is no exception to this story . typically , the amplified output is suffered from inevitable excess noise . this limitation is quantum - mechanically imposed , thus does not depend on the specific realization methods . caves classified general linear amplification into phase - insensitive amplification ( pia ) and phase - sensitive amplification ( psa ) @xcite . he also systematically derived the quantum limit of excess noise for such general linear amplification with arbitrary gain from the requirement to preserve commutation relations . this excess noise originates from quantum fluctuations in the auxiliary system required to keep energy conservation . we concentrate on pia , supposing the target of amplification to be optical wave amplitude of a single mode , which is denoted by the term `` signal '' . classical counterpart of pia is a conversion of arbitrary complex wave amplitude @xmath1 into @xmath2 , where @xmath3 is the gain of amplification . as is found in ordinary textbooks , annihilation operators in quantum optics correspond to complex amplitudes in classical optics . therefore , we describe the amplifying process by the transformation of annihilation operators . quantum - mechanically optimal pia in the sense that the excess noise is minimized can be achieved by the following transformation @xcite : @xmath4 where @xmath5 and @xmath6 are the signal mode s annihilation operators before and after the amplification , respectively . there is an extra term @xmath7 which is introduced in order to meet the commutation relation of @xmath8=1 $ ] for both the input and output signal modes . here , @xmath9 is an arbitrary phase factor , and @xmath10 is another mode s annihilation operator in the auxiliary system . throughout this paper , the ancilla mode represented by @xmath10 is denoted by the term `` idler '' and distinguished from other ancilla modes . . becomes the input - output relation of optimal pia when the idler is in a vacuum state . the quantum fluctuation of the idler contaminates the amplified signal . this is the inevitable excess noise of pia . note that this penalty prevents amplification from being a loophole of the uncertainty relation in joint measurements @xcite . at the limit of high - gain amplification , we can see the famous @xmath11 db cost of the noise figure for pia of coherent states . in addition to this intrinsic excess noise , further nonintrinsic excess noise may be caused by other ancilla modes in nonoptimal pia . there are numerous practical realizations of optical amplification . doped fiber amplifiers ( dfas ) and semiconductor optical amplifiers ( soas ) utilize stimulated emissions @xcite , and raman amplifiers ( ras ) and optical parametric amplifiers ( opas ) utilize nonlinear optical effects . in principle , there is no quantum - mechanical reason to prevent these realizations from achieving the optimal pia in the form of eq . . however , the real devices with current technology are accompanied by further excess noises . recently , pia operating almost at the optimal level is experimentally demonstrated by josse _ et al . _ by utilizing feedforward @xcite . the reason for the high efficiency of josse s pia is that it does not require inefficient nonclassical operations or nonclassical ancillas . it uses a vacuum state as an ancilla which is present everywhere , and linear optics and homodyne measurements followed by feedforward which are highly efficient . although josse s pia is a good attainment , it is not the end of the story . the signal transformation in eq . is an irreversible thermalizing process . complete pia should have _ unitary _ realization on an expanded hilbert space . in order to unitarize pia , two - mode description is sufficient . the full input - output relation becomes as follows : [ eq : piaunitary ] @xmath12 note that the roles of the signal and idler are symmetric in this relation . the significance of unitarization must be the reversibility . the inverse transformation is easily derived when we take notice of the fact that eq . is equivalent to two - mode squeezing operation . a two - mode squeezing operation parametrized by @xmath13 is canceled by another two - mode squeezing operation where the squeezing direction is opposite , i.e. , @xmath14 . nonetheless , in many amplification schemes including josse s experimental demonstration @xcite , the idler output is lost in the inextractable environment , making the process irreversible . in order to realize idler - preserving and close - to - optimal pia , we require some nonclassicality for the amplifier . this is contrastive to josse s idler - nonpreserving pia which does not require any nonclassicality . a typical strategy to introduce nonclassicality into feedforward - based quantum circuits is to use nonclassical states as ancillas . continuous - variable ( cv ) quantum teleportation @xcite and cv error correction @xcite are good examples . in these examples , squeezed states are used as ancillas that support the performance beyond the classical limit , and the complex operations after the state preparation stage are efficiently implemented by linear optics . in this paper , by employing the feedforward - based scheme proposed in ref . @xcite , we demonstrate pia of coherent states which preserve the idler output . the scheme basically relies on linear optics including homodyne measurements and feedforward . squeezed vacuum states are used as ancillas , which inject nonclassicality into our pia . only for generating the nonclassical ancilla states , we resort to nonlinear optical effects . our demonstration is for the amplification gain of @xmath0 , which is tuned via passive optical devices and feedforward electric circuits . combining pia for @xmath0 with a half beamsplitter , we also demonstrate @xmath15 approximate cloning of coherent states , where an `` anticlone '' remains at the output . ( anticlone will be explained in sec . [ sec : clone ] . ) in principle , our amplifier and cloner becomes quantum - mechanically optimum at the limit of infinite squeezing of the ancillas . for the case of finite squeezing , as is the real situation in experiments , further excess noise invades in accordance to the level of the squeezing . however , the degradation is small enough to retain nonclassical features . the behaviors of our amplifier and cloner are fully characterized by using several coherent states as inputs . furthermore , we also pay much attention to the output correlations , because nonclassical properties clearly appear in them . for the pia experiment , we check the einstein - podolsky - rosen ( epr ) correlation between the signal and idler outputs . for the cloning experiment , we check bipartite entanglement between each clone and the anticlone , which as a whole proves tripartite entanglement of class i @xcite . moreover , for both experiments , the reversibility is investigated from the output correlations . our idler - preserving pia is significant in several respects . first of all , the reversibility will pave the way to new schemes . recently , there is a proposal of a cv quantum interface that enables in principle a unit fidelity of transfer using such reversible pia @xcite . moreover , the reversibility in cloning is also advantageous . cloning of unknown states is distribution of information , and its reversibility reserves the option to recover the distributed fragments of the information . this will be further discussed in sec . [ sec : clone ] . secondly , our pia would have some applications as two - mode squeezing operation . note that one - mode squeezing operation is already demonstrated successfully in ref . @xcite with similar approach . in this introduction , pia has been described together with a brief historical review . especially , the nonclassical property of pia is discussed , which is obscure in many amplification processes because the idler output is lost in the inextractable environment . the subsequent contents of this paper are as follows . in sec . [ sec : ffpia ] , feedforward - based pia is described , explicitly showing the excess noise due to finite squeezing of ancillas . in sec . [ sec : clone ] , cv quantum state cloning and its connection with pia are described . in sec . [ sec : setup ] , the experimental setup is described . in sec . [ sec : resultspia ] , the experimental results for pia of coherent states with @xmath0 are shown . in sec . [ sec : resultsclone ] , the experimental results for @xmath15 approximate cloning of coherent states are shown . in sec . [ sec : summary ] , our experimental achievements are summarized . in our definition , feedforward means that the operations after some measurements are changed depending on the measurement outcomes which in general are obtained randomly . in particular , in this paper it indicates phase space displacement operations whose amounts are proportional to the results of homodyne measurements . we know two specific schemes for feedforward - based pia that preserves the idler at the output . one scheme is proposed by filip _ et al . _ in ref . @xcite , in which pia is composed of two feedforward - based single - mode squeezers proposed in the same paper . the other scheme is proposed by josse _ et al . _ in ref . @xcite as a modification of the idler - nonpreserving pia . note that josse s idler - preserving pia is just a theoretical proposal and the idler - nonpreserving pia alone is experimentally demonstrated . both of filip s scheme and josse s scheme rely on linear optics including homodyne measurements and feedforward , and require offline - prepared nonclassical states as ancillas . moreover , in both schemes , the gain of amplification is accurately and stably determined via the choice of passive optical devices and correspondingly feedforward gains . as for the nonclassical ancillas , filip s scheme requires two single - mode squeezed states , on the other hand , josse s scheme requires a two - mode squeezed state . since two single - mode squeezed states can be converted to a two - mode squeezed state and vice versa by a half beamsplitter interaction , the amounts of the nonclassical resources required for the two distinct schemes are the same . for both schemes , the feedforward - based pia coincides with the quantum - mechanically optimal pia only at the limit of infinite squeezing of the ancillas . for the case of finite squeezing , excess noise contaminates the output to some extent . note that this is a common matter of feedforward - based cv deterministic processing @xcite . the difference between the two schemes proposed by filip and josse solely arises in this excess noise . for filip s scheme , it appears symmetrically in the signal and idler outputs . on the other hand , for josse s scheme , it appears only in the idler output . the better choice between different schemes depends on the specific application . we have chosen the symmetrical one . in the demonstration in sec . [ sec : resultspia ] , we confirm the symmetry of pia by swapping the roles of the signal and idler . [ sfig : ampschematic ] shows the schematic of our pia , from which the symmetry of the signal and idler is obvious . its details will be described in sec . [ sec : setup ] . here we give the input - output relation . in the following , the quadrature phase amplitudes of each optical mode are denoted by @xmath16 and @xmath17 , which correspond to the real and imaginary parts of the mode s annihilation operator @xmath18 , i.e. , @xmath19 . the phase factor @xmath20 in eq . can be arbitrarily changed by pre- and post - processing of phase rotation of the idler . therefore , we consider the case of @xmath21 without loss of generality . explicitly showing the excess noise coming from finitely squeezed ancillas , the input - output relation becomes as follows @xcite : @xmath22 the subscripts ` 1 ' and ` 2 ' represent the two main modes . they correspond to the signal and idler , though we do not specify which is which because the relation is symmetric . @xmath23 and @xmath24 denote the squeezed quadratures of the two ancilla modes . at the limit of infinite squeezing , these terms vanish , and the transformation above strictly coincides with the optimal pia . the amplification gain @xmath25 is determined via one parameter @xmath26 , which is a common reflectivity of two beamsplitters and in fig . [ sfig : ampschematic ] , with the relation of @xmath27 one - to - one correspondence of @xmath28 and @xmath29 is easily checked . note that the feedforward gain is also parameterized by @xmath26 . it is chosen so that the antisqueezed noises from the ancillas are canceled out at the output . for the demonstration of @xmath30 , the value of @xmath26 should be chosen as @xmath31 . the resulting input - output relation becomes as follows : [ eq : in - out_g2 ] @xmath32 it is known as the no - cloning theorem that an unknown quantum state @xmath33 can not be perfectly duplicated as @xmath34 @xcite . however , approximate cloning is possible , which can go beyond some classical limit in general . in this section , we will discuss cv cloning , and make its connection with pia . furthermore , its reversibility is discussed by introducing the notion of anticlone . in general , cloning can be described as a unitary operation supported by ancilla systems . the ancilla output system generally depends on the cloned state , from which anticlones are obtained . we show the equations for @xmath15 cloning , which corresponds to the experimental demonstration in sec . [ sec : resultsclone ] . however , pia allows general @xmath35 cloning in principle , which will be described in appendix [ sec : gencln ] . here , the notation @xmath35 means that @xmath36 clones are created from @xmath37 identical originals . first , we would like to say that pragmatic cloning for cv is not the universal cloning @xcite with respect to the infinite - dimensional hilbert space , because it is an unnatural situation that all the states in the noncompact space appears with equal probability . in general , the choice of the appropriate cloner depends on how the information is embedded in the infinite - dimensional hilbert space . the typical situation is that the cv information is embedded as a displacement on some quantum state @xmath33 . here , @xmath33 , which we refer to as a core state , is either known or unknown . then , the set of possible original states is @xmath38 , where @xmath39 $ ] is the displacement operator . [ the probability density @xmath40 is omitted because we consider for simplicity the case where @xmath41 is uniformly distributed . ] as a special case of this , the set @xmath42 becomes all coherent states when the core state @xmath33 is known to be a vacuum state . this way of embedding is found in ordinary cv quantum key distribution ( qkd ) protocols @xcite . for such protocols , the role of cloning is distribution of the information , rather than duplication of a quantum state . therefore , the measure of the cloning precision should be related to the estimation of @xmath41 , instead of the traditional fidelity . furthermore , asymmetric cloner is significant as well as symmetric cloner . arbitrary share ratio of the information is achieved by cloner with tunable asymmetry . we suppose a simple picture of cloning where some noise is added to the original state as the penalty of cloning . then , the quality of cloning is totally determined by this noise . for simplicity , we impose rotational symmetry on the noise added to each clone . this is naturally justified when the core state @xmath33 is either known to be symmetric or unknown . the added noise is characterized by its variance @xmath43 @xcite , where @xmath44 for @xmath15 cloning . note that @xmath45 corresponds to the mean photon number of thermalization in the clone . the variances @xmath45 are directly connected to the mean square errors in the estimation of @xmath41 . therefore , a measure can be constructed from them . given the desired asymmetry , the cost function is determined @xcite : @xmath46 the positive parameters @xmath47 determine the asymmetry . ( @xmath48 corresponds to symmetric cloning . ) the cloner that minimize the cost function is the optimal . it is obvious that the optimal cloner is gaussian when the cost function is set as a function of the noise variances as in eq . . in order to minimize it , the ancillas that support cloning are chosen in minimum uncertainty states , which are gaussian . this contrasts with the evaluation by the fidelity . non - gaussian cloning can slightly go beyond the gaussian fidelity for coherent states @xcite . we emphasize that the evaluation by the variances is more practical . it is pointed out that optimal attack in qkd is gaussian @xcite . there is a restriction on the excess noises @xmath45 which is imposed by quantum mechanics @xcite : @xmath49 the optimal cloner with respect to the cost function in eq . satisfies the equality in eq . by necessity . this noise penalty comes from consistency with the uncertainty relation . the attainable information of the original state does not increase by cloning due to this noise . recall that the inevitable noise in pia comes from the same reason . indeed , the optimal cloner can be constructed from the optimal phase - insensitive amplifier and beamsplitters . for example , @xmath15 cloning with arbitrary asymmetry is achieved by putting an amplifier in one of the arms of a mach - zehnder interferometer @xcite . especially , for symmetric cloning , the reflectivity of the first beamsplitter becomes unity , i.e. , it is achieved by first amplifying the original with @xmath30 and then splitting the amplified signal in half . this procedure can be extended to @xmath35 cloning @xcite . the optimality of this realization is proven with respect to the cost function in eq . @xcite . for gaussian cloning of coherent states , the added noise variance @xmath45 and the fidelity @xmath50 have correspondence as @xmath51 . by using eq . , the upper limit of fidelity is obtained for arbitrarily asymmetric gaussian cloning . in particular , it becomes @xmath52 for the symmetric case . this is significantly higher than the classical limit of @xmath53 , where we regard the limit of state estimation as the classical limit of symmetric cloning because the estimated state is classical information which can be copied any number of times . note that the sameness of state estimation and asymptotic cloning where the number of clones tends to infinity is proven for a general set @xmath42 of possible original states @xcite . we refer to these fidelities only for the consistency with previous works . we stress again that our actual interest is the variances . we have seen above that the clones are made of the signal output of the amplifier . when cloning is unitarily realized , we still have the idler output , whose state is affected by the original state . now we pay attention to this ancilla output system . as mentioned at the beginning of this section , anticlones are byproducts of cloning which are obtained from the ancilla systems . especially , for @xmath15 cloning , the idler output itself is an anticlone . it is an approximation of the phase - conjugated original state , or in other words , the output of an approximate not gate . qubit version of this gate is demonstrated in ref . @xcite . anticlones are important when we are concerned with the reversibility . the originals can in principle be perfectly reproduced only when all the clones and anticlones are present . for the reversibility , the essential resource is nonclassical correlations , or entanglement . conceptually , the excess noises in cloning is canceled by using the nonclassical correlations . therefore , we discuss the existence of entanglement in the three - mode output system of @xmath15 cloning . clones are obtained by splitting the amplified signal , thus there is no entanglement among clones . however , there is entanglement between each clone and the anticlone . we stress that the resource for the recovery is not the anticlones themselves but the entanglement . we can obtain anticlones without entanglement with clones as follows . suppose the situation where two independent cloners are running , and the same states are used as the inputs for them . then , the clones obtained from one cloner do not have entanglement with the anticlones from the other . in this case , the originals can not be recovered from the noncorrelated outputs . for the recovery of the originals , the inverse unitary operation is not required . the optimal cloning can be fully reversed by the bell measurement on a clone and an anticlone and subsequent feedforward to the remaining single clone @xcite . note that this recovery scheme works not only for coherent states but also for an arbitrary core state @xmath33 . this scheme is efficient from two points of view . one is on a technical level that the homodyne measurements and feedforward displacement operations are quite efficient with current technology . the other is on a conceptual level that the performer of the bell measurement and the owner of the remaining clone who is willing to recover the original can be spatially separated . for this case , they only need classical channels for communication , and never quantum channels . note that even partial reversal is possible with a similar scheme based on local operations and classical communication ( locc ) , which converts , e.g. , symmetric clones to asymmetric clones @xcite . we would like to discuss practical aspects of cloning and its reversibility assisted by classical communication . as described above , cloning of a quantum state is regarded as distribution of information among plural participants . the information of the original is to some extend accessible to individual participants . this situation is clearly distinguished from that found in usual quantum error correcting codes where the quantum information is mapped on a larger hilbert space so that no information about the original is accessible from a localized system . such share of information would play important roles in several scenarios , in which the reversibility would give a tactical aspect to information exchange . for example , cloning is a possible attack by an eavesdropper in qkd . in this example , the reversibility of cloning provides the opportunity for the communicators to negotiate with the eavesdropper when they know the attack @xcite . since coherent states are a strong candidate for the information carrier in quantum communication , cloning of coherent states is especially of great significance . there are several experimental previous works which demonstrate cloning of coherent states beyond the classical limit of @xmath53 in non - reversible ways , i.e. , their anticlones are lost in the environment . in ref . @xcite , using feedforward - based pia of ref . @xcite , almost quantum - limited @xmath15 cloning is demonstrated . in ref . @xcite , telecloning is demonstrated where the original coherent state is teleported and cloned at the same time . in sec . [ sec : resultsclone ] , we demonstrate @xmath15 symmetric gaussian cloner which preserves an anticlone at the output . as is shown in fig . [ sfig : clnschematic ] , we apply a half beamsplitting to the signal output of the feedforward - based pia with the gain @xmath30 described in sec . [ sec : ffpia ] . in the demonstration , the reversibility is checked from the output correlations . to our knowledge , there is no previous experiment of this kind even in qubit regime . although our demonstration is only for coherent states , our cloner should equally work for arbitrary core state @xmath33 as discussed above . we close this section by giving the input - output relation of the optimal @xmath15 symmetric cloning . by substituting @xmath30 and @xmath21 into the input - output relation in eq . and splitting the signal output in half , we obtain , @xmath54 where the subscripts ` org ' , ` ' , ` ' , and ` ' denote the original , first clone , second clone , and anticlone , respectively . the subscript ` idl ' denotes the idler input for pia which is in a vacuum state . the annihilation operators with the subscript ` vac ' indicate another ancilla in a vacuum state , which invades from the empty port of the final half beamsplitter . for the excess noise of the two clones , @xmath55 is easily checked . therefore , this cloner is optimum when evaluated by the cost function in eq . with @xmath48 . when pia is realized with a feedforward - based scheme as is found in sec . [ sec : resultsclone ] , further excess noise contaminates the output in accordance with the squeezing levels of the ancillas . schematic of the experimental setup for pia is illustrated in fig . [ sfig : ampschematic ] , and that for approximate cloning is illustrated in fig . [ sfig : clnschematic ] . the light source is a ti : sapphire laser , which has a continuous - wave single - mode output of @xmath56 nm in wavelength and about @xmath57 w in power . we treat the quantum states of narrow sidebands located at @xmath58 mhz apart from the optical carrier frequency . two main beams that go from and to and in fig . [ sfig : ampschematic ] carry the quantum states which are targets of pia . the setup has a form of a mach - zehnder interferometer that holds a single - mode squeezer ( or ) in each arm . this decomposition of unitary pia into squeezers and beamsplitters is derived from the bosonic version of bloch - messiah reduction shown in ref . we note that this setup is almost the same as that for quantum nondemolition ( qnd ) interaction demonstrated in ref . this fact shows the capability of our setup to realize many types of two - mode gaussian interaction . combining pia for @xmath0 with another half beamsplitter as is shown in fig . [ sfig : clnschematic ] , @xmath15 approximate cloning of coherent states is achieved . and are feedforward - based squeezers , which are theoretically proposed in ref . @xcite and experimentally demonstrated in ref . @xcite . each squeezer consumes an ancilla in a squeezed state , which is generated by an optical parametric oscillator ( opo ) . note that several essential optical elements are omitted from fig . [ fig : schematic ] , such as a second harmonic generation ( shg ) cavity to generate pump beams for opos , and three spatial - mode cleaning cavities ( mccs ) . one mcc is used for local oscillators ( los ) for homodyne measurements and auxiliary beams for feedforward displacements . the other two mccs are used for individual input beams . the experimental procedure is divided into three steps : firstly , we prepare input coherent states and ancilla squeezed vacuum states . secondly , we implement pia and cloning via feedforward . finally , the output states are homodyne measured for verification . in the following , we describe the experimental details of each step . at this step , we generate coherent states which are used as inputs , and squeezed vacuum states which are used as ancillas . the nonzero mean values of the sideband coherent states at @xmath58 mhz are produced by appropriately modulating the optical carriers . in our setup , the relative phase of interference at each beamsplitter is designed to be fixed with active feedback control . therefore , in order to make an arbitrary phase space displacement in the input modes , both amplitude modulation ( am ) and phase modulation ( pm ) are utilized . am and pm make non - zero mean values of @xmath59 and @xmath60 for the first input mode , and those of @xmath61 and @xmath62 for the second input mode , respectively . each of four electro - optic modulators ( eoms ) before pia in fig . [ sfig : ampschematic ] corresponds to one of these four quadratures . on the other hand , in fig . [ sfig : clnschematic ] , only two eoms are depicted before pia which are both located at the first input beam path . therefore , the symmetry of the two input modes is broken in the cloning experiment . one input mode is the target of cloning , and the other input mode is set in a vacuum state throughout . for both experiments , the modulations are switched on and off in order to use several coherent states as inputs . after these eoms , there are mccs though they are omitted from fig . [ fig : schematic ] . squeezed vacuum states are each generated by an opo which is driven below the threshold . our opo has a bow - tie shaped configuration with a round - trip length of about 500 mm . it contains a periodically - poled ktiopo@xmath63 ( ppktp ) crystal as a nonlinear optical medium , which is commercially available from raicol and has 10 mm length and 1 mm by 1 mm cross section . the experimental datails of our opo squeezing are found in ref . @xcite . the squeezing level with the pump of about @xmath64 mw is about @xmath65 db relative to the shot noise level at @xmath58 mhz . the pump beams for the opos are the second harmonic of a fundamental beam , generated by a shg cavity . most of the ti : sapphire laser output is sent to the shg cavity , whose output of about @xmath66 mw is divided into two to pump the individual opos . the shg cavity has almost the same configuration as that of opos , whereas a knbo@xmath67 ( kn ) crystal is used instead of the ppktp crystal . modulation sidebands other than @xmath58 mhz are exploited for active feedback control of optical interferences . a modulation at @xmath68 mhz is utilized for locking cavities , including the shg cavity , the two opos , and the three mccs . on the other hand , lower frequency modulations at @xmath69 khz and @xmath70 khz are utilized at the opos to lock the phases of the pump beams . furthermore , the two input beams are modulated at @xmath71 khz and @xmath72 khz . these four low - frequency modulations contribute to the lock of the downstream interferometric system in the subsequent steps , as mentioned later . the two input beams are combined at a preceding half beamsplitter ( ) and then sent to and . after the squeezing operations , the two beams interfere again at another half beamsplitter ( ) , which completes pia . by splitting one of the two output beams by another half beamsplitter ( ) , @xmath15 cloner is obtained . the squeezing procedure goes as follows . first , the main beam is combined with an ancilla beam coming from an opo at a beamsplitter ( or ) . next , one of the two beams after the beamsplitter is homodyne measured . finally , the measurement outcome is fed forward to the remaining beam . and have the common reflectivity of @xmath26 . this parameter @xmath26 determines the degree of the feedforward - based squeezing and thus the gain of amplification @xmath25 with the relation shown in eq . . as is already mentioned , @xmath73 for our demonstration of @xmath0 . the feedforward operation is a phase space displacement whose amount is proportional to the random outcome of the homodyne measurement . the electric signal from the homodyne detector is sent to an eom to be converted into an optical signal , where the gain and phase at @xmath58 mhz are carefully chosen . the auxiliary beam which is modulated by the feedforward eom has the power of @xmath74 @xmath75w , @xmath76 of which subsequently enters the mainstream via an asymmetric beamsplitter ( 99:1 ) . the powers of the two input beams are @xmath77 @xmath75w , and those of the two ancilla beams are @xmath78 @xmath75w . these powers are considerably smaller than @xmath11 mw of los used for homodyne detections . the four beamsplitters of pia ( , , , and ) are actually composed of two polarization beam splitters and a half - wave plate in the same manner as the qnd experiment in ref . @xcite . their reflectivities are variable by rotating half - wave plates . they enable us to measure the input states as well as the output states with the same homodyne detectors for verification . the propagation losses of two main beams are measured to be @xmath79 on average , which mostly come from these variable beam splitters . in order to control the relative phases at beamsplitters with active feedback , interferences between the carriers and the low - frequency modulations are monitored . this is typically done by picking up @xmath76 of the beam after the interference , though such details are omitted from fig . [ fig : schematic ] . for each locking point , an appropriate modulation sideband is chosen , and the error signal is extracted from the interference between the carrier and the sideband by demodulation . however , and are exceptions , where the interference between two modulation sidebands , namely @xmath71 khz and @xmath72 khz , is exploited . the beat frequency of @xmath80 khz is chosen for the reference signal of demodulation to obtain the error signals . pia is characterized by measuring two - mode input states as well as two - mode output states using two homodyne detections . in the cloning experiment , on the other hand , three - mode output states are compared with single - mode input states . the input states are measured by setting the reflectivities of the four variable beamsplitters to unity and disabling the feedforward . the quantum efficiency of a homodyne detector is about @xmath81 , and the dark noise is about @xmath82 db below the optical shot noise produced by the lo . the interference visibilities to the los are @xmath83 on average . the outcomes of the final homodyne measurements are analyzed in either of the two ways below . in one way of analysis , the quadrature data are directly treated , which are obtained by lock - in detection of @xmath58 mhz components of the homodyne outputs . a signal from a homodyne detector is mixed with the reference signal at @xmath58 mhz , and then low - pass filtered with the cutoff of @xmath84 khz . subsequently it is analog - to - digital ( a / d ) converted for storage with the sampling rate of @xmath66 khz and the resolution of @xmath85 bits ( , national instruments corporation ) . in this analysis , the phase of the homodyne detection is slowly scanned . the phase information is stored simultaneously with the quadrature values using the same a / d board . from the resulting marginal distributions , phase space distributions ( i.e. , wigner functions ) are reconstructed , where we assume that all the quantum states obtained in the experiments are gaussian . the first and second moments are computed so that the likelihoods are maximized . the other way is the power analysis at @xmath58 mhz using a spectrum analyzer . in this analysis , the measured quadratures are set to either @xmath16 or @xmath17 . not only the powers of the output quadratures but also those of their correlations are measured for several input coherent states . the resolution bandwidth is @xmath84 khz , the video bandwidth is @xmath66 hz , the sweep time is @xmath86 s , and @xmath87 times averaging is taken for each trace . note that we can easily see the effect of the hermitian conjugate term in eq . as a mirror image with the former way of analysis , whereas we can not do this with the latter . the two main modes are denoted by `` '' and `` '' . one of them is the `` signal '' and the other is the `` idler '' , which are initially in a coherent state and a vacuum state , respectively . by swapping the role of the signal and idler , we check the symmetry of our pia . + + we first show the results of the lock - in detection , because it is intuitively easier to see . figs . [ fig : amp_md1 ] and [ fig : amp_md2 ] show the experimental quadrature values at various phases of los . for fig . [ fig : amp_md1 ] , the is the signal and the is the idler , whereas for fig . [ fig : amp_md2 ] , the is the signal and the is the idler . there are three subfigures corresponging to the signal input ( a ) , the signal output ( b ) , and the idler output ( c ) . horizontal axes are the measurement phases @xmath88 which are scanned from @xmath89 to @xmath90 . the quadrature at @xmath91 corresponds to @xmath16 and that at @xmath92 does to @xmath17 . vertical axes are normalized quadrature values where the standard deviation of vacuum fluctuation is @xmath93 . each set of data is taken for about @xmath94 seconds . quadrature data are plotted every @xmath77 points in the figures , whereas the whole data are used for the analysis . the sinusoidal curve of the signal input represents the nonzero mean amplitude of a coherent state , and the fluctuation around the sinusoid represents the quantum noise . with regard to the fluctuation , it grows uniformly at both the signal and idler outputs . this uniformity is an evidence of the phase - insensitivity of our amplifier . on the other hand , with regard to the sinusoidal curve , the two output modes show different behaviors . at the signal output , the amplitude of the sinusoid is amplified from that of the signal input , maintaining the phase . at the idler output , the amplitude of the sinusoid is the same as the signal input , whereas the phase is flipped . this flip is due to the hermitian conjugate term in eq . . for both figures , the same qualitative behaviors as mentioned above are observed . [ fig : amp_ps1 ] and [ fig : amp_ps2 ] are phase space diagrams , which are computed from the quadrature data shown in figs . [ fig : amp_md1 ] and [ fig : amp_md2 ] , respectively . the experimental results ( a ) and the theoretical calculations for the optimal pia ( b ) are depicted next to each other . in the theoretical calculations , the experimental value is used for the amplitude of the signal input . the first and second moments are expressed by ellipses , which correspond to the cross sections of the wigner functions . note that the theoretical ellipses in ( b ) are strictly circles . in each phase space diagram , there are three ellipses . the ellipse in green is the signal input . its radius is almost @xmath93 which corresponds to the standard deviation of vacuum fluctuation . the ellipse in red is the signal output . its center is about @xmath95 times farther away from the origin and its radius is about @xmath96 times larger than those of the signal input . the ellipse in blue is the idler output . its radius is almost the same as that of the signal output , whereas its center is flipped around the @xmath97-axes from that of the signal input , which again represents the hermitian conjugate term in eq . . db squeezed ancillas . green : theory for our pia outputs with vacuum ancillas . ] in principle , we can fully characterize our pia with only the above way of analysis that treats quadrature values directly . however , such treatment requires a large amount of data for good accuracy . thus , in the following , we resort to power measurements using a spectrum analyzer . not only output quadratures ( shown in figs . [ fig : amp_v ] and [ fig : amp_c ] ) but also their correlations ( shown in figs . [ fig : amp_epr ] , [ fig : amp_rv ] , and [ fig : amp_rc ] ) are taken for several input states . in each figure , results of each quadrature are contained in one of boxes . vertical axes are powers in db scale which are normalized by corresponding shot noises . [ fig : amp_v ] shows the experimental results for vacuum inputs ( fluctuating traces ) , together with their theoretical expectations ( straight lines ) . there are four boxes corresponding to the four output quadratures , namely , @xmath98 , @xmath99 , @xmath100 , and @xmath101 . the traces in blue are the powers of the output quadratures . the traces in cyan around @xmath89 db are the powers of the shot noises , which are used for normalization . since the inputs are in vacuum states , the powers of the shot noises correspond to those of the input quadratures @xmath59 , @xmath60 , @xmath62 , and @xmath61 . we put three kinds of theoretical lines corresponding to three different conditions . for the optimal pia of a coherent state with @xmath0 , the output quadrature variances become three times larger than the initial shot - noise - limited variance , where two from amplification and one from contamination by the other mode . the corresponding @xmath102 db is marked by the lines in red . our pia with finite ancilla squeezing is suffered from further excess noise . assuming @xmath65 db of squeezing for the ancilla states , we calculate theoretical values which are marked by the lines in magenta . we also show them with vacuum ancillas by the lines in green . the lines in red , magenta , and green are very close to each other , thus other experimental errors are dominant rather than the ancilla squeezing levels in these results . therefore , it is hard to discuss nonclassicality of our pia only from these results . as is shown later , the effects of ancilla squeezing more clearly appear in the output correlations . next we use several coherent states as inputs . the results are shown in fig . [ fig : amp_c ] . the four input quadratures @xmath59 , @xmath60 , @xmath62 , and @xmath61 are displaced from zero mean values by turns , leaving the other three quadratures at the vacuum level . there are four subfigures labeled from ( a ) to ( d ) corresponding to such four excitations . for each subfigure , there are five boxes . the trace in magenta in the leftmost box shows the measured power of the excited input quadrature . the other four boxes correspond to the four output quadratures , namely , @xmath98 , @xmath99 , @xmath100 , and @xmath101 . the traces in red show the output quadrature powers with the input excitation . they are compared to those without the excitation shown by the blue traces , which are replottings of the blue traces in fig . [ fig : amp_v ] . the obtained results show the following features . when a quadrature @xmath16 or @xmath17 of an input mode is excited , the same quadratures of both of the output modes are excited , whereas the conjugate quadratures do not change from the nonexcited levels . the two increased output powers differ by about @xmath103 db , where the larger one corresponds to the amplified signal and the smaller one corresponds to the phase - conjugated idler output . these features are exactly what are expected from eq . for @xmath30 and @xmath21 . note that the coefficients @xmath95 correspond to the @xmath103 db . the results in figs . [ fig : amp_v ] and [ fig : amp_c ] are only for the five specific input states . however , the results for other input states can be predicted on the assumption of linearity . more precisely speaking , the absolute values of the coefficients of @xmath59 , @xmath60 , @xmath62 , and @xmath61 in eq . are determined from these results . on the other hand , the signs of the coefficients are not determined from them . however , they are checked from the phase space diagrams shown in figs . [ fig : amp_ps1 ] and [ fig : amp_ps2 ] . in the above sense , the results shown so far give full information of the input - output relation when output modes are separately concerned . in order to fully characterize our amplifier , the individual behaviors of the output modes are not enough . in the following , we are concerned with the output correlations . since unitary pia is equivalent to two - mode squeezing , the two output modes should be entangled , and have an epr type of correlation . the results of epr correlation are shown in fig . [ fig : amp_epr ] . here the two input modes are both in vacuum states . there are two boxes corresponding to @xmath97 and @xmath104 correlations . the lower traces in blue show the two - mode squeezing of @xmath105 and @xmath106 , on the other hand , the upper traces in blue show the two - mode antisqueezing of @xmath107 and @xmath108 , respectively . they are compared with the summed shot noises of the two homodyne detections shown by the traces in cyan . several theoretical lines are plotted together . the lower and upper lines in red are the theoretical values of two - mode squeezing and antisqueezing for the optimal pia , respectively . our results of two - mode squeezing are degraded from the ideal case due to finite squeezing of ancillas . assuming @xmath65 db squeezing for ancillas , theoretical expectation is marked by the lines in magenta . that for vacuum ancillas is also marked by the lines in green , which is exactly equal to the shot noise level . in contrast , the theoretical two - mode antisqueezing is always ideal for arbitrary ancillas . the experimental results well agree with the theory assuming @xmath65 db squeezing of ancillas . since the lower traces in blue are both below the traces in cyan , existence of entanglement is verified between the two output modes via the duan - simon criterion @xcite . and @xmath106 . upper blue : two - mode antisqueezing in @xmath107 and @xmath108 . cyan : summed shot noises . lower red : theory of two - mode squeezing for optimal pia outputs . upper red : theory of two - mode antisqueezing for optimal pia outputs . magenta : theory of two - mode squeezing for our pia outputs with @xmath65 db squeezed ancillas . green : theory of two - mode squeezing for our pia outputs with vacuum ancillas . ] from the nonclassical correlation between the two output modes , we investigate the reversibility of our pia . for this purpose , we virtually realize the inverse transformation electrically and reconstruct the initial quadratures . neglecting the excess noise from finite ancillas , pia that we demonstrate has the input - output relation as @xmath109 , @xmath110 , which is obtained by substituting @xmath30 and @xmath21 into eq . . the inverse transformation becomes as @xmath111 , @xmath112 , or equivalently , [ eq : piainv ] @xmath113 therefore , by adding or subtracting the two homodyne outcomes with @xmath103 db difference of gains , the initial quadratures are reconstructed . the reconstructed quadratures are denoted by the superscripts `` rec '' in the following . note that the initial quantum state is not recovered in the experiment . in addition , note also that only one of two quadratures @xmath16 or @xmath17 can be reconstructed in each moment , and never both simultaneously . the recovery of the initial state is possible only when either a quantum channel is between the signal and idler outputs @xcite or linear pre - processing is applied @xcite . however , the demonstrated reconstruction of the initial quadratures can show that necessary correlations for the recovery of the initial state are present . db squeezed ancillas . green : theory for our pia with vacuum ancillas . ] in fig . [ fig : amp_rv ] , the results of such reconstruction of initial vacuum fluctuations are shown . there are four boxes corresponding to the four reconstructed quadratures @xmath114 , @xmath115 , @xmath116 , and @xmath117 . the powers of the reconstructed quadratures are shown by the traces in blue . the traces in cyan are the powers of the summed shot noises of the two homodyne detections , which are taken with the same electric gains as the traces in blue . note that the lower levels of the blue traces than the cyan traces are due to the nonclassical correlation . from eq . , the summed shot noises should have three times larger variances than those corresponding to the initial vacuum fluctuations . thus , we infer the original vacuum level to be @xmath102 db below the measured sum of the shot noises . all results shown here are normalized by the inferred vacuum level . for the optimal pia , vacuum fluctuations are perfect reconstructed , thus the theoretical expectation coincides with the vacuum level of @xmath89 db , which is marked by the lines in red . the increases of the blue traces from @xmath89 db show the imperfection of our pia . the theoretical values for @xmath65 db squeezing of and no squeezing of ancillas are marked by the lines in magenta and green , respectively . the experimental results of the traces in blue are in good agreement with the lines in magenta . next , we pay attention to the reconstruction of mean amplitude , using coherent states as inputs . the results are shown in fig . [ fig : amp_rc ] . the four input quadratures , namely @xmath59 , @xmath60 , @xmath62 , and @xmath61 , are excited one by one , as shown in the leftmost boxes of four subfigures . for each excitation , the powers of the four reconstructed quadratures are measured , the results of which are put in the other four boxes on the right side . for the leftmost box , the trace in magenta is the power of the excited input quadrature , and the trace in cyan is the power of the shot noise of the corresponding homodyne detection . the increase of the magenta trace from the cyan trace indicates the excitation . for the other four boxes , the traces in red and blue show the powers of the reconstructed quadratures with and without the input excitation respectively , and the traces in cyan are the summed shot noise powers of the two homodyne detections . similarly to fig . [ fig : amp_rv ] , one third of the summed shot noise power is used for normalization . the reconstructed quadratures are excited almost to the same levels as the input quadratures at the excited quadratures , whereas they remain unchanged from the nonexcited levels at the nonexcited quadratures . all the results shown above prove the success of our demonstration of pia . they agree well with the theoretical calculations assuming @xmath65 db of squeezing for ancillas . we show next the results of the cloning experiment in a manner similar to the pia experiment , i.e. , quadrature data and the phase space diagrams reconstructed from them are used for intuitive understanding and also for check of the mirror image that is found in the anticlone , and the full verification is given by the power analysis for various input states . in the following , since the representation and interpretation of results are almost the same as those for the pia experiment , we give only short description of them . the signal input of pia is denoted by the term `` original '' , and the resulting two clones and one anticlone are denoted by `` clone-1 '' , `` clone-2 '' , and `` anticlone '' , respectively . there are also abbreviation of these terms as `` org '' , `` cln-1 '' , `` cln-2 '' , and `` a - cln '' in the figures and mathematical expressions . occasionally , the term `` input '' is used indicating the original , and `` output '' indicating the two clones and one anticlone . + fig . [ fig : cln_md ] shows the quadrature data . from an original ( a ) in a coherent state , two clones ( b ) and ( c ) and an anticlone ( d ) are produced . we see that the original and the two clones have almost the same sinusoidal curves of the mean amplitudes , though the fluctuations are uniformly increased in the clones . on the other hand , the anticlone has the same sinusoid when the phase is flipped . [ fig : cln_ps ] is the phase space diagram computed from the quadrature data shown in fig . [ fig : cln_md ] . the first and second moments of each distribution are represented by an ellipse . next to the experimental diagram ( a ) , theoretical calculations for the optimal cloning is depicted ( b ) . the two ellipses of the clones ( red and magenta ) in the experimental diagram are almost overlapped . the centers of the two clones are almost the same as that of the original ( green ) , whereas the radii of the clones are larger than that of the original . on the other hand , the anticlone ( blue ) has a different center , where the sign of @xmath104 quadrature is opposite to that of the original . we move on to the power analysis . first we show the powers of the output quadratures ( shown in figs . [ fig : cln_ov ] and [ fig : cln_oc ] ) , and then their correlations ( shown in figs . [ fig : cln_epr ] , [ fig : cln_rv ] , and [ fig : cln_rc ] ) . fig . [ fig : cln_ov ] shows the cloning of a vacuum state , and fig . [ fig : cln_oc ] shows the cloning of several coherent states . in fig . [ fig : cln_ov ] , there are six boxes corresponding to six output quadratures , namely , @xmath118 , @xmath119 , @xmath120 , @xmath121 , @xmath122 , and @xmath123 . for each box , there are two experimental traces . the traces in blue are the powers of the output quadratures . the traces in cyan are the shot noise powers used for normalization , which are equal to the powers of the input quadratures . there are also three kinds of theoretical lines . the lines in red are for the optimal cloning , and the lines in magenta and green are for our cloning using @xmath65 db squeezed and vacuum ancillas for pia , respectively . note that the lines in red at @xmath103 db for clones correspond to the cloning limit . from these results , cloning fidelities are estimated at @xmath124 for a vacuum original , which is higher enough than the classical limit of @xmath53 and very close to the cloning limit of @xmath52 . in fig . [ fig : cln_oc ] , there are two subfigures ( a ) and ( b ) corresponding to excitations in @xmath125 and @xmath126 , respectively . each subfigure is composed of seven boxes , where the leftmost one shows the input excitation in @xmath125 or @xmath126 , and the other six boxes show the output quadratures of @xmath118 , @xmath119 , @xmath120 , @xmath121 , @xmath122 , and @xmath123 . the trace in magenta shows the power of the excited input quadrature . the traces in red and blue show the output powers with and without the input excitation , respectively . the traces in cyan are the shot noise powers used for normalization . when we excite one quadrature of the original , the same quadratures in the three output modes are excited to almost the same level , whereas the conjugate quadratures do not change . db squeezed ancillas . green : theory for our cloner with vacuum ancillas . ] our remaining concerns are the output correlations and the reversibility . first , in fig . [ fig : cln_epr ] , we show epr correlation between each clone and the anticlone as a sufficient condition for entanglement . the correlation between the clone-1 and the anticlone is shown in fig . [ sfig : cln_epr_1 ] , and that between the clone-2 and the anticlone is shown in fig . [ sfig : cln_epr_2 ] . by electrically adding or subtracting the two homodyne signals with the same electric gains , four observables are measured , namely , @xmath127 , @xmath128 , @xmath129 , and @xmath130 , which are separately contained in boxes and shown as the blue traces . these traces are all below the summed shot noises of the traces in cyan . from these results , we verify bipartite entanglement between each clone and the anticlone from duan - simon criterion @xcite , and eventually , tripartite entanglement of class i where none of three partial systems is separable from the others @xcite . theoretical lines in red , magenta , and green are plotted together , corresponding to infinite squeezing , finite squeezing of @xmath65 db , and no squeezing of ancillas for pia , respectively . the experimental results of the traces in blue well agree with the lines in magenta . db squeezed ancillas . green : theory for our cloner with vacuum ancillas . ] using the nonclassical correlations , we reconstruct the original quadratures . the reconstructed quadratures are denoted by @xmath131 and @xmath132 . the results for a vacuum state are shown in fig . [ fig : cln_rv ] , and those for coherent states are shown in fig . [ fig : cln_rc ] . for the reconstruction , three homodyne signals are added with the same electric gains and appropriate signs . for the similar reason to the pia experiment , the summed shot noise has three times larger variance than that corresponding to the vacuum fluctuation of the original . thus , one third of the summed shot noise power is used for normalization . in fig . [ fig : cln_rv ] , the powers of the reconstructed vacuum fluctuations are plotted as the traces in blue , and compared to that of the summed shot noise plotted as the traces in cyan . the blue traces are below the cyan traces for both @xmath131 and @xmath132 due to nonclassical correlations . theoretical expectations are also shown as the lines in red , magenta , and green corresponding to the three different conditions of infinite squeezing , finite squeezing of @xmath65 db , and no squeezing of ancillas , respectively . the perfect reconstruction corresponding to @xmath89 db is marked by the red lines , which is not achieved in the experiment due to the finite squeezing of ancillas . the results are degraded almost to the level of the magenta lines , as expected from the theory . however , they are still slightly below @xmath103 db which corresponds to the cloning limit . from these results , the fidelity of reconstruction is calculated for a vacuum state . a perfect unitary cloning allows the reconstruction fidelity of @xmath133 . the experimental value is calculated as @xmath134 , which is higher than the cloning limit of @xmath52 . the cloning limit can be considered as the classical limit for the reproduction of the original state , because one can never obtain a better approximation of the original state than the clones if the nonclassical correlations between the clones and anticlones can not be utilized . in fig . [ fig : cln_rc ] , the input quadratures @xmath125 and @xmath126 are excited one by one , and for each case they are reconstructed from the output correlations . the trace in magenta in the leftmost box is the power of the excited original quadrature , whereas the trace in cyan in the same box is the power of the corresponding shot noise . the traces in red and blue in the other two boxes on the right side are the powers of the reconstructed quadratures with and without the excitation , respectively , and the traces in cyan at @xmath102 db are the powers of the summed shot noises of the three homodyne detections . at the excited quadrature , almost the same level of excitation is reconstructed . in contrast , at the conjugate quadrature , no effect of the excitation is observed . we succeeded in phase - insensitive optical amplification in a reversible manner . our amplifier preserves the idler output . the entanglement between the signal and idler is responsible for the reversibility . the scheme is basically based on linear optics , homodyne measurements and feedforward . offline - prepared squeezed states which are used as ancillas provide nonclassical properties for our pia . we demonstrated for the amplification gain of @xmath0 . by splitting the amplified output in half , we also demonstrated @xmath15 approximate cloning of coherent states , where the remaining idler output was interpreted as the anticlone . for both experiments , the full demonstration was given in the following sequence . first , we characterized the individual output modes . by treating the quadrature data directly , they were visualized as phase space diagrams to help intuitive understanding . especially , the mirror image in the idler output or the anticlone is shown . then the input - output relation was examined more strictly by using several different coherent states as inputs . finally , the output correlations were examined . they are important because the nonclassical properties are only accessible via them . not only the ordinary epr correlations were shown , but also the possibility of the reverse operation was directly presented by appropriate measurements of the correlations . our results are a good demonstration that shows the properties of an amplification process , which have been theoretically known for decades but not fully demonstrated experimentally . especially , it is reversible when the idler is present . such reversible amplification is significant from practical respects , as is shown in refs . we did not demonstrate the reverse operation but showed its possibility from the correlations . the recovery of the signal state only requires the bell measurement and feedforward @xcite , which would be much less lossy and noisy than implementing the inverse transformation . full and partial recovery of the distributed information via such a feedforward scheme is left for future experiments . this work was partly supported by scf , gia , g - coe , pfn and first commissioned by the mext of japan , the research foundation for opt - science and technology , scope program of the mic of japan , and , jsps and ascr under the japan - czech republic research cooperative program . r. f. acknowledges projects : msm 6198959213 and me10156 of the czech ministry of education , grant 202/08/0224 of ga r and eu grant fp7 212008 compas . in this appendix , the discussion in sec . [ sec : clone ] is extended to @xmath35 cloning . the procedure of @xmath35 symmetric cloning can be decomposed into three steps as follows @xcite . firstly , all the information of @xmath41 are put together into a single mode by a beamsplitter network . a state with larger amplitude @xmath135 is created from the @xmath37 identical originals @xmath136 in this step . secondly , the combined signal is amplified with the gain @xmath137 . finally , the amplified signal is combined with @xmath138 ancillas by another beamsplitter network , creating @xmath36 clones . for asymmetric cloning , the procedure is essentially the same , but the amplification in the second step is applied to a part of the combined signal , and the gain is changed correspondingly as @xmath139 @xcite . on the other hand , @xmath140 anticlones are obtained from the idler output by combining it with @xmath141 ancillas by another beamsplitter network . therefore , as a whole , @xmath36 clones and @xmath140 anticlones are obtained , whose complex mean amplitudes are @xmath142 and @xmath143 respectively , from @xmath37 originals with the complex mean amplitude of @xmath142 . it is evident that there is no entanglement among clones or anticlones , however , there is entanglement between a clone and an anticlone . the asymmetric clones have the following form : @xmath144 where @xmath145 is the annihilation operator of the clone , @xmath146 is that of the combined original after the first step , @xmath147 is that of the idler input , and @xmath148 is that of the ancilla input for the latter beamsplitter network . the coefficients @xmath149 and @xmath150 are not independent in order to preserve the commutation relations . by setting all the ancillas in vacuum states , the added noises become rotationally symmetric and gaussian , and their variances correspond to the parameters @xmath45 @xcite . they satisfy the relation below @xcite : @xmath151 the optimality of eq . is proven with respect to the cost function constructed from the variances @xcite : @xmath152 in particular , for the symmetric cloning , i.e. , @xmath153 , eq . saturates the following inequality which is obtained in ref . @xcite from the consistency with the uncertainty relation : @xmath154 the limit fidelity for symmetric gaussian cloning is calculated as @xmath155 . on the other hand , taking the limit that @xmath36 goes to infinity , the classical limit of cloning ( i.e. , the limit of state estimation ) is obtained as @xmath156 . as with @xmath15 cloning , optimal @xmath157 cloning can be fully reversed by @xmath138 bell measurements performed on each set of a clone and an anticlone and subsequent feedforward to the remaining single clone @xcite . even with the smaller number of bell measurements , the original is partially recovered accordingly . braunstein , v. buzek , and m. hillery , * 63 * , 052313 ( 2001 ) . for example , f. grosshans and p. grangier , * 88 * , 057902 ( 2002 ) ; f. grosshans and p. grangier , arxiv : quant - ph/0204127 ( 2002 ) . j. fiurek and n.j . cerf , * 75 * , 052335 ( 2007 ) . n.j . cerf , o. krger , p. navez , r.f . werner , and m.m . wolf , * 95 * , 070501 ( 2005 ) . f. grosshans and n.j . cerf , * 92 * , 047905 ( 2004 ) . a. leverrier and p. grangier , * 81 * , 062314 ( 2010 ) . n.j . cerf , a. ipe , and x. rottenberg , * 85 * , 1754 ( 2000 ) . j. fiurek , * 86 * , 4942 ( 2001 ) . braunstein , n.j . cerf , s. iblisdir , p. van loock , and s. massar , * 86 * , 4938 ( 2001 ) . j. bae and a. acn , * 97 * , 030402 ( 2006 ) . f. de martini , v. buek , f. sciarrino , and c. sias , nature * 419 * , 815 ( 2002 ) . r. filip , j. fiurek , and p. marek , * 69 * , 012314 ( 2004 ) . u.l . andersen , v. josse , and g. leuchs , * 94 * , 240503 ( 2005 ) . s. koike , h. takahashi , h. yonezawa , n. takei , s.l . braunstein , t. aoki , and a. furusawa , * 96 * , 060504 ( 2006 ) . braunstein , * 71 * , 055801 ( 2005 ) . j. yoshiakawa , y. miwa , a. huck , u.l . andersen , p. van loock , and a. furusawa , * 101 * , 250501 ( 2008 ) . s. suzuki , h. yonezawa , f. kannari , m. sasaki , and a. furusawa , * 89 * , 061116 ( 2006 ) . duan , g. giedke , j.i . cirac , and p. zoller , * 84 * , 2722 ( 2000 ) . more strictly speaking , the story becomes as follows : the simple picture of cloning that some noise is added to the original @xmath136 is valid for @xmath157 cloning , irrespective of the core state @xmath33 . however , for @xmath35 cloning with @xmath158 , it is not always valid . when the core state @xmath33 is known , there would be some choice of the ancilla states which make the picture valid . indeed , when the core state is known to be a vacuum state ( i.e. , the set @xmath42 of the possible original states is all coherent states ) , we can choose vacuum states as the ancillas . when the core state is unknown , a part of the original quantum fluctuation is replaced by that of the ancillas , therefore , the added noises @xmath45 become ill - defined . nonetheless , the cloner would be still optimal in some sense from the standpoint of the share of the information of @xmath41 .

we propose a method of free energy calculation for a system of interacting particles arranged in a bravais lattice . it will be shown how to treat divergences for infinite unbounded systems with `` catastrophic '' potentials like coulomb and compare two of them . besides , we show that current method may be used not only for essentially classical systems , but for some quantum as well . two systems are considered : grains in dusty plasma and electrons on the liquid - helium surface . for dusty - plasma parameters of grain s lattice are calculated by numerical solution of obtained equations . electrons on the liquid - helium surface are analyzed to get localization distance . besides , conditions for existence of electron lattice are found .

1505.00408

there are many soft - matter systems , such as grains in dusty plasma , colloids in various solvents , surfactant solutions , etc . , that exhibit self - organization and rearrangement in crystalline structures under certain conditions . often their inter - particle potential is long - range and similar to the coulomb one . usually , this kind of potentials cause divergences ( that s why they are often called `` catastrophic '' ) and complicates consideration @xcite . on the other hand , mentioned systems are very interesting due to their application to the studies of a variety of peculiar phenomena in different fields of science @xcite . it seems to be the most challenging problem to treat coulomb - like systems with high concentrations of interacting particles @xcite . when concentration increases , one can observe crystallization - like phenomena and transition between different lattice symmetries @xcite . in current contribution we will treat concrete systems along with general formalism development . they were chosen as manifesting a lot of interesting effects . for example , dusty plasma may serve as perfect media for the experimental investigation of classical fluids and solids along with colloids @xcite . the second system is `` electrons on the liquid - helium surface '' @xcite . it may undergo wigner crystallization @xcite and transition between triangular and square lattices @xcite . moreover , homogeneous distribution of electron density is not always stable , and there are critical parameters when spatial structures , especially periodic deformations and multi - electron dimples , are formed @xcite . but our original goal is some general description of particles arranged in a bravais lattice when interparticle potential may be `` catastrophic '' . it is rather difficult , since traditional methods of statistical mechanics can not be applied to systems with coulomb - like interactions because of free - energy divergences . moreover , partition function can be exactly evaluated only for few model systems of interacting particles in the thermodynamical limit @xcite . thus , basing on @xcite , we will use alternative approach that employs hubbard - stratonovich representation of the partition function @xcite . besides , we use approach involving probability distribution function for the whole lattice as in @xcite , but construct it from single - particle distribution functions with a special procedure . namely , we will expand and significantly supplement results obtained in @xcite for two - dimensional systems . new results for three - dimensional case , that is far more complicated than two - dimensional , will be obtained . to be more rigorous : our goal is to find expression for free energy of particles arranged in a bravais lattice even if inter - particle potential is `` catastrophic '' . general equations will be applied to two concrete systems : three - dimensional grains in dusty plasma , and two - dimensional electrons on the liquid - helium surface . in first case we ll minimize this energy to find out which lattice should be observed experimentally . and for electrons localization distance will be found . besides , this will show that presented method allows easily reduce dimension number from three to two . present article is organized as follows . it can be contingently divided into two parts : development of technique and its application . sections [ sec : stat_descr ] and [ sec : lattice ] focus on general expression for free energy for any bravais lattice and any potential . the former shows that not only classical , but lots of quantum systems may be treated this way . besides , an expression for `` entropy part '' of free energy is calculated . and in section [ sec : lattice ] interaction energy for paricles arranged in a bravais lattice is calculated . both this sections are equipped with examples ( [ subsec : two_dim_fermi ] , [ subsec : pdf_for_sp][subsec : capillary_interaction ] ) that show how to use developed formalism . this examples are rather useful , because they are later reused when treating physical systems . concrete applications of developed formalizm start with section [ sec : dust_crystal ] . here crystal of dust particles is considered . in [ sec : dust_crystal ] it is shown that hexagonal close packing seems to be the lattice we expect to be seen in experiment . this is in agreement with experiments @xcite and computer simulations @xcite . system of electrons on the liquid - helium surface is analyzed in [ sec : electrons_on_helium ] . in [ sec : electrons_on_helium ] localization distance is calculated and compared to classical result @xcite . for low temperatures @xmath0 they coincide . for not to be confused with lots and lots of complicated mathematical computations we moved them to appendices [ sec : appendices ] . thus article mostly contains some physical explanations and results . but using references to appendices , one may find here rigorous exposition of presented ideas as well . first of all let s outline class of systems we will be able to consider . since everything is clear with classical systems , we will give heed to quantum ones . in this section we aim some general consideration of inhomogeneous system of interacting particles @xcite . nor system , neither interaction potential will be specified to keep generosity and extendability . as result , we will get free energy in some general form and ready for `` applying '' translation symmetry ( section [ sec : lattice ] ) . of course , along this section we will do some assumptions about system under consideration . together they will form restrictions on quantum systems we can treat with proposed method . in current model macroscopic states of the system are described by occupation numbers . besides , we suppose that it s hamiltonian has form : @xmath1 here @xmath2 is the additive part of particle energy ( usually it is kinetic energy , but it can be particle s energy in external field as well . ) , @xmath3 indicates particle state , @xmath4 is interaction energy between particles in states @xmath3 and @xmath5 , @xmath6 is the occupation number of state @xmath3 . we neglect quantum correlations , so only essentially classical systems are considered . partition function for this kind of system will be : @xmath7 where summation is performed over all possible states @xmath8 of the system . now we use some properties of gaussian integrals over auxiliary fields , i.e. hubbard - stratonovich transformation @xcite : @xmath9 with @xmath10 . second - order dependence on occupation numbers is avoided by carrying them to introduced field and partition function can be written : @xmath11\!.\ ] ] for canonical ensemble number of particles can be fixed by using cauchy equation : @xmath12 that leads to @xmath13-particle partition function : @xmath14 \prod_s \sum_{\{n_s\ } } \left [ \xi \exp(i \varphi_s - \beta \varepsilon_s ) \right]^{n_s}. \end{aligned}\ ] ] performing summation over occupation numbers : @xmath15,\\ \beta f&(\varphi,\xi ) = \frac1{2 } \sum_{s , s ' } v_{ss'}^{-1 } \varphi_s \varphi_{s ' } + \\ & + \delta \sum_s \ln \left ( 1 - \delta \xi e^{-\beta \varepsilon_s+i \varphi_s } \right ) + ( n+1 ) \ln \xi . \end{aligned}\ ] ] here type of statistics is incorporated into @xmath16 it equals to @xmath17 for bose - einstein , @xmath18 for maxwell - boltzmann and @xmath19 for fermi - dirac statistics . now we have expression for free energy in auxiliary field representation with @xmath20 for chemical activity . equation contains all information about probable states of the system and corresponds to the sequence of equilibrium states with regard to their weights . we are interested in asymptotic value of partition function , but want to avoid using perturbation theory . thus domain is extended to complex plane and saddle - point method is applied . dominant contribution is made by states satisfying extrema condition : @xmath21 variation of yields expression for saddle - point states : @xmath22 for a certain state expression @xmath23 from can be treated as an average occupation number . now saddle - point states ( that may be interpreted as thermodynamically stable distributions ) can be obtained . equation contains inverse matrix , that is quite inconvenient for us , because its calculation is a challenging mathematical problem even for simple potentials @xcite . this problem can be avoided if we use and perform inverse transformation @xmath24 than free energy can be written @xmath25 = - \frac{\beta}{2 } \sum_{s , s ' } v_{ss ' } f_s f_{s ' } & - \delta \sum_{s } \ln ( 1 + \delta f_s ) \\ & + ( n+1 ) \ln \xi ( f ) . \end{aligned}\ ] ] in terms of canonical ensemble we can write from @xmath26,\\ e_s & = \sum_{s ' } v_{ss ' } f_{s ' } , \end{aligned}\ ] ] and substitute last expression into free energy to rewrite its expression without chemical potential . @xmath27 & = \sum_s f_s \varepsilon_s + \frac{1}{2 } \sum_{s , s ' } v_{ss ' } f_s f_{s ' } + \\ & + \underbrace{\frac{1}{\beta } \sum_s \left[f_s \ln f_s - ( f_s+\delta ) \ln ( 1 + \delta f_s ) \right]}_{f_{ent}}. \end{aligned}\ ] ] first two terms in are kinetic and potential energy respectively , the last one is a contribution due to entropy @xmath28 ( should be equal to zero if @xmath29 ) . now let s do a quick test of obtained results . if a grand canonical ensemble with fixed chemical potential is considered , we can get from @xmath30 which is generalization of the well - known distribution with a chemical potential @xmath31 it is obvious that the `` saddle point '' and `` mean field '' approximations are equivalents in this case . if an ideal gas ( @xmath32 ) is considered , we can obtain classical statistical distributions . before winding up this section let s review the results . we aimed to find some general expression for free energy in a mean - field approximation . as the one , equation should be pointed out ( and in some sense that can be used to obtain particles distribution in grand canonical ensemble ) . besides , we have got some restrictions on the class of systems we will be able to consider : special form of hamiltonian and negligible quantum cross - correlations . before we shift our attention to potential energy of particles arranged in a lattice , let s do some example . with this subsection we aim two things : give some idea on how to use the above formalism and secondly calculate contribution due to entropy for electrons on the liquid - helium surface ( will be used in [ sec : electrons_on_helium ] ) . electrons on the liquid - helium surface have two degrees of freedom only @xcite . though we set @xmath33 ( fermi particles ) and @xmath34 ( system is two - dimensional ) . then equation can be written @xmath35 = \int \frac{d^2 \vec{p}\ , d^2 \vec{r}}{(2 \pi \hbar)^2 } \frac{\beta \vec{p}^{\,2}}{2 m } \frac{1}{e^{\beta(\vec{p}^{\,2 } / ( 2 m ) - \mu(\vec{r}\ , ) ) } + 1 } + \\ & { } + \frac{\beta}{2 } \iint \frac{d^2 \vec{p}\ , d^2 \vec{r}}{(2 \pi \hbar)^2 } \frac{d^2 \vec{p}^{\,\prime}\ , d^2 \vec{r}^{\,\prime}}{(2 \pi \hbar)^2 } \times \\ & { } \times \frac{v\left(|\vec{r}-\vec{r}^{\,\prime}|\right ) } { \left(e^{\beta(\vec{p}^{\,2 } / ( 2 m ) - \mu(\vec{r}\ , ) ) } + 1\right)\!\ ! \left(e^{\beta(\vec{p}^{\,\prime 2 } / ( 2 m ) - \mu(\vec{r}^{\,\prime}\ , ) ) } + 1\right ) } -\\ & { } - \int \frac{d^2 \vec{p}\ , d^2 \vec{r}}{(2 \pi \hbar)^2 } \left ( \frac{\ln\left(e^{\beta(\vec{p}^{\,2 } / ( 2 m ) - \mu(\vec{r}\ , ) ) } + 1\right ) } { e^{\beta(\vec{p}^{\,2 } / ( 2 m ) - \mu(\vec{r}))}+1 } \right . + \\ & { } + \left . \frac{\ln\left(e^{-\beta(\vec{p}^{\,2 } / ( 2 m ) - \mu(\vec{r}\ , ) ) } + 1\right ) } { e^{-\beta(\vec{p}^{\,2 } / ( 2 m ) - \mu(\vec{r}\ , ) ) } + 1 } \right ) . \end{aligned}\ ] ] here integration sign @xmath36 means integration over the whole phase space . since last expression is quite cumbersome we should simplify it somehow . as first approximation we suppose @xmath37 actually , in the presence of an external field , dispersion relation should be a bit different . but we know , that system of fermi particles , we will be interested in , is highly degenerated @xcite . so quadratic form for dispersion relation looks quite reasonable . using last equation and introducing thermal length @xmath38 we get from following expression @xmath39 which connects chemical potential @xmath40 and particles density @xmath41 now we return to simplification of . performing integration of over momentum and taking and into account one may get ( see appendix [ subsec : appendix_momentum_integration ] ) @xmath42 & = \frac{1}{2 } \iint d^2 \vec{r}\ , d^2 \vec{r}^{\,\prime } v(|\vec{r}-\vec{r}^{\,\prime}| ) \rho(\vec{r}\ , ) \rho(\vec{r}^{\,\prime } ) + \\ & + \frac{\pi}{\beta\lambda_t^2 } \int d^2 \vec{r } \left [ \text{li}_2\left(e^{- \pi^{-1 } \lambda_t^2 \rho(\vec{r}\,)}\right ) - \frac{\pi^2}{6 } \right ] + \\ & + \frac{\lambda_t^2}{2\pi\beta } \int d^2 \vec{r } \rho^2(\vec{r}\ , ) . \end{aligned}\ ] ] one may notice that in case of bose statistics we will loose the bose - condensation effects due to this integration . but we aim to use this equations for fermi particles , so there is nothing to be worried about . we know that electrons on the liquid - helium surface exist in forms of fluid or wigner crystal @xcite . we are interested in case when electrons are strongly localized , i.e. their coordinates are quite determined . but since temperature differs from zero , we expect that particle s position will fluctuate near its equilibrium location . though we assume following distribution function @xmath43 for _ * * s**ingle * * p**article _ ( somewhat analogous to form factor in qft ) @xmath44 where @xmath3 is dispersion , or localization distance by physical meaning , and @xmath34 because system is two - dimensional ( dimensionality ) even if it is redundant at the point we introduce the equation first time . ] . equation is normal distribution which seems to be quite reasonable . at least this distribution is valid for ground state of quantum harmonic oscillator which is the simplest approximation for particle fluctuating near its equilibrium location . last assumption we do is that `` gaussian '' is very `` sharp '' and every particle lays outside localization radius of any other particle . this means that @xmath3 is much smaller then other characteristic distances in this system . so free energy per one particle can be written : @xmath45 here @xmath28 is contribution to free energy by _ * * ent**ropy _ and temperature . first summand in @xmath28 is obtained from direct integration of @xmath46 ( see last term in ) . to get second summand ( penultimate term in ) we use `` sharpness '' of and approximate @xmath47 with equations and we finish our consideration . we fulfilled our goal to show an example of using section s [ sec : stat_descr ] formalism . moreover , now we have expression for entropy part @xmath28 of free energy , which we use later in [ sec : electrons_on_helium ] . in section [ sec : stat_descr ] we found some expressions for free energy of inhomogeneous system of interacting particles . in this section we will be interested in second term of , i.e. potential energy . more accurately , we aim to find method of potential energy calculation for system of identical ( from inter - particle s potential point of view ) particles arranged in a bravais lattice . during this section we will show how to use translation symmetry to get in some cases more convenient expressions for potential energy . last remark before we dive in . we aim to find appropriate expressions for both two- and three - dimensional cases . but to avoid reduplication all along the section [ sec : lattice ] only three - dimensional case will be treated as far more complicated . for two - dimensional case we will only provide results and short description where difference comes from . let @xmath48 designates one subset of exact cover of @xmath49 ( figure [ fig : vectors ] ) with similar domains containing strictly one particle ( relative position of the particle supposed to be the same in every domain ) ) . ] . then free energy of the particle localized near bravais lattice site can be written as ( we neglect kinetic energy in since particles are localized and potential energy should be bigger ) [ freeenergyseparated ] @xmath50 equation needs some explanations . interaction energy is presented here with @xmath51 . this `` splitting '' is just another way of saying we want to integrate over @xmath52 . if one collects both expressions together and uses @xmath53 he will get regular integration over @xmath52 for @xmath54 . in other words , since particle is localized near lattice site , we take some `` part of space '' @xmath48 around it . then by integration we calculate interaction energy between particle in @xmath48 and all the rest particles in @xmath52 . one may get a good intuitive idea about integration domain @xmath52 through discrete case . when we calculate coulomb energy for @xmath55-th particle we write @xmath56 . integration over @xmath52 is somewhat analogous to @xmath57 for discrete case . we will further call @xmath58 _ * * int**eraction energy _ and @xmath59 _ * * s**elf - interaction energy_. this `` splitting '' has one more convenience . basing on physical background of considered potential we may want or not want to compensate self - interaction . obviously , coulomb potential should have self - interaction compensated potential energy is zero if there is only one particle in a system ( considered later in example [ subsec : coulomb_potential ] ) . on the other hand , self - interaction for effective potentials caused by influence through medium should be uncompensated , because even in absence of other particles medium will influence this one . surface distortion is a good example . if particle has distorted a surface it has already done some contribution to system s free energy independently on other particles ( considered later in example [ subsec : capillary_interaction ] ) . of course , it interacts with distortions created by other particles . but at the same time it interacts with `` its own '' surface distortion and this may be treated as `` self - interaction '' . now lets consider more carefully . all 3d bravais lattices can be treated as constructed from parallelepipeds . we will use this property and use parallelepiped as integration region @xmath48 . . vectors @xmath60 , @xmath61 and @xmath62 are base vectors for bravais lattice . angle between @xmath60 and @xmath61 is supposed to be @xmath63 . angle between @xmath62 and @xmath64 plane is @xmath65 and its projection on @xmath64 plane and @xmath60 is @xmath66 . one particle is shown in the center . , width=302 ] next thing we want to do is expressing probability distribution function @xmath67 for the whole @xmath49 through single - particle probability distribution @xmath43 . basing on the inverse lattice vectors ( appendix [ subsec : appendix_properties_of_f_k ] ) we will provide following decomposition [ pdf_in_fourier ] @xmath68 0 & \cfrac{\csc(\alpha)}{b } & -\cfrac{\cot(\beta_c)\sin(\alpha_c)}{b\sin(\alpha ) } \\[12pt ] 0 & 0 & \cfrac{\csc(\beta_c)}{c } \end{pmatrix}. \label{pdf_in_fourier_g_operator}\end{aligned}\ ] ] here @xmath69 designates complex conjugation . it may be noticed that all vectors are treated as column vectors . @xmath70 is mean particle density here . to get better intuitive understanding how to compute one may check example [ subsec : pdf_for_sp ] for @xmath43 from . it may be shown that presented series are actually fourier series for @xmath43 and @xmath67 is periodic along vectors @xmath60 , @xmath61 and @xmath62 ( appendix [ subsec : appendix_properties_of_f_k ] ) . so we may treat @xmath71 as function on @xmath49 `` composed '' from single particle distributions `` arranged in a lattice '' . this principle is demonstrated by fig . [ fig : one_to_many ] . to many particle @xmath41 . function @xmath41 is defined for all @xmath72 with two - dimensional analogy of . , title="fig:",width=75 ] to many particle @xmath41 . function @xmath41 is defined for all @xmath72 with two - dimensional analogy of . , title="fig:",width=226 ] since @xmath67 has the same symmetry as the lattice and locally is a good approximation for @xmath43 it will be treated as probability distribution function for the whole lattice . matrix @xmath73 comprises all _ * * g**eometry _ of the lattice under consideration . performing minimization of @xmath74 over entries of @xmath73 , one may find which lattice can be formed under certain conditions . we will use later this type of analysis for dust crystal in section [ sec : dust_crystal ] . if we use equations we can express in terms of @xmath43 ( appendix [ subsec : appendix_expression_f_int ] ) [ f_int ] @xmath75 where @xmath70 is mean particle density . before moving further , one more thing should be emphasized . this equations are useful because they allow us to deal with `` catastrophic '' potentials like coulomb . one may notice that for any descending potential only @xmath76 can be diverging and only if potential is `` catastrophic '' . but this term does not contain `` lattice geometry '' , it depends on mean particle density only . this means , we can compare two lattices with equal mean particle density even if inter - particle potential is `` catastrophic '' ( somewhat analogous to renormalization in qft ) . with equations , and we achieve our goal and finish consideration of interaction energy for particles arranged in a lattice . following examples show how to use obtained equations for screened coulomb ( example [ subsec : coulomb_potential ] ) and capillary interaction ( example [ subsec : capillary_interaction ] ) . they both use @xmath67 obtained with @xmath43 from ( example [ subsec : pdf_for_sp ] ) . we suppose that all particles are arranged in a lattice and their positions are quite determined . but since temperature differs from zero we can expect that particle position fluctuates near its lattice site . using considerations from [ subsec : two_dim_fermi ] we suppose it to be with @xmath77 . introduced constant @xmath3 means dispersion of presented distribution . from physical point of view it can be treated as localization distance . later , in section [ sec : electrons_on_helium ] , we ll find its value and compare it with localization distance obtained in different works . using that @xmath78 is a very `` sharp '' gaussian , we expand integration limits to @xmath49 in and perform integration ( appendix [ subsec : appendix_rho_k ] ) @xmath79 as mentioned previously , we are interested in two - dimensional case as well . one may check , that there exists some sort of analogy of for two dimensions @xcite . difference will appeare in changing summations and integrations @xmath80 , @xmath81 and usage of different matrix @xmath82 0 & \csc(\alpha)/b \\[12pt ] \end{pmatrix}.\ ] ] performing calculations as in appendix [ subsec : appendix_rho_k ] with and @xmath34 , one will get @xmath83 where @xmath70 is two- , not three - dimensional density as in . equations and are aimed result of this example . later they will be used in calculations of potential energy for coulomb and capillary interactions . in section [ sec : dust_crystal ] grains in dusty plasma will be considered and in section [ sec : electrons_on_helium ] we ll set our eyes on electrons on the liquid - helium surface . in both cases we have to deal with coulomb potential screened for dust particles and regular for electrons . the latter will be treated as screened coulomb with screening distance approaching infinity . during this example we aim to find @xmath84 for screened coulomb potential using methods of section [ sec : lattice ] . consideration of physics of this two systems will be postponed until sections [ sec : dust_crystal ] and [ sec : electrons_on_helium ] . screened coulomb potential can be written as @xmath85 where @xmath86 is debye screening distance . it may be shown ( appendix [ subsec : appendix_coulomb_v_k ] ) that expression for @xmath87 should be @xmath88 at this point we already have @xmath87 and @xmath89 . so we can write @xmath58 from equation @xmath90 compared to straight - forward method of computing , this one has better convergence . for big values of @xmath86 we will see , that first summands in descend as @xmath91 . but if we write series containing interaction energy between this particle and any other , first summands will descend only as @xmath92 . in section [ sec : lattice ] we have mentioned that there is one compensatory term we need to take into account . this term will exclude coulomb self - interaction . if we omit technical details ( see appendix [ subsec : appendix_f_s ] ) explicit form of @xmath59 can be obtained @xmath93 or substituting screened coulomb potential as @xmath94 and approximating for the case when @xmath3 is much smaller then screening distance @xmath86 @xmath95 expressions we ve got for @xmath58 and @xmath59 are a bit inconvenient . so we use that @xmath3 is small compared to distances in the lattice ( `` sharp peak '' approximation , section [ sec : lattice ] ) and perform some approximation . first of all we introduce mean distance calculated from particle s density @xmath96{\bar{\rho}}},\ ] ] and split sum into two parts . further we will suppose lattice under consideration as not too much degenerated . in other words , we will assume that it may be described as some `` deformation '' of cubic lattice ( appendix [ subsec : approximation_f_int ] ) @xmath97 one may see that last term is equal to energy in case of evenly distributed charge . other terms are corrections due to charge pointness and comprise all information on lattice geometry . it is very useful to notice , that this is the only summand that approaches infinity if @xmath98 . this means that can be used for different lattice comparison even without screening . we will only need to measure energy starting from @xmath99 as `` ground - level '' . one more thing to notice . coulomb is a long - range interaction and localization distance @xmath3 should not play any significant role in approximation . here we see that it is really absent . rather different may be a picture for short - range interactions or effective interactions with uncompensated @xmath59 ( example [ subsec : capillary_interaction ] ) . last example we will consider is a rough estimation of capillary interaction energy . this results will be used in section [ sec : electrons_on_helium ] . more discussion on physics of the phenomenon and references we postpone until [ sec : electrons_on_helium ] . it is known , that electrons deform liquid - helium surface when pressed against with electric field and this causes effective interaction with potential @xcite @xmath100 where @xmath101 is force caused by electric field , @xmath102 surface tension of liquid helium , @xmath103 distance between particles , @xmath104 modified bessel function , and @xmath105 @xcite capillary length that depends on the fluid properties only . since we know that capillary interaction is essentially two - dimensional ( particles should be localized on some surface , no three - dimensional analogue ) we will use with and . performing calculations ( appendix [ subsec : appendix_capillary_v_k ] ) we get @xmath106 so we can immediately write @xmath107 for our purposes ( see section [ sec : electrons_on_helium ] ) a very rough approximation would be enough . so we use the same trick as in appendix [ subsec : approximation_f_int ] , but perform integration instead of summation and dealing with theta functions ( see appendix [ subsec : approximation_f_cap ] ) @xmath108 one should notice , that @xmath109 , so logarithm value is less then zero and thus whole expression is less then zero . it coincides with the fact that capillary interaction makes particles attract each other @xcite . it is known that grains in dusty plasma interact and even show self - organization in form of melting and crystallization of dust crystalls @xcite . we aim to apply methods developed in section [ sec : lattice ] to this system . as a first approximation we suppose that grains interact only as charged particles ( screened coulomb potential ) . obtained equation ( example [ subsec : coulomb_potential ] ) allows to calculate energy of any lattice just inserting correct matrix @xmath73 and performing summation . but much more interesting is finding the lattice with minimal energy when particles density @xmath70 is fixed . obtained result can be verified experimentally . both sums in are highly convergent and this is very convenient for numeric calculations . thus we will limit summation with @xmath110 . this limit was found experimentally : taking more summands does not change calculated minimal lattice and free energy . on the other hand , taking less results in wrong lattice . and one more thing to mension . searching for lattice with minimal energy does not need calculation of full expression for @xmath84 . we need to minimize only parts with @xmath73 . moreover , they both contain @xmath111 , which means , that results of minimization do not depend of particle charge . minimization was performed with standard function findminimum of wolfram mathematica v.9 . we use as function to be minimized . particles density is fixed in a following way . we consider @xmath112 suppose @xmath113 , @xmath114 and express @xmath115 through other parameters to keep constant charge density . performing this one may notice that @xmath116 cancel out in exponents and denominators . before considering results , one more remark should be done . mapping between set of all possible parameter values ( translation vectors ) and all possible bravais lattices is not a bijection . it is rather a surjection one lattice may be described with different sets of parameters ( different translation vectors ) . for minimization algorithms to be not confused , we provide some restrictions . and @xmath61 . upper figure shows different choice for second translation vector @xmath61 and @xmath117 . using reflections one may show that lattice on bottom figure is the same as on the upper one . so @xmath118 is possible choice as well . when performing numerical calculations additional restrictions are used to overcome this uncertainty . , width=226 ] figure [ fig : lattice_uncertainty ] shows uncertainty when choosing parameters for lattice description for 2d case . one may see that it can be eliminated if we claim that projection of @xmath61 on @xmath60 is less then a half of @xmath60 . in terms of parameters this may be expressed as @xmath119 restrictions for 3d case are more sophisticated . but they may be achieved similarly as . actually should be kept for 3d as is . we only need to add analogous considerations for @xmath62 . first of all we introduce projection of vector @xmath62 on @xmath64 plane ( see figure [ fig : vectors ] ) and take particles density into account ( substituting @xmath115 from ) @xmath120 obviously this projection should fall into parallelogram constructed on @xmath60 and @xmath61 . this condition is an obvious extension of two - dimensional case to third dimension . but we can strenghten this condition by using reflection with respect to @xmath64 plane ( figure [ fig : lattice_uncertainty_3d ] ) . particle from layer above @xmath64 plane is shown with circle and below the plane with filled circle . particles in @xmath64 plane are not explicitly shown . if particle from above the plane is projected into the upper half of @xmath121 parallelogram a reflection with respect to @xmath64 plane can be used to get representation where this particle is in lower half of parallelogram . , width=302 ] we can always assume that @xmath62 is projected into lower part of parallelogram constructed with @xmath60 and @xmath61 . as result we get following conditions @xmath122 now we can present results of numerical minimization in form of figure [ figure_hcp ] and following table . looking at figure [ figure_hcp ] , one my see , it is very similar to hexagonal close packing ( hcp ) lattice . hcp consists of two interleaving planes with triangular 2-d lattice in each as well . so we may check , if this is really hcp lattice by comparing their parameters . [ cols="^,^,^ " , ] results show that lattice obtained by numerical calculations is really hcp . this complies with experimental data and computer simulation @xcite . in this section we consider electrons on the liquid helium surface , calculate localization distance and compare it to obtained in other works . in the presence of an external field electrons can be pressed against the helium surface with significant force . but they can not go through this surface , because quantum effects push them out @xcite . thus they deform surface and this deformation , in its turn , changes interaction potential between electrons . the problem of finding explicit expression of this potential is solved by adding to coulomb interaction effective capillary interaction . the lateral for two electrons on the helium surface was calculated in @xcite . we aim to find localization distance for electrons on the liquid helium surface . to achieve this goal we will minimize single - particle s free energy @xmath74 over @xmath3 ( localization distance , see example [ subsec : two_dim_fermi ] ) . so first of all we construct @xmath74 and find @xmath123 . in section [ sec : stat_descr ] we developed a method for treating some quantum systems . example [ subsec : two_dim_fermi ] shows how to use it for two - dimensional system of fermi - particles . so we consider @xmath28 from as free energy of one particle caused with entropy in the system . potential energy is more complicated . in section [ sec : lattice ] we developed methods for its calculation . there are two parts of potential energy for electrons on the liquid helium surface @xcite : coulomb interaction and capillary interaction . for coulomb interaction we can find a shortcut it was shown in @xcite that this part approximately does not depend on localization distance @xmath3 . besides , one may think on approximate expression for coulomb interaction from example [ subsec : coulomb_potential ] . it does not contain @xmath3 , and the same behavior we may expect from coulomb interaction for 2d lattice . since we are interested in derivative @xmath123 , consideration of this part of potential energy may be omitted . more interesting is capillary interaction presented in example [ subsec : capillary_interaction ] , approximation . it contains explicit dependence on @xmath3 and thus should be taken into account . bringing it all together we get @xmath124 independently on lattice we are considering . substituting electron charge @xmath125 and solving equation @xmath126 we get @xmath127 if temperature tends to be small @xmath0 ( e.g. @xmath128 ) @xmath129 this result does not differ from the classical one presented in @xcite . if we consider more carefully , some interesting result may be obtained . let s consider low densities so that any value of @xmath3 may be treated as much lower than interelectron distance @xmath116 . then we may notice that solution for @xmath3 exists if @xmath130 last equation means that there is no lattice if temperature @xmath131 is too high , or if field @xmath132 is too weak . this statement is in agreement with our physical intuition . in conclusion we want to make some overview of results we have obtained . in * section [ sec : stat_descr ] * we started with a statistical description of inhomogeneous system of interacting particles . it was shown that variety of quantum systems may be described with no much difference from classical ones . we found an expression for free energy . the main feature of this equation appeares in expression for `` entropy part '' of free energy . obtained result is applicable to a wide variety of systems with different number of dimensions . to make a step aside from pure theory we considered an example [ subsec : two_dim_fermi ] of two - dimensional system of fermi particles . this result was later reused in section [ sec : electrons_on_helium ] . in * section [ sec : lattice ] * we stated that particles are arranged in a lattice . it seems , that adding translation symmetry to consideration we can get some simplification for potential energy expressions and . one may notice that for any `` catastrophic '' potential there is only one diverging term . but this term does not contain information on `` lattice geometry '' , it depends on mean particle density only . this means , we can compare two lattices with equal mean particle density even if inter - particle potential is `` catastrophic '' . in this approach all potentials are classified into two groups : ones with self - interaction ( often effective interactions through medium ) and without . examples [ subsec : coulomb_potential ] and [ subsec : capillary_interaction ] demonstrate how to use proposed technique in both cases . later they are reused for treating grains in dusty plasma ( section [ sec : dust_crystal ] ) and electrons on the liquid - helium surface ( section [ sec : electrons_on_helium ] ) . in * section [ sec : dust_crystal ] * we considered grains in dusty plasma . we started with potential energy obtained in examle [ subsec : coulomb_potential ] and minimized it over all possible lattices . since there are no good approximations for theta - function it was too complicated to perform this minimization analytically . thus we used numerical computations . this computations have some features presented in a real - life case [ sec : dust_crystal ] . results obtained in [ sec : dust_crystal ] show that we should expect hexagonal close packing ( hcp ) lattice for grains in dusty plasma . this coincides with numerical simulations in @xcite and experiments @xcite . last * section [ sec : electrons_on_helium ] * shows how to use developed methods for electrons on the liquid - helium surface . it is shown that in some cases much more simple ( but more rough as well ) compared to section [ sec : dust_crystal ] approximation may be obtained . this approximation is used to calculate localization distance and then compared to classical result @xcite . when temperature @xmath0 they fully coincide . besides , basing on this results some `` guess '' when lattice can exist is done . obtained equation can be verified experimentally . during this article we tried to provide rigorous exposition of presented ideas , but without physical sole being lost in lots of equations . thus , calculations that heavily rely on mathematical transfomations only , were moved to appendices . reader , interested only in `` physical part '' , may omit them . we start with . first two summands can be easily integrated if we mention @xmath133 but last summand should be integrated in terms of special functions , namely dilogarithm @xcite @xmath134 if we mention that @xmath135 we would get @xmath136 & = \frac{m^2}{8 \pi^2 \hbar^4 \beta } \iint d^2 \vec{r}\ , d^2 \vec{r}^{\,\prime } v(|\vec{r}-\vec{r}^{\,\prime}| ) \times\\ & \times \ln\left(1+e^{\beta \mu(\vec{r}\,)}\right ) \ln\left(1+e^{\beta \mu(\vec{r}^{\,\prime})}\right ) + \\ & + \frac{m}{2 \pi \hbar^2 \beta } \int d^2 r \left [ \beta\mu(\vec{r}\,)\ln\left(1+e^{\beta \mu(\vec{r}\,)}\right ) -\right.\\ & \left . - \frac{\beta^2\mu^2(\vec{r}\,)}{2 } - \text{li}_2 \left(-e^{-\beta \mu ( \vec{r}\ , ) } \right ) - \frac{\pi^2}{6 } \right ] , \end{aligned}\ ] ] which can be simplified further . first of all we will use landen s identity @xcite @xmath137-\infty;0\right ] \end{aligned}\ ] ] with @xmath138 and expand @xmath139 . this leads to @xmath136 & = \frac{m^2}{8 \pi^2 \hbar^4 \beta } \int \int d^2 \vec{r}\ , d^2 \vec{r}^{\,\prime } v(|\vec{r}-\vec{r}^{\,\prime}| ) \times\\ & \times \ln\left(1+e^{\beta \mu(\vec{r}\,)}\right ) \ln\left(1+e^{\beta \mu(\vec{r}^{\,\prime})}\right ) + \\ & + \frac{m}{2 \pi \hbar^2 \beta } \int d^2 r \left [ \text{li}_2\left(\frac{1}{1+e^{\beta\mu(\vec{r}\,)}}\right ) - \frac{\pi^2}{6 } + \right.\\ & \left . + \frac{1}{2}\ln^2\left(1+e^{\beta\mu(\vec{r}\,)}\right ) \right ] . \end{aligned}\ ] ] in we present how to calculate probability distribution function for the whole system basing on single - particle probability distribution function . it is based on the decomposition of @xmath43 using @xmath140 as basis functions in inverse lattice space . here we add some consideration about connection between fourier series and presented one . let s consider a bijection @xmath73 from @xmath48 to @xmath144 and its inverse @xmath145 ( see fig . [ fig : vectors ] for geometrical reasoning ) @xmath146 it may be seen that rows of @xmath73 are components of inverse lattice basis vectors . we are interested in case @xmath147 . since we know that @xmath148 one can immediately write @xmath149 and same way we consider second equation from @xmath150},\ ] ] where @xmath151 is jacobian . here @xmath152 is treated as new variable with domain @xmath48 . jacobian @xmath153 = abc\sin(\alpha)\sin(\beta_c ) = 1/\bar{\rho}$ ] is actually the volume of @xmath48 or inverse mean particle density . designating @xmath154 as @xmath89 in and , changing variable in @xmath155 and designating exponent in as @xmath156 ( it can be seen that exponent in is @xmath157 , where @xmath69 designates complex conjugation ) we immediately get equations moreover , obtained result means that all properties of fourier series can be applied to decomposition . last thing to mention are periodical properties of presented series . we are going to expand domain to @xmath49 so that @xmath41 will be defined everywhere in the space . so now we need to explore behavior of this function . from one may see , if @xmath158 , @xmath159 and @xmath160 : @xmath161 where @xmath162 , @xmath163 and @xmath164 are unit vectors along coordinate axis . with regard to @xmath165 and from @xmath166 last equation justifies our view of connection between @xmath67 and @xmath43 as it is presented on fig.[fig : one_to_many ] . at this point we are interested in specific form of @xmath43 , thus it is explicitly substituted into . besides we designate @xmath168 and rewrite expression in cartesian coordinates changing multiple integral to the product of integrals @xmath169 lets start with . inner integral has infinite borders so we may rewrite this expression as follows @xmath173 since @xmath71 is real it can be replaced with complex conjugate @xmath174 without changing @xmath58 . now we can substitute @xmath67 from and mention that @xmath175 . so if we designate @xmath87 as in it can be written @xmath176 last equation can be simplified if we perform integration . from appendix [ subsec : appendix_properties_of_f_k ] we expect orthogonality of @xmath156 functions . let s mention from @xmath177 and we will get @xmath178 here @xmath179 product of three kronecker s delta . we will consider screened coulomb potential , which means integration over @xmath103 can be performed . one may use following relation @xcite @xmath186 and combine it with equation for @xmath87 to get @xmath187 changing variable @xmath188 and performing integration over @xmath189 we will get @xmath190 last expression is equivalent to . one may check this by expanding @xmath154 and @xmath191 . we will start with equation and assume that @xmath43 is a very `` sharp '' function ( section [ subsec : two_dim_fermi ] ) . with this assumption we can change integration limits to infinite and this , in turn , will allow us to perform variable exchange as we did in @xmath192 first of all we perform some mathematical transformations with ( in cartesian coordinate system ) and write the following @xmath193 ^ 2/s^2 - \vec{r}^{\,\prime\,2}/\left(4s^2\right)}.\ ] ] now we can substitute this expression into and perform integration over @xmath194 . writing result in spherical coordinate system and integrating over angles we immediately get . since we know explicit expression for @xmath94 we can substitute it into and perform integration using definition of the complementary error function @xcite @xmath195 as result we will get @xmath196 since we may obviously suppose that @xmath3 is very small compared to screening distance , then @xmath197 . so we can approximate last equation and as result we will get . we start with expression for @xmath58 . but one may see that it contains expressions @xmath198 . components of @xmath199 matrix in are proportional to different products of inverse distances in a lattice , e.g. @xmath200 , @xmath201 , and so on ( see ) . since @xmath202 , @xmath203 and @xmath204 ( assumption about `` sharp gaussian '' ) we expect @xmath205 . this means , that exponent `` starts acting '' only for terms with very large @xmath206 . in this section we will try to rewrite equations to get even better convergence then we have . we notice , that variable @xmath208 takes values much smaller then mean distance as well as much bigger . thus we split this integral into two . first integration we will provide from @xmath209 to @xmath116 and second from @xmath116 to infinity . we assume that lattice is not too degenerated ( section [ subsec : coulomb_potential ] ) , which means that angles @xmath63 and @xmath65 should not significantly differ from @xmath210 . so we may expect elements of @xmath211 to be less than @xmath212 for first integral and greater than @xmath212 for second . if so , then second integration may be performed and we end up with the sum that converges much better than . @xmath213 ^ 2 + \vec{k}^{\,t } \hat{g } \hat{g}^{t } \vec{k } \right ) } } { 1/\lambda_d^2 + 4 \pi^2 \vec{k}^{\,t } \hat{g } \hat{g}^{t } \vec{k}}. \end{aligned}\ ] ] from one may check with sylvester s criterion @xcite that @xmath73 is a positive - definite matrix . obviously , we expect @xmath214 and their product @xmath215 to be positive - definite matrices as well . since @xmath208 takes only positive values @xmath211 is positive as well . this means highlighted sum in the last equation is a riemann theta function @xcite at point @xmath216 , so we appropriately designate it with @xmath217 . we used the fact that @xmath222 and that multiplication by @xmath223 is equal to multiplication by diagonal matrix which has all elements equal to @xmath223 . explicit expression for @xmath145 may be used from . besides , we may notice that @xmath224 and rewrite expression for @xmath58 as follows @xmath225 ^ 2 + \left| \hat{g}^{t } \vec{k } \right|^2 \right ) } } { 1/\lambda_d^2 + 4 \pi^2 \vec{k}^{\,t } \hat{g } \hat{g}^{t } \vec{k}}. \end{aligned}\ ] ] now we can perform integration and find @xmath226 . since result is very complicated we first introduce two helper functions @xmath227 , \end{aligned}\ ] ] and present result in terms of this functions @xmath228 our previous calculations do not rely on any approximation , but expression for @xmath230 is very complicated , so we need to perform one . before we do , one should show it is acceptable to approximate this sum . subsequent calculations are divided into two parts : _ uniform convergence proof _ and _ approximation _ itself . * proof of uniform convergence*. we will use cauchy criterion for sequence of functions @xmath231 in domain @xmath132 to achieve this goal : @xmath232 . further @xmath233 acts as a sequence for convergence checking and entries of @xmath145 as well as @xmath3 , @xmath116 and @xmath86 are treated as variables from domain of admissible parameters . we have already examined properties of @xmath145 and will not get into details again . first of all we use @xmath234 . this relation holds for physical system under consideration and we will use it in all subsequent calculations . this means @xmath235 and thus all terms in sum will be positive and we can avoid using absolute value . the second thing that appears is following bounding @xmath236 then we use relation @xmath237 for @xmath238 @xcite and replace @xmath239 in @xmath240 . expanding squares in exponents and performing simplification we get @xmath241 besides @xmath242 was used . now we can use known relation @xmath243 , where @xmath244 is minimal eigenvalue of @xmath245 and @xmath69 designates hermitian adjoint @xcite . we know that @xmath246 and thus @xmath247 has nonzero eigenvalues . taking the minimal one @xmath244 and designating @xmath248 we claim @xmath249 . now we should change multidimensional summation to one - dimensional . since all summands are positive we can rearrange them in any convenient order . thus we will perform summation over `` cube surfaces '' . first of all let s designate @xmath250 . then total sum can be expressed as @xmath251 where number of summands in @xmath252 can be easily calculated as @xmath253 . minimal length for index vector in this set of indices is @xmath254 . collecting together previous calculations we get @xmath255 further proof is supposed to be obvious and thus we avoid it . proof of uniform convergence for first summand in @xmath58 can be performed much more easily by presented scheme and thus is avoided as well . instead one useful remark shoud be done . series converges better when @xmath256 is bigger . taking physical meaning of this variables into account we get following statement : if we project vectors @xmath60 , @xmath61 and @xmath62 on axis @xmath257 , @xmath258 and @xmath259 respectively ( see figure [ fig : vectors ] ) and divide it by mean distance in lattice @xmath260{\bar{\rho}}$ ] , we will get measure of how good current series converges , or by physical meaning how close is current lattice to cubic . * approximation . * we take into account @xmath261 , neglect all terms containing @xmath262 and @xmath263 in @xmath239 , and approximate @xmath264 @xcite . obviously @xmath265 and thus all @xmath266 with @xmath209 arguments are neglected . besides , we neglect all small terms in exponent and denominators . to make this procedure more clear we state @xmath267 . it means screening distance is big enough not to feel lattice deviation from cubic . this leads us to equations @xmath268 to get the first one , identity @xmath269 and approximation @xmath270 @xcite were used . integrating over @xmath183 we get @xmath273 then we substitute last equality and expression for @xmath274 from into equation for @xmath87 . it is known @xcite , if @xmath275 and @xmath276 @xmath277 with this equation we easily obtain @xmath278 for capillary interaction part @xmath279 comparison with shows that they are equal . for two - dimension density it is true @xmath283 . besides , we perform integration over @xmath208 and get @xmath284 where @xmath285 is gamma - function . since w know that @xmath109 last equation can be approximated . we use known approximation @xcite @xmath286 , where @xmath287 is euler mascheroni constant . we neglect this constant and exponent . the rest can be written as . 99 y.bilotsky , advances in materials science and applications , vol.2 , iss.4 , pp.127137 , dec.(2013 ) . l.n.kantorovich and i.i.tupitsyn , j.phys . : matter , `` coulomb potential inside a large finite crystal , '' vol.11 , pp.61596168 . l. p. buhler , r. e. crandal , j. phys . a : math . gen . , `` on the convergence problem for lattice sums , '' vol.23 , pp.25232528 , ( 1990 ) . d.borwein , j.m.borwein , r.shail and i.j.zucker , `` energy of static electron lattices , '' j.phys.a : math.gen . , vol.20 , pp.15191531 , ( 1988 ) . v.e.fortov et al . , phys.rep . * 421 * , 1 ( 2005 ) . h.lowen , phys.rep . * 237 * , 249 ( 1994 ) . g.e.morfill , h.m.thomas , u.konopka , and m.zuzic , phys.plasmas * 6 * , 1769 ( 1999 ) . e.helpand and f.h.stilinger , j.chem.phys . * 49 * , 1232 ( 1968 ) . b.klumov , g.joyce , c.rth , p.huber et all , _ structural properties of 3d complex plasmas under microgravity conditions _ , epl , vol.92 , no.1 , p.15003 , ( 2010 ) . b. a. klumov , g. e. morfill , _ structural properties of complex ( dusty ) plasma upon crystallization and melting _ , jetp letters , vol.90 , no.6 , pp . 444448 , ( 2009 ) . b.i.lev and a.g.zagorodny , phys.lett.a * 373 * , 158 ( 2009 ) . b.i.lev , v.b.tymchyshyn , and a.g.zagorodny , condens.phys . * 12 * , 593 ( 2009 ) . h.thomas , g.e.morfill , v.demmel , j.goree , b.feuerbacher , and d.mohlmann , phys.rev.lett . * 73 * , 652 ( 1994 ) . j.h.chu and lini , phys.rev.lett . * 72 * , 4009 ( 1994 ) . a.melzer , t.trottenberg , and a.piel , phys.lett.a * 191 * , 301 ( 1994 ) . s. v. vladimirov , s. a. khrapak , m. chaudhuri , and g.e.morfill , phys.rev.lett . * 100 * , 055002 ( 2008 ) . h.ikezi , phys . fluids . * 29 * , 1764 ( 1986 ) . a.melzer , a.homann , and a.piel , phys.rev.e * 53 * , 2757 ( 1996 ) . a.g.sitenko , a.g.zagorodny , and v.n.tsytovich , aip conf.proc . * 345 * , 311 ( 1995 ) . s.a.brazovsky , sov.phys . jetp * 68 * , 715 ( 1975 ) . b.i.lev and h.yokoyama , ijmp b ( international journal of modern physics ) * 17 * , 4913 ( 2003 ) . h.totsuji , t.kishimoto , and c.totsuji , phys.rev.lett . * 78 * , 3113 ( 1997 ) . p.leiderer two - dimensional electron systems by ed . e.y . andrei , ( springer netherlands 1997 ) . v.b.shikin and p.leiderer , sov.phys . jetp lett . , * 54 * , 92 , ( 1981 ) . d.c.tsui , h.l.stormer , and a.c.gossard , phys.rev . lett . * 48 * , 1559 ( 1982 ) . r.b.laughlin , phys.rev.lett . * 50 * , 1395 ( 1983 ) . v. s. edelman , uspekhi fizicheskikh nauk * 23 * , 227 ( 1980 ) . t.ando , a.fowler , and f.stern , rev . phys . * 54 * , 437 ( 1982 ) . w.wigner , phys . rev . * 46 * , 1002 ( 1934 ) . c.c.grimes , and g.adams , phys . lett . * 42 * , 795 ( 1979 ) . p.m.platzman and h.fukuyama , phys . b * 10 * , 3150 ( 1974 ) . i.skachko , _ phase diagram of a 2-dimensional electron system on the surface of liquid helium _ , thesis , rutgers , the state university of new jersey ( 2006 ) . d.k.lambert , _ electrons on the surface of liquid helium _ , ph.d . thesis , lawrence berkeley laboratory ( 1979 ) . m.haque , i.paul , and s.pankov , phys . b * 68 * , 045427 ( 2003 ) . b. i. lev , v. p. ostroukh , v. b. tymchyshyn , and a.g.zagorodny , _ statistical description of the system electrons on the liquid helium surface _ , eur.phys.j . b , 87:253 ( 2014 ) . l.p.gorkov and d.m.chernikova , jetp lett . * 18 * , 119 ( 1973 ) . d. ruelle , _ statistical mechanics : rigorous results _ , w.a.benjamin , new york , ( 1969 ) . a.isihara , _ statistical mechanics _ , academic press , new york - london , ( 1971 ) . k.huang , _ statistical mechanics _ , j. wiley and sons , new york , ( 1963 ) . r.baxter , _ exactly solved models in statistical mechanics _ , academic press , new york , ( 1982 ) . y.d.bilotsky and b.i.lev , teor . fiz . * 60 * , 120 ( 1984 ) . b.i.lev and a.y.zhugaevych , phys . e , * 57 * , 6460 ( 1998 ) . b.i.lev and a.g.zagorodny , statistical description of coulomb - like systems , phys.rev.e * 84 * , 061115 ( 2011 ) . r. l. stratonovich , soviet physics doklady , vol . 2 , ( 1958 ) . j.hubbard , phys . * 3 * , 77 ( 1959 ) . s.f.edwards and a.lennard , j. math . phys . * 3 * , 778 ( 1962 ) . s.samuel , phys . d * 18 * , 1916 ( 1978 ) . a.erdlyi , w.magnus , f.oberhettinger , and f.g.tricomi , _ higher transcendental functions _ , vol.1 . , mcgraw - hill , new york , ( 1953 ) . i.s.gradshteyn and i.m.ryzhik , _ table of integrals , series , and products _ , amsterdam : elsevier / academic press , ( 2007 ) . a.erdlyi , w.magnus , f.oberhettinger , and f.g.tricomi , _ higher transcendental functions _ , malabar , krieger , ( 1981 ) . r.bellman , _ introduction to matrix analysis _ , society for industrial and applied mathematics , ( 1970 ) . r.bellman , _ a brief introduction to theta functions _ , holt , rinehart and winston , ( 1961 ) . m.chiani , d.dardari , m.k.simon , new exponential bounds and approximations for the computation of error probability in fading channels , ieee transactions on wireless communications , 4(2 ) , pp 840845 , ( 2003 ) . c.meyer , _ matrix analysis and applied linear algebra _ , siam , ( 2000 ) . m.abramowitz , i.stegun , _ handbook of mathematical functions with formulas , graphs , and mathematical tables _ , tenth printing , ( 1972 ) .

inspired by the recent creation of the honeycomb optical lattice and the realization of the mott insulating state in a square lattice by shaking , we study here the shaken honeycomb optical lattice . for a periodic shaking of the lattice , a floquet theory may be applied to derive a time - independent hamiltonian . in this effective description , the hopping parameters are renormalized by a bessel function , which depends on the shaking direction , amplitude and frequency . consequently , the hopping parameters can vanish and even change sign , in an anisotropic manner , thus yielding different band structures . here , we study the merging and the alignment of dirac points and dimensional crossovers from the two dimensional system to one dimensional chains and zero dimensional dimers . we also consider next - nearest - neighbor hopping , which breaks the particle - hole symmetry and leads to a metallic phase when it becomes dominant over the nearest - neighbor hopping . furthermore , we include weak repulsive on - site interactions and find the density profiles for different values of the hopping parameters and interactions , both in a homogeneous system and in the presence of a trapping potential . our results may be experimentally observed by using momentum - resolved raman spectroscopy .

1111.2210

the study of dirac points , i.e. the contact points between different energy bands with an approximate linear dispersion relation , has become a major issue since the experimental breakthrough in graphene - based electronics @xcite . indeed , the low - energy electronic properties of graphene are governed by a pseudo - relativistic 2d dirac equation for massless fermions situated at the @xmath0 and @xmath1 corners of the brillouin zone @xcite . the dirac points are topologically protected and a gap is opened only when the inversion symmetry of the lattice or the time - reversal symmetry are broken . the possibility to generate topological phase transitions in graphene - like systems has recently attracted a great deal of attention . within a tight - binding description , an anisotropy in the nearest - neighbor hopping parameters makes the dirac points move away from the high - symmetry @xmath0 and @xmath1 points and , under appropriate conditions , merge at time - reversal invariant points in the first brillouin zone @xcite . most saliently , this merging of dirac points is associated with a topological phase transition between a semimetallic phase and a gapped band - insulating phase . an experimental investigation of the merging transition in graphene turns out to be problematic , since in order to appropriately modify the hopping parameters , an unphysically large strain needs to be applied to the graphene sheet @xcite . an alternative system for the study of such topological transitions is that of ultracold atoms trapped in a honeycomb optical lattice . since the seminal realization of the superfluid - mott - insulator transition in the bose - hubbard model , ultracold atoms in optical lattices have become promising systems to emulate condensed - matter physics . indeed , the lattice geometry , the dimensionality , the atomic species , as well as the interactions can be engineered with a high degree of precision . the more involved triangular and honeycomb geometries were recently realized experimentally and exotic correlated states of matter have been observed experimentally @xcite or predicted theoretically @xcite . the application of a time - periodic perturbation on the optical lattice introduces yet another parameter scale into the system . a periodic shaking of the optical lattice , up to the khz frequency range , has been implemented by placing one of the mirrors used to create the optical lattice on a piezoelectric material , such that the mirror can be moved back and forth in the direction of the beam @xcite . the floquet formalism shows that the hopping energy of the atoms in the shaken lattice is renormalized by a bessel function , as a function of the shaking frequency and amplitude , thus allowing both the magnitude and the sign of the hopping parameter to change . this rather counter - intuitive phenomenon , as compared to the standard tight - binding physics , has been experimentally observed in a one - dimensional cold - atomic system @xcite . in this paper , we consider ultracold fermions trapped in a _ shaken _ honeycomb optical lattice . within the floquet formalism , we derive an effective hamiltonian that generalizes that of a graphene - like material under strain . in particular , we find that the alignment and merging of dirac points in momentum space are now accessible with ultracold fermions in the shaken optical lattice and the phase diagram consists of various phases of the corresponding solid - state system that are otherwise difficult to realize . furthermore , by taking into account a hubbard - like interaction for spinful fermions , we study the density profiles for the homogenous and the trapped gas within a hartree - fock theory . the outline of this paper is the following : in sec.[sec01a ] we introduce the time - dependent hamiltonian and in sec.[sec01b ] we derive the time - independent one , by applying the floquet formalism . in sec.[sec02 ] we investigate the merging and alignment of dirac points , when the optical lattice is shaken along specific directions . the description is extended to include interactions in sec.[sec03 ] , where we derive the dependence of the density on the chemical potential . implications of our results for experiments are discussed in sec.[sec04 ] . finally , our conclusions are presented in sec.[conc ] . in this section , we derive a time - independent effective description for ultracold atoms trapped in a periodically shaken honeycomb optical lattice by utilizing a floquet theory . for simplicity , we focus on a system of single - component fermionic atoms and consider only single - particle terms in this section . the results from the floquet theory are valid for fermionic atoms with internal degrees of freedom as well as for bosonic atoms . in particular , the hyperfine state of fermionic atoms , playing the role of an effective spin-@xmath2 degree of freedom for electrons , will be considered when interaction effects are taken into account in sec . iv . in the tight - binding limit , the system of ultracold fermionic atoms trapped in a 2d _ shaken _ honeycomb optical lattice can be described by the hamiltonian @xmath3 which consists of two distinct parts . the static part @xmath4 is simply the tight - binding hamiltonian in the honeycomb lattice , where @xmath5 ( @xmath6 ) and @xmath7 ( @xmath8 ) are , respectively , fermionic creation and annihilation operators on the lattice site @xmath9 in the @xmath10 ( @xmath11 ) sublattice . the three vectors @xmath12 connect an @xmath10-lattice site with its three nearest - neighbor ( nn ) @xmath11-lattice sites and are given in terms of the distance @xmath13 between nn sites , where @xmath14 is the laser wave number ( see fig.[fig : lattice ] ) . here , @xmath15 characterize the energy gained in hopping to the nn and next - nearest - neighbor ( nnn ) sites , respectively , and @xmath16 is the on - site energy . we remark that the nnn hopping is taken into account because the nn hopping may be rendered vanishingly small in the effective time - independent description . in this regime , the nnn may become the dominant kinetic term . in a square lattice , where the potential is separable in independent @xmath17 and @xmath18 components , the nnn hopping is identically zero @xcite . however , the nnn hopping can be nonzero in the honeycomb lattice , since its potential is not separable in @xmath17 and @xmath18 components . nevertheless , it may be expressed as the sum of two triangular lattices . , red dots , and @xmath11 , blue dots ) . the vectors @xmath19 , @xmath20 , and @xmath21 connect a site on the @xmath10 sublattice to its nearest neighbors on the @xmath11 sublattice . ] the time - dependent part of the hamiltonian ( [ basich ] ) , @xmath22 describes the harmonic shaking of the lattice in the direction @xmath23 with a driving frequency @xmath24 in the co - moving frame of ref . @xcite . as a consequence of the transformation to the co - moving frame , @xmath25 describes atoms of mass @xmath26 experiencing a position - dependent sinusoidal force . the unavoidable complexity that arises when dealing with a quantum many - body system out of equilibrium has recently motivated the development of new theoretical tools , for example time - dependent density matrix renormalization group @xcite , time - dependent dynamical mean field theory @xcite , and exact diagonalization @xcite . however , for a periodically driven quantum system , the floquet theory offers a simplified description of the system , in the form of a time - independent effective hamiltonian , if the period @xmath27 is the shortest time scale in the problem @xcite . in this limit , the atoms can not follow the shaking motion adiabatically and remain thus at their average lattice position , albeit with renormalized hopping parameters . the system is thus considered to be in a stationary state and the knowledge of equilibrium physics can be employed . let us consider the floquet hamiltonian defined by @xmath28 , where @xmath29 is periodic in time @xcite . the eigenvalue equation is then given by @xmath30 where @xmath31 is the quasienergy defined uniquely up to a multiple of @xmath32 . any solution @xmath33 is part of a set of solutions @xmath34 with integer @xmath35 , which all correspond to the same physical solution . hence , the spectrum of the floquet hamiltonian possesses a brillouin - zone - like structure @xcite . the interest therefore lies with the states in the first brillouin zone , i.e. states with quasi - energies @xmath36 . the space in which the states @xmath33 are defined is the composite of the hilbert space spanned by square integrable functions on configuration space , @xmath37 , and the space of @xmath38-periodic functions . the state @xmath33 may be written down in an orthonormal basis in the composite space according to @xmath39 |\alpha ( q ) \rangle,\ ] ] where @xmath40 are coefficients to normalize @xmath33 and the operator @xmath41 can be any @xmath38-periodic hermitian operator . therefore , we can conveniently choose @xmath41 to be @xmath42 , such that @xmath43 . if the condition @xmath44 \ , \exp [ i ( n - n ' ) \omega t ] \ , h_0 \exp[-i \hat{f}(t ) ] \rangle_t | \ , \alpha ( q ) \rangle \ll \hbar \omega \label{eq : condition}\ ] ] is satisfied for any two states @xmath45 and @xmath46 , then the eigenvalues @xmath47 are approximately @xmath48 \ , h_0 \ , \exp[-i \hat{f}(t ) ] \rangle_t \,| \,\alpha ( q ) \rangle . \label{eq : floquet - approximation}\ ] ] here , @xmath49 denotes the time average of the operator @xmath50 over the period @xmath38 . the condition ( @xmath51 ) will hold for @xmath52 , if @xmath53 is nearly constant during the period @xmath38 , which is small if @xmath24 is large . in this case , states with different @xmath35 do not mix . if @xmath24 is large enough , such that the condition ( @xmath51 ) also holds for @xmath54 , then the energy spectrum will split up into energy bands labelled by an index @xmath35 , where the details within the energy band are determined by @xmath53 . because the states with a different index @xmath35 are separated by an energy which is a multiple of @xmath32 and because the spectrum possesses a brillouin - zone - like structure , only the terms with @xmath55 need to be taken into account . the effective hamiltonian @xmath56 , which gives rise to the same spectrum as the floquet hamiltonian , is then defined by @xcite @xmath57 \ , h_{0 } \ , \exp[-i \hat{f}(t ) ] \bigg>_t \nonumber \\ & = \bigg < \sum^\infty_{n=0 } \frac{i^n}{n ! } [ \hat{f}(t ) , h_{0}]_n \bigg>_t . \label{eq : effective - ham}\end{aligned}\ ] ] here , @xmath58_n$ ] denotes the multiple commutator , which is defined by @xmath58_{n+1 } = [ \hat{f},[\hat{f},\hat{g}]_n]$ ] and @xmath58_0 = \hat{g}$ ] . effective hamiltonians corresponding to eq.([eq : effective - ham ] ) have been derived for linear shaking of a one - dimensional lattice @xcite and for elliptical shaking of a triangular lattice @xcite . for the shaken honeycomb lattice studied here , the condition ( @xmath51 ) is satisfied if @xmath59 , and the effective hamiltonian becomes @xmath60 where the renormalized nn hopping parameters @xmath61 are given by @xmath62 and the renormalized nnn hopping parameters are given by @xmath63 ( see appendix [ app1 ] for detailed calculations ) . in these expressions , @xmath64 denotes the zeroth order bessel function of the first kind , which shows a damped oscillation around zero . in terms of the renormalized nn and nnn hopping parameters , the diaginalization of the effective hamiltonian ( [ hamiltonian - eff ] ) yields the dispersion relation @xmath65 where @xmath66 is the band index , and we have defined the functions @xmath67 and @xmath68.\ ] ] in this section , the honeycomb lattice with anisotropic hopping is studied . in the first two subsections , only nn hopping is considered for illustration reasons . indeed , this allows for a simple understanding of the main consequences of shaking on dirac - point motion and dimensional crossover . in subsec . [ sec02c ] , we discuss how the picture evolves when nnn hopping is included , and the sign of the hopping parameters is investigated in subsec . [ sec02d ] . since the system with two nn equal hopping parameters and a single independent one captures the essential features of the systems with three independent nn hopping parameters , we will focus on this system . the numbering of the @xmath61s is chosen such that @xmath69 , which can be achieved by shaking in a direction parallel or perpendicular to @xmath19 . although the atoms in the optical lattice are charge - neutral objects , we shall adopt the language from condensed - matter physics and call a zero - gap phase with a pair of dirac cones and a vanishing density of states at the band - contact points a _ semimetal _ , whereas a gapped phase is called _ band insulator_. furthermore , nnn hopping induces a _ metallic _ phase for small values of @xmath61 because of an overlap between the two bands that yields a non - vanishing density of states at the energy level of the band - contact points . if the latice is shaken in the direction perpendicular to @xmath19 ( direction 1 in fig.[fig : lattice ] ) , @xmath70 remains equal to @xmath71 , whereas @xmath72 and @xmath73 are renormalised to a smaller value . an increase in the shaking amplitude results in a decrease in @xmath74 , which is depicted by the arrow @xmath75 in fig.[fig : phase - diagram - abs](a ) . when the hopping parameters change according to this arrow @xmath75 , the energy spectrum evolves from fig.[fig : merging - energy - dispersion](a ) to fig.[fig : merging - energy - dispersion](b ) . the dirac points , originally situated at the corners @xmath0 and @xmath1 of the first brillouin zone , start to move in the @xmath76-direction along the vertical edges of the latter . this motion is depicted by the arrows in fig.[fig : phase - diagram - abs](b ) . even if the two dirac points are no longer located at the high - symmetry points @xmath0 and @xmath1 , they remain related by time - reversal symmetry , such that their berry phases @xmath77 and @xmath78 are opposite . this non - zero berry phase topologically protects each of the dirac points and thus the semimetallic phase remains robust until @xmath79 , where the two points merge at a time - reversal invariant momentum , i.e. half of a reciprocal lattice vector @xcite . in the present example , this point is situated at the center of the vertical edges of the first brillouin zone , and the band dispersion becomes parabolic in the @xmath80-direction while remaining linear in the @xmath81-direction [ see fig.[fig : merging - energy - dispersion](b ) ] . the merged dirac points are no longer topologically protected due to the annihilation of the opposite berry phases . consequently , a further increase of the shaking amplitude , which results in a further decrease of @xmath82 , leads to the opening of a gap between the two bands . thus , the system undergoes a topological phase transition from a semimetal to a band insulator . this merging transition was also studied in a static setup in ref . @xcite , where the hopping amplitudes @xmath71 were proposed to be modified by a change in the intensity of one of the lasers used to create the optical lattice . in contrast to this static setup , shaking the honeycomb lattice allows one to completely annihilate some of the nn hopping parameters and to even change their sign . this sign change occurs at the zeros of the bessel function [ see eq.([eq : renormalized - gamma ] ) ] . for an example system of @xmath83k atoms in a lattice created by lasers with a wavelength of 830 nm , which is shaken in the direction perpendicular to @xmath19 , the situation @xmath84 is encountered for @xmath85 which corresponds to the first zero of the bessel function . at this particular point , and if @xmath86 in addition , the system consists of a set of effectively decoupled horizontal bonds along which the atoms are solely allowed to hop . this yields two flat bands at @xmath87 [ see fig.[fig : merging - energy - dispersion](c ) ] that may be viewed as the extreme limit of the band - insulating phase . alternatively , one may view this situation upon decreasing the value of @xmath82 as a dimensional crossover from a 2d band insulator to a zero - dimensional ( 0d ) system . a small non - zero value of @xmath82 simply provides a weak dispersion of these decoupled bands ( not shown ) . , @xmath88 , and @xmath86 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the isotropic case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ] , @xmath88 , and @xmath86 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the isotropic case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ] , @xmath88 , and @xmath86 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the isotropic case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ] , @xmath88 , and @xmath86 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the isotropic case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ] . here , we have chosen @xmath86 . ( b ) dirac - point motion in the first brillouin zone for a shaking direction perpendicular to @xmath19 [ direction @xmath75 in the phase diagram ( a ) ] . ( c ) dirac - point motion in the first brillouin zone for a shaking direction parallel to @xmath19 [ direction @xmath10 in the phase diagram ( a ) ] . the contour plots depict the dispersion of the isotropic system with an arbitrary color scale . the area with higher contrast is the first brillouin zone . ] another dimensional crossover , from 2d to 1d , may be obtained if the lattice is shaken in the direction parallel to one of the nn vectors ( direction 2 in fig.[fig : lattice ] ) . here , we choose @xmath93 to maintain the symmetry @xmath94 . in this case , both @xmath70 and @xmath82 are renormalized by bessel functions , albeit with different arguments . since all hopping parameters are renormalized , the trajectory of the system in the phase space upon increasing the shaking amplitude is not a straight line , as was the case for shaking perpendicular to a hopping direction , and has a new feature : the alignment of dirac points , which occurs for @xmath95 . the first zero of @xmath70 is found at @xmath96 for the same system of @xmath83k atoms mentioned above . here , for illustrative purposes , a simplified trajectory of the system is depicted in fig.[fig : merging - energy - dispersion ] by arrow @xmath10 , which corresponds to the motion of the dirac points in reciprocal space as shown in fig.[fig : phase - diagram - abs](c ) . as @xmath70 approaches zero , the dirac points align in lines parallel to the @xmath81-axis at @xmath97 and the energy barriers between the aligning points are lowered . consequently , when @xmath95 , the energy spectrum contains lines where the two energy band meet and the dispersion is linear , as is shown in fig.[fig : merging - energy - dispersion](d ) . the dispersion relation ( [ dispersionrel ] ) then reads simply @xmath98 and one clearly sees the 1d character . indeed , there is no dispersion in the @xmath89-direction , as is also evident from fig.[fig : merging - energy - dispersion](d ) , and the system may be viewed as completely decoupled 1d chains in which the zig - zag arrangement is of no importance . in this particular limit , the sites a and b are therefore no longer inequivalent such that the unit cell is effectively divided by two , and the size of the first brillouin zone is consequently doubled . the aligned dirac points may thus , alternatively , be viewed as due to an artificial folding of the second ( outer ) half of the first brillouin zone into its inner half . however , this aspect is very particular in that the brillouin zone immediately retrieves its original size when @xmath70 is small , but non - zero , or if nnn hoppings are taken into account . in both cases , one needs to distinguish the two different sublattices and one obtains a dispersion in the @xmath89-direction . the actual behavior of the system for an increasing shaking amplitude is discussed in sec.[sec04 ] . this behavior is more complicated because all three nn hopping parameters are renormalized , which , beyond the alignment , leads to the merging of dirac points and the opening of a gap also in the case of shaking parallel to one of the nn vectors . in the absence of nnn hopping , the 0d limit can be reached in addition . the major consequence of nnn hopping is to break particle - hole symmetry , as may be seen from eq . ( [ dispersionrel ] ) , where non - zero values of @xmath99 yield @xmath100 . its relevance depends sensitively on the shaking direction , because of the different renormalization of the nn hopping parameters . the band structure with nnn hopping included is depicted in fig.[fig : nnn - energy - dispersion ] for different shaking directions . in the case of a shaking perpendicular to @xmath19 , only @xmath82 are decreased , whereas @xmath101 remains the leading energy scale in the band structure . as a reference direction , but it may naturally be replaced by any other direction @xmath102 , in which case @xmath103 remains constant . ] the band structure for the unshaken lattice is depicted in fig.[fig : nnn - energy - dispersion](a ) for @xmath104 , and one notices that the main features of the band structure , namely the dirac points , are unaltered with respect to the case @xmath86 in fig.[fig : merging - energy - dispersion](a ) , apart from the flattening of the upper band as compared to the lower one . when approaching the merging transition @xmath79 , the value of which is determined by the zeros of @xmath105 in eq . ( [ fq ] ) and that therefore does not depend on the nnn hopping parameters , the band width remains dominated by the largest hopping parameter @xmath70 , such that the band structure [ fig.[fig : nnn - energy - dispersion](b ) ] at the transition is essentially the same as in fig.[fig : merging - energy - dispersion](b ) for @xmath86 . in the 0d limit , with @xmath106 the originally flat bands [ fig.[fig : merging - energy - dispersion](c ) ] acquire the weak dispersion of a triangular lattice as a consequence of the non - zero nnn hopping parameters . however , as expected from the above arguments , the dispersion is on the order of @xmath107 and thus small as compared to the energy separation @xmath108 between the two bands . in contrast to a shaking direction perpendicular to @xmath19 , nnn hopping has more drastic consequences if the lattice is shaken in the direction parallel to @xmath19 . in this case , all nn hopping parameters are decreased , and the relative importance of nnn hopping is enhanced . notice further that the nnn lattice vectors @xmath109 are now perpendicular to the shaking direction such that @xmath110 remains unrenormalized . also in this case , the system is approaching the 1d limit , with @xmath111 [ see fig.[fig : nnn - energy - dispersion](d ) ] . however , in contrast to fig.[fig : merging - energy - dispersion](d ) , the chains remain coupled by nnn hopping that yields a dispersion in the @xmath89-direction . furthermore , as mentioned above , the a and b sites are now not equivalent from a crystallographic point of view , such that the outer parts of the first brillouin zone can not be folded back into the inner one , as may be seen from fig.[fig : nnn - energy - dispersion](d ) . finally , for particular values of the shaking amplitude in the direction parallel to @xmath19 , the nn hopping parameters can be decreased in such a manner as to render @xmath82 more relevant . in this case , the two bands can overlap in energy , as depicted in figs.[fig : nnn - energy - dispersion](e ) and [ fig : nnn - energy - dispersion](f ) for @xmath112 ( in which case @xmath113 ) and @xmath114 ( with @xmath115 ) , respectively . in the latter example there are no band contact points , in spite of the overlap between the two bands , and the system would be in an insulating phase if nnn hopping terms were not taken into account . this overlap in energy between the two bands yields a non - zero density of states at any energy , such that the semi - metallic ( or insulating ) phase vanishes and yields , at half - filling , a metallic phase with particle and anti - particle pockets in the first brillouin zone . , @xmath88 , and @xmath116 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the homogeneous case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ( e ) an example of the metallic phase with @xmath112 . ( f ) another example of the metallic phase with @xmath114 . ] , @xmath88 , and @xmath116 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the homogeneous case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ( e ) an example of the metallic phase with @xmath112 . ( f ) another example of the metallic phase with @xmath114 . ] , @xmath88 , and @xmath116 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the homogeneous case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ( e ) an example of the metallic phase with @xmath112 . ( f ) another example of the metallic phase with @xmath114 . ] , @xmath88 , and @xmath116 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the homogeneous case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ( e ) an example of the metallic phase with @xmath112 . ( f ) another example of the metallic phase with @xmath114 . ] , @xmath88 , and @xmath116 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the homogeneous case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ( e ) an example of the metallic phase with @xmath112 . ( f ) another example of the metallic phase with @xmath114 . ] , @xmath88 , and @xmath116 . the labels @xmath89 and @xmath76 represent the @xmath81 and @xmath80 components of the momentum , respectively . the @xmath81 and @xmath80 axes have been chosen such that the nn vectors are given by eq.([eq : nn - vec ] ) . ( a ) the homogeneous case , where @xmath90 . ( b ) the merged dirac points , where @xmath91 . ( c ) the zero dimensional case , where @xmath84 . ( d ) the aligned dirac points , where @xmath92 . ( e ) an example of the metallic phase with @xmath112 . ( f ) another example of the metallic phase with @xmath114 . ] as already alluded to in the previous sections , the shaking of a honeycomb lattice can lead to a sign change of the hopping parameters . quite generally , fig.[fig : phase - diagram - sign ] shows that changing the relative signs of the nn hopping parameters results in a translation of the energy spectrum in momentum space . this effect was also mentioned in ref . @xcite . indeed , the relative signs determine at which of the four time - reversal invariant momenta in the first brillouin zone the merging of dirac points and the semimetal - insulator transition take place when @xmath117 . however , the sign change of the nn hopping parameters can be transformed away by a gauge transformation @xcite . nevertheless , the sign of the nnn hopping parameter is important , since it determines whether the upper or the lower band is flattened . . ( a ) to ( h ) contour plots of the energy bands , where the color scaling is arbitrary and the first brillouin zone is the area with higher contrast . the value of the nn hopping parameters for each contour plot are given by the position of the corresponding letter in the phase diagram and @xmath86 . the dark regions indicate energies close to zero , whereas brighter regions are further away in energy from the fermi level at half filling . ] until now , we have considered single - component fermionic atoms and , due to the pauli principle , the absence of @xmath118-wave interaction naturally results in an ideal fermi lattice gas , albeit with an unusual band structure . by trapping two hyperfine states of the fermionic atoms , hubbard - like interaction terms arise , @xmath119 where the fermionic operators now acquire an additional spin index @xmath120 , which is summed over , and @xmath121 is the interaction energy . naturally , in order to be able to apply the floquet theory in the presence of interactions , the hamiltonian @xmath53 in eq . ( [ eq : condition ] ) must now be replaced by @xmath122 . this is the case in our study because we investigate the weak - coupling limit with @xmath123 . since the interaction term commutes with the shaking , it is not renormalized , similarly to the on - site energy term proportional to @xmath16 in eq.([eq : h0 ] ) . however , it has been shown that complications may arise when a multiple of the energy @xmath121 is in resonance with a harmonic of @xmath124 , @xmath125 , for integer @xmath26 and @xmath35 . whereas the limit @xcite @xmath126 is not considered here because it is in contradiction with the small-@xmath121 large - frequency limit , critical resonances may occur for @xmath127 @xcite . nevertheless , it has been shown in ref . @xcite that these resonances , which occur in higher - order perturbation theory , are strongly suppressed in the large-@xmath26 limit . in the weakly - interacting regime considered here , the ground state is adiabatically connected to that of the non - interacting system , with no broken symmetry . first , we use the fourier transform of the creation and annihilation operators , @xmath128 , to find the hamiltonian in momentum space . within a hartree - fock theory , we introduce a mean - field decoupling of the interaction terms , @xmath129 such that the expectation values of both sides are equal . for the mean value we take @xmath130 where @xmath131 is the number of sites per sublattice , @xmath132 is the density of atoms with momentum @xmath133 and spin index @xmath134 , and @xmath135 is the kronecker delta . we then obtain the mean - field hamiltonian @xmath136 where the total density is defined by @xmath137 . the hamiltonian ( [ eq : ham - mf1 ] ) may be rewritten in a matrix form : @xmath138 where we have introduced the functions @xmath139,\ ] ] and @xmath105 is defined in eq . ( [ fq ] ) the hamiltonian ( @xmath140 ) can then be diagonalized by the unitary operator @xmath141 which yields @xmath142 because the @xmath143 and @xmath144 quasiparticles are free , the partition function corresponding to the hamiltonian ( @xmath145 ) reads @xmath146 \bigg\ } + \log\bigg\ { 1 + \exp\big[-\beta \big(h(\mu , \textbf{q } ) + |f(\textbf{q})| \big ) \big ] \bigg\ } \bigg ) \bigg],\ ] ] where @xmath147 with @xmath148 denoting the boltzmann s constant and @xmath38 the temperature . the total number of particles @xmath149 is given by @xmath150 , and one obtains @xmath151 } + \frac{1}{1 + \exp\big[\beta \big(h(\mu , \textbf{q } ) + |f(\textbf{q})| \big ) \big ] } \bigg ) . \label{eq : nrparticles}\ ] ] since the expression inside the sum does not depend on spin , summing over @xmath134 yields a factor 2 . one recognizes in eq.([eq : nrparticles ] ) the fermi - dirac distribution function @xmath152^{-1}$ ] . the number of particles @xmath149 is related to the density @xmath35 , which is defined here as the number of particles per lattice site , i.e. @xmath153 . converting the sum over @xmath133 into an integral , the following self - consistent equation for the density is derived @xmath154 + n_{fd}\big[\beta \big(h(\mu , \textbf{q } ) + |f(\textbf{q})| \big ) \big ] \bigg\ } , \label{eq : density - interaction - os}\ ] ] where the integral is restricted to the first brillouin zone , the surface of which is @xmath155 . + in fig.[fig : density - mu](a ) , the density @xmath156 is plotted for several values of @xmath157 . for the isotropic case , @xmath158 , the result of zhu _ et al_. is reproduced @xcite . for the 0d limit , the flat line due to the gap in the spectrum is clearly visible at the chosen temperature . fig.[fig : density - mu](b ) confirms that repulsive interactions lead to a lower density than in a system without interactions for the same chemical potential . fig.[fig : density - mu](c ) agrees with the observation that the nnn hopping breaks the particle - hole symmetry . this effect is also visible in fig.[fig : density - mu](d ) , where the dependence of the density on the chemical potential is calculated for a shaking vector where the system is in the zero - gapped semimetallic phase for @xmath86 and in the metallic phase for @xmath116 . as a function of the chemical potential @xmath16 . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] as a function of the chemical potential @xmath16 . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] as a function of the chemical potential @xmath16 . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] as a function of the chemical potential @xmath16 . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] honeycomb optical lattices have recently been realized experimentally , although the existing setups have only been used to investigate bosonic atoms @xcite . shaking of a lattice has been experimentally implemented in a one - dimensional one by a periodic modulation on the position of the reflecting mirrors @xcite . for a honeycomb lattice , the shaking could be realized by means of an acousto - optical device , as proposed for a triangular lattice in ref . @xcite . the magnitude of the nn hopping parameter @xmath71 in a honeycomb optical lattice has been evaluated in ref . @xcite , @xmath162 in terms of the recoil energy @xmath163 and the magnitude of the potential barrier between nearest - neighbor lattice sites @xmath164 . the magnitude of the nnn hopping parameter @xmath107 is not yet known , but could be determined from numerical band structure calculations . in a typical experimental situation , we expect the ratio @xmath165 to be in the @xmath166 range , in agreement with the parameter chosen in the discussion of sec.[sec02c ] . in a typical experiment , the shaking amplitude would be increased from zero to a finite value . fig.[fig : phases - linear ] shows in which order the system goes through the different phases and dimensions upon increasing the shaking amplitude . here , also the values of the shaking amplitude required for the dimensional crossovers are given for the same system as discussed in sec.[sec02c ] and @xmath167 . if the shaking direction is perpendicular to one of the nn vectors , the system will be in the gapped insulating phase beyond a certain value of the shaking amplitude , since the bessel function crosses the value 0.5 only once and never obtains the value -0.5 . if the shaking is parallel to one of the nn vectors , the system will be in the metallic phase beyond a certain value of the shaking amplitude . nevertheless , it is still possible to induce a merging of dirac points and to open up a gap in the spectrum , since the nn hopping parameters are renormalized such that the value of one of them will in general differ from that of the other two . however , whether the system is actually driven into an insulating phase or remains metallic depends on the precise value of the ratio @xmath165 . in nm correspond to the optical lattice discussed in sec.[sec02c ] with @xmath168 , @xmath169khz , @xmath170 nm , and containing @xmath83k atoms . ( a ) shaking perpendicular to one of the nn hopping directions . ( b ) shaking parallel to one of the nn hopping directions . ] in experiments , an overall harmonic trapping potential is imposed to confine the atoms . it is described by @xmath171 where @xmath172 is the trapping frequency , and @xmath9 is the position measured from the center of the trap . by applying the local density approximation ( lda ) , one finds that the chemical potential evolves radially according to @xmath173 . fig.[fig : density - r](a ) shows the density profile for several ratios of @xmath157 , without nnn hopping or interactions . the case with @xmath174 , when the system is in the extreme limit of the band insulating phase , can be well distinguished from the other cases . fig.[fig : density - r](b ) shows that stronger interactions lead to a higher density away from the center of the trap . this effect becomes visible when the density starts to deviate from one particle per lattice site . next - nearest - neighbor hopping leads to a higher density at the edge of the cloud compared to the case without nnn hopping , which can be seen from comparing figs.[fig : density - r](a ) and ( c ) and from fig.[fig : density - r](d ) . the latter shows the effect of nnn hopping on the density profile for the case where the nnn hopping gives rise to the metallic phase for two different shaking vectors . in the first case , @xmath112 , which gives @xmath113 , such that without nnn hopping , the system is in the zero - gapped semi - metallic phase and the dirac points are located very close to the corners of the first brillouin zone . in the second case , @xmath114 , which results in @xmath115 , such that without nnn hopping the system is in the insulating phase and the two energy bands are almost flat . as a function of the distance from the trap s centre @xmath175 , which is expressed in units of the nearest - neighbor distance @xmath144 . the trapping frequency has been chosen such that the trapping potential is given by @xmath176 . the chemical potential @xmath16 for each case has been chosen such that the density at the trap s centre @xmath35 is one particle per site . this corresponds to half - filling , since we consider 2 species of fermions . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] as a function of the distance from the trap s centre @xmath175 , which is expressed in units of the nearest - neighbor distance @xmath144 . the trapping frequency has been chosen such that the trapping potential is given by @xmath176 . the chemical potential @xmath16 for each case has been chosen such that the density at the trap s centre @xmath35 is one particle per site . this corresponds to half - filling , since we consider 2 species of fermions . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] as a function of the distance from the trap s centre @xmath175 , which is expressed in units of the nearest - neighbor distance @xmath144 . the trapping frequency has been chosen such that the trapping potential is given by @xmath176 . the chemical potential @xmath16 for each case has been chosen such that the density at the trap s centre @xmath35 is one particle per site . this corresponds to half - filling , since we consider 2 species of fermions . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] as a function of the distance from the trap s centre @xmath175 , which is expressed in units of the nearest - neighbor distance @xmath144 . the trapping frequency has been chosen such that the trapping potential is given by @xmath176 . the chemical potential @xmath16 for each case has been chosen such that the density at the trap s centre @xmath35 is one particle per site . this corresponds to half - filling , since we consider 2 species of fermions . unless specified otherwise in the figure , the nn hopping parameters @xmath159 , the nnn hopping parameter @xmath86 , the interaction strength @xmath160 , and the inverse temperature @xmath161 . ( a ) effect of the renormalization of the nn hopping parameters . ( b ) effect of the interaction strength @xmath121 for the isotropic case . ( c ) effect of the nnn hopping parameter @xmath116 in the shaken lattice . ( d ) the metallic phase . for the isotropic cases , @xmath112 , whereas for the 0d cases @xmath114 . these systems are in the metallic phase for @xmath116 , whereas for @xmath86 they are in the semi - metallic and the insulating phase , respectively . ] we emphasize that , in the present paper , we only discuss weak correlations that adiabatically affect the density . however , when increasing further the onsite interaction , one may expect correlated phases with inhomogeneous density , even at half - filling . a detailed study of these correlated phases is a vast research issue that is yet ongoing and that is beyond the scope of the present paper . here , we only provide a glimpse on how the density , which we discussed above in the weak - coupling limit , may evolve in view of some phases studied in the literature . mean - field calculations indicate a transition to an antiferromagnetic state above a value of @xmath177 @xcite , whereas more sophisticated quantum monte - carlo calculations indicate an intermediate spin - liquid phase between the semimetal and the anti - ferromagnetic phase @xcite . the spin - liquid phase may be viewed as a mott insulator with a charge localization on the lattice sites , and recent slave - rotor calculations indicate that such spin - liquid phases dominate the phase diagram for @xmath178 @xcite , which is the parameter range where the dirac points would merge in the absence of interactions . the precise transition between the weakly - interacting liquid phases and these strongly - correlated mott insulators could in principle be determined with the help of the above - mentioned density measurements . a more promising technique for detecting dirac - point motion is momentum - resolved raman spectroscopy . this technique has been proposed as an equivalent of angle - resolved photoemission spectroscopy for cold atom systems @xcite . it has not yet been realized experimentally , while momentum - resolved radio - frequency spectroscopy , which is a very similar technique , has already been implemented @xcite . notice further that another very similar technique , momentum - resolved bragg spectroscopy , has been applied to ultracold bosonic atoms in a static optical lattice by ernst _ et al_. @xcite . momentum - resolved spectroscopy can allow us to indirectly visualize the band structure . in momentum - resolved raman spectroscopy , two laser pulses with frequencies @xmath179 and @xmath180 are irradiated upon the system . if the frequency difference is in resonance with a transition @xmath181 between atomic hyperfine states , @xmath182 , some atoms are excited in a second - order process to the higher hyperfine state . then , with state - selective time - of - flight measurements , the dispersion of the atoms in the new state are measured , from which the dispersion of the original atoms can be derived . when the atoms are confined in a trapping potential and the laser pulses are focussed on the center of the trap , the quality of the results obtained by raman spectroscopy is comparable to those of a homogeneous system @xcite . furthermore , raman spectroscopy yields better results for a system with strong interactions compared to standard time - of - flight measurements @xcite . notice that momentum resolved raman spectroscopy was originally proposed to be applied to a gas of ultracold fermionic atoms at equilibrium and not for a shaken lattice . we therefore discuss , in this final paragraph , why we think that this technique may also be applied to the present case . naturally , as long as the frequencies of the additional lasers in the raman - spectroscopy setup are small with respect to the shaking frequency , @xmath183 , even the full system satisfies the condition ( [ eq : condition ] ) for the validity of floquet theory . as in the case of interactions , one needs , however , to avoid resonances between the different laser frequencies that could become critical @xcite . the opposite limit , in which the laser frequencies and that of the hyperfine transition are larger than the shaking frequency , is more delicate . however , even then , the shaken system remains at quasi - equilibrium as long as the intensities of the lasers used in raman spectroscopy are weak , such that they only constitute a small perturbation . the atomic dynamics probed even at high frequencies is therefore still that of the atoms at quasi - equilibrium , with the band strucure obtained from floquet theory . furthermore , in the experimental studies by zenesini _ et al_. @xcite , time - of - flight measurements were used to determine the momentum distribution of bosonic atoms in a shaken lattice . apart from the time - scale considerations , there are also some length scales that need to be taken into account . there are indeed two requirements for the correct size of the focus of the laser beams . on the one hand , it needs to be larger than the lattice spacing , such that sufficiently many atoms can be excited , while on the other hand the focus of the beams should be small enough , in order to have an approximately flat trapping potential inside the focus area . in addition , choosing the length of the pulses could possibly be a problem , since for shorter pulses , the excited atoms will be less affected by the lattice potential , whereas for longer pulses more atoms can be excited , leading to a stronger signal . in conclusion , we have investigated the band engineering of fermionic atoms in an optical honeycomb lattice with the help of a periodic shaking of the lattice . if the shaking frequency @xmath24 is large enough , i.e. if @xmath124 constitutes the largest energy scale in the system , the floquet theory may be applied and the system is at quasi - equilibrium in the sense that the atoms can not follow the rapid motion associated with the shaking . depending on the direction of the shaking , one may render the hopping amplitudes in the quasi - static lattice anisotropic , due to a renormalization of the nn and nnn hopping parameters by bessel functions that go through zero and change sign . as a consequence , dimensional crossovers can be induced in absence of the nnn hopping . for a shaking direction parallel to one of the nn vectors ( such as e.g. @xmath19 ) , one can make one of the nn hopping parameters vanish , @xmath184 . the system then undergoes a transition from 2d to 1d , while the dirac points align simultaneously . shaking in the perpendicular direction ( @xmath185 ) allows one to decrease two nn hopping amplitudes simultaneously while maintaining @xmath70 unrenormalized . in this case , a dimensional crossover from 2d to 0d is induced for @xmath186 , leading to two flat energy bands , beyond the merging of dirac points @xcite , which occurs at @xmath187 . a nonzero value of @xmath107 breaks the particle - hole symmetry and leads to a coupling among the 1d chains and the 0d dimers , for the @xmath95 and @xmath106 cases , respectively , and thus to a weak 2d dispersion . the merging and the alignment of dirac points , however , are not affected . moreover , for a shaking direction parallel to @xmath19 , one pair of nnn hopping amplitudes [ @xmath109 ] remains unrenormalized , and its relative importance is thus enhanced when compared to the decreasing nn hopping amplitudes . in this limit , beyond the semi - metallic and the band - insulating phases , a novel metallic phase can appear that consists of particle and hole pockets with a non - vanishing density of states even at half - filling . furthermore , we have investigated the role of weak repulsive on - site interactions . the resulting ground state is then adiabatically connected to that of the non - interacting system , and we have self - consistently calculated the dependence of the atomic density on the ( local ) chemical potential . the density profiles of the different phases , e.g. the gapless semimetal or the gapped band insulator , and the different dimensionality may be measured experimentally by in - situ density measurements . moreover , momentum - resolved raman spectroscopy may be a promising technique to measure the band structure associated with these different phases . we thank gilles montambaux , guangquan wang , andreas hemmerich , and marco di liberto for fruitful discussions . we also thank christoph lschlger for informing us about an error in the calculations . this work was financially supported by the anr project nanosim graphene under grant no . anr-09-nano-016 and by the netherlands organization for scientific research ( nwo ) . for the studied case , @xmath189 , where @xmath53 was given in eq.([eq : h0 ] ) . since the nn hopping is usually larger than the nnn hopping , @xmath190 , and since we let the chemical potential be in the range @xmath191 , the dominant energy scale in the hamiltonian @xmath53 is @xmath71 . therefore , if @xmath59 , the condition ( [ eq : condition ] ) is satisfied and the floquet theory may be applied . in the general case , the effective hamiltonian is given by @xcite @xmath192_n \bigg \rangle_t , \label{eq : app - h - eff}\ ] ] where for the shaken honeycomb lattice , we choose @xmath193 using the ( nonvanishing ) commutation relations @xmath194 = & a^\dagger_{\textbf{r } ' } \ , b_{\textbf{r}+\textbf{d}_j } \ , \delta_{\textbf{r } ' , \textbf{r } } \ , , \nonumber \\ [ a^\dagger_{\textbf{r}'}\,a_{\textbf{r } ' } , b^\dagger_{\textbf{r}+\textbf{d}_j } \ , a_{\textbf{r } } ] = & - b^\dagger_{\textbf{r}+\textbf{d}_j } \ , a_{\textbf{r } ' } \ , \delta_{\textbf{r } ' , \textbf{r } } , \nonumber \\ [ a^\dagger_{\textbf{r}'}\,a_{\textbf{r } ' } , a^\dagger_{\textbf{r } } \ , a_{\textbf{r}+\textbf{d}_i -\textbf{d}_j } ] = & a^\dagger_{\textbf{r } ' } \ , a_{\textbf{r}+\textbf{d}_i -\textbf{d}_j } \ , \delta_{\textbf{r}',\textbf{r } } \nonumber \\ & - a^\dagger_\textbf{r } \ , a_{\textbf{r } ' } \ , \delta_{\textbf{r}',\textbf{r}+\textbf{d}_i -\textbf{d}_j } , \label{eq : commutators}\end{aligned}\ ] ] which are valid for both fermionic and bosonic creation and annihilation operators , and equivalent ones for the creation and annihilation operators on the @xmath11 sublattice , the multiple commutator in eq.([eq : app - h - eff ] ) becomes @xmath195_n & = \left[\frac{m \omega^2}{\hbar \omega } \sin(\omega t ) \right]^n \nonumber \\ \bigg\ { & - \gamma \sum_{j=1}^3 \sum_{\textbf{r } \in a } ( \textbf{d}_j \cdot \boldsymbol{\rho})^n \left [ ( -1)^n a^\dagger_\textbf{r } b_{\textbf{r } + \textbf{d}_j } + b^\dagger_{\textbf{r } + \textbf{d}_j } a_\textbf{r } \right ] \nonumber \\ & - \gamma ' \sum^3_{i=1 } \sum^3_{j=1,j\neq i } \bigg ( \sum_{\textbf{r } \in a } [ ( \textbf{d}_j - \textbf{d}_i ) \cdot \boldsymbol{\rho}]^n \ , a^\dagger_\textbf{r } \ , a_{\textbf{r } + \textbf{d}_i - \textbf{d}_j } \nonumber \\ & + \sum_{\textbf{r } \in b } [ ( \textbf{d}_j - \textbf{d}_i ) \cdot \boldsymbol{\rho}]^n \ , b^\dagger_\textbf{r } b_{\textbf{r } + \textbf{d}_i - \textbf{d}_j } \bigg ) \bigg\ } \nonumber \\ & - \mu \bigg ( \sum_{\textbf{r } \in a } \ , a^\dagger_\textbf{r } a_\textbf{r } + \sum_{\textbf{r } \in b } \ , b^\dagger_\textbf{r } b_\textbf{r } \bigg ) .\end{aligned}\ ] ] finally , after performing the time average and evaluating the sum over @xmath35 , the effective hamiltonian ( [ hamiltonian - eff ] ) is obtained . c. becker , p. soltan - panahi , j. kronjger , s. drscher , and k. bongs , and k. sengstock , new j. phys . * 12 * , 065025 ( 2010 ) ; a. klinger , s. degenkolb , n. gemelke , k .- a . brickman soderberg , c. chin , rev . instrum . * 81 * , 013109 ( 2010 ) ; j. struck , c. lschlger , r. le targat , p. soltan - panahi , a. eckardt , m. lewenstein , p. windpassinger , and k. sengstock , science * 333 * , 996 ( 2011 ) ; p. soltan - panahi , d .- s . lhmann , j. struck , p. windpassinger , and k. sengstock , nature physics * 8 * , 71 ( 2012 ) . white and a.e . feiguin , phys . lett . * 93 * , 076401 ( 2004 ) ; a.j . daley , c. kollath , u. schollwck , and g. vidal , j. stat . : theor . exp . * p04005 * ( 2004 ) ; d. gobert , c. kollath , u. schollwck , and g. schutz , phys . e * 71 * , 036102 ( 2005 ) ; u. schollwck , rev . phys . * 77 * , 259 ( 2005 ) ; d. gobert , c. kollath , u. schollw ck , and g. schtz , phys . e * 71 * , 036102 ( 2005 ) ; c. kollath , a. m. luchli , and e. altman , phys . . lett . * 98 * , 180601 ( 2007 ) .

we present a _ chandra _ and _ xmm - newton _ study of the supernova remnant ( snr ) kes 73 hosting the anomalous x - ray pulsar 1e 1841 . the chandra image reveals clumpy structures across the remnant with enhanced emission along the western rim . the x - ray emission fills the radio shell and spatially correlates with the infrared image . the global x - ray spectrum is described by a two - component thermal model with a column density = 2.6 and a total luminosity of = 3.3 ergs s ( 0.510 kev , at an assumed distance of 8.5 kpc ) . the soft component is characterized by a temperature = 0.5 kev , a high ionization timescale , and enhanced si and s abundances suggesting emission that is dominated by shocked ejecta . the hard component has a temperature = 1.6 kev , a relatively low ionization timescale , and mostly solar abundances suggesting emission that is dominated by interstellar / circumstellar shocked material . a spatially resolved spectroscopy study reveals no significant variations in the spectral properties . we infer an snr age ranging between 750 yr and 2100 yr , an explosion energy of 3.0 ergs and a shock velocity of ( 1.2 0.3) km s ( under the sedov phase assumption ) . we also discuss the possible scenario for kes 73 expanding into the late red supergiant wind phase of its massive progenitor . comparing the inferred metal abundances to core - collapse nucleosynthesis model yields , we estimate a progenitor mass , adding a candidate to the growing list of highly magnetized neutron stars proposed to be associated with very massive progenitors .

1311.6515

the galactic supernova remnant ( snr ) kes 73 ( g27.4 + 0.0 ) has been classified as a shell - type snr and hosts the anomalous x - ray pulsar ( axp ) 1e 1841@xmath0045 ( helfand et al . 1994 ; vasisht & gotthelf 1997 ) . radio studies of the remnant show an incomplete , small diameter ( @xmath205@xmath21 ) , shell characterized by a steep spectral index ( @xmath22 @xmath20 0.68 ) , a flux density of 6 jy at 1 ghz , and with no detection of the central compact source with a flux limit @xmath23 @xmath24 0.45 mjy and @xmath25 @xmath24 0.60 mjy ( green 2009 ; kriss et al . 1985 ; helfand et al . 1994 ) . the infrared studies , carried out as part of the galactic snr surveys , clearly detected the remnant in the 24 @xmath26 m band ( wachter et al . 2007 ; carey et al . 2009 ; pinheiro goncalves et al . the infrared emission originating from the snr with an estimated dust mass of 0.11 @xmath19 has been explained by pinheiro goncalves et al . ( 2011 ) as likely due to grains heated by collisions in the hot plasma , while carey et al . ( 2009 ) suggested that the 24 @xmath26 m emission may be mostly due to nebular emission lines such as [ o iv ] ( 25.89 @xmath26 m ) and [ fe ii ] ( 25.99 @xmath26 m ) with a small contribution to the continuum emission from dust produced in the remnant . the snr kes 73/axp system was previously observed and studied using many x - ray observatories including _ einstein _ , _ rosat _ , _ asca _ , _ chandra _ , and _ xmm - newton_. the _ asca _ spectrum was described by a thermal bremsstrahlung model with @xmath27 @xmath20 0.6 kev plus gaussian emission lines with evidence for enhanced mg , and possibly o and ne , abundances , with the snr age estimated to be @xmath282000 years ( gotthelf & vasisht 1997 ) . observation of the snr was briefly studied by morii et al . ( 2010 ) in connection with the _ suzaku _ studies of axp 1e 1841@xmath0045 , where the entire snr spectrum was modeled by a vsedov model . the _ xmm - newton _ study of the remnant was performed by vink & kuiper ( 2006 ) using the mos data . they fitted the spectrum using either a two - component non - equilibrium ionization ( nei ) model under the spex fitting package ( kaastra & mewe 2000 ) or a one - component vsedov model , both of which yielded solar metal abundances . their study , primarily targeted to determine the supernova explosion energy , argued against the millisecond proto - neutron star model for magnetar formation . recently , lopez et al . ( 2011 ) also reported x - ray results on kes 73 as part of a survey study aimed at typing snrs using their x - ray morphology as well as to set observational constraints on the hydrodynamical models . their study , which made use of one of the available _ chandra _ observations of kes 73 , yielded a mean temperature of @xmath27 = 0.84 @xmath16 0.49 kev using a one - component vpshock model and detected enhanced abundances from mg , si , and s for some of the small scale regions extracted from within the remnant . the above mentioned x - ray studies lack a detailed imaging and spectroscopic analysis of the remnant using all available data . hence , in the following work , we extend the earlier studies to include multi - wavelength data and in particular , the previously unpublished _ chandra _ and _ xmm - newton _ observations . we provide a detailed x - ray imaging - spectroscopic analysis to investigate the snr s multi - wavelength morphology ( radio , infrared , and x - rays ) , map the spectral parameters across the remnant , revisit the supernova explosion properties , and address the mass of its progenitor star . the latter goal is motivated by addressing the nature of the progenitors and the environment of highly magnetized neutron stars ( with a surface dipole magnetic field @xmath29 @xmath18 4@xmath310@xmath30 g ) through studying their hosting snrs ( see safi - harb & kumar 2012 ) . kes 73 is one of only a few snrs associated with high-@xmath29 pulsars . the x - ray emission from its associated axp , the slowest ( @xmath31 = 11.8 s ) known magnetar with a characteristic age of @xmath204.7 kyr , was first decoupled from the diffuse emission from the snr using _ asca _ ( helfand et al . 1994 ; gotthelf & vasisht 1997 ) . this source has been manifesting itself as a quiescent magnetar since the time of its discovery until recently , when it exhibited the first magnetar - like burst caught by the _ swift _ x - ray observatory ( beardmore et al . 2010 ; kumar & safi - harb 2010 ) followed by a few more bursts ( lin et al . 2011 ) , further establishing its magnetar nature with @xmath29 @xmath20 7@xmath310@xmath32 g. although the axp s emission will not be further discussed in this paper , the study of snr kes 73 , because of its relative youth and brightness , provides a unique opportunity to shed light on the progenitor of highly magnetized neutron stars . the paper is organized as follows : section 2 describes the _ chandra _ and _ xmm - newton _ observations and data reduction . in sections 3 and 4 , we present the details of the x - ray imaging analysis and spectral fitting of the remnant , respectively . section 5 discusses the results and finally , we summarize our study in section 6 . the snr kes 73 was observed with the advanced ccd imaging spectrometer ( acis ) onboard the _ chandra x - ray observatory _ on 2000 july 23 ( obsid : 729 ) and 2006 july 30 ( obsid : 6732 ) . the snr coordinates were positioned at the aimpoint of the back - illuminated acis - s3 chip . data were taken in full - frame timed - exposure mode and the ccd temperature was @xmath0120@xmath33 c with a ccd frame readout time of 3.2 s. the data were reduced using the standard _ chandra _ interactive analysis of observations ( ciao ) version 4.3 routines . we reprocessed the event files ( from level 1 to level 2 ) to remove pixel randomization and to correct for ccd charge transfer efficiencies . the bad grades were filtered out and good time intervals were reserved . the resulting total effective exposure for the two observations after data processing is 54.1 ks ( see table 1 ) . _ xmm - newton _ observed kes 73 on 2002 october 5 and 7 ( obsids : 0013340101 and 0013340201 ) with the european photon imaging camera ( epic ) , which has one pn ( strder et al . 2001 ) and two mos cameras ( turner et al . 2001 ) covering the energy range between 0.2 and 12 kev with a energy resolution of 0.15 kev at 1 kev . their on - axis angular resolution is around 6@xmath34 fwhm and 15@xmath34 half - power diameter with a field of view of 30@xmath35 . the epic - pn and mos cameras were operated in full frame mode with a time resolution of 73 ms and 2.6 s , respectively . we analyzed the data from both the mos and pn instruments using the @xmath36 science analysis system ( sas ) version 10.0.0 and the most recent calibration files . the event files were created from observational data files ( odfs ) using the sas tasks _ epchain _ and _ emchain_. the events were then filtered to retain only patterns 0 to 4 for the pn data ( 0.215 kev energy band ) and patterns 0 to 12 for the mos data ( 0.212 kev energy range ) . the data were screened to remove spurious events and time intervals with heavy proton flaring by inspecting the light curves for each instrument separately at energies above 10 kev . the light curves were created with 100 s bins and the bins with count - rates greater than 0.35 and 0.4 counts s@xmath9 were rejected for mos1/2 and pn , respectively . the resulting total effective exposure for the mos1 + 2 and pn cameras for the two observations is @xmath2020.3 ks and @xmath206.7 ks , respectively ( see table 1 ) . [ cols="<,<,<,<,<,^ " , ] from both the _ chandra _ and _ xmm - newton _ observations , we defined 9 regions to characterize the diffuse emission from within the remnant . the coordinates and the sizes of the extracted regions are summarized in table 2 and shown in figure 2 . the source regions , marked by the green ellipses , were selected in such a way that the entire diffuse emission from within the remnant could be explored thoroughly . the western side of the snr is covered by regions 1 and 2 , southern side by regions 3 and 4 , eastern side by region 5 , and the northern part by regions 8 and 9 . regions 6 and 7 lie more toward the interior of the remnant . for the elliptical regions 19 , the background was selected from a nearby source - free region to the northeast of the snr on the _ chandra _ and _ xmm - newton _ chips and is shown by the cyan - colored dashed ellipse in figure 2 . we further note here that the spectral fits for all small - scale regions were explored with different backgrounds ( northwest and southwest regions near the snr , and the snr annular ring ) and binning , and the spectral parameters obtained were consistent with those shown in table 3 . in order to study the global snr properties , we extracted the whole snr region of radius 150@xmath34 with the emission from the pulsar and other point sources removed . this is shown in figure 2 , where the global snr encloses the area inside the dashed cyan - colored annulus and the background extends from 150@xmath34180@xmath34 . for the _ xmm - newton _ data , to maximize the snr signal while minimizing any contamination by the axp s spectrum , we excluded a circle centered at the axp with a radius of 25@xmath34 . we have investigated any possible contamination of the snr s spectrum by the axp s spectrum ( especially in the hard x - ray band where its emission dominates ) and found that it is negligible . in particular , we checked the pulsar s spectrum out to a larger radius of @xmath2035@xmath34 ( which encompasses more than 90% of the encircled energy for a point source ) and found that the additional axp flux ( within 25@xmath3435@xmath34 ) is negligible in comparison to the snr s flux , with a relative axp to snr hard ( 410 kev ) x - ray flux of only @xmath202% , which is within the error on the estimated snr flux . furthermore , the small - scale inner snr regions extracted closest to the pulsar are at a radius @xmath3735 " , implying no contamination of these regions spectra by the axp . all the extracted spectra were analyzed using the x - ray spectral fitting package , xspec version 12.6.0 , and were grouped by a minimum of 25 and 50 counts per bin for regions 19 and the global snr , respectively . the errors quoted are at the 90@xmath38 confidence level . the spectra extracted from different regions within the snr clearly showed the presence of line features . hence , we started analyzing the emission from the remnant using thermal models of varying abundances . these include collisional ionization equilibrium ( cie ) models such as vmekal ( mewe et al . 1985 ; liedahl et al . 1995 ) and vraymond ( raymond & smith 1977 ) used for modeling snrs with optically thin thermal plasma that has reached collisional ionization equilibrium , and nei models such as vpshock , vnei , and vsedov ( borkowski et al . 2001 ) appropriate for modeling young snrs in which the plasma is still being ionized . we find that the spectral analysis using cie models did not yield acceptable fits ( reduced chi - squared @xmath39 @xmath40 2 where @xmath41 is the number of degrees of freedom ) while the nei models with varying metal abundances resulted in much better fits . vpshock is a plane - parallel non - equilibrium ionization plasma model with variable abundances , characterized by a constant electron temperature and a range of ionization timescales from 0 to @xmath42 = @xmath43 , where @xmath44 is the post - shock electron density and @xmath45 is the time since the passage of the shock . the vnei model also assumes variable abundances , but is characterized by a constant temperature and a single ionization parameter . vsedov is another nei model with variable abundances based on the sedov self - similar dynamics ( sedov 1959 ) . this model is characterized by mean and electron temperatures immediately behind the shock and by the ionization timescale . l l l l l l l l l l l l parameter & reg 1 & reg 2 & reg 3 & reg 4 & reg 5 & reg 6 & reg 7 & reg 8 & reg 9 & snr + @xmath46 ( @xmath47 @xmath5 ) & 3.3@xmath48 & 3.2@xmath49 & 2.8@xmath7 & 2.7@xmath50 & 2.9@xmath51 & 2.6@xmath52 & 2.8@xmath53 & 2.9@xmath54 & 2.7@xmath55 & 2.6@xmath2 + hard component & + @xmath56 ( kev ) & 1.1@xmath57 & 1.4@xmath58 & 1.7@xmath59 & 1.7@xmath59 & 1.1@xmath60 & 1.3@xmath48 & 1.6@xmath61 & 1.7@xmath2 & 1.3@xmath62 & 1.6@xmath13 + fe@xmath63 & 1.5@xmath64 & 2.3@xmath65 & 1.6@xmath66 & 2.4@xmath67 & 0.9@xmath68 & 2.0@xmath69 & 1.2@xmath70 & 1.2@xmath71 & 1.1@xmath72 & 1.4@xmath73 + @xmath74 ( 10@xmath75 @xmath76s ) & 2.8@xmath77 & 1.4@xmath7 & 0.7@xmath73 & 0.5@xmath48 & 2.7@xmath78 & 3.2@xmath79 & 1.0@xmath80 & 0.8@xmath81 & 1.5@xmath82 & 0.8@xmath57 + @xmath83 ( 10@xmath84 ) & 9.8@xmath85 & 6.6@xmath86 & 4.3@xmath87 & 3.2@xmath88 & 3.7@xmath89 & 1.7@xmath90 & 2.4@xmath91 & 3.6@xmath92 & 2.8@xmath93 & 181@xmath94 + soft component & + @xmath10 ( kev ) & 0.3@xmath95 & 0.4@xmath57 & 0.5@xmath48 & 0.5@xmath57 & 0.3@xmath57 & 0.5@xmath96 & 0.4@xmath95 & 0.4@xmath95 & 0.5@xmath11 & 0.5@xmath11 + o & 1.6@xmath97 & 1.9@xmath98 & 1.1@xmath99 & 1.2@xmath100 & 1.4@xmath101 & 1.6@xmath102 & 1.0@xmath103 & 1.0@xmath104 & 1.0@xmath105 & 1.8@xmath106 + mg & 1.7@xmath107 & 0.9@xmath108 & 0.9@xmath109 & 0.8@xmath110 & 1.2@xmath111 & 1.2@xmath112 & 0.5@xmath57 & 0.6@xmath113 & 1.0@xmath114 & 1.0@xmath115 + si & 5.9@xmath107 & 2.3@xmath116 & 1.4@xmath117 & 1.6@xmath118 & 5.3@xmath119 & 2.8@xmath120 & 1.1@xmath121 & 1.4@xmath122 & 1.9@xmath123 & 1.7@xmath124 + s & 13.2@xmath125 & 4.3@xmath126 & 2.5@xmath127 & 2.8@xmath128 & 10.2@xmath129 & 4.1@xmath130 & 2.6@xmath131 & 3.5@xmath132 & 3.0@xmath133 & 2.9@xmath134 + fe@xmath135 & 1.5@xmath64 & 0.6@xmath136 & 0.4@xmath137 & 0.3@xmath2 & 1.0@xmath138 & 0.6@xmath139 & 0.2@xmath140 & 0.7@xmath91 & 0.6@xmath59 & 0.3@xmath96 + @xmath141 ( 10@xmath142 @xmath76s ) & @xmath371.2 & 0.5@xmath143 & @xmath37 4.4 & @xmath371.8 & @xmath37 0.6 & @xmath371.0 & 0.4@xmath60 & @xmath370.8 & @xmath370.7 & @xmath376.1 + @xmath83 ( 10@xmath144 ) & 2.3@xmath115 & 3.4@xmath64 & 0.9@xmath114 & 4.4@xmath145 & 1.1@xmath2 & 3.7@xmath146 & 2.2@xmath147 & 1.0@xmath148 & 0.5@xmath59 & 20.6@xmath149 + @xmath150 ( 10@xmath84 ) & 0.4@xmath95 & 0.3@xmath95 & 0.4@xmath57 & 0.2@xmath151 & 0.2@xmath96 & 0.2@xmath57 & 0.2@xmath151 & 0.2@xmath11 & 0.2@xmath48 & 12.8@xmath152 + @xmath153 & 1.29/708 & 1.11/652 & 1.29/753 & 1.14/605 & 1.15/516 & 1.26/372 & 1.29/375 & 1.32/410 & 1.06/504 & 1.51/1743 + the _ chandra _ and _ xmm - newton _ spectra of the snr regions were first investigated with single component vpshock models , but none of them gave a satisfactory fit ( @xmath39 @xmath40 2 ) and did not account for residuals above @xmath204 kev unlike previous studies with poorer statistics ( gotthelf & vasisht 1997 ; vink & kuiper 2006 ) . since the x - ray spectrum of a young snr is expected to show distinct components associated with the swept - up ism and ejecta , the spectra were subsequently examined using two - component thermal models . the addition of a second thermal component ( e.g. , a vpshock+vpshock model ) significantly improved the fits for all the regions . throughout the paper , we use the subscripts ` @xmath154 ' and ` @xmath155 ' to denote the soft and hard components , respectively . the abundances of all the elements were initially fixed at their solar values given by anders & grevesse ( 1989 ) . all the models included the morrison & mccammon ( 1983 ) interstellar absorption in @xmath156 ) and the spectral parameters were consistent with those given in table 3 . ] and ni was tied to fe throughout . a gaussian line near 3.1 kev was added to the two - component models since the vpshock model used in xspec does not account for the ar emission line visible in the spectra . the spectra were first fit by treating the column density , temperature , and normalization as free parameters and then letting the abundances vary freely as needed . we obtained nearly solar values for all the hard component elements except for a marginally enhanced fe , while the soft component showed enhanced abundances for mg , si , and s , suggesting that the soft component is associated with shock - heated ejecta . allowing fe@xmath63 to vary improved the @xmath39 value with an @xmath157-test probability @xmath15810@xmath159 . the emission lines from o lie in the soft x - ray band ( @xmath240.6 kev ) and are hard to detect due to the large column density of the snr ( @xmath160 @xmath20 2.6@xmath310@xmath4 @xmath5 ) . however , when allowed to vary , the o abundance improved the spectral fits with an @xmath157-test probability of @xmath2010@xmath161 for some regions and for the global snr s spectrum . we also note a slightly enhanced ca abundances for some regions , but allowing ca to vary did not improve the fits significantly and posed some difficulty in constraining the errors . therefore , the ca abundance was frozen to its solar value for the small scale regions during spectral fitting . in summary , the spectral fitting for all regions was performed by leaving the hard component fe abundance ( fe@xmath63 ) and the soft component o , mg , si , s , and fe@xmath135 abundances as free parameters.table 3 summarizes the best fit spectral parameters obtained with a vpshock+vpshock model for all regions , and figure 3 shows the spectra of some of the snr regions of interest : the bright western limb ( region 1 ) , bright interior regions 3 , 6 , and 7 , the shell - like feature on the southern side ( region 4 ) , and the relatively faint northern edge ( region 9 ) . for the global snr spectrum , we performed spectral fitting using the vsedov+vpshock model plus a gaussian line near 3.1 kev ( table 3 ) . here , the vsedov model is associated with the hard component and yielded the same electron and ion temperatures , while the vpshock model is associated with the soft component . the snr spectrum was further explored using a vpshock+vpshock model and the spectral parameters were found to be in good agreement with those obtained using the vsedov+vpshock model ( table 3 ) , with a slightly different abundance values for si = 2.3@xmath162 , s = 4.2@xmath163 , and fe@xmath135 = 0.6@xmath59 . in figure 4 , we show the _ xmm - newton _ pn spectrum extracted from the whole snr fitted with a vsedov+vpshock model , displaying the different model contributions as dotted lines . in order to check for any underlying synchrotron component originating from the acceleration of particles at the supernova shock , we explored the spectra of all extracted regions by adding a non - thermal model ( e.g. , power - law ) to the one- and two - components thermal models . the addition of a power - law component to the one - component thermal models ( vpshock or vnei ) resulted in a large @xmath39 @xmath40 1.8 value for all regions , confirming that the second component is better described by a thermal model . when added to the best fit two - component thermal models ( vpshock+vpshock or vnei+vnei ) , the additional power - law component slightly improved the fits for some of the regions . for the global snr spectrum , the additional power - law component added to the best - fit vsedov+vpshock model yielded an f - test probability of @xmath16410@xmath159 with a photon index @xmath165 = 3.4@xmath166 and an upper limit on the non - thermal luminosity of @xmath205@xmath310@xmath167 erg s@xmath9 ( at an assumed distance of 8.5 kpc ) . adding a power - law component to the best fit vpshock+vpshock model for the global snr spectrum also resulted in similar spectral parameters . we note a caveat for fitting a synchrotron component with a simple power - law model . more appropriate models include the @xmath168 model in xspec ; however due to uncertainties on the radio properties of our selected regions and the marginal need for an additional non - thermal component , we did not explore this further . a more detailed radio and x - ray spatially resolved study would require deeper observations . as seen from table 3 , there is no significant variation in the spectral parameters across the remnant , except for the column density ( @xmath160 ) which appears slightly elevated on the western side for region 1 . the estimated difference in @xmath160 between the average snr and the western side value , although model dependent , was not affected by varying the o abundance , as discussed above . ] is @xmath207@xmath310@xmath169 @xmath5 . to the best of our knowledge , there is no evidence of any observed molecular cloud towards the western side of the remnant that might account for the higher absorption . tian & leahy ( 2008 ) suggest the existence of a co cloud ( at 89 @xmath16 2 km s@xmath9 ) in the direction of kes 73 , but located behind the remnant and therefore , the cloud is not absorbing the snr . moreover , we explored the environment of kes 73 in the radio and infrared wavelengths ( discussed in section 5.1 ) and do not find any evidence of a cloud or nearby sources that appear to interact with the snr . this is consistent with the overall circular morphology of the snr . we note that lopez et al . ( 2011 ) analyzed spectra from a large number of small scale regions ( @xmath2015@xmath170 radius ) , and reported good fits for one - component models using data from only the earliest _ chandra _ observation of kes 73 ( obsid 729 ) , providing evidence for enhanced elemental abundances across the snr and an elevated value of @xmath160 for the western side . we obtain similar results using all of the available _ chandra _ observations of kes 73 , and using somewhat larger spectral regions , but find that two - component thermal models are required for all of our extracted regions . to check this apparent discrepancy , we extracted spectra from a representative @xmath2015@xmath170 radius region using only data from obsid 729 . we found that a single - component vpshock adequately describes the spectrum , and that the improvement in the fit from addition of another vpshock model is not statistically significant ( f - test probability of @xmath200.05 ) . on the other hand , for slightly larger region sizes ( e.g. , region 7 in figure 2 ) using the same single data set , the addition of a second vpshock model is required to adequately fit the spectrum , with an f - test chance probability of @xmath206@xmath310@xmath171 . in summary , the differences in spectral results between our work and that of lopez et al . ( 2011 ) are primarily due to the larger region sizes and the inclusion of all available data for kes 73 . we investigate the multi - wavelength morphology of kes 73 using the radio , infrared , and x - ray data . the radio image was extracted from the multi - array galactic plane imaging survey ( magpis ) at a wavelength of 20 cm for the region 5@xmath33 @xmath24 @xmath172 @xmath24 [email protected] , @xmath173 @xmath24 [email protected] ( helfand et al . 2006 ) and the infrared image from the survey of the inner galactic plane using _ spitzer _ s multiband imaging photometer ( mips ) at a wavelength of 24 @xmath26 m ( mipsgal ) for the region 4@xmath33 @xmath24 @xmath172 @xmath24 50@xmath33 , @xmath173 @xmath24 [email protected] ( carey et al . 2009 ) . in figure 5 ( top ) , we show the radio ( left ) and infrared ( right ) images of the snr , with a field of view of @[email protected]@[email protected] centered around the pulsar . we estimate an apparent diameter of @xmath205@xmath35 for the remnant in radio and infrared wavelengths in comparison with the slightly smaller diameter of @[email protected] in x - rays . we have also constructed a composite color image of the radio ( red ) , infrared ( green ) , and x - ray ( blue ) emission of snr kes 73 as shown in figure 5 ( bottom ) . the x - ray image shown here corresponds to the broadband ( 0.57.0 kev ) _ chandra _ image , adaptively smoothed ( using the _ csmooth _ command in ciao ) using a gaussian kernel of @xmath174 = 1@xmath342@xmath34 and for a significance of detection 2 to 5 . as mentioned earlier , kes 73 has been classified as a shell - type remnant in the radio ( kriss et al . now , we compare the radio and x - ray morphologies of the remnant and observe a few similarities . the remnant exhibits a spherical morphology with the western limb ( which has the highest @xmath160 and lowest @xmath12 ) appearing the brightest in both wavelengths , possibly due to some local density enhancements in that region . the radio shell extending from the eastern limb to the southern edge of the snr matches with the corresponding quarter shell - like structure ( i.e. , regions 4 and 5 in figure 2 ) observed in the x - ray image , thus appearing purple in figure 5 ( bottom ) . the enhanced si and s abundances obtained from the spectral fits ( table 3 ) also suggest the presence of reverse shocked ejecta in these regions . we also notice some faint radio clumps on the northern side of the snr spatially correlating with the x - ray image . apart from these few similarities , the remnant s morphology also differs in these wavelengths with the apparent lack of bright radio emission from the snr interior and the pulsar ( as expected due to the nature of their emission ) . the radio shell appears more extended on the northeastern side of the snr and displays a thin filamentary structure along the edges in pure red ( figure 5 ) . this feature is probably the expanding outer blast wave , which is not detected in x - rays . in summary , we see more diffuse x - ray emission originating from the center of the snr due to the reverse shocked ejecta and filling the radio shell , as seen from the blue color at the snr interior in figure 5 ( bottom ) . the _ spitzer _ mips image ( figure 5 , top right ) reveals a morphology more comparable to that of the x - ray image with arc - like features and bright infrared emission filling the snr . the western limb of the snr is very bright in the infrared as well , precisely outlining the radio and x - ray emission , and so appears white in the multi - wavelength image of kes 73 . the interior structures partially coincide with the x - ray image , but no significant infrared emission is seen from the northeastern boundary of the snr . the infrared image shows that the remnant lies in a complex region of the galactic plane , as previously noted by other authors ( e.g. , wachter et al . 2007 ) , with several saturated foreground or background sources , appearing in green . pinheiro goncalves et al . ( 2011 ) also suggested a strong morphological association of the 24 @xmath26 m infrared emission and x - ray emission in snr kes 73 , although it is not clear whether the strong 24 @xmath26 m emission is the result of line emission or hot dust . we discuss here the origin of the soft and hard components derived from our spectral fits . our spatially resolved spectroscopic study confirms the need for a two - component thermal model to describe the x - ray emission originating from the small scale diffuse emission regions . as discussed in section 4.2 , the soft component , with plasma temperatures @xmath200.30.5 kev , shows enhanced metal abundances while the hard component , with plasma temperatures @xmath201.11.7 kev , is characterized by solar abundances ( except for a slightly higher fe@xmath63 for some regions ) . these results indicate that the soft component arises primarily from the reverse shocked ejecta and the hard component is dominated by emission from the forward shock . furthermore , the soft component has reached ionization equilibrium ( @xmath141 @xmath18 10@xmath142 @xmath76 s ) for most of the regions as opposed to the under - ionized plasma ( @xmath74 @xmath20 ( 0.53.2)@xmath310@xmath75 @xmath76 s ) characterizing the hard component . interestingly , the results we obtained for kes 73 are also very similar to those obtained for the mixed - morphology remnant 3c 397 ( safi - harb et al . in particular , 3c 397 also needed a two - component model to fit the x - ray emission from all small scale regions , with the hard component showing a lower ionization timescale while the soft component was closer to ionization equilibrium . as well , the soft component was dominated by ejecta requiring enhanced abundances while the hard component showed solar abundances except for fe@xmath63 and was interpreted as the presence of reverse shocked fe bubbles ( safi - harb et al . 2005 ) . however , the abundance of fe@xmath63 in kes 73 is only marginally enhanced unlike in 3c 397 whose spectrum is characterized by a very strong fe - k line not detected in kes 73 . as mentioned above , we detect enhanced metal abundances in the soft spectral component ( see table 3 ) . s appears to be highly enhanced across the remnant followed by si ( for regions 1 , 2 , 5 , 6 , and 9 ) . mg and o seem slightly enhanced for regions 1 and 2 , respectively . fe@xmath135 is seen enhanced only for region 1 while fe@xmath63 is slightly enhanced for regions 1 , 2 , 3 , 4 , and 6 . most notably , the western and eastern x - ray limbs of the remnant ( regions 1 and 5 ) exhibit very similar spectral properties ( although @xmath160 appears slightly higher for the brighter region 1 ) with elevated and comparable si and s abundances . the emission from these regions correlates well with their radio counterpart and appears to fall within the outermost radio shell , the red emission seen along the outer edges in figure 6 ( bottom ) . these regions likely represent the reverse shocked ejecta seen in x - rays . the outermost fainter region seen in the radio image , which most likely corresponds to the forward shock , is not clearly detected in x - rays . in summary , we observe enhanced si and s abundances for all regions across the remnant suggesting the presence of ejecta . in the following , we estimate the physical properties of the diffuse emission regions across the remnant and the supernova explosion parameters using the spectral fit parameters obtained with the vpshock / vsedov models summarized in table 3 . the derived quantities are shown in table 4 . here , we take the distance to the snr as 8.5 kpc ( tian & leahy 2008 ) , noting that an earlier , slightly smaller , distance estimate of 67.5 kpc was obtained by sanbonmatsu & helfand ( 1992 ) . we introduce a scaling factor @xmath175 = @xmath176/8.5 kpc to account for the distance uncertainty and take a radius of [email protected] for the snr ( as estimated from the extent of the radio emission , see figure 5 ) , which corresponds to a physical size @xmath177 = 6.2 @xmath175 pc = 1.9@xmath310@xmath178 @xmath175 cm . the volume of the x - ray emitting regions ( @xmath179 ) is estimated by assuming that the plasma fills an ellipsoid with the semi - major and semi - minor axes equivalent to those of the extracted snr regions ( table 2 ) and the depth along the line of sight equal to the long axes of the extracted regions . we then introduce the volume filling factors for the cold ( @xmath180 ) and hot ( @xmath181 ) plasma and assume that @xmath180 + @xmath181 = 1 . the emission measure for each component is related to the amount of plasma available to produce the observed flux and is given by @xmath182 = @xmath183@xmath184 @xmath20 @xmath185 where @xmath44 is the electron density of the shocked gas and @xmath186 is the proton density ( @xmath44 = 1.2@xmath186 assuming cosmic abundances ) . by assuming pressure equilibrium between the soft and hard thermal components ( @xmath187 @xmath188 @xmath189 ) , the relative filling factors can then be derived as @xmath180 = [ 1 + @xmath190@xmath9 ; @xmath181 = 1 @xmath0 @xmath180 ( see table 4 ) . for cosmic abundances and strong shock rankine - hugoniot jump conditions , the ambient density @xmath191 can be estimated from the electron density @xmath44 as @xmath44 = 4.8@xmath191 ; here @xmath191 includes only hydrogen ( borkowski et al . 2001 ) . using the normalization value @xmath192 = ( 10@xmath193/4@xmath194d@xmath195@xmath184 determined from the x - ray spectral fits in _ xspec _ , we estimate the ems of the two thermal components as well as the corresponding densities @xmath44 and @xmath191 ( derived from @xmath196 of the hard component ) of the x - ray emitting gas , as shown in table 4 . the electron densities inferred for the soft component are higher than those associated with the hard component . the ambient density , obtained from the global fit to the snr using the vsedov component @xmath20 0.6 @xmath76 using the vpshock+vpshock model , consistent with the value obtained using the vsedov+vpshock model . ] , is @xmath191= 0.5@xmath54 @xmath197@xmath198 @xmath76 . the total unabsorbed flux of the snr in the 0.510.0 kev energy range is @xmath199 = 3.9@xmath99@xmath310@xmath200 ergs @xmath5 s@xmath9 , which corresponds to an x - ray luminosity of @xmath6 = 3.3@xmath7@xmath310@xmath8 @xmath201 ergs s@xmath9 . l l l l l l l l l l parameters & reg 1 & reg 2 & reg 3 & reg 4 & reg 5 & reg 6 & reg 7 & reg 8 & reg 9 & snr + _ soft component _ & + em@xmath135 ( 10@xmath202@xmath203 @xmath76 ) & 5.4@xmath204 & 6.2@xmath205 & 5.4@xmath206 & 2.7@xmath121 & 4.4@xmath207 & 1.6@xmath208 & 6.0@xmath209 & 4.8@xmath210 & 2.4@xmath211 & 106@xmath212 + @xmath213 ( @xmath214@xmath198 @xmath76 ) & 8.4@xmath215 & 10.4@xmath216 & 8.1@xmath217 & 4.8@xmath121 & 6.3@xmath218 & 10.7@xmath219 & 28.6@xmath220 & 15.2@xmath221 & 4.6@xmath222 & 6.6@xmath223 + @xmath224 ( @xmath225@xmath226 kyr ) & @xmath374.5 & 1.4@xmath227 & @xmath3717.3 & @xmath3711.8 & @xmath372.9 & @xmath373.0 & 0.4@xmath228 & @xmath371.7 & @xmath375.0 & @xmath3735.8 + @xmath180 & * 0.21 * & 0.43 & 0.55 & 0.52 & 0.30 & 0.47 & 0.56 & 0.38 & 0.46 & 0.47 + _ hard component _ & + em@xmath63 ( 10@xmath202@xmath203 @xmath76 ) & 1.6@xmath53 & 0.7@xmath229 & 0.4@xmath151 & 0.2@xmath95 & 0.8@xmath2 & 0.3@xmath228 & 0.3@xmath57 & 0.4@xmath57 & 0.4@xmath54 & 11.7@xmath230 + @xmath196 ( @xmath231@xmath198 @xmath76 ) & 4.5@xmath114 & 3.4@xmath232 & 2.2@xmath233 & 1.4@xmath229 & 2.7@xmath234 & 4.4@xmath235 & 6.3@xmath71 & 4.3@xmath106 & 1.9@xmath137 & 2.2@xmath236 + @xmath237 ( @xmath231@xmath198 @xmath76 ) & 0.9@xmath11 & 0.7@xmath11 & 0.4@xmath11 & 0.3@xmath95 & 0.6@xmath57 & 0.9@xmath238 & 1.3@xmath61 & 0.9@xmath239 & 0.4@xmath57 & 0.5@xmath54 + @xmath240 ( @xmath241@xmath226 kyr ) & 1.9@xmath13 & 1.3@xmath100 & 1.0@xmath242 & 1.2@xmath55 & 3.2@xmath243 & 2.3@xmath244 & 0.5@xmath61 & 0.6@xmath245 & 2.5@xmath207 & 1.2@xmath68 + @xmath181 & * 0.79 * & 0.57 & 0.45 & 0.48 & 0.70 & 0.53 & 0.44 & 0.62 & 0.54 & 0.53 + it has been previously suggested that the remnant could be either in its early evolutionary stage or in a transition to its sedov phase ( gotthelf & vasisht 1997 ; vink & kuiper 2006 ) . under the assumption of a uniform ambient density medium , the swept - up mass is calculated as @xmath246 = ( @xmath247@xmath248)@xmath31.4@xmath249@xmath191 = 16@xmath250 m@xmath251 . for a filling factor @xmath181 = 0.53 ( table 4 ) , @xmath246 @xmath20 1440 @xmath252 m@xmath251 . this range implies that the remnant can be in its late free expansion phase or in a transition to the sedov phase . it is also plausible that the snr may still be in the dense wind in an @xmath253 profile , which would be more realistic for kes 73 if the explosion originated from a massive star ( e.g. , sn iil / b ) as proposed in section 5.4 . we discuss below the evolutionary properties of kes 73 under both scenarios . assuming a free expansion ( or sedov ) phase gives a lower ( or higher ) estimate on the snr age . for an initial expansion velocity of 5000 km s@xmath9 appropriate for a core - collapse snr ( reynolds 2008 ) , we infer a free expansion age of @xmath158750 @xmath175 kyr for the remnant . when the swept - up mass ( @xmath246 ) becomes comparable to or exceeds the ejected mass ( @xmath254 ) , an snr enters the adiabatic phase . if kes 73 is in its early sedov phase , we can compute an upper limit to the sedov age as @xmath45 = @xmath255@xmath177/@xmath256 using the vsedov spectral parameters ( table 3 ) , where @xmath255 depends on the evolutionary stage of the snr and is equal to 0.4 assuming a uniform blast wave expansion ( sedov 1959 ) . the blast wave velocity @xmath256 is estimated as @xmath256 = ( 16@xmath257@xmath258/3@xmath26@xmath259)@xmath260 , where @xmath26 = 0.604 is the mean mass per free particle for a fully ionized plasma , @xmath257 = 1.38@xmath310@xmath261 ergs k@xmath9 is boltzmann s constant and @xmath258 is the post - shock temperature ( sedov 1959 ) . assuming full equilibration between the electrons and nuclei ( a valid assumption here since our vsedov fit gives similar temperatures for the mean temperature of the particles and the electron temperature right behind the shock ) , we determine the shock velocity using the tabulated temperature from our sedov fit ( table 3 ) of @xmath12 = 1.6@xmath13 kev . the inferred velocity of the blast wave is @xmath256 = ( [email protected])@xmath310@xmath262 km s@xmath9 , and the sedov age is estimated as @xmath45 = [email protected] @xmath175 kyr for kes 73 . this age is consistent with the previous estimated age of @xmath282.2 kyr ( gotthelf & vasisht 1997 ; vink & kuiper 2006 ; lopez et al . however , incomplete equilibration at the shock would lead to a higher shock velocity than that estimated above , yielding an even younger age . studies of electron and ion temperatures in snr shocks indicate that equilibration is far from complete at such high velocities ( ghavamian et al . 2007 ) , suggesting that kes 73 is considerably younger than the characteristic age of the associated axp , @xmath263 kyr , which provides an upper limit for the system age . to the best of our knowledge , there is no measurement of the braking index which would provide an accurate age estimate for 1e 1841@xmath0045 . based on the sedov blast wave model in which a supernova with an explosion energy @xmath264 , expands into an ism of uniform density @xmath191 , we can estimate the snr explosion energy as @xmath264 = @xmath265@xmath266@xmath267@xmath191@xmath268 = 3.0@xmath14@xmath310@xmath269 ergs , where @xmath267 = 1.4@xmath249 is the mean mass of the nuclei and @xmath249 is the mass of the proton . this value is slightly lower than the canonical value of @xmath270 ergs , but is consistent with the explosion energy of ( 5 @xmath16 3)@xmath310@xmath15 ergs derived by vink & kuiper ( 2006 ) assuming a distance of 7.0 kpc . if full electron - ion equilibration has not been achieved in the shock for the sedov phase , the shock velocity @xmath256 determined from the electron temperature will be a lower limit to the shock temperature @xmath258 , which will lead to an underestimation of the snr s explosion energy . now , we discuss the model treatment of an snr expanding into the late red - supergiant ( rsg ) phase wind of its massive progenitor . this is motivated by the progenitor mass estimate inferred from our study ( section 5.4 ) and follows the expansion of the snr in an @xmath253 wind profile as discussed in chevalier ( 2005 ) and applied to other snrs such as g296.10.5 ( castro et al . 2011 ) . in the following discussion however , we will focus on the snr s expansion in a wind profile in the adiabatic phase , and infer the snr s expansion velocity and age in comparison to the values estimated from its expansion into a uniform medium ( i.e. the sedov phase described above ) . the sedov profiles for an adiabatic blast wave expansion in a power - law density distribution have been solved analytically by cox and franco ( 1981 ) . chevalier ( 2005 ) describes the circumstellar wind density of a sn iil / b progenitor as @xmath271 = @xmath272 @xmath273 @xmath274 and defines the dimensionless quantity @xmath275 = @xmath276/@xmath277 , where @xmath277 = 1@xmath310@xmath32 g cm@xmath9 is the value of the coefficient of the density profile , @xmath279 is the mass loss and @xmath280 is the rsg wind velocity . the circumstellar mass swept - up by the snr shock to a radius @xmath281 is then given by @xmath246 = 9.8@xmath282(@xmath281/5 pc ) @xmath19 , with the snr s blast wave expanding to a radius @xmath281 = ( 3@xmath264/2@xmath283)@xmath284@xmath285 during the adiabatic phase of its evolution . the corresponding velocity in this sedov approximation is then given by @xmath256 = 2@xmath281/3@xmath45 ( chevalier 1982 ; chevalier 2005 ; castro et al . 2011 ) . for kes 73 , we assume here that the hot component arises entirely from the shocked csm and that the shocked wind fills the full volume of @xmath179 = 2.9@xmath310@xmath202 cm@xmath262 ( which is a reasonable assumption for the adiabatic phase ) . for an emission measure em@xmath63 = 11.7@xmath230@xmath310@xmath202 @xmath287 @xmath76 , we obtain @xmath246 = 62@xmath288 @xmath197@xmath289 m@xmath251 . this value will be subsequently used below to determine the supernova s explosion energy using the equation above for @xmath281 as a function of @xmath264 , @xmath276 ( or @xmath246 ) , and @xmath45 . next , we derive the snr s age ( @xmath45 ) and shock speed ( @xmath256 ) in the rsg wind phase . using the sedov profiles for an adiabatic blast wave in an @xmath253 wind density profile ( cox & franco 1981 ) , the ratio between the average and shock temperature ( weighted by @xmath290 ) in the rsg wind scenario is @xmath291 = 5/7 , whereas in the uniform case @xmath291 = 1.276 . assuming that the x - ray spectrum depends mostly on the average temperature , we approximate the ratio between the shock temperatures and shock speeds in the wind and uniform cases as @xmath292/@xmath293 = 1.786 and @xmath294/@xmath295 = @xmath296 , respectively . now , since @xmath256 = @xmath255 @xmath281/@xmath45 where @xmath255=2/5 for the sedov uniform case and @xmath255=2/3 for the wind adiabatic case , we estimate the ages ratio as @xmath297/@xmath298 = ( 5/3 ) / @xmath296 = 1.247 . using the above - estimated sedov velocity and age of ( 1.2@xmath1610@xmath262)@xmath310@xmath262 km s@xmath9 and [email protected] kyr , respectively , we obtain @xmath256 = ( [email protected])@xmath310@xmath262 km s@xmath9 and @xmath299 = [email protected] @xmath175 kyr . by substituting @xmath299 in the equations for @xmath246 as a function of the snr s age and explosion energy , we can infer @xmath264 = 1.1@xmath300@xmath310@xmath301 @xmath197@xmath289 ergs , which corresponds to @xmath201.5@xmath310@xmath301 ergs for @xmath181=0.53 and a distance of 8.5 kpc . subsequently , the shock age @xmath302 ( the time since the passage of shock ) of the two thermal components was calculated using the ionization timescale ( @xmath43 ) values derived from the fits ( table 4 ) , noting that the soft - component is approaching ionization equilibrium in some of the regions ( table 3 ) . we obtained a shock age of @xmath303 = ( 0.53.2 ) @xmath304@xmath226 kyr for the hard component , which is more or less consistent with the snr age . nucleosynthesis studies through x - ray spectroscopy of shock - heated ejecta associated with young snrs can reveal important information on the nature of their progenitor stars . it is possible to directly measure the abundances of heavy elements obtained from x - ray spectral fitting . by comparing them to the models of explosive nucleosynthesis yields such as those of woosley & weaver 1995 ( hereafter ww95 ) and nomoto et al . 2006 ( hereafter n06 ) , we can estimate the mass of the progenitor star . such studies are also valuable for testing and improving the nucleosynthesis models commonly used to fit x - ray spectra . we use the abundance ratios relative to si given by ( x / si)/(x / si)@xmath251 , where x is the ejected mass of any element with respect to si and with solar values obtained from anders & grevesse ( 1989 ) . in figure 6 , we show the ratios ( with respect to si ) for o , mg , s , and fe , obtained from the global snr spectral fitting , in comparison to the predicted ratios for a range of progenitor masses using the ww95 and n06 models . ww95 specify models a , b , and c for progenitor stars with masses @xmath4030 @xmath19 , where the final kinetic energies at infinity of the ejecta for models b and c are enhanced by a factor of @xmath201.5 and @xmath202 , respectively , with respect to model a ( @xmath305 @xmath20 1.2@xmath310@xmath301 ergs ) . the yields from ww95 and n06 are different due to the different treatment of the convective boundary and since mass loss is not included in ww95 models , whereas n06 includes mass loss as a function of metallicity ( k. nomoto 2013 , private communication ) . as a result , the inferences on the progenitor masses are highly dependent on the models used . when comparing with the ww95 and n06 yields , the abundance ratios are consistent with a 2030 m@xmath251 progenitor star ( figure 6 ) . as mentioned before , s and si are overabundant in most of the regions and while the nucleosynthesis models predict a s / si ratio @xmath1581 , the data here seem to show a much higher s / si ratio ( although with a large error bar ) . a plausible reason could be that , the remnant being young , the reverse shock may not have reached the center yet and hence , the current abundances may not reflect accurately the actual abundances of the heavy elements produced in the supernova explosion . another factor contributing to the uncertainty in the s abundance is that the s line lies in the energy band where the two components have nearly the same flux . we have also compared our values with lower progenitor masses ( @xmath15818 m@xmath251 ) , however , the ratios do not match with such lower mass stars . nevertheless , keeping in mind the error bars and uncertainty in the models , as well as the caveats mentioned above , the existing data and models suggest a progenitor mass @xmath1820 m@xmath251 for kes 73 . chevalier ( 2005 ) suggests a supernova type sn iil / b for kes 73 , with the remnant running into the rsg wind lost from the progenitor star . we discussed this scenario in section 5.3 . sn iil / b stars are believed to have progenitor masses in the range of 2535@xmath19 ( heger et al . 2003 ) with x - ray luminosities typically in the range of @xmath2010@xmath8 ergs s@xmath9 . they end their lives as rsgs with extensive mass loss ( @xmath279 @xmath40 3@xmath310@xmath306 @xmath19 yr@xmath9 for a wind velocity @xmath280 = 15 km s@xmath9 ) from the h envelope , leading to the formation of a dense circumstellar region that extends out to 57 pc from the progenitor ( chevalier 2005 ) . interestingly , the radius of kes 73 ( @xmath206.2 pc ) also closely matches the expected size of the sn iil / b wind bubble , and together with our inferred x - ray luminosity ( @xmath203@xmath310@xmath8 ergs s@xmath9 ) and progenitor mass of @xmath1820@xmath19 , the prediction of a sn iil / b progenitor star for kes 73 seems reasonable . observational studies show that several magnetars are associated with star clusters and by studying their association with these clusters , it is possible to estimate the initial mass of their progenitors . a recent study by clark et al . ( 2009 ) , on the discovery of a third massive rsg cluster ( rsgc3 ) located at the base of the scutum - crux arm , suggests that the snr kes 73 could be physically associated with the rsgc13 star formation complex with the location of 1e 1841@xmath0045 approximately equidistant from the rsg clusters 2 and 3 . therefore , they speculate that the progenitor mass of 1e 1841@xmath0045 can not be @xmath4020@xmath307 by associating the axp with the star formation that yielded rsgc13 . together with another study performed by davies et al . ( 2009 ) which reported a low mass progenitor star of 17 @xmath16 2@xmath307 for the soft gamma repeater ( sgr ) 1900 + 14 , these authors suggest that magnetars do not necessarily descend from massive progenitors . however , as mentioned above , based on our x - ray spectroscopic analysis as well as the nucleosynthesis yield models used here for comparison , we estimate a massive progenitor @xmath4020@xmath307 for kes 73 . next , we compare the estimated progenitor mass for kes 73 to that determined for a few other magnetars and high - magnetic field radio pulsars . the progenitor masses of sgr 180620 associated with the cluster ci 180620 , cxou j164710.2455216 associated with the cluster westerlund 1 ( wd 1 ) , and the axp 1e 1048.15937 were estimated as 48@xmath308 @xmath19 ( bibby et al . 2008 ) , 40 @xmath16 5 @xmath19 ( muno et al . 2006 ) , and 3040 @xmath19 ( gaensler et al . 2005 ) , respectively . the progenitor masses obtained for the two high - b pulsars ( j11196127 and j18460258 ) also suggests massive progenitors . the x - ray spectroscopic study of snr g292.20.5 associated with the high - b radio and x - ray pulsar j11196127 suggests a progenitor mass @xmath2030 @xmath19 ( kumar et al . the snr kes 75 , associated with the high - magnetic field x - ray pulsar j18460258 ( which recently showed magnetar - like bursts and spectral softening ; gavriil et al . 2008 ; kumar & safi - harb 2008 ) , was suggested to have a wolf - rayet progenitor ( morton et al . 2007 ) . in summary , the above mentioned studies imply a range of progenitor masses for the highly magnetized neutron stars , with the majority implying very massive progenitors . however , any conclusion on the factors deciding the origin and evolutionary properties of these sources requires further studies in x - rays and other wavelengths . in particular , in the x - rays , future high - resolution spectroscopic studies are needed to obtain accurate abundance measurements and to perform plasma diagnostics of the individual lines to confirm the temperature of the ejecta as well as that inferred for the interstellar / circumstellar environment in which kes 73 is expanding . in this paper , we have performed the first detailed imaging and spectral analysis of the snr kes 73 , using all archival _ chandra _ and _ xmm - newton _ data , to determine the intrinsic properties of the supernova explosion and the physical properties of the remnant . the main results are summarized as follows : 1 . the high resolution x - ray images confirm a spherical morphology of @xmath205@xmath35 size with several clumpy and knotty structures , and bright diffuse emission originating from the snr interior . the multi - wavelength morphology of the remnant in the radio , infrared , and x - rays shows the western limb to be much brighter than all other regions . the infrared image spatially correlates well with the x - ray image while the radio shell appears to extend slightly beyond the x - ray emission and displays a thin filamentary feature along the edges , which likely represents the location of the forward shock . the spectra obtained from different diffuse emission regions are best fit with a vpshock+vpshock model . the western limb of the snr showed a slightly higher column density given by @xmath160 = 3.3@xmath48@xmath310@xmath4 @xmath5 . the soft component is dominated by enhanced metal abundances mainly for si and s in all the regions , with plasma temperatures of 0.30.5 kev and ionization timescales @xmath3710@xmath142 @xmath76 s ( except for regions 2 and 7 ) . the hard component is dominated by solar abundances , with plasma temperatures of 1.11.7 kev and ionization timescales of ( 0.53.2)@xmath310@xmath75 @xmath76 s. this indicates that the soft component plasma for most regions has reached ionization equilibrium earlier than the plasma associated with the hard component . 3 . the global x - ray emission from kes 73 is best described by a two - component vsedov+vpshock model , with a column density @xmath160 = 2.6@xmath2@xmath310@xmath4 @xmath5 and the soft and hard components characterized by plasma temperatures of 0.5@xmath11 kev ( @xmath141 @xmath37 6.1@xmath310@xmath142 @xmath76 s ) and 1.6@xmath13 kev ( @xmath74 = 0.8@xmath57 @xmath3 10@xmath75 @xmath76 s ) , respectively . the presence of enhanced abundances in the soft component suggests that this component is dominated by shock - heated ejecta , while the hard component characterized mostly by solar abundances is dominated by shocked interstellar / circumstellar material . we have also refined the snr age ranging between 750 @xmath175 yr for the free expansion phase ( assuming an expansion velocity of 5000 km s@xmath9 ) and 2100 @xmath175 yr assuming a sedov phase of evolution . the sedov phase yields a shock velocity of ( 1.2 @xmath16 0.3)@xmath310@xmath17 km s@xmath9 , an explosion energy of @xmath264 = 3.0@xmath14@xmath310@xmath269 ergs , and a swept - up mass of 16@xmath309 m@xmath251 under the assumption of an explosion in a uniform ambient medium . considering kes 73 to be still expanding into the dense wind of its late phase rsg evolution , we infer a shock velocity of ( [email protected])@xmath310@xmath262 km s@xmath9 , an age of [email protected] @xmath175 kyr , and explosion energy @xmath264 = ( 1.1@xmath310@xmath310@xmath301 @xmath197@xmath289 ergs . these derived values are consistent with the predictions of a very massive progenitor for kes 73 . 4 . although the abundances are not well - constrained , the abundance ratios , when compared to core - collapse nucleosynthesis models , suggest a progenitor mass @xmath1820@xmath311 for kes 73 . a much deeper exposure with existing x - ray missions , and high - resolution spectroscopy with the soft x - ray spectrometer of _ astro - h _ ( takahashi et al . 2012 ) , are needed for an accurate measurement of the abundances and shock velocities , and to further motivate the development of the existing nucleosynthesis models . we thank ken@xmath21ichi nomoto for helpful discussions on the nucleosynthesis yield models for core - collapse supernovae and the referee for useful comments that helped improved the paper . this research made use of nasa s astrophysics data system ( ads ) and of nasa s heasarc maintained at the goddard space flight center ( gsfc ) . s. safi - harb acknowledges support by a discovery grant from the natural sciences and engineering research council of canada ( nserc ) , the canada research chairs ( crc ) program , the canada foundation for innovation , the canadian institute for theoretical astrophysics , and the canadian space agency . pos acknowledges support from nasa contract nas8 - 03060 .

message - passing iterative decoders for low - density parity - check ( ldpc ) block codes are known to be subject to decoding failures due to so - called pseudo - codewords . these failures can cause the large signal - to - noise ratio performance of message - passing iterative decoding to be worse than that predicted by the maximum - likelihood decoding union bound . in this paper we address the pseudo - codeword problem from the convolutional - code perspective . in particular , we compare the performance of ldpc convolutional codes with that of their `` wrapped '' quasi - cyclic block versions and we show that the minimum pseudo - weight of an ldpc convolutional code is at least as large as the minimum pseudo - weight of an underlying quasi - cyclic code . this result , which parallels a well - known relationship between the minimum hamming weight of convolutional codes and the minimum hamming weight of their quasi - cyclic counterparts , is due to the fact that every pseudo - codeword in the convolutional code induces a pseudo - codeword in the block code with pseudo - weight no larger than that of the convolutional code s pseudo - codeword . this difference in the weight spectra leads to improved performance at low - to - moderate signal - to - noise ratios for the convolutional code , a conclusion supported by simulation results . * index terms * convolutional codes , quasi - cyclic codes , low - density parity - check ( ldpc ) codes , linear programming decoding , message - passing iterative decoding , pseudo - codewords , pseudo - weights .

cs0609148

although low - density parity - check block codes ( ldpc - bcs ) have very good performance under message - passing iterative ( mpi ) decoding , they are known to be subject to decoding failures due to so - called pseudo - codewords . these are real - valued vectors that can be loosely described as error patterns that cause non - convergence in iterative decoding due to the fact that the algorithm works locally and can give priority to a vector that fulfills the equations of a graph cover rather than the graph itself . [ example : introduction ] consider the trivial length-@xmath0 and dimension-@xmath1 code @xmath2 defined by the @xmath3 parity - check matrix @xmath4 and whose tanner graph is shown in fig . [ fig : smallgraph:1 ] ( left ) . a possible cubic cover of this tanner graph is also depicted in fig . [ fig : smallgraph:1 ] ( right ) . because @xmath5 is a valid configuration in this cubic cover , the vector @xmath6 is a pseudo - codeword of the code @xmath7 defined by @xmath8 ( see @xcite ) . it has been shown @xcite that the performance of mpi decoding schemes for ldpc - bcs is largely dominated _ not _ by minimum hamming weight considerations but by minimum _ pseudo - weight _ considerations , where the minimum pseudo - weight in the case of an additive white gaussian noise channel ( awgnc ) is defined as @xmath9 where @xmath10 and @xmath11 are , respectively , the @xmath12-norm and @xmath13-norm , and @xmath14 is the set of all pseudo - codewords for the code @xmath7 defined by @xmath8 . the minimum pseudo - weight is a measure of the effect that decoding failures have on the performance of the code . , the minimum hamming weight may still be a key factor in performance analysis ; however , the larger the gap between @xmath15 and @xmath16 , the greater role the minimum pseudo - weight plays . in any case , the minimum hamming weight is important for quantifying the impact of undetectable errors . ] as a consequence , the large signal - to - noise ratio ( snr ) performance of mpi decoding can be worse than that predicted by the maximum - likelihood decoding union bound , which constitutes a major problem when trying to determine performance guarantees . addressing this problem from the convolutional - code perspective , i.e. , studying the pseudo - codeword problem described above for ldpc convolutional codes ( ldpc - ccs ) , constitutes the major topic of this paper . we investigate a class of time - invariant ldpc convolutional codes derived by `` unwrapping '' certain classes of quasi - cyclic ( qc ) ldpc block codes that are known to have good performance @xcite . unwrapping a qc block code to obtain a time - invariant convolutional code represents a major link between qc block codes and convolutional codes . this link was first introduced in a paper by tanner @xcite , where it was shown that the free distance of the unwrapped convolutional code , if non - trivial , can not be smaller than the minimum distance of the underlying qc code . this idea was later extended in @xcite . more recently , a construction for ldpc convolutional codes based on qc - ldpc block codes was introduced by tanner et al . @xcite , and a sliding - window mpi decoder was described . in that paper it was noted that the ( non - trivial ) convolutional versions of these codes significantly outperformed their block code counterparts in the waterfall region of the bit error rate ( ber ) curve , even though the graphical representations of mpi decoders were essentially equivalent . throughout this paper we mainly take the approach of @xcite , which connects the presence of pseudo - codewords in mpi decoding and linear programming ( lp ) decoding . lp decoding was introduced by feldman , wainwright , and karger @xcite and can be seen as a relaxation of the maximum - likelihood decoding problem . more precisely , maximum - likelihood decoding can be formulated as the solution to an optimization problem , where a linear cost function is minimized over a certain polytope , namely the polytope that is spanned by the set of all codewords . for general codes , there is no efficient description of this polytope and so feldman et al . suggested replacing it with an ( efficiently describable ) relaxed polytope , which in the following will be called the `` fundamental polytope '' . in other words , the decoding result of the lp decoder is the point in the fundamental polytope that minimizes the above - mentioned linear cost function . in order to analyze the behavior of unwrapped ldpc convolutional codes under lp decoding , we need to examine the fundamental polytope @xcite of the underlying qc - ldpc block codes . ( because of symmetries , it is actually sufficient to study the structure of the fundamental polytope around the zero codeword , i.e. , it is sufficient to study the so - called fundamental cone . ) our goal is to formulate analytical results ( or at least efficient procedures ) that will allow us to bound the minimum pseudo - weight of the pseudo - codewords of the block and convolutional codes . the paper aims at addressing this question and related issues . in the following sections , we will study the connections that exist between pseudo - codewords in qc codes and pseudo - codewords in the associated convolutional codes and show that this connection mimics the connection between the codewords in qc codes and the associated convolutional codes . as a motivational example we simulated a rate @xmath17 @xmath18-regular ldpc - cc with syndrome former memory @xmath19 , together with three wrapped block code versions : a @xmath20 $ ] @xmath18-regular qc - ldpc block code , a @xmath21 $ ] code , and a @xmath22 $ ] code , with parity - check matrices of increasing circulant sizes @xmath23 , @xmath24 , and @xmath25 , respectively , while keeping the same structure within each @xmath26 circulant @xcite ( see section [ sec : link : qc : and : conv : codes:1 ] ) . ( note that increasing the circulant size of the qc code increases its complexity , i.e. , its block length . also note that each of the three block codes has rate slightly greater than @xmath27 . ) a sliding - window mpi decoder as in @xcite was used to decode the convolutional code . conventional ldpc - bc mpi decoders were employed to decode the qc - ldpc block codes . all decoders were allowed a maximum of @xmath28 iterations . the resulting ber performance of these codes on a binary - input awgn channel is shown in fig . [ fig : compare:1 ] . we note that , particularly in the low - to - moderate snr region , where the complete pseudo - weight spectrum plays an important role , the unwrapped ldpc - cc performs between @xmath29 and @xmath30 better than the associated qc - ldpc block codes . also , as the circulant size increases , the performance of the block codes approaches that of the convolutional code . these performance curves suggest that the pseudo - codewords in the block code that result in decoding failures may not cause such failures in the convolutional code , which suggests that ldpc - ccs may have better iterative decoding thresholds than comparable ldpc - bcs ( see also @xcite and @xcite ) . in order to underline the influence of pseudo - codewords under mpi decoding , we consider the following experiment . let @xmath31 be a _ minimal pseudo - codeword _ @xcite for the above - mentioned @xmath17 @xmath18-regular ldpc - cc , i.e. , a pseudo - codeword that corresponds to an _ edge of the fundamental cone _ of that ldpc - cc . moreover , we define the log - likelihood ratio vector @xmath32 to be is not squared , therefore this ratio does not correspond to the awgnc pseudo - weight of @xmath31 ( see sec . [ sec : pseudo : weight : comparison:1 ] ) , although it is closely related to that value . ] @xmath33 we then run the mpi decoder that is initialized with @xmath34 and count how many iterations it takes until the decoder decides for the all - zero codeword as a function of @xmath35 and @xmath36 . the results are shown in fig . [ fig : alphabetaplot:1 ] . the meaning of @xmath32 is the following . if @xmath37 , then @xmath38 corresponds to the log - likelihood ratio vector that the receiver sees when the communication system operates at a signal - to - noise ratio @xmath39 and when the noise vector that is added by the binary - input awgn channel happens to be the all - zero vector ( see , e.g. , the discussion in ( * ? ? ? 3 ) ) . for non - zero @xmath35 the expression for @xmath40 has been set up such that the lp decoder has a decision boundary at @xmath41 : for @xmath42 the all - zero codeword wins against the pseudo - codeword @xmath31 whereas for @xmath43 the all - zero codeword loses against the pseudo - codeword @xmath31 under lp decoding . the simulations in fig . [ fig : alphabetaplot:1 ] were obtained using a search algorithm that looked for a low - pseudo - weight minimal pseudo - codeword @xmath31 in the fundamental cone of the above - mentioned ldpc - cc . the @xmath31 that we found has awgnc pseudo - weight @xmath44 , which happens to be smaller than the free distance . secondly , we ran the mpi decoder for various choices of @xmath35 and @xmath36 : fig . [ fig : alphabetaplot:1 ] left shows the number of iterations needed using a sum - product - algorithm - type mpi decoder whereas fig . [ fig : alphabetaplot:1 ] right shows the number of iterations needed using a min - sum - algorithm - type mpi decoder . ( note that the decisions reached by the latter are independent of the choice of @xmath36 , @xmath45 . ) because of the more oscillatory behavior of the min - sum - algorithm - type mpi decoder close to decision boundaries of the lp decoder , observed empirically , it is advantageous to run that decoder for many iterations in our scenario , whereas in the case of the sum - product - algorithm - type mpi decoder it hardly pays to go beyond @xmath46 iterations . in this paper , we provide a possible explanation for the performance difference observed in the motivational example above . based on the results of @xcite that relate code performance to the existence of pseudo - codewords , we examine the pseudo - codeword weight spectra of qc - ldpc block codes and their associated convolutional codes . we will show that for a non - trivial ldpc - cc derived by unwrapping a non - trivial qc - ldpc block code , the minimum pseudo - weight of the convolutional code is at least as large as the minimum pseudo - weight of the underlying qc code@xcite , i.e. , @xmath47 this result , which parallels the well - known relationship between the free hamming distance of non - trivial convolutional codes and the minimum hamming distance of their non - trivial quasi - cyclic counterparts @xcite , is based on the fact that every pseudo - codeword in the convolutional code induces a pseudo - codeword in the block code with pseudo - weight no larger than that of the convolutional code s pseudo - codeword . this difference in the weight spectra leads to improved ber performance at low - to - moderate snrs for the convolutional code , a conclusion supported by the simulation results presented in fig . [ fig : compare:1 ] . the paper is structured as follows . in sec . [ sec : ps : qc : and : conv : codes:1 ] we develop the background necessary to describe the connection between pseudo - codewords in unwrapped ldpc convolutional codes and those in the associated qc - ldpc codes . thus , in sec . [ sec : link : qc : and : conv : codes:1 ] we briefly discuss the connection between convolutional codes and their associated qc codes , especially how codewords in the former can be used to construct codewords in the latter , and in sec . [ sec : fundam : cone : par : check : matrix:1 ] we define the fundamental polytope / cone of a matrix and show how we can describe the fundamental cone of a polynomial parity - check matrix through polynomial inequalities . we end the section by showing how pseudo - codewords in unwrapped ldpc convolutional codes yield pseudo - codewords in the associated qc - ldpc codes . in sec . [ sec : pseudo : weight : comparison:1 ] , we compare the pseudo - weights of unwrapped convolutional and their associated qc block codes . sec . [ sec : pseudo : weight : definitions:1 ] introduces various channel pseudo - weights and sec . [ sec : pseudo : weight : inequality:1 ] presents the main result , namely that the minimum awgn pseudo - weight , the minimum binary erasure channel ( bec ) pseudo - weight , the minimum binary symmetric channel ( bsc ) pseudo - weight , and the minimum max - fractional weight of a convolutional code are at least as large as the corresponding minimum pseudo - weights of a wrapped qc block code . [ sec : analysis : problematic : pseudo : codewords : conv : codes:1 ] discusses a method to analyze problematic pseudo - codewords , i.e. , pseudo - codewords with small pseudo - weight . the method addresses the convolutional code case . it introduces two sequences of `` truncated '' pseudo - weights and , respectively , `` bounded pseudo - codeword '' pseudo - weights , that play an important role in identifying the minimum pseudo - weight of the convolutional code , similar to the role that column distances and row distances play in identifying the free distance . we end with some conclusions in sec . [ sec : conclusions:1 ] . throughout the paper we will use the standard way of associating a tanner graph with a parity - check matrix and vice - versa . we will use the following notation . we let @xmath48 , @xmath49 , @xmath50 , and @xmath51 be the galois field of size @xmath13 , the field of real numbers , the set of non - negative real numbers , and the set of positive real numbers , respectively . if @xmath52 is a polynomial over some field and @xmath53 is some positive integer then @xmath54 denotes the polynomial @xmath55 of degree smaller than @xmath53 such that @xmath56 . we say that a polynomial @xmath57 with real coefficients is non - negative , and we write @xmath58 , if all its coefficients @xmath59 satisfy @xmath60 . similarly , a polynomial vector @xmath61 is non - negative , and we write @xmath62 , if all its polynomial components @xmath63 satisfy @xmath58 for all @xmath64 . moreover , a polynomial matrix @xmath65 is non - negative , and we write @xmath66 , if all its entries are non - negative polynomials . finally , for any positive integer @xmath53 and for any @xmath67 , @xmath68 will represent the @xmath69-times cyclically left - shifted identity matrix of size @xmath70 . we begin this section by presenting an important link between qc block codes and convolutional codes that was first introduced in a paper by tanner @xcite and later extended in @xcite . similar to the connection between codewords in an unwrapped convolutional code and codewords in the underlying qc code , we will show that pseudo - codewords in the convolutional code give pseudo - codewords when projected onto an underlying qc code . in this section we introduce the background needed for the later development of the paper . note that all codes will be binary linear codes . any length @xmath71 quasi - cyclic ( qc ) code @xmath72 with period @xmath73 can be represented by an @xmath74 _ scalar _ block parity - check matrix @xmath75 that consists of @xmath76 circulant matrices of size @xmath26 . using the isomorphism between the ring of binary circulant matrices of size @xmath26 and the ring of polynomials @xmath77 / \langle x^r - 1 \rangle}$ ] of degree less than @xmath53 , we can associate with the scalar parity - check matrix @xmath75 the _ polynomial _ parity - check matrix @xmath78 / \langle x^r - 1 \rangle}\big)^{j \times l}$ ] , with polynomial operations performed modulo @xmath79 . due to the existence of this isomorphism , we can identify two descriptions , scalar and polynomial , and use either of the two depending on their usefulness . [ cubic : matrices ] the cubic cover @xmath80 of the trivial code in depicted in fig . [ fig : smallgraph:1 ] is a quasi - cyclic code with scalar parity - check matrix @xmath81 and corresponding polynomial parity - check matrix @xmath82 / \langle x^r - 1 \rangle}\big)^{3 \times 3}$ ] ( @xmath83 ) given by @xmath84 by permuting the rows and columns of the scalar parity - check matrix @xmath75 ( i.e. , by taking the first row in the first block of @xmath53 rows , the first row of the second block of @xmath53 rows , etc . , then the second row in the first block , the second row in the second block , etc . , and similarly for the columns ) , we obtain the parity - check matrix @xmath85 of a code that is equivalent to @xmath86 . the scalar parity - check matrix @xmath85 has the form @xmath87 where the scalar @xmath88-matrices @xmath89 satisfy @xmath90 . [ cubic : matrices : decomposition ] the code in example [ cubic : matrices ] has @xmath91 , where @xmath92 given the polynomial parity - check matrix of a qc - code , it is easy to see the natural connection that exists between quasi - cyclic codes and convolutional codes ( see , e.g. , @xcite ) . briefly , with any qc block code @xmath86 of length @xmath93 , given by a @xmath94 polynomial matrix parity - check matrix @xmath95 , with polynomial operations performed modulo @xmath79 , we can associate a rate @xmath96 convolutional code @xmath97 given by the same @xmath94 polynomial parity - check matrix @xmath98 , where the change of variables indicates the lack of modulo @xmath99 operations @xcite . let the _ syndrome former memory _ @xmath100 of @xmath97 be the largest integer in @xmath101 such that @xmath102 . then the polynomial parity - check matrix @xmath103 has the following scalar description ( see , e.g. , @xcite ) @xmath104 ( note that @xmath105 is a semi - infinite parity - check matrix . ) [ conv : matrix ] unwrapping the code in example [ cubic : matrices ] , we obtain a convolutional code with polynomial parity - check matrix @xmath106 given by @xmath107 the syndrome former memory is equal to 2 and the decomposition leading to the scalar matrix @xmath105 is @xmath108 where @xmath109 are the same matrices as in . in the same way as we obtained @xmath97 from @xmath86 , we can , upon choosing a positive integer @xmath53 , obtain a code @xmath110 from a given code @xmath111 . , i.e. , @xmath112 . some of the results can be stated for an arbitrary @xmath53 . ] it easily follows that , similar to obtaining @xmath105 by unwrapping @xmath85 , @xmath85 is obtained from @xmath105 by suitably wrapping @xmath105 . this unwrapping / wrapping can also be observed in the tanner graph representations of @xmath113 and @xmath103 , as illustrated in fig . [ fig : tanner : graph:1 ] . we turn now our attention to codewords in these codes . codewords in @xmath86 will be denoted by a polynomial vector like @xmath114 and codewords in @xmath97 will be denoted by a polynomial vector like @xmath115 ( when the code is represented by a scalar parity - check matrix @xmath105 , then it will be denoted by a vector like @xmath116 ) . ( we will later use a similar notation for pseudo - codewords . ) for any non - zero codeword @xmath115 with finite support in the convolutional code , its @xmath53 _ wrap - around _ , defined by the vector @xmath117 / \langle x^r - 1 \rangle}\big)^l$ ] , is a codeword in the associated qc - code , since @xmath118 in @xmath119 / \langle x^r - 1 \rangle}\big)^l$ ] . in addition , the hamming weight of the two codewords is linked by the following inequality @xmath120 , which gives the inequality @xcite ( we assume all codes are non - trivial ) @xmath121 moreover , we have the following inequalities . @xmath122 and @xmath123 let @xmath124 be a parity - check matrix of @xmath125 and @xmath126 be the corresponding parity - check matrix of @xmath86 . let @xmath114 be a nonzero codeword in @xmath127 of weight equal to the minimum distance @xmath128 . we have @xmath129 in @xmath119 / \langle x^{2r } - 1 \rangle}\big)^l$ ] and since @xmath130 , it follows that @xmath131 in @xmath119 / \langle x^r - 1 \rangle}\big)^l$ ] . if @xmath132 then we obtain the inequality @xmath133 if @xmath134 we can write @xmath135 with @xmath136 ( otherwise @xmath137 . ) hence @xmath138 or equivalently , @xmath139 , which implies that @xmath140 is a nonzero codeword in @xmath86 . the hamming weight @xmath141 mimicking this proof we also obtain @xmath142 and hence the desired inequality @xmath143 from the way we construct the semi - infinite sliding matrix of the convolutional by unwrapping the scalar parity - check matrix of the qc block code versions , we can see that there exists a qc code of circulant size @xmath53 large enough , so that its minimum distance is equal to the free distance of the convolutional code . this assures the limit equality above . if we denote by @xmath144 , @xmath145 , @xmath146 , @xmath147 the tanner graphs associated with the parity - check matrices of @xmath86 , @xmath148 , @xmath149 , @xmath147 of increasing size ( which are associated with the same polynomial matrix @xmath150 ) , it is easy to see that we obtain a _ tower of covers _ of the graph @xmath151 . the above relationship between the minimum distances of the codes in the tower is then easily verified in graph language , since a codeword @xmath115 in a larger graph , say @xmath149 , when projected onto the graphs of @xmath86 and @xmath148 using the formula @xmath152 , respectively , @xmath153 , gives another codeword . finally , the graph of the associated convolutional code is an infinite cover of each of the graphs in the tower . [ codewords ] using the graphs associated with the trivial code @xmath154 and its cubic cover @xmath80 having parity - check matrix @xmath81 ( see example [ cubic : matrices ] ) corresponding to the codeword @xmath155 in @xmath7 , we can identify the codeword @xmath156 in @xmath80 . the polynomial description of @xmath116 is @xmath157 the wrap - around of @xmath114 @xmath158 , @xmath159 / \langle x^r - 1 \rangle}\big)^3 , r=1 $ ] , gives the codeword @xmath160 in @xmath161 the associated convolutional code is trivially zero . in this section we introduce our main object of study , the fundamental cone of a matrix , a set that contains all relevant pseudo - codewords @xcite . [ def : fundamental : cone:1 ] let @xmath8 be a binary matrix of size @xmath162 , let @xmath163 be the set of column indices of @xmath8 , and let @xmath164 be the set of row indices of @xmath8 , respectively . for each @xmath165 , we let @xmath166 be the support of the @xmath167-th row of @xmath8 . the _ fundamental polytope _ @xmath168 of @xmath8 is then defined as @xcite @xmath169 where @xmath170 is the @xmath167-th row of @xmath8 , and @xmath171 is the convex hull of @xmath172 , defined as the set of convex combinations of points in @xmath172 when seen as points in @xmath173 . the _ fundamental cone _ @xmath174 of @xmath8 is defined as the conic hull of the fundamental polytope @xmath175 , which includes the vertex @xmath176 , stretched to infinity . note that if @xmath177 , then also @xmath178 for any real @xmath179 . moreover , for any @xmath177 , there exists an @xmath179 ( in fact , a whole interval of @xmath35 s ) such that @xmath180.@xmath181 vectors in @xmath182 are called _ pseudo - codewords of @xmath8 _ , and we will call any vector in @xmath174 a pseudo - codeword and two pseudo - codewords that are equal up to a positive scaling constant will be considered to be equivalent . clearly , not all pseudo - codewords are codewords , but all codewords are pseudo - codewords . the fundamental cone and polytope can both be described by certain sets of inequalities that are computationally useful @xcite . since our study of pseudo - codewords will rely heavily on the fundamental cone , we now describe it in more detail . a vector @xmath183 is in the fundamental cone @xmath184 if and only if @xmath185 let @xmath186 $ ] be a parity - check matrix for the code @xmath187 @xmath188 . the fundamental polytope @xmath182 associated with @xmath8 is ( see fig . [ fig : polytope:1 ] ) @xmath189 from which we obtain the fundamental cone @xmath174 associated with @xmath8 : @xmath190 we see that the pseudo - codeword @xmath191 satisfies the inequalities of both the fundamental polytope and the fundamental cone . also , the pseudo - codeword @xmath192 and its positive multiples are in the fundamental cone but not necessarily in the fundamental polytope . as explained briefly in sec . [ sec : introduction ] , pseudo - codewords can also be described using graph theory language as vectors corresponding to codewords in a cover graph . looking again at the cover tower @xmath193 , @xmath145 , @xmath146 , @xmath147 , a codeword @xmath115 of a larger graph , say @xmath149 , seen as a vector with real components , when projected over @xmath49 onto the graphs of @xmath86 and @xmath148 , using the formula @xmath194 , respectively , @xmath153 , gives a pseudo - codeword . similarly , a codeword @xmath115 of the unwrapped convolutional code @xmath97 , when projected onto the graphs of @xmath86 @xmath148 , @xmath195 using the formulas @xmath152 , @xmath153 , @xmath196 , gives pseudo - codewords in the qc codes . let @xmath197 be the codeword identified in example [ codewords ] for the code @xmath80 , with corresponding polynomial description @xmath198 . when projecting @xmath114 over @xmath49 onto the graph of @xmath154 we obtain @xmath199 , which is a pseudo - codeword in @xmath7 , equivalent to the pseudo - codeword @xmath200 presented in example [ example : introduction ] . the fundamental cone of the convolutional code given by @xmath103 in example [ conv : matrix ] is trivially zero , i.e. , it does not contain any non - zero pseudo - codewords . in the above example we made the following important connection between two graphical ways of representing pseudo - codewords @xmath31 of a qc code @xmath7 of length @xmath201 as vectors corresponding to codewords @xmath114 in some qc code ( or convolutional code ) associated with an @xmath202-cover ( or an infinite cover ) of the graph of @xmath7 . we can see them as vectors with each entry representing an average of the ones attributed to the variable nodes of the cover graph that are in the `` cloud '' of that entry , e. g. , @xmath200 , corresponding to pseudo - codewords in the polytope of @xmath7 , or we can see them as vectors that are the projections of @xmath114 onto the code @xmath7 by the formula @xmath152 , e.g. , @xmath203 , corresponding to pseudo - codewords in the fundamental cone of @xmath7 that are not necessarily in the polytope . in fact a similar statement can be made about pseudo - codewords in any block code @xmath7 of length @xmath201 , not having necessarily a qc structure . a pseudo - codeword @xmath31 in @xmath7 is a vector corresponding to some codeword @xmath116 in some block code ( or convolutional code ) associated with an @xmath202-cover ( or an infinite cover ) of the graph of @xmath7 with each entry representing an average of the ones attributed to the variable nodes of the cover graph that are in the `` cloud '' of that entry . however , this is the same , under the equivalence between different representations of a pseudo - codeword in the fundamental cone , as defining a pseudo - codeword @xmath31 as a vector for which there exists a cover corresponding to a block code or to a convolutional code , @xmath204 , of the code @xmath7 , and there exists some codeword , @xmath116 in @xmath204 , such that @xmath205 where the vector @xmath206 is equal to a permutation of the vector @xmath116 ( where the components of @xmath116 are taken in the following order @xmath207 ) . the vector @xmath208 it is seen as a vector with real entries and the modular operation @xmath209 is performed over the real numbers . the above description of the fundamental polytope and cone provides computationally easy descriptions of the pseudo - codewords , especially in the case of qc and convolutional codes . from and it follows that for any parity - check matrix @xmath8 there is a matrix @xmath210 such that a vector @xmath31 is a pseudo - codeword in the fundamental code @xmath211 if and only if @xmath212 . in the case of qc and convolutional codes , the fundamental cone can be described with the help of polynomial matrices . namely , for any @xmath113 there is a polynomial matrix @xmath213 such that @xmath214 if and only if @xmath215 . similarly , for any @xmath103 there is a polynomial matrix @xmath216 such that @xmath217 if and only if @xmath218 . this fundamental - cone description is particularly simple and hence useful in the case of * monomial * parity - check matrices @xmath106 ( see @xcite ) , since the fundamental - cone inequalities can then be stated very simply as follows . a vector @xmath31 is a pseudo - codeword in the fundamental cone of a monomial matrix @xmath219 if and only if the associated polynomial vector @xmath220 satisfies : @xmath221 equivalently , if @xmath222 is an @xmath223 matrix with @xmath224-th row vector @xmath225 for @xmath226 , then @xmath227 is a pseudo - codeword if and only if @xmath228 for all @xmath229 and @xmath230 [ ex : smallexample:1 ] let @xmath231 be a polynomial parity - check matrix of a rate-@xmath232 convolutional code . the following vector @xmath233 corresponding to the scalar vector @xmath234 is a pseudo - codeword for the convolutional code since @xmath235 and the matrix @xmath236 gives @xmath237 i.e. , @xmath238 has all polynomial entries with non - negative coefficients . similarly , for a qc code that is described by a * monomial * @xmath88 parity - check matrix @xmath113 , a vector @xmath239 is a pseudo - codeword in the corresponding fundamental cone if and only if the associated polynomial vector @xmath240 satisfies @xmath241 and @xmath242 [ ex : small : example : qc:1 ] let @xmath243 be the @xmath244 parity - check matrix obtained from @xmath103 in example [ ex : smallexample:1 ] for a qc block code of length @xmath245 . then , for the polynomial vector @xmath246 we obtain @xmath247 since @xmath248 and @xmath249 , we know that @xmath250 is a pseudo - codeword . the pseudo - codeword @xmath251 in example [ ex : smallexample:1 ] was obtained by using a method to be described in sec . [ sec : analysis : problematic : pseudo : codewords : conv : codes:1 ] . projecting this pseudo - codeword in the convolutional code onto the @xmath53-wrapped qc code , for @xmath244 , we obtained the pseudo - codeword @xmath252 in the qc code . this is not a mere coincidence as the following lemma shows . [ lemma : qc : vs : conv : code : pseudo : codeword:1 ] let @xmath227 be a pseudo - codeword in the convolutional code defined by @xmath103 , i.e. , @xmath253 . then its @xmath53 wrap - around polynomial vector is a pseudo - codeword in the associated qc - code defined by @xmath113 , i.e. , @xmath254 . we have seen that for any @xmath103 describing a convolutional code there is a matrix @xmath216 such that a polynomial vector @xmath227 is a pseudo - codeword in the fundamental cone @xmath255 if and only if @xmath256 . by reducing @xmath216 modulo @xmath99 , we obtain a matrix @xmath257 with the property that a polynomial vector @xmath239 is a pseudo - codeword in the fundamental cone @xmath258 if and only if @xmath259 . reducing @xmath256 modulo @xmath99 , we obtain @xmath260 , which proves the claim . this result can be easily deduced using the graph - theory language at the end of sec . [ sec : link : qc : and : conv : codes:1 ] . indeed , looking again at the cover tower @xmath261 , a pseudo - codeword @xmath262 of a larger graph , say @xmath149 , when projected onto the graphs of @xmath86 and @xmath148 , using the formula @xmath263 , respectively , @xmath264 , gives another pseudo - codeword . similarly , a pseudo - codeword @xmath262 of the unwrapped convolutional code @xmath97 , when projected onto the graphs of @xmath86 @xmath148 , @xmath195 using the formulas @xmath265 , @xmath264 , @xmath266 gives pseudo - codewords in the qc codes . [ projecting : conv : pseudo - codewords ] the pseudo - codeword @xmath267 of the convolutional code in example [ ex : smallexample:1 ] projects onto the following pseudo - codewords in the qc wrapped block code versions of parity - check matrix @xmath268 , for all @xmath269 : , as lemma [ lemma : qc : vs : conv : code : pseudo : codeword:1 ] holds for each @xmath270 . ] this section begins with an introductory subsection in which various channel pseudo - weights are defined and continues with the main result that the minimum awgnc , bec , bsc , and max - fractional pseudo - weights of a convolutional code are at least as large as the corresponding pseudo - weights of a wrapped qc block code . @xcite let @xmath272 @xmath273 be a _ nonzero _ vector in @xmath274 . the awgnc pseudo - weight and the bec pseudo - weight of the vector @xmath31 are defined to be , respectively , @xmath275 where @xmath276 and @xmath277 are the @xmath12-norm , respectively @xmath13-norm , of @xmath31 , and @xmath278 is the set of all indices @xmath279 corresponding to nonzero components @xmath280 of @xmath31 . in order to define the bsc pseudo - weight @xmath281 , we let @xmath282 be the vector of length @xmath283 with the same components as @xmath31 but in non - increasing order . now let @xmath284 then the binary symmetric channel ( bsc)-pseudo - weight @xmath281 is @xmath285 . finally , the fractional weight of a vector @xmath286^n$ ] and the max - fractional weights of a vector @xmath287 are defined to be , respectively , @xmath288 where @xmath289 is the infinite or max norm . for @xmath290 we define all of the above pseudo - weights , fractional weights , and max - fractional weights to be zero . @xmath181 a discussion of the motivation and significance of these definitions can be found in @xcite . note that whereas the fractional weight has an operational meaning only for vertices of the fundamental polytope , the other measures have an operational meaning for any vector in the fundamental polytope or cone . note also that here @xmath291 and @xmath292 are defined for any vector in @xmath274 , whereas @xmath291 and @xmath292 in @xcite are what we will call @xmath293 and @xmath294 . let @xmath295 be the pseudo - codeword in example [ ex : smallexample:1 ] and let @xmath296 @xmath297 be its scalar vector description . then @xmath298 in order to compute @xmath281 , we let @xmath299 be the vector that lists the components of @xmath31 in non - increasing order . we obtain @xmath300 , since we need to add up @xmath301 ordered components of @xmath282 to obtain @xmath302 . finally @xmath303 , from which it follows that @xmath304 . a measure of the effect that the pseudo - codewords have on the performance of a code is given by the minimum _ pseudo - weight _ @xcite @xmath305 where @xmath306 is the set of all non - zero vertices in the fundamental polytope @xmath182 and the pseudo - weights are the appropriate ones for each channel ( awgnc , bsc , and bec pseudo - weights ) and the minimum fractional and max - fractional weights . computing these values can be quite challenging , since the task of finding the set of vertices of @xmath182 is in general very complex . however , in the case of four of the above pseudo - weights ( the minimum awgnc , bsc , and bec pseudo - weights and the minimum max - fractional weight ) there is a computationally simpler description given by @xmath307 for the appropriate pseudo - weight . ( note that there is no such statement for the minimum fractional weight ; see , e.g. , @xcite ) . let @xmath243 be the matrix of example [ ex : small : example : qc:1 ] and let @xmath86 be the qc block code of length @xmath308 , with @xmath309 . the minimum distance is 6 and it is equal to the @xmath310 . ( this was obtained using a vertex enumeration program for polytopes that lists all the minimal pseudo - codewords of a code , see @xcite . ) [ remark : relationship : minium : weights:1 ] in @xcite it was shown that for any code defined by a parity - check matrix @xmath8 , the following inequalities hold : @xmath311 therefore , @xmath312 and @xmath313 can serve as lower bounds for @xmath314 , @xmath315 , and @xmath316 . in what follows , we compare the minimum pseudo - weights and minimum max - fractional weight of a qc block code to those of its corresponding convolutional code which we assume to have a fundamental cone containing non - zero vectors . in order to analyze the minimum pseudo - weight and minimum max - fractional weight , it is sufficient to analyze the weights of the non - zero vectors in the fundamental cone . throughout this section , without loss of generality , all pseudo - codewords @xmath227 are assumed to have finite support . with @xmath317 . note that such polynomial vectors also fulfill @xmath318 . ] [ th : qc : vs : conv : code : pseudo : weight:1 ] for the awgnc , bec , and bsc pseudo - weights , if @xmath319 , then @xmath320 therefore , if the fundamental cone of the convolutional code is not trivial ( i.e. , it contains non - zero vectors ) we obtain @xmath321 in the following , we need to analyze separately the awgnc , bec , and bsc pseudo - weights of @xmath227 and of its @xmath53 wrap - around @xmath322 . let @xmath323 be a pseudo - codeword . by assumption @xmath227 has finite support , i.e. , there exists an integer @xmath324 such that the maximal degree of any @xmath63 , @xmath64 , is smaller than @xmath324 . * case 1 ( awgnc ) : * since @xmath325 and @xmath326 we obtain @xmath327 . * case 2 ( bec ) : * since the components of the vector @xmath322 are obtained by adding in @xmath50 certain non - negative components of @xmath227 , it follows that @xmath328 and we obtain @xmath329 . * case 3 ( bsc ) : * in order to compare the bsc - pseudo - weight of the two vectors , we first need to arrange the components in decreasing order . let @xmath330 and @xmath331 be listings of all the coefficients of all the components of @xmath227 and @xmath332 , respectively , in non - increasing order . since @xmath333 , we obtain that @xmath334 , which gives @xmath335 . hence the two sequences of non - negative integers form two partitions , @xmath336 and @xmath337 , respectively , of @xmath338 . we fill the shorter partition with zeros so that both partitions have the same length , say @xmath339 . it is enough to show that @xmath340 for all @xmath341 , i.e. , that @xmath337 majorizes @xmath336 @xcite . we show first that @xmath342 . suppose the contrary , i.e. , @xmath343 . since @xmath344 for all @xmath345 , we obtain that @xmath346 for all @xmath345 . but @xmath347 , @xmath348 , was obtained by adding over @xmath50 a certain subset of the set @xmath349 . so there should be at least one @xmath350 that has @xmath351 in its composition , and hence @xmath352 . this is a contradiction , from which we obtain @xmath342 . we finish the proof by induction . namely , we want to show that from @xmath353 for some @xmath354 , it follows that @xmath355 . if @xmath356 then this induction step clearly holds . so , assume that @xmath357 . since @xmath358 , we can deduce that @xmath359 , and in fact all @xmath347 with @xmath360 , can not contain any @xmath361 with @xmath362 in its composition . hence all possible @xmath361 , @xmath363 , have occurred in the composition of @xmath347 for @xmath364 , which gives @xmath365 . this proves that @xmath337 majorizes @xmath336 and we obtain @xmath366 theorem [ th : qc : vs : conv : code : pseudo : weight:1 ] implies that low - pseudo - weight vectors in the block code may correspond to higher pseudo - weight vectors in the convolutional code , but the opposite is not possible . this suggests that the pseudo - codewords in the block code that result in decoding failures may not cause such failures in the convolutional code , thereby resulting in improved performance for the convolutional code at low - to - moderate signal - to - noise ratios ( snrs ) . a similar bound also holds for the max - fractional weight , as shown in the next theorem . [ th : qc : vs : conv : code : max : frac : weight:1 ] if @xmath253 , then @xmath367 therefore , @xmath368 we have @xmath333 and @xmath369 which leads to @xmath370 it now follows that @xmath371 in the case of the fractional weight , it is easy to see that for any @xmath372 , we have @xmath333 , and hence @xmath373 . when comparing the minimum fractional weight of the convolutional and qc codes , we encounter a computatinally harder case , as these values must be computed over the set of nonzero pseudo - codewords that are vertices of the fundamental polytope . this is not an easy task , because a vertex pseudo - codeword in the convolutional code might not map into a vertex pseudo - codeword in the qc code . the theorem below , however , can be established . for its better understanding , we recall that @xmath374 has the following meaning @xcite . let @xmath375 be the set of positions where bit flips occurred when using @xmath86 for data transmission over a bsc with crossover probability @xmath376 , @xmath377 . if @xmath378 , then lp decoding is correct . similarly , @xmath379 implies the following . if @xmath380 is the set of positions where bit flips occurred when using @xmath97 for data transmission over a bsc , then @xmath381 guarantees that lp decoding is correct . [ th : qc : vs : conv : code : frac : weight:1 ] assume that we are using @xmath97 for data transmission over a bsc with cross - over probability @xmath376 , where @xmath377 , and that bit flips occur at positions @xmath380 . if @xmath382 , then lp decoding is correct . ( note that on the right - hand side of the previous inequality we have @xmath374 and not @xmath379 . ) we know that @xmath383 where step @xmath384 follows from @xcite ( see also @xcite ) and step @xmath385 follows from theorem [ th : qc : vs : conv : code : max : frac : weight:1 ] . remember that @xmath386 has the following meaning @xcite . let @xmath380 be the set of positions where the bit flips occured when using @xmath97 for data transmission over a bsc . if @xmath387 , then lp decoding is correct . now , because the theorem statement assumes that @xmath388 , using we have @xmath389 and so , according to the meaning of @xmath390 , lp decoding is correct . .the pseudo - weights of the pseudo - codewords in example [ projecting : conv : pseudo - codewords ] . [ cols="^,^,^,^,^,^ " , ] studying pseudo - codewords of small pseudo - weight , and , in particular ( since the minimum pseudo - weight is upper bounded by the minimum hamming weight ) , studying pseudo - codewords of pseudo - weight smaller than the minimum hamming weight , represents an important problem in the performance analysis of ldpc codes because it allows us to identify potential failures in mpi decoding . upper and lower bounds on the minimum pseudo - weight of a convolutional code can be obtained by exploiting the `` sliding '' structure of its semi - infinite parity - check matrix @xmath106 and some of its sub - matrices , which allows relatively easy computations by taking advantage of the increased sparseness compared to the corresponding parity - check matrix of an underlying qc code . on the one hand , this technique allows us to find certain existing low - weight pseudo - codewords , and on the other , it illustrates the advantage of using a convolutional code structure over a block code structure in pseudo - codeword analysis . in addition , similar to the possible increase in minimum distance expected when going from a qc code to its unwrapped convolutional version , we expect an increase in the minimum pseudo - weight of the convolutional code , leading to better performance compared to the original qc code . our theoretical results and experimental observations point strongly in this direction . we now explain briefly our technique . similar to associating with a convolutional code @xcite an increasing sequence of _ column distances _ @xmath391 , and a decreasing sequence of _ row distances _ @xmath392 having the property that @xmath393 we define two sequences of pseudo - weights that prove helpful in identifying the overall minimum pseudo - weight . we recall that an encoder polynomial generator matrix @xmath394 of a rate @xmath395 convolutional code with encoder memory @xmath396 has associated with it a semi - infinite sliding generator matrix @xmath397 . let @xmath398 denote the @xmath399 submatrix of @xmath397 with rows indexed by the first @xmath167 block rows of @xmath397 and columns indexed by the first @xmath279 block columns of @xmath397 , and let @xmath400 the following two sequences of matrices @xmath401 and @xmath402 give us an increasing sequence of distances @xmath391 , @xmath403 , commonly called _ column distances _ , associated with the first sequence , and a decreasing sequence of distances @xmath392 @xmath404 , commonly called _ row distances _ , associated with the second sequence . the column distance @xmath405 is a `` truncation '' distance , i.e. , it measures the minimum of the hamming weights of the vectors of length @xmath406 that constitute the first @xmath406 components of some codeword with non - zero first component . the row distance @xmath407 is a `` bounded codeword '' distance , i.e. , it measures the minimum of the hamming weights of the codewords with non - zero first component and of length @xmath406 or smaller , or equivalently , of polynomial degree @xmath408 or smaller . the column distances and row distances represent valuable lower bounds and , respectively , upper bounds , on the free distance that become increasingly tight with increasing @xmath408 , and , in the limit , become equal to the free distance . if similar sequences could be defined for pseudo - weights , they would prove helpful in identifying the overall minimum pseudo - weight . with this in mind , we define corresponding sequences of `` truncated '' pseudo - weights and `` bounded pseudo - codeword '' pseudo - weights . let @xmath150 be a polynomial parity - check matrix for a convolutional code @xmath97 with syndrome former memory @xmath409 , and let @xmath105 be its semi - infinite sliding parity - check matrix . similar to the notation above , let @xmath410 be the @xmath411 submatrix of @xmath105 with rows indexed by the first @xmath167 block rows of @xmath105 and columns indexed by the first @xmath279 block rows of @xmath105 . we will consider two sequences of such sub - matrices : @xmath412 in which @xmath413 is the @xmath414 submatrix of the @xmath415 matrix @xmath416 formed by its first @xmath417 columns , for all @xmath418 , and @xmath419 in which @xmath420 is the @xmath421 submatrix of the @xmath422 matrix @xmath423 formed by its first @xmath424 rows , for all @xmath418 . in the first sequence , the first matrix that has with certainty a nonzero nullspace is @xmath425 , since there is a nonzero polynomial codeword of degree @xmath396 ( associated with a scalar codeword of length @xmath426 ) . since @xmath427^{\mathsf{t}}=0 $ ] , these matrices act like parity - check matrices in computing the row distances , with @xmath428 giving the @xmath408-th row distance , for all @xmath429 . there might be nonzero nullspaces earlier in the sequence , so we denote by @xmath430 the first matrix with a non - zero nullspace . obviously this would mean that @xmath431 is the first matrix in the second sequence @xmath432 , to have a nonzero nullspace . similarly , @xmath433 will act like parity - check matrices in computing the @xmath408th column distances . so by computing the nullspaces of these parity - check matrices we get upper and lower bounds on @xmath434 for the convolutional code that are similar to the column and row distances defined from the generator matrix . we remark also that if we wrap the convolutional code modulo @xmath79 , with @xmath435 , then the matrix @xmath430 is a submatrix of the parity - check matrix @xmath75 of the qc code that remains unchanged after the wrapping . hence a codeword of minimum weight for the matrix @xmath430 will be , if extended by zeros , a codeword in the qc code . if this codeword has weight equal to the minimum distance of the qc code , then the free distance of the convolutional code is equal to the weight of this codeword . the minimum distance of the qc code could be smaller , however , and in this case the free distance will only be upper bounded by the weight of this codeword and lower bounded by the minimum distance of the qc code . in what follows we will mimic the theory of row distances and column distances of a convolutional codes to bound the minimum pseudo - weight of the convolutional code . we obtain upper and lower bounds on the minimum pseudo - weight of the convolutional code as follows : @xmath436 as above , we will denote by @xmath437 and @xmath438 , respectively , the first matrices in the above sequences whose fundamental cones contain a non - zero vector . so by computing vectors in the fundamental cones of these parity - check matrices we get upper and lower bounds on @xmath439 for the convolutional code that are similar to the column and row distances defined from the generator matrix . the pseudo - codeword in example [ ex : smallexample:1 ] was obtained by attempting to compute small degree non - zero vectors in the fundamental cone of @xmath440 using the above technique . the first nonzero `` row pseudo - weight '' is @xmath441 , and the vector @xmath442 is in the fundamental cone of @xmath443 . its awgn pseudo - weight is @xmath444 , which is an upper bound on the minimum pseudo - weight of the convolutional code . the free distance of this code is @xmath445 . the reduced pseudo - codeword @xmath446 modulo @xmath447 , for @xmath448 has the same weight @xmath444 , larger than the minimum distance of the @xmath449 $ ] code , which makes this pseudo - codeword irrelevant , but smaller than the minimum distances of the codes @xmath450 $ ] , @xmath451 $ ] , @xmath452 $ ] . the upper bound @xmath453 in example [ smallexample2 ] therefore becomes @xmath454 based on this pseudo - codeword . this computational method has been applied successfully to larger codes as well . an example is the rate @xmath17 ldpc - cc with syndrome former memory @xmath19 that was simulated in fig . [ fig : compare:1 ] . the code was constructed by unwrapping a @xmath455 $ ] @xmath18-regular qc - ldpc block code with minimum hamming distance @xmath456 . the convolutional code has free distance @xmath457 , which already suggests a possible performance improvement compared to the qc code . following the approach described above , we constructed a class of pseudo - codewords for which we obtained a minimum pseudo - weight of @xmath458 . thus this class of pseudo - codewords contains vectors of weight less than the free distance , which makes them relevant to the performance analysis of iterative decoding . consequently , an upper bound on the minimum pseudo - weight of the convolutional code is @xmath458 , and , from the way we constructed this class of pseudo - codewords , we believe it is a very tight bound . projecting this pseudo - codeword onto the qc codes obtained by wrapping the convolutional code gives upper bounds on the minimum pseudo - weight of these codes as well ( in some cases tighter than the ones obtained using the methods of @xcite ) . the upper bound in @xcite for the minimum pseudo - weight of the @xmath20 $ ] code is @xmath459 . these upper bounds together with our simulation curves suggest that the minimum pseudo - weight of the convolutional code is strictly greater than the minimum pseudo - weight of the [ 155,64 ] qc code . an evaluation of the exact values of the minimum pseudo - weight in this cases is not possible , however , due to the large complexity of such a task . also note that if an upper bound on the minimum pseudo - weight of the convolutional code smaller than @xmath459 could be found , it would decrease the upper bound on the minimum pseudo - weight of the @xmath20 $ ] @xmath18-regular qc - ldpc block code as well . for an ldpc convolutional code derived by unwrapping an ldpc - qc block code , we have shown that the free pseudo - weight of the convolutional code is at least as large as the minimum pseudo - weight of the underlying qc code . this result suggests that the pseudo - weight spectrum of the convolutional code is `` thinner '' than that of the block code . this difference in the weight spectra leads to improved ber performance at low - to - moderate snrs for the convolutional code , a conclusion supported by the simulation results presented in figs . [ fig : compare:1 ] and [ fig : compare : small:1 ] . we also presented three methods of analysis for problematic pseudo - codewords , i.e. , pseudo - codewords with small pseudo - weight . the first method introduces two sequences of `` truncated '' pseudo - weights and , respectively , `` bounded pseudo - codeword '' pseudo - weights , which lower and upper bound the minimum pseudo - weight of the convolutional code , similar to the role that column distances and row distances play in bounding from below and above the free distance . the other two methods can be applied to any qc or convolutional code and consist of projecting codewords with small weight in qc codes or convolutional codes onto tanner graph covers of the code . r. smarandache , a. e. pusane , p. o. vontobel , and d. j. costello , jr . , `` pseudo - codewords in ldpc convolutional codes , '' in _ proc . ieee intern . symp . on inform . theory _ , seattle , wa , usa , july 914 2006 , pp . 13641368 . o. vontobel and r. koetter , `` graph - cover decoding and finite - length analysis of message - passing iterative decoding of ldpc codes , '' _ submitted to ieee trans . inform . theory , available online under _ ` http://www.arxiv.org/abs/cs.it/0512078`_ _ , dec . 2005 . g. d. forney , jr . , r. koetter , f. r. kschischang , and a. reznik , `` on the effective weights of pseudocodewords for codes defined on graphs with cycles , '' in _ codes , systems , and graphical models ( minneapolis , mn , 1999 ) _ , ser . i m a vol . b. marcus and j. rosenthal , eds.1em plus 0.5em minus 0.4emspringer verlag , new york , inc . , 2001 , vol . 101112 . r. m. tanner , d. sridhara , a. sridharan , t. e. fuja , and d. j. costello , jr . , `` ldpc block and convolutional codes based on circulant matrices , '' _ ieee trans . on inform . theory _ , vol . it50 , no . 29662984 , dec . 2004 . a. sridharan , d. j. costello , jr . , d. sridhara , t. e. fuja , and r. m. tanner , `` a construction for low density parity check convolutional codes based on quasi - cyclic block codes , '' in _ proc . ieee intern . symp . on inform . theory _ , lausanne , switzerland , june 30july 5 2002 , p. 481 . r. m. tanner , `` convolutional codes from quasi - cyclic codes : a link between the theories of block and convolutional codes , '' _ university of california , santa cruz , tech report ucsc - crl-87 - 21 _ , nov . 1987 . y. levy and d. j. costello , jr . , `` an algebraic approach to constructing convolutional codes from quasi - cyclic codes , '' in _ coding and quantization ( piscataway , nj , 1992 ) _ , ser . . discrete math . sci.1em plus 0.5em minus 0.4emprovidence , ri : amer . . , 1993 , vol . 14 , pp . 189198 . a. sridharan , m. lentmaier , k. s. zigangirov , and d. j. costello , jr . , `` convergence analysis of ldpc convolutional codes on the erasure channel , '' in _ proc . 42nd allerton conf . on communications , control , and computing _ , allerton house , monticello , illinois , usa , sep . 29oct . 1 2004 . o. vontobel and r. koetter , `` on the relationship between linear programming decoding and min - sum algorithm decoding , '' in _ proc . intern . symp . on inform . theory and its applications ( isita ) _ , parma , italy , oct . 1013 2004 , pp . 991996 . j. feldman , `` decoding error - correcting codes via linear programming , '' ph.d . dissertation , massachusetts institute of technology , cambridge , ma , 2003 , available online under ` http://www.columbia.edu/~jf2189/pubs.html ` . j. feldman , t. malkin , c. stein , r. a. servedio , and m. j. wainwright , `` lp decoding corrects a constant fraction of errors , '' in _ proc . ieee intern . symp . on inform . theory _ , chicago , il , usa , june 27july 2 2004 , p. 68 . d. avis , `` a revised implementation of the reverse search vertex enumeration algorithm '' , in _ polytopes combinatorics and computation _ , birkhuser - verlag , pp . 177198 , 2000 , programs are available online under _ ` http://cgm.cs.mcgill.ca/~avis/c/lrs.html ` _

we obtain some restrictions on the topology of infinite volume hyperbolic manifolds . in particular , for any and any closed negatively curved manifold of dimension , only finitely many hyperbolicmanifolds are total spaces of orientable vector bundles over .

math9807189

a _ hyperbolic manifold _ is , by definition , a quotient of a negatively curved symmetric space by a discrete isometry group that acts freely . recall that negatively curved symmetric spaces are hyperbolic spaces over the reals , complex numbers , quaternions , or cayley numbers . the homotopy type of a hyperbolic manifold is determined by its fundamental group . conversely , for finite volume hyperbolic manifolds , fundamental group determines the diffeomorphism type , thanks to the mostow rigidity theorem . by contrast infinite volume hyperbolic manifolds with isomorphic fundamental groups can be very different topologically . for example , the total spaces of many plane bundles over closed surfaces carry hyperbolic metrics ( this and other examples are discussed in the section [ s : examples ] ) . in this paper we attempt to count hyperbolic manifolds up to intersection preserving homotopy equivalence . a homotopy equivalence @xmath3 of oriented @xmath0manifolds is called _ intersection preserving _ if , for any @xmath4 and any pair of ( singular ) homology classes @xmath5 and @xmath6 , their intersection number in @xmath7 is equal to the intersection number of @xmath8 and @xmath9 in @xmath10 . for example , any map that is homotopic to an orientation - preserving homeomorphism is an intersection preserving homotopy equivalence . conversely , oriented rank two vector bundles over a closed oriented surface are isomorphic iff their total spaces are intersection preserving homotopy equivalent . [ intro : alg ] let @xmath11 be the fundamental group of a finite aspherical cell complex and let @xmath12 be a negatively curved symmetric space . let @xmath13 be a sequence of discrete injective representations of @xmath11 into the group of orientation - preserving isometries of @xmath12 . suppose that @xmath13 is precompact in the pointwise convergence topology . then the sequence of manifolds @xmath14 falls into finitely many intersection preserving homotopy equivalence classes . the space of conjugacy classes of faithful discrete representations of a group @xmath11 into the isometry group of a negatively curved symmetric space is compact provided @xmath11 is finitely presented , not virtually nilpotent and does not split over a virtually nilpotent group ( see [ compactness thm ] ) . since all the intersections in a hyperbolic manifold with virtually nilpotent fundamental group are zero ( see [ nilpotent groups ] ) , we get the following . [ no splitting implies finiteness of intersection classes ] let @xmath11 be the fundamental group of a finite aspherical cell complex that does not split as an hnn extension or a nontrivial amalgamated product over a virtually nilpotent group . then , for any @xmath0 , the class of orientable hyperbolic @xmath0manifolds with fundamental group isomorphic to @xmath11 breaks into finitely many intersection preserving homotopy equivalence classes . a particular case of [ no splitting implies finiteness of intersection classes ] was proved by kapovich in @xcite . namely , he gave a proof for _ real _ hyperbolic @xmath15manifolds homotopy equivalent to a closed orientable negatively curved manifold of dimension @xmath16 . if @xmath11 _ does _ split over a virtually nilpotent group , the space of representations is usually noncompact . for instance , this happens if @xmath11 is a surface group . yet , for real hyperbolic @xmath17manifolds homotopy equivalent to closed surfaces , kapovich @xcite proved a result similar to [ no splitting implies finiteness of intersection classes ] . more generally , kapovich @xcite proved that there is a universal function @xmath18 such that for any incompressible singular surfaces @xmath19 , @xmath20 in an oriented real hyperbolic @xmath17manifold we have @xmath21 , [ \sigma_{g_2}]\rangle surfaces @xmath19 , @xmath20 in a closed oriented negatively curved @xmath17manifold @xmath1 with the sectional curvature pinched between @xmath22 and @xmath23 there is a bound @xmath21 , [ \sigma_{g_2}]\rangle |\le c(g_1 , g_2 , k , k , \chi(m))$ ] for some universal function @xmath24 . another way to classify hyperbolic manifolds is up to tangential homotopy equivalence . recall that a homotopy equivalence of smooth manifolds @xmath3 is called tangential if the vector bundles @xmath25 and @xmath26 are stably isomorphic . for example , any map that is homotopic to a diffeomorphism is a tangential homotopy equivalence . conversely , a tangential homotopy equivalence of open @xmath0manifolds is homotopic to a diffeomorphism provided @xmath27 and each of the manifolds is the total space of a vector bundle over a manifold of dimension @xmath28 ( * ? ? ? * pp 226228 ) . an elementary argument ( based on finiteness of the number of connected components of representation varieties and on finiteness of the number of symmetric spaces of a given dimension ) yields the following . [ intr : tangent hom ] let @xmath11 be a finitely presented group . then , for any @xmath0 , the class of complete locally symmetric nonpositively curved riemannian @xmath0manifolds with the fundamental group isomorphic to @xmath11 falls into finitely many tangential homotopy types . knowing both the intersection preserving and tangential homotopy types sometimes suffices to recover the manifold up to finitely many possibilities . [ intr : vector ] let @xmath1 be a closed orientable negatively curved manifold of dimension @xmath2 and @xmath29 be an integer . let @xmath30 be a sequence of smooth embeddings of @xmath1 into orientable hyperbolic @xmath0manifolds such that @xmath31 induces monomorphisms of fundamental groups . then the set of the normal bundles @xmath32 of the embeddings breaks into finitely many isomorphism classes . in particular , only finitely many orientable hyperbolic @xmath0manifolds are total spaces of vector bundles over @xmath1 . in some cases it is easy to decide when there exist _ infinitely _ many rank @xmath33 vector bundles over a given base . for example , by a simple k theoretic argument the set of isomorphism classes of rank @xmath33 vector bundles over a finite cell complex @xmath34 is infinite provided @xmath35 and @xmath36 . furthermore , oriented rank two vector bundles over @xmath34 are in one - to - one correspondence with @xmath37 via the euler class . note that many arithmetic closed real hyperbolic manifolds have nonzero betti numbers in all dimensions @xcite . any closed complex hyperbolic manifold has nonzero even betti numbers because the powers of the khler form are noncohomologous to zero . similarly , for each @xmath4 , closed quaternion hyperbolic manifolds have nonzero @xmath38th betti numbers . examples of vector bundles with hyperbolic total spaces are given in the section [ s : examples ] . note that according to a result of anderson @xcite the total space of any vector bundle over a closed negatively curved manifold admits a complete metric with the sectional curvature pinched between two negative constants . i am grateful to bill goldman , misha kapovich , and jonathan rosenberg for many helpful discussions . i would like to thank heinz helling and sfb-343 at the university of bielefeld for support and hospitality . [ [ outline - of - the - paper . ] ] outline of the paper . + + + + + + + + + + + + + + + + + + + + + the section [ s : examples ] is a collection of examples of hyperbolic manifolds . some invariants of maps and representations are defined in sections [ s : invariants maps ] and [ s : invariants of representations ] . sections [ s : spaces repr ] , [ s : stable range ] , and [ s : loc symm ] are devoted to a proof of the theorem [ intr : tangent hom ] and other related results . sections [ s : geom conv ] , [ s : alg conv ] , and [ s : comp thm ] contain background on algebraic and geometric convergence needed for the main theorem which is proved in section [ s : main thm ] . theorems [ intro : alg ] and [ no splitting implies finiteness of intersection classes ] are proved in section [ s : intersection ] . finally , theorem [ intr : vector ] is proved in section [ s : vector ] . to help the reader appreciate the results stated above we collect some relevant examples of hyperbolic manifolds . * ( plane bundles over closed surfaces)total spaces of rank two vector bundles over closed surfaces often admit hyperbolic metrics . for instance , an orientable @xmath39bundle over a closed oriented surface of genus @xmath40 admits a real hyperbolic structure provided the euler number @xmath41 of the bundle satisfies @xmath42 @xcite ( cf @xcite , @xcite , @xcite ) . complex hyperbolic structures exist on orientable @xmath39bundles over closed oriented surfaces when @xmath43 @xcite . * note that the euler number is equal to the self - intersection number of the zero section , hence the total spaces of bundles with different euler classes are not intersection homotopy equivalent . for nonorientable bundles over nonorientable surfaces the condition @xmath44 $ ] on the twisted euler number implies the existence of a real hyperbolic structure @xcite . * ( plane bundles over closed hyperbolic @xmath45manifolds)total spaces of plane bundles over closed hyperbolic @xmath45manifolds sometimes carry hyperbolic metrics @xcite . * in fact , it can be deduced from @xcite that for every @xmath4 there exists a closed oriented real hyperbolic @xmath45manifold @xmath46 and oriented real hyperbolic @xmath47manifolds @xmath48 that are total spaces of plane bundles over @xmath1 and such that no two of them are intersection preserving homotopy equivalent . * ( fundamental group at infinity)there are real hyperbolic @xmath17manifolds that are intersection homotopy equivalent but not homeomorphic to plane bundles over closed surfaces @xcite . the invariant that distinguishes these manifolds from vector bundles is the fundamental group at infinity . * even more surprising examples were given in @xcite and @xcite . namely , there are orientable real hyperbolic @xmath17manifolds that are homotopy equivalent but not homeomorphic to handlebodies ; these manifolds have nontrivial fundamental group at infinity . note that if @xmath7 and @xmath10 are orientable @xmath17manifolds that are homotopy equivalent to a handlebody , then each homotopy equivalence @xmath49 is both tangential and intersection preserving . * ( nonorientable line bundles)here is a simple way to produce homotopy equivalent hyperbolic manifolds that are not tangentially homotopy equivalent . * first , note that via the inclusion @xmath50 , any discrete subgroup of @xmath51 can be thought of as a discrete subgroup of @xmath52 that stabilizes a subspace @xmath53 . the orthogonal projection @xmath54 is an @xmath51equivariant line bundle over @xmath55 . in particular , given a real hyperbolic manifold @xmath1 , the manifold @xmath56 carries a real hyperbolic metric . if @xmath57 , this construction can be twisted to produce nonorientable line bundles over @xmath58 with real hyperbolic metrics . indeed , a nonzero element @xmath59 defines an epimorphism @xmath60 . make @xmath61 act on @xmath62 as the reflection in @xmath55 . then let @xmath63 act on line bundle @xmath64 by @xmath65 . the quotient @xmath66 is the total space of a line bundle over @xmath1 with the first stiefel whitney class @xmath67 . in particular , line bundles with different first stiefel whitney classes have total space that are homotopy equivalent but _ not _ tangentially homotopy equivalent . ( tangential homotopy equivalences preserve stiefel whitney classes yet @xmath68 . ) thus , we get many tangentially homotopy inequivalent manifolds in a given homotopy type . let @xmath69 be a topological space and @xmath70 be a set . let @xmath71 be a map that , given a smooth manifold @xmath7 , and a continuous map from @xmath69 into @xmath7 , produces an element of @xmath70 . we call @xmath71 an _ invariant of maps of @xmath69 _ if the two following conditions hold : ( 1)homotopic maps @xmath72 and @xmath73 have the same invariant . ( 2)let @xmath74 be a diffeomorphism of @xmath7 onto an open subset of @xmath10 . then , for any continuous map @xmath75 , the maps @xmath75 and @xmath76 have the same invariant . there is a version of this definition for maps into oriented manifolds . namely , we require that the target manifold is oriented and the diffeomorphism @xmath77 preserves orientation . in that case we say that @xmath71 is an _ invariant of maps of @xmath69 into oriented manifolds_. * ( tangent bundle)[tangent bundle ] assume @xmath69 is paracompact and @xmath70 is the set of isomorphism classes of real vector bundles over @xmath69 . given a continuous map @xmath75 , set @xmath78 , the isomorphism class of the pullback of the tangent bundle to @xmath7 under @xmath79 . clearly , @xmath80 is an invariant . * * ( intersection number in oriented @xmath0manifolds ) [ intersection number in oriented @xmath0manifolds ] assume @xmath69 is compact . fix two cohomology classes @xmath81 and @xmath82 . ( in this paper we always use singular ( co)homology with integer coefficients unless stated otherwise . ) * let @xmath75 be a continuous map of a compact topological space @xmath69 into an oriented @xmath0manifold @xmath7 where @xmath83 . set @xmath84 to be the intersection number @xmath85 of @xmath8 and @xmath9 in @xmath7 . we next show that @xmath86 is an integer - valued invariant of maps into oriented manifolds . recall that the intersection number @xmath85 can be defined as follows . start with an arbitrary compact subset @xmath34 of @xmath7 that contains @xmath87 . let @xmath88 and @xmath89 be the poincar duals of @xmath90 and @xmath91 , respectively . then , set @xmath92\rangle$ ] where @xmath93 $ ] is the fundamental class of @xmath7 near @xmath34 ( * ? ? ? * vii.13.5 ) . note that @xmath94\rangle= \langle a , b\cap [ n , n\backslash k]\rangle=\langle a , f_*\beta\rangle.\ ] ] the following commutative diagram shows that @xmath85 is independent of @xmath34 . @xmath95 note that @xmath86 is an invariant . indeed , property @xmath96 holds trivially ; property @xmath97 is verified in ( * ? ? ? * viii.13.21(c ) ) . we say that an invariant of maps is _ liftable _ if in part @xmath97 of the definition the word `` diffeomorphism '' can be replaced by a `` covering map '' . for example , tangent bundle is a liftable invariant . intersection number is not liftable . the following proposition shows to what extent it can be repaired . [ liftable prop ] let @xmath98 be a covering map of manifolds and let @xmath69 be a finite connected cw complex . suppose that @xmath99 is a map such that @xmath100 is an embedding ( ie a homeomorphism onto its image ) . then @xmath101 for any invariant of maps @xmath71 . since @xmath102 is an embedding , so is @xmath79 . then , the map @xmath103 @xmath104 is a homeomorphism . using compactness of @xmath87 , one can find an open neighborhood @xmath105 of @xmath87 such that @xmath106 is a diffeomorphism . since invariants @xmath107 and @xmath108 can be computed in @xmath105 and @xmath109 , respectively , we conclude @xmath101 . assume @xmath12 is a smooth contractible manifold and let @xmath110 be the group of all self - diffeomorphisms of @xmath12 equipped with compact - open topology . let @xmath11 be the fundamental group of a finite - dimensional cw complex @xmath34 with universal cover @xmath111 . we refer to a group homomorphism @xmath112 as a _ representation_. to any representation @xmath112 , we associate a continuous @xmath113equivariant map @xmath114 as follows . consider the @xmath12bundle @xmath115 over @xmath34 where @xmath115 is the quotient of @xmath116 by the following action of @xmath11 @xmath117 since @xmath12 is contractible , the bundle has a section that is unique up to homotopy through sections . any section can be lifted to a @xmath113equivariant continuous map @xmath118 . projecting to @xmath12 , we get a @xmath113equivariant continuous map @xmath114 . note that any two @xmath113equivariant continuous maps @xmath119 , are @xmath113equivariantly homotopic . ( indeed , @xmath120 and @xmath121 descend to sections @xmath122 that must be homotopic . this homotopy lifts to a @xmath113equivariant homotopy of @xmath120 and @xmath121 . ) assume now that @xmath123 acts freely and properly discontinuously on @xmath12 . then the map @xmath120 descends to a continuous map @xmath124 . we say that @xmath113 is _ induced _ by @xmath79 . let @xmath71 be an invariant of continuous maps of @xmath34 . given a representation @xmath113 such that @xmath123 acts freely and properly discontinuously on @xmath12 , set @xmath125 to be @xmath107 where @xmath113 is induced by @xmath79 . we say @xmath113 is an _ invariant of representations of @xmath126_. similarly , any invariant @xmath71 of continuous maps of @xmath34 into oriented manifolds defines an invariant of representations into the group of orientation - preserving diffeomorphisms of @xmath12 . note that representations conjugate by a diffeomorphism @xmath127 of @xmath12 have the same invariants . ( indeed , if @xmath128 is a @xmath113equivariant map , the map @xmath129 is @xmath130equivariant . ) the same is true for invariants of orientation - preserving representations when @xmath127 is orientation - preserving . * ( tangent bundle ) [ invariant tau]let @xmath80 be the invariant of maps defined in [ tangent bundle ] . then , for any representation @xmath113 such that @xmath123 acts freely and properly discontinuously on @xmath12 , let @xmath131 be the pullback of the tangent bundle to @xmath132 via a map @xmath133 that induces @xmath113 . * in fact , @xmath80 can be defined for any representation as follows . look at the `` vertical '' bundle @xmath134 over @xmath115 where @xmath135 is the tangent bundle to @xmath12 . set @xmath131 to be the pullback of the vertical bundle via a section @xmath136 . thus , to every representation @xmath137 we associated a vector bundle @xmath131 of rank @xmath138 over @xmath34 . * ( intersection -0.7pt number for orientation -0.7pt preserving actions)assume the cell complex @xmath34 is finite and choose an orientation on @xmath12 ( which makes sense because , like any contractible manifold , @xmath12 is orientable ) . * fix two cohomology classes @xmath139 and @xmath140 . let @xmath86 is an invariant of maps defined in [ intersection number in oriented @xmath0manifolds ] where @xmath141 . let @xmath113 be a representation of @xmath11 into the group of orientation - preserving diffeomorphisms of @xmath12 such that @xmath123 acts freely and properly discontinuously on @xmath12 . then let @xmath142 be the intersection number @xmath85 of @xmath8 and @xmath9 in @xmath132 where @xmath133 is a map that induces @xmath113 . let @xmath12 be a smooth contractible manifold and let @xmath110 be the group of all self - diffeomorphisms of @xmath12 equipped with compact open topology . we equip the space of representations @xmath143 with the pointwise convergence topology , ie a sequence of representations @xmath13 converges to @xmath113 provided @xmath144 converges to @xmath145 for each @xmath146 . in the next two sections we explore the consequences of the following observation . [ first prop in spaces of representations and tangential homotopy equivalence ] let @xmath147 and @xmath148 be injective representations of @xmath11 into@xmath110 such that the groups @xmath149 and @xmath150 act freely and properly discontinuously on @xmath12 . suppose that @xmath147 and @xmath148 can be joined by a continuous path of representations @xmath151 ( where continuous means that , for every @xmath146 , the map @xmath152\to\mathrm{diffeo}(x)$ ] is continuous ) . then the homotopy equivalence of manifolds @xmath153 and @xmath154 induced by @xmath155 is tangential . since @xmath151 is a continuous path of representations , the covering homotopy theorem implies that the bundles @xmath156 and @xmath157 are isomorphic ( the invariant @xmath80 is defined in [ invariant tau ] ) . let @xmath158 and @xmath159 be homotopy equivalences that induce @xmath147 and @xmath148 , respectively . for @xmath160 the bundle @xmath161 is isomorphic to the pullback of the tangent bundle to @xmath162 via @xmath163 . thus @xmath164 is a tangential homotopy equivalence . in fact , the covering homotopy theorem implies that @xmath80 is constant on any path - connected component of the space @xmath165 . [ smooth conjugacy ] under the assumptions of [ first prop in spaces of representations and tangential homotopy equivalence ] , suppose that @xmath166 and that each of the manifolds @xmath153 and @xmath154 is homeomorphic to the total space of a vector bundle over a manifold of dimension @xmath167 . then @xmath147 and @xmath148 are smoothly conjugate on @xmath12 . according to ( * ? ? ? * pp 226228 ) , the homotopy equivalence @xmath164 is homotopic to a diffeomorphism . hence , @xmath147 and @xmath148 are smoothly conjugate on @xmath12 . more precisely , the result proved in ( * ? ? ? * pp 226228 ) is as follows . let @xmath168 be a tangential homotopy equivalence of smooth @xmath0manifolds with @xmath169 . if each of the manifolds is homeomorphic to the interior of a regular neighbourhood of a simplicial complex of dimension @xmath28 , then @xmath79 is homotopic to a diffeomorphism . suppose @xmath12 is a symmetric space of nonpositive sectional curvature . note that for any discrete torsion - free subgroup @xmath170 that stabilizes a totally geodesic submanifold @xmath171 , the exponential map identifies the quotient @xmath172 and the total space of normal bundle of @xmath173 in @xmath172 . applying [ first prop in spaces of representations and tangential homotopy equivalence ] , we deduce the following . let @xmath12 be a nonpositively curved symmetric space of dimension @xmath174 and let @xmath11 be a group . let @xmath148 and @xmath175 be injective discrete representations of @xmath11 into the isometry group of @xmath12 that lie in the same path - connected component of the space @xmath176 . suppose that each of the representation @xmath148 and @xmath175 stabilizes a totally geodesic subspace of dimension @xmath167 . then @xmath148 and @xmath175 are smoothly conjugate on @xmath12 . of course , the above argument works in other geometries as well . here is a sample result for complete affine manifolds . let @xmath147 and @xmath148 be injective representations of a group @xmath11 into @xmath177 that lie in the same path - connected component of the space of representations @xmath178 @xmath179 . assume that the groups @xmath149 and @xmath150 act freely and properly discontinuously on @xmath180 and , furthermore , suppose that @xmath149 and @xmath150 are contained in @xmath181 . since the coordinate projection @xmath182 is @xmath183equivariant , the manifolds @xmath184 and @xmath185 are the total spaces of vector bundles over the manifolds @xmath186 and @xmath187 , respectively . then [ smooth conjugacy ] implies that @xmath147 and @xmath148 are conjugate by a diffeomorphism of @xmath180 provided @xmath169 and @xmath188 . let @xmath189 be a subgroup of @xmath110 such that the space of representations @xmath190 has finitely many path - connected components . then [ first prop in spaces of representations and tangential homotopy equivalence ] implies that the class of manifolds of the form @xmath132 where @xmath113 is injective and @xmath123 acts freely and properly discontinuously on @xmath12 falls into finitely many tangential equivalence classes . for example , @xmath190 has finitely many path - connected components if @xmath11 is finitely presented and @xmath189 is either real algebraic , or complex algebraic , or semisimple with finite center @xcite . in particular , the following is true . [ thm on finitely many tan hom types ] let @xmath11 be a finitely presented torsion - free group and let @xmath12 be a nonpositively curved symmetric space . then the class of manifolds of the form @xmath132 , where @xmath191 is a faithful discrete representation , falls into finitely many tangential homotopy types . represent @xmath12 as a riemannian product @xmath192 where @xmath171 is a nonpositively curved symmetric space without euclidean factors . by de rham s theorem this decomposition is unique , so @xmath193 . the group @xmath194 is semisimple with trivial center , hence the analytic variety @xmath195 has finitely many path - connected components ( * ? ? ? * p 567 ) . the same is true for @xmath196 because @xmath197 is real algebraic ( * ? ? ? * p 567 ) . hence the analytic variety @xmath198 has finitely many connected components . let @xmath11 be a finitely presented torsion - free group . then , for any @xmath0 , the class of locally symmetric complete nonpositively curved @xmath0manifolds with the fundamental group isomorphic to @xmath11 falls into into finitely many tangential homotopy types . for any @xmath0 there exist only finitely many nonpositively curved symmetric spaces of dimension @xmath0 . hence [ thm on finitely many tan hom types ] applies . in this section we discuss some basic facts on geometric convergence . the notion of geometric convergence was introduced by chaubaty ( see @xcite ) . more details relevant to our exposition can be found in @xcite , @xcite and @xcite . in this section we let @xmath189 be a lie group equipped with some left invariant riemmanian metric . let @xmath199 be the set of all closed subgroups of @xmath189 . define a topology on @xmath199 as follows . we say that a sequence @xmath200 converges to @xmath201 geometrically if the following two conditions hold : ( 1)if @xmath202 converges to @xmath203 , then @xmath204 . ( 2)if @xmath204 , then there is a sequence @xmath205 with @xmath206 in @xmath189 . ( @xcite)the space @xmath199 is compact and metrizable . ( @xcite)@xmath207 iff for every compact subset @xmath208 the sequence @xmath209 in the hausdorff topology ( ie for any @xmath210 , there is @xmath7 such that , if @xmath211 , then @xmath212 lies in the @xmath213neihgborhood of @xmath214 and @xmath214 lies in the @xmath213neihgborhood of @xmath212 ) . [ lok : discrete groups are open ] ( @xcite)let @xmath215 is a discrete subgroup . then there is @xmath210 such that , for any @xmath207 in @xmath199 and any compact @xmath208 , there is @xmath7 such that , if @xmath211 and @xmath216 , then there is a unique @xmath205 that is @xmath213close to @xmath217 . in particular , @xmath218 is discrete for @xmath211 , since @xmath219 is the only element of @xmath218 , that is in the @xmath213neighbourhood of the identity . [ limit is torsion free ] let @xmath220 be a sequence of torsion - free groups converging to a discrete group @xmath221 in @xmath199 . then @xmath221 is torsion - free . indeed , choose @xmath222 with @xmath223 . find a sequence @xmath206 . then @xmath224 . by [ lok : discrete groups are open ] we have @xmath225 for large @xmath0 . since @xmath218 are torsion - free , @xmath226 as desired . [ lok : lipschitz convergence ] let @xmath12 be a simply connected homogeneous riemannian manifold and @xmath189 be a transitive group of isometries . let @xmath218 be a sequence of torsion - free subgroups of @xmath189 converging geometrically to a discrete subgroup @xmath221 . let @xmath227 be a relatively compact open set . then , if @xmath0 is large enough , there exists a relatively compact open set @xmath228 with @xmath229 and smooth embeddings @xmath230 such that @xmath231 descend to embeddings @xmath232 . @xmath233 the embeddings @xmath231 converge @xmath234uniformly to the inclusion . let @xmath235 , so @xmath34 is compact . by [ lok : discrete groups are open ] , if we take @xmath0 sufficiently large , then for any @xmath236 there exists a unique @xmath205 that is close to @xmath217 . it defines an injective map @xmath237 of @xmath214 into @xmath218 . our goal is to construct embeddings @xmath238 that are close to the inclusion and @xmath237equivariant ( in the sense that @xmath239 for @xmath222 ) . the construction is non - trivial and we refer to @xcite for a complete proof . [ i d component is compact ] let @xmath189 be a lie group and @xmath220 be a sequence of discrete groups that converges geometrically to @xmath240 . if the identity component @xmath241 of @xmath221 is compact , the sequence @xmath242 has unbounded torsion ( ie for any positive integer @xmath7 , there is @xmath243 of finite order which is greater than @xmath7 ) . choose @xmath244 so small that the closed @xmath245neighborhood @xmath105 of @xmath241 is disjoint from @xmath246 ( this is possible since @xmath241 is compact ) . given @xmath247 , for large @xmath0 , @xmath218 and @xmath248 are @xmath249hausdorff close on the compact set @xmath105 . take an arbitrary element @xmath250 in @xmath251neighborhood of the identity @xmath41 and choose @xmath252 so that it is @xmath249close to @xmath40 . first , show that @xmath253 has finite order . suppose not . since @xmath105 is compact and @xmath254 is discrete and infinite , there is a smallest @xmath255 with @xmath256 . so , @xmath257 . the metric is left invariant , hence , @xmath258 since @xmath218 and @xmath248 are @xmath259hausdorff close on @xmath105 and since @xmath260 , we get @xmath261 . this implies @xmath262 and , thus , @xmath263 , a contradiction . thus @xmath253 has finite order . we have just showed that any neighborhood of @xmath40 contains an element of finite order @xmath252 ( for @xmath0 large enough ) . so we get a sequence @xmath264 converging to @xmath40 . assume the torsion of @xmath265 is uniformly bounded by @xmath7 . then , for any finite order element @xmath266 , we have @xmath267 . in particular @xmath268 converges to @xmath269 . we , thus , proved that any element @xmath40 in @xmath241 that is @xmath251close to the identity satisfies @xmath270 . this is absurd . ( indeed , it would mean that the map @xmath271 that takes @xmath40 to @xmath269 maps the @xmath251neighborhood of the identity to the identity . but the map is clearly a diffeomorphism on some neighborhood of the identity ) . thus , @xmath265 has unbounded torsion as desired . [ i d component is nilpotent ] let @xmath189 be a lie group and let @xmath220 be a sequence of discrete groups that converges geometrically to @xmath240 . then the identity component @xmath241 of @xmath221 is nilpotent . denote the identity component by @xmath272 . the group @xmath221 is closed , so it is a lie group . so the identity component @xmath241 of @xmath221 is a lie group ; we are to show that @xmath273 is nilpotent . let u be a neighborhood of the identity that lies in a zassenhaus neighborhood in @xmath272 ( see @xcite ) . being a lie group @xmath241 is generated by any neighborhood of the identity , in particular by @xmath274 . to show @xmath241 is nilpotent , it suffices to check that , for some @xmath4 , @xmath275 , that is , any iterated commutator of weight @xmath4 with entries in @xmath228 is trivial @xcite . fix an iterated commutator @xmath276\dots]$ ] with @xmath277 and choose a sequences @xmath278 where @xmath279 . for @xmath0 large enough , the elements @xmath280 lie in a zassenhaus neighborhood . then by zassenhaus kazhdan margulis lemma @xcite the group @xmath281 lies in a connected nilpotent lie subgroup of @xmath272 . the class of any connected nilpotent lie subgroup of @xmath272 is bounded by @xmath282 . thus , for @xmath283 , the @xmath4iterated commutator @xmath284\dots]$ ] is trivial , for large @xmath0 . this implies @xmath276\dots]$ ] is trivial and , therefore , @xmath241 is nilpotent . the set of representations @xmath190 of a group @xmath11 into a topological group @xmath189 can be given the so - called _ algebraic _ topology ( also called pointwise convergence topology ) . namely , a sequence of representations @xmath13 converges to @xmath113 provided @xmath144 converges to @xmath123 for each @xmath146 . in this section we discuss how algebraic and geometric convergences interact . it follows from definitions that geometric limit always contains the algebraic one . more precisely , if @xmath13 converges to @xmath113 algebraically and @xmath285 converges to @xmath221 geometrically , then @xmath286 . in particular , if @xmath221 is discrete , so is @xmath123 . [ geometric limit discrete ] let @xmath11 be a finitely generated group and let @xmath12 be a negatively curved symmetric space . let @xmath287 be a sequence of representations converging algebraically to a representation @xmath113 where , for every @xmath4 , the group @xmath285 is discrete and the sequence @xmath288 has uniformly bounded torsion . suppose that @xmath123 is an infinite group without nontrivial normal nilpotent subgroups . then the closure of @xmath288 in the geometric topology consists of discrete groups . in particular , the group @xmath123 is discrete . passing to subsequence , we can assume that @xmath288 converges geometrically to @xmath221 . being a closed subgroup @xmath221 is a lie group . suppose , arguing by contradiction , that the identity component @xmath241 of @xmath221 is nontrivial . by lemma [ i d component is nilpotent ] , @xmath241 is a nilpotent lie group . since @xmath123 does not have a nontrivial normal nilpotent subgroup , @xmath289 , so @xmath123 is discrete . using the selberg lemma we find a torsion - free subgroup @xmath290 of finite index . the group @xmath291 is infinite since @xmath123 is . notice that the group @xmath291 may have at most two fixed points at infinity ( any isometry that fixes at least three points is elliptic ( * ? ? ? * p 84 ) ) . the next goal is to show that @xmath291 is virtually nilpotent . according to @xcite , the nilpotence of @xmath241 implies one the following mutually exclusive conclusions : _ case 1_@xmath241 has a fixed point in @xmath12 ( hence @xmath241 is compact which is is impossible by [ i d component is compact ] ) . _ case 2_@xmath241 fixes a point @xmath292 and acts freely on @xmath293 . since @xmath291 normalizes @xmath241 ( in fact , @xmath241 is a normal subgroup of @xmath221 ) it has to fix @xmath294 too . if @xmath294 is the only fixed point of @xmath291 , then every nontrivial element of @xmath291 is parabolic @xcite . any parabolic preserves horospheres centered at @xmath294 ( * ? ? ? * p 84 ) , therefore , according to @xcite @xmath291 is virtually nilpotent . if @xmath291 fixes two points at infinity , then it acts freely and properly discontinuously on the geodesic joining the points . hence @xmath295 , the fundamental group of a circle . _ case 3_@xmath241 has no fixed points in @xmath12 and preserves setwise some bi - infinite geodesic . in this case the fixed point set of @xmath241 is the endpoints @xmath294 and @xmath296 of the geodesic . indeed , @xmath241 fixes each of the endpoints , since @xmath241 is connected . assume @xmath241 fixes some other point of @xmath297 . then any element of @xmath241 is elliptic ( see ( * ? ? ? * p 85 ) ) and , as such , it fixes the bi - infinite geodesic pointwise . thus @xmath241 has a fixed point in @xmath12 which contradicts the assumptions of case 3 . since @xmath291 normalizes @xmath241 , it leaves the set @xmath298 invariant . moreover , @xmath291 preserves @xmath298 pointwise because @xmath291 contains no elliptics ( any isometry that flips @xmath294 and @xmath296 has a fixed point on the geodesic joining @xmath294 and @xmath296 ) . therefore , @xmath295 as before . hence , @xmath291 is virtually nilpotent . thus , @xmath123 has a nilpotent subgroup of finite index and , therefore , has a normal nilpotent subgroup of finite index . this contradicts the assumption that @xmath123 is an infinite group without nontrivial normal nilpotent subgroups . [ remark on no normal nilpotent subgroup ] let @xmath248 be a _ torsion - free discrete _ subgroup of @xmath299 that has a nontrivial normal nilpotent subgroup . then @xmath248 is , in fact , virtually nilpotent . indeed , repeating the arguments above , we see that @xmath248 fixes a point at infinity and , hence , is virtually nilpotent . conversely , any virtually nilpotent group clearly has a normal nilpotent subgroup of finite index . [ algebraic limit is faithful ] let @xmath189 be a lie group and let @xmath11 be a finitely generated group without nontrivial normal nilpotent subgroups . let @xmath287 be a sequence of discrete faithful representations that converges algebraically to a representation @xmath113 . then @xmath113 is faithful . denote the identity component of @xmath189 by @xmath272 . it suffices to prove that the group @xmath300 is nilpotent . let @xmath228 be a set of generators for @xmath34 ( maybe infinite ) . to show @xmath34 is nilpotent , it suffices to check that , for some @xmath33 , @xmath301 , that is , any iterated commutator of weight @xmath33 with entries in @xmath228 is trivial @xcite . fix an iterated commutator @xmath302\dots]$ ] with @xmath303 . for @xmath4 large enough , the elements @xmath304 lie in a zassenhaus neighborhood of @xmath272 . then by zassenhaus kazhdan margulis lemma @xcite the group @xmath305 lies in a connected nilpotent lie subgroup of @xmath272 . the class of any nilpotent subgroup of @xmath272 is bounded by @xmath282 . thus , for @xmath306 , the @xmath33iterated commutator @xmath307\dots]\ ] ] is trivial , for large @xmath4 . since @xmath13 is faithful @xmath302\dots]$ ] is trivial and we are done . [ injective , discrete is closed ] let @xmath12 be a negatively curved symmetric space and let @xmath11 be a finitely generated , torsion - free , discrete subgroup of @xmath299 that is not virtually nilpotent . then the set of faithful discrete representations of @xmath11 into @xmath299 is a closed subset of @xmath176 . let a sequence @xmath13 of faithful discrete representations converge to @xmath308 . according to [ remark on no normal nilpotent subgroup ] the group @xmath11 has no normal nilpotent subgroup . then [ algebraic limit is faithful ] implies @xmath113 is faithful . by compactness @xmath285 has a subsequence that converges geometrically to @xmath221 which is a discrete group by [ geometric limit discrete ] . therefore , @xmath309 is also discrete as wanted . [ geom limit of discrete injective is discrete ] let @xmath12 be a negatively curved symmetric space and let @xmath11 be a finitely generated , torsion - free , discrete subgroup of @xmath299 that is not virtually nilpotent . suppose @xmath13 is a sequence of injective , discrete representations of @xmath11 into @xmath299 that converges algebraically . then the closure of @xmath288 in the geometric topology consists of discrete groups . combine [ geometric limit discrete ] , [ remark on no normal nilpotent subgroup ] , and [ injective , discrete is closed ] . in this section we state a compactness theorem for the space of conjugacy classes of faithful discrete representations of a group @xmath11 into the isometry group of a negatively curved symmetric space . the proof follows from the work of bestvina and feighn @xcite based on ideas of rips ( see the review of @xcite by paulin in mr@xmath310 ) . earlier versions of the theorem have been proved by thurston , morgan and shalen @xcite . let @xmath12 be a negatively curved symmetric space and @xmath299 be the isometry group of @xmath12 . we equip @xmath176 with the algebraic topology . [ compactness thm ] let @xmath12 be a negatively curved symmetric space and let @xmath11 be a discrete , finitely presented subgroup of the isometry group of @xmath12 . suppose that @xmath11 is not virtually nilpotent and does not split as an hnn extension or a nontrivial amalgamated product over a virtually nilpotent group . then the space of conjugacy classes of faithful discrete representations of @xmath11 into @xmath299 is compact . here is an example of a group that does not split over a virtually nilpotent group . [ aspherical do not split ] let @xmath1 be a closed aspherical @xmath0manifold such that any nilpotent subgroup of @xmath63 has cohomological dimension @xmath311 . then @xmath63 does not split as a nontrivial amalgamated product or hnn extension over a virtually nilpotent group . assume , by contradiction , that @xmath312 is of the form @xmath313 or @xmath314 where @xmath7 is a virtually nilpotent group and @xmath315 is a proper subgroup of @xmath11 , for @xmath316 . first , suppose that both @xmath317 and @xmath318 have infinite index in @xmath11 . note that by the definition of hnn extension , @xmath241 has infinite index in @xmath11 . then it follows from the mayer vietoris sequence ( * ? ? ? * vii.9.1 , viii.2.2.4(c ) ) that @xmath319 the cohomological dimension of @xmath312 is @xmath0 because @xmath1 is a closed aspherical @xmath0manifold . since @xmath320 , @xmath315 is the fundamental group of a noncompact manifold of dimension @xmath0 , hence @xmath321 . finally , @xmath322 and we get a contradiction . second , assume that @xmath323 and , say , @xmath317 has finite index in @xmath11 . look at the map @xmath324 induced by the inclusion @xmath325 in the homology with @xmath326twisted integer coefficients . it is proved in ( * ? ? ? * iii.9.5(b ) ) that the image of @xmath327 is generated by @xmath328 $ ] where @xmath329 $ ] is the fundamental class of @xmath1 . since @xmath317 has finite index in @xmath11 , the group @xmath7 is of finite index in @xmath318 because @xmath330 . in particular , @xmath331 ( * ? ? ? * viii.3.1 ) . look at the @xmath0th term of the mayer vietoris sequence with @xmath326twisted integer coefficients ( * ? ? ? * vii.9.1 ) : @xmath332 using the cohomological dimension assumption on @xmath7 , we get @xmath333 , and hence the inclusion @xmath325 induces an isomorphism of @xmath0th homology . thus @xmath334 which contradicts the assumption the @xmath317 is a proper subgroup of @xmath11 . [ word - hyperbolic dimension at least three ] let @xmath1 be a closed aspherical manifold of dimension @xmath2 with word - hyperbolic fundamental group . then @xmath63 does not split as a nontrivial amalgamated product or hnn extension over a virtually nilpotent group . it is well known that any virtually nilpotent subgroup of a word - hyperbolic group is virtually cyclic . since any virtually cyclic group has cohomological dimension one , proposition [ aspherical do not split ] applies . throughout this section @xmath71 is an invariant of continuous maps of @xmath34 into oriented manifolds . we also use the letter @xmath71 for the corresponding invariant of discrete representations . let @xmath11 be the fundamental group of a finite cw complex @xmath34 . [ main theorem ] let @xmath12 be a contractible homogeneous riemannian manifold and @xmath189 be a transitive group of orientation - preserving isometries . let @xmath335 be a sequence of representations such that * the groups @xmath285 are torsion - free , and * @xmath13 converges algebraically to @xmath113 , and * @xmath285 converges geometrically to a discrete group @xmath221 . let @xmath336 be a continuous map that induces the homomorphism @xmath337 . then @xmath338 for all large @xmath4 . according to [ lok : discrete groups are open ] , we can assume that the groups @xmath285 are discrete . being a limit of torsion - free groups the discrete group @xmath221 is itself torsion - free by [ limit is torsion free ] . thus @xmath221 acts freely and properly discontinuously on @xmath12 , so @xmath339 is a manifold . we consider the universal covering @xmath340 . since @xmath34 is a finite complex , one can choose a finite connected subcomplex @xmath341 with @xmath342 . ( pick a representing cell in every orbit , the union of the cells is a finite subcomplex that is mapped onto @xmath171 by @xmath296 . adding finitely many cells , one can assume the complex is connected . ) according to section [ s : invariants of representations ] , the representation @xmath113 defines a continuous @xmath113equivariant map @xmath343 which is unique up to @xmath113equivariant homotopy . the map @xmath120 descends to a continuous map @xmath344 . the identity map @xmath345 is equivariant with respect to the inclusion @xmath346 , therefore , @xmath120 is equivariant with respect to the homomorphism @xmath347 . we denote by @xmath79 the composition of @xmath348 and the covering @xmath349 induced by the inclusion @xmath350 . let @xmath227 be an open relatively compact neighborhood of @xmath351 . we are in position to apply theorem [ lok : lipschitz convergence ] . thus , if @xmath4 is large enough , there is a sequence of embeddings @xmath352 that converges to the inclusion and is @xmath353equivariant ( recall that @xmath353 was defined in the proof of [ lok : lipschitz convergence ] ) . so the map @xmath354 is @xmath355equivariant . by the very definition of @xmath353 we have @xmath356 whenever the left hand side is defined . we now extend the map @xmath354 by equivariance to a @xmath13equivariant map @xmath357 . the map @xmath358 descends to @xmath359 . notice that by construction @xmath360 . since @xmath361 converge to the inclusion , @xmath361 is orientation - preserving for large @xmath4 . thus @xmath362 for large @xmath4 because @xmath71 is an invariant of maps into oriented manifolds . notice that @xmath363 since the map @xmath364 is @xmath13equivariant ( according to section [ s : invariants of representations ] any @xmath13equivariant map can be used to define @xmath365 ) . therefore , for large @xmath4 , @xmath366 as desired . let @xmath12 be a negatively curved symmetric space . then , according to [ geometric limit discrete ] , the assumption `` @xmath221 is discrete '' of the theorem [ main theorem ] can be replaced by `` @xmath285 are discrete and @xmath123 is an infinite group without nontrivial normal nilpotent subgroups '' . in some cases the conclusion of the theorem [ main theorem ] can be improved to `` @xmath367 for all large @xmath4 '' . since @xmath368 , it suffices to show that @xmath369 . this is clearly true when @xmath370 in which case it is usually said that @xmath13 converges to @xmath113 strongly . another obvious example is when @xmath71 is a liftable invariant . finally , according to [ liftable prop ] , @xmath369 provided @xmath79 is homotopic to an embedding ( ie a homeomorphism onto its image ) . [ cor of main thm ] let @xmath12 be a negatively curved symmetric space and let @xmath34 be a finite cell complex such that @xmath371 is a torsion - free , not virtually nilpotent , discrete subgroup of @xmath299 . let @xmath13 be a sequence of discrete injective representations of @xmath11 into the group of orientation - preserving isometries of @xmath12 . suppose that @xmath13 converges in the pointwise convergence topology . then @xmath372 for all large @xmath4 . the result follows from [ main theorem ] and [ geom limit of discrete injective is discrete ] . the results of this section certainly hold if @xmath71 is any invariant of maps rather that an invariant of maps into _ oriented _ manifolds . for such a @xmath71 we do not have to assume that the isometry groups preserve orientation . a homotopy equivalence @xmath3 of oriented @xmath0manifolds is called _ intersection preserving _ if , for any @xmath33 and any pair of ( singular ) homology classes @xmath373 and @xmath374 , their intersection number in @xmath7 is equal to the intersection number of @xmath8 and @xmath9 in @xmath10 . for example , any map that is homotopic to an orientation - preserving homeomorphism is an intersection preserving homotopy equivalence @xcite . [ nilpotent groups ] let @xmath7 be an oriented hyperbolic manifold with virtually nilpotent fundamental group . then the intersection number of any two homology classes in @xmath7 is zero . it suffices to prove that @xmath7 is homeomorphic to @xmath375 for some space @xmath171 . first note that any torsion - free , discrete , virtually nilpotent group @xmath248 acting on a hyperbolic space @xmath12 must have either one or two fixed points at infinity ( see the proof of [ geometric limit discrete ] ) . if @xmath248 has only one fixed point , @xmath248 is parabolic and , hence , it preserves all horospheres at the fixed point . therefore , if @xmath376 is such a horosphere , @xmath172 is homeomorphic to @xmath377 . if @xmath248 has two fixed points , @xmath248 preserves a bi - infinite geodesic . hence @xmath172 is the total space of a vector bundle over a circle and the result easily follows . [ alg precompact implies finiteness of intersections ] let @xmath11 be the fundamental group of a finite aspherical cell complex and let @xmath12 be a negatively curved symmetric space . let @xmath13 be a sequence of discrete injective representations of @xmath11 into the group of orientation - preserving isometries of @xmath12 . suppose that @xmath13 is precompact in the pointwise convergence topology . then the sequence of manifolds @xmath14 falls into finitely many intersection preserving homotopy equivalence classes . according to [ nilpotent groups ] , we can assume that @xmath126 is not virtually nilpotent . argue by contradiction . pass to a subsequence so that no two of the manifolds @xmath14 are intersection preserving homotopy equivalent and so that @xmath13 converges algebraically . being the fundamental group of a finite aspherical cell complex , @xmath11 has finitely generated homology . choose a finite set of generators of @xmath378 . using [ cor of main thm ] we pass to a subsequence so that @xmath379 for any pair of generators @xmath373 and @xmath374 . hence , the homotopy equivalence that induces @xmath380 is intersection preserving . [ bound on intersections ] let @xmath34 be a finite connected cell complex such that @xmath126 does not split as an hnn extension or a nontrivial amalgamated product over a virtually nilpotent group . then , given a nonnegative integer @xmath0 and homology classes @xmath381 and @xmath140 , there exists @xmath382 such that , ( 1)for any continuous map @xmath383 of @xmath34 into an oriented hyperbolic @xmath0manifold @xmath7 that induces an isomorphism of fundamental groups , and ( 2)for any embedding @xmath383 of @xmath34 into an oriented hyperbolic @xmath0manifold @xmath7 that induces a monomorphism of fundamental groups , the intersection number of @xmath8 and @xmath9 in @xmath7 satisfies @xmath384 . first , we prove @xmath385 . arguing by contradiction , consider a sequence of maps @xmath386 that induce @xmath387isomorphisms of @xmath34 and oriented hyperbolic @xmath0manifolds @xmath388 and such that @xmath389 is an unbounded sequence of integers . pass to a subsequence so that no two integers @xmath389 are the same . each map @xmath31 induces an injective discrete representation @xmath13 of @xmath126 into the group of orientation - preserving isometries of a negatively curved symmetric space ; in particular @xmath390 . since in every dimension there are no more than four negatively curved symmetric spaces , we can pass to a subsequence to ensure that all the symmetric spaces where @xmath391 acts are isometric . according to [ nilpotent groups ] , we can assume that @xmath126 is not virtually nilpotent . by [ compactness thm ] there exists a sequence of orientation - preserving isometries @xmath392 of @xmath12 such that the sequence @xmath393 is algebraically precompact . as we explained in the section [ s : invariants of representations ] , @xmath394 . thus [ cor of main thm ] provides a contradiction . finally , according to [ liftable prop ] , @xmath385 implies @xmath395 . let @xmath11 be the fundamental group of a finite aspherical cell complex that does not split as an hnn extension or a nontrivial amalgamated product over a virtually nilpotent group . then , for any @xmath0 , the class of hyperbolic @xmath0manifolds with fundamental group isomorphic to @xmath11 breaks into finitely many intersection preserving homotopy equivalence classes . by [ nilpotent groups ] , we can assume that @xmath126 is not virtually nilpotent . thus the result follows from [ alg precompact implies finiteness of intersections ] and [ compactness thm ] . i am most grateful to jonathan rosenberg for explaining to me the following nice fact . [ p : classifying vector bundles ] let @xmath34 be a finite cw complex and @xmath33 be a positive integer . then the set of isomorphism classes of oriented real ( complex , respectively ) rank @xmath33 vector bundles over @xmath34 with the same rational pontrjagin classes and the rational euler class ( rational chern classes , respectively ) is finite . to simplify notation we only give a proof for complex vector bundles and then indicate necessary modifications for real vector bundles . we need to show that the `` chern classes map '' @xmath396\to h^*(k,\bbb q)$ ] is finite - to - one . first , notice that @xmath397 classifies line bundles ( * ? ? ? * i.3.8 , i.4.3.1 ) , so we can assume @xmath398 . the integral @xmath399th chern class @xmath400 $ ] can be represented by a continuous map @xmath401 such that @xmath402 where @xmath403 is the fundamental class of @xmath404 . ( recall that , by the hurewicz theorem @xmath405 and @xmath406 ; the class in @xmath407 corresponding to the identity homomorphism is called the fundamental class and is denoted @xmath408 . ) it defines `` chern classes map '' @xmath409 we now check that the map induces an isomorphism on rational cohomology . according to @xcite for even @xmath0 , @xmath410 $ ] . by the knneth formula @xmath411\cong \bbb q[\alpha_2,\dots \alpha_{2m}]\ ] ] and under this ring isomorphism @xmath412 corresponds to @xmath413 . it is well known that @xmath414 $ ] @xcite . thus the homomorphism @xmath415 defines a homomorphism @xmath416 \cong \bbb q[c_1,\dots , c_m]$ ] that takes @xmath403 to @xmath417 . since this is an isomorphism , so is @xmath418 as promised . therefore @xmath419 induces an isomorphism on rational homology . then , since @xmath420 is simply connected , the map @xmath419 must be a rational homotopy equivalence @xcite . in other words the homotopy theoretic fiber @xmath421 of the map @xmath419 has finite homotopy groups . consider an oriented rank @xmath33 vector bundle @xmath422 with characteristic classes @xmath423 . our goal is to show that the map @xmath424 has at most finitely many nonhomotopic liftings to @xmath420 . look at the set of liftings of @xmath424 to @xmath420 and try to construct homotopies skeleton by skeleton using the obstruction theory . the obstructions lie in the groups of cellular cochains of @xmath34 with coefficients in the homotopy groups of @xmath421 . ( note that the fibration @xmath425 has simply connected base and fiber ( since @xmath398 ) , so the coefficients are not twisted . ) since the homotopy groups of @xmath421 are finite , there are at most finitely many nonhomotopic liftings . this completes the proof for complex vector bundles . for oriented real vector bundles of odd rank the same argument works with @xmath426})$ ] , where @xmath427 is the @xmath399th pontrjagin class . similarly , for oriented real vector bundles of even rank we set @xmath428 , where @xmath41 is the euler class . ( the case @xmath429 can be treated separately : since @xmath430 any oriented rank two vector bundle has a structure of a complex line bundle with @xmath41 corresponding to @xmath397 . thus , according to ( * ? ? ? * i.3.8 , i.4.3.1 ) , oriented rank two vector bundles are in one - to - one correspondence with @xmath37 . ) a similar result is probably true for nonorientable bundles . however , the argument given above fails due to the fact @xmath431 is not simply connected ( ie the map @xmath426})$ ] of @xmath431 to the product of eilenberg maclane spaces is _ not _ a rational homotopy equivalence even though it induces an isomorphism on rational cohomology ) . for simplicity , we only deal with oriented bundles . [ pontrjagin classifies ] note that , if either @xmath33 is odd or @xmath432 , the rational euler class is zero , and , hence rational pontrjagin classes determine an oriented vector bundle up to a finite number of possibilities . [ normal bundles of immersions ] let @xmath1 be a closed smooth manifold and let @xmath33 be an integer such that either @xmath33 is odd or @xmath433 . let @xmath434 be a sequence of smooth immersions of @xmath69 into complete locally symmetric nonpositively curved riemannian manifolds @xmath388 with orientable normal bundle and @xmath435 . according to section [ s : invariants of representations ] , the immersions @xmath31 induce representations @xmath437 such that @xmath438 . the sequence @xmath439 of vector bundles breaks into finitely many isomorphism classes because the representation variety @xmath440 has finitely many connected components ( see [ thm on finitely many tan hom types ] ) . in particular , there are only finitely many possibilities for the total pontrjagin class of @xmath439 . the normal bundle of the immersion @xmath31 satisfies @xmath441 . applying the total pontrjagin class , we get @xmath442 . the total pontrjagin class of any bundle is a unit , hence we can solve for @xmath443 . thus , there are only finitely many possibilities for @xmath443 . finally , [ p : classifying vector bundles ] and [ pontrjagin classifies ] imply that there are only finitely many possibilities for @xmath32 . let @xmath1 be a closed negatively curved manifold of dimension @xmath2 and let @xmath29 be an integer . let @xmath434 be a sequence of smooth embeddings of @xmath1 into hyperbolic @xmath0manifolds such that for each @xmath4 then the set of the normal bundles @xmath436 falls into finitely many isomorphism classes . in particular , up to diffeomorphism , only finitely many hyperbolic @xmath0manifolds are total spaces of orientable vector bundles over @xmath1 . arguing as in the proof of [ normal bundles of immersions ] , we deduce that there are only finitely many possibilities for the total pontrjagin classes of @xmath446 . thus , according to [ p : classifying vector bundles ] , we can pass to a subsequence so that the ( rational ) euler classes of @xmath446 are all different . denote the integral euler class by @xmath447 . first , assume that @xmath1 is orientable . recall that , by definition , the euler class @xmath447 is the image of the thom class @xmath448 under the map @xmath449 . according to ( * ? ? ? * viii.11.18 ) the thom class has the property @xmath450=f_{k*}[m]$ ] where @xmath451 $ ] is the fundamental class of the pair @xmath452 and @xmath329 $ ] is the fundamental class of @xmath1 . therefore , for any @xmath453 , the intersection number of @xmath454 and @xmath455 $ ] in @xmath388 satisfies @xmath456 , f_{k*}\alpha)=\langle\tau(\nu_k ) , f_{k*}\alpha\rangle= \langle f^*\tau(\nu_k ) , \alpha\rangle=\langle e(\nu_k),\alpha\rangle.\ ] ] since @xmath1 is compact , @xmath457 is finitely generated ; we fix a finite set of generators . the ( rational ) euler classes are all different , hence the homomorphisms @xmath458 are all different . then there exists a generator @xmath453 such that @xmath459 is an infinite set of integers . hence @xmath460 , f_{k*}\alpha)\}$ ] is an infinite set of integers . combining [ bound on intersections ] and [ word - hyperbolic dimension at least three ] , we get a contradiction . assume now that @xmath1 is nonorientable . let @xmath461 be the orientable two - fold cover . any finite cover of aspherical manifolds induces an injection on rational cohomology ( * ? ? ? * iii.9.5(b ) ) . hence @xmath462 implies that the rational euler classes of the pullback bundles @xmath463 are all different , and there are only finitely many possibilities for the total pontrjagin classes of @xmath463 . furthermore , the bundle map @xmath464 induces a smooth two - fold cover of the total spaces , thus the total space of @xmath463 is hyperbolic . finally , @xmath465 is a closed orientable negatively curved manifold . thus , we get a contradiction as in the oriented case .

modern technology often generates data with complex data structures in which both response and explanatory variables are matrix - valued . existing methods in the literature are able to tackle matrix - valued predictors but are rather limited for matrix - valued responses . in this article , we study matrix - variate regressions for such data , where the response on each experimental unit is a random matrix and the predictor can be either a scalar , a vector , or a matrix , treated as non - stochastic in terms of the conditional distribution . we propose models for matrix - variate regressions and then develop envelope extensions of these models . under the envelope framework , redundant variation can be eliminated in estimation and the number of parameters can be notably reduced when the matrix - variate dimension is large , possibly resulting in significant gains in efficiency . # 1 * key words : * matrix - valued response ; matrix - variate regression ; reducing subspace ; sufficient dimension reduction .

1605.01485

data with a matrix - valued response for each experimental unit are commonly encountered in contemporary statistical applications . for example , a longitudinal multivariate response can be treated integrally as a matrix - valued variable by designating rows and columns to be time and variates . temporal and spatial data , multivariate growth curve data , image data and data from cross - over designs also generate matrix - valued responses . for example , in a twin cross - over bioassay of insulin by the rabbit blood sugar method ( v@xmath3lund , 1980 ) , each rabbit received two different treatments on two days . blood sugar was measured at hourly intervals for six hours each day . in this case , the response for each rabbit is a @xmath4 matrix , with rows and columns indicating treatments and time respectively . the eeg data ( li et al . 2010 ) is another example that contains temporal and spatial matrix - formed variates from 77 alcoholic subjects and 45 non - alcoholic subjects . the electrical records of each subject form a matrix of dimensions @xmath5 that can be treated as a matrix - valued response variable when we investigate the association between the brain signals and alcoholism . in these examples , the components of the matrix - variates are dependent among rows and columns . this dependence distinguishes longitudinal data in a matrix - valued response from traditional longitudinal modeling of vector - valued responses in which independent units are each measured over time . vectorizing a matrix - valued response , or modeling the row or column vectors separately , typically loses dependency information and fails to capture the data structure . tackling matrix - variates directly can circumvent this issue . research into this topic has gained considerable interest in recent years . ( 2010 ) proposed a class of sufficient dimension reduction ( sdr ) methods , called dimension folding sdr , for data with matrix - valued predictors . pfeiffer et al . ( 2012 ) extended sliced inverse regression ( sir ) to longitudinal predictors . ding and cook ( 2014 ) developed model - based dimension folding methods for matrix - valued predictors . xue and yin ( 2014 ) introduced dimension folding sdr for conditional mean functions . on another track , hung and wang ( 2013 ) , zhou et al . ( 2013 ) , and zhou and li ( 2014 ) extended generalized linear models ( glm ) to matrix- and tensor - valued predictors for analyzing image data . all these methods , however , address data with matrix or tensor - valued predictors . methods for dealing directly with matrix - valued responses are relatively limited in the literature . viroli ( 2012 ) proposed special matrix - variate regressions assuming independent rows for error terms or covariates . yet in many applications such assumptions are violated as dependency can exist among both rows and columns of a matrix - valued response . we recently became aware that li and zhang ( 2015 ) studied tensor response regression where a linear relationship between a tensor - valued response and a predictor vector is considered . their method is more suited for tensor responses and vector - valued predictors . however , our motivation , rationale , model formulation , and estimation algorithms are different from those of li and zhang ( 2015 ) . firstly , our work was motivated by contemporary matrix - variate data and was geared to develop demanding yet limited matrix - variate analysis tools for such data . thus , it is tailored for matrix - valued responses and can handle matrix - valued predictors simultaneously . secondly , our method can fully utilize intrinsic matrix - variate data structure and allow a matrix linear model to reduce number of parameters in matrix - variate regressions . moreover , our method of estimation is different than that used by li and zhang ( 2015 ) . as discussed in section [ sec:7.2 ] this can have important consequences in applications . in this article , we propose matrix - variate regressions in a general framework , where the response @xmath0 on each experimental unit is a random matrix and the predictor @xmath1 can be either a scalar , a vector , or a matrix , treated as non - stochastic in terms of the conditional distribution @xmath2 . neither the rows nor the columns of the matrix - valued variables are required to be independent . thus , the proposed methods can capture intrinsic dependent data structures . moreover , they can reduce the number of parameters and improve efficiency in estimation compared to conventional methods . as matrix - variate responses may have sizable dimensions , extraneous error variation may often occur beyond that anticipated by the model . to allow for such situations , we further propose envelope methods for efficient estimation in matrix - variate regressions . by applying the idea of enveloping , one can achieve dimension reduction in the analysis , by extracting only the useful information and eliminating redundant information from estimation . this can lead to substantial efficiency gains in estimation . the remainder of this article is organized as follows . in section [ sec:2 ] we propose a new class of matrix - variate regressions and connect them with conventional regression models . section [ sec:3.1 ] reviews the idea of enveloping and illustrates it with a real example . section [ sec:3 ] is devoted to the development of envelope methods for matrix - variate regressions . section [ sec:5 ] studies theoretical properties of matrix regression models and envelopes . sections [ sec:7 ] and [ sec:8 ] are devoted to illustrations with simulations and real data analyses . technical details and proofs are included in a supplement . generally , a two - way measurement layout can be treated integrally as a matrix - valued variable , denoted as @xmath6 . in application , modeling the relationship between @xmath0 and certain covariates is often of interest , where the covariates can be matrix - valued @xmath7 , vector - valued @xmath8 , or univariate , depending on the specific data . in the insulin assay data , for example , the covariates are formed as a @xmath9 matrix with elements indicating different treatment and dose levels . the goal is to investigate how the treatment and dose levels influence blood sugar concentration , a matrix - valued response . as an obvious first step , we might use the @xmath10 operator that stacks the columns of a matrix to transform the problem into a standard vector - variate regression : @xmath11 where @xmath12 , @xmath13 , and the error vector @xmath14 has mean 0 and covariance matrix @xmath15 . in view of the potential for many parameters , this model may be hard to estimate well and difficult to interpret , unless perhaps some type of sparse structure is imposed . for problems in which a sparsity assumption is untenable , it seems to be a rather blunt instrument that neglects the matrix structure and may be grossly over - parameterized . we propose instead a relatively parsimoniously parameterized model that reflects the underlying structure of the matrix - variate response and predictor . in preparation , the following operators for matrix - valued variables are predefined . for @xmath6 , the expectation of @xmath0 is @xmath16 , the covariance matrix of @xmath0 is assumed to be of the form @xmath17 = { \mathrm{cov}}_r({\textbf{y}})\otimes { \mathrm{cov}}_c({\textbf{y}})$ ] , where ` @xmath18 ' stands for the kronecker product , and @xmath19/\mathrm{tr}[{\mathrm{cov}}_c({\textbf{y}})]$ ] and @xmath20/\mathrm{tr}[{\mathrm{cov}}_r({\textbf{y}})]$ ] are defined as the row and column covariance matrices of @xmath0 ( de waal 1985 ) . here ` tr ' means the trace operation . the two covariance matrices @xmath21 and @xmath22 are uniquely defined up to a proportionality constant . for @xmath23 , we correspondingly define the covariances between @xmath24 and @xmath25 in terms of the row covariance and column covariance , designated @xmath26 $ ] and @xmath27 $ ] . given a matrix - variate predictor @xmath28 , we define the matrix regression of @xmath6 on @xmath1 as the bilinear model @xmath29 where @xmath30 is the overall mean , and @xmath31 and @xmath32 are the row and column coefficient matrices . here @xmath33 and @xmath34 are uniquely defined only up to a proportionality constant . for identifiability , the column coefficient matrix @xmath34 is defined to have frobenius norm 1 and positive element in its first row and column . the distribution of the matrix - valued random error @xmath35 is assumed be independent of @xmath1 , and have zero mean and covariance matrix @xmath36 = { \mathrm{\boldsymbol\sigma}}_2\otimes { \mathrm{\boldsymbol\sigma}}_{1}$ ] , where @xmath37 and @xmath38 are the column and row covariance matrices of @xmath35 . similarly , we require @xmath39 to have unit frobenius norm and positive first diagonal element in order to uniquely identify the two covariance matrices @xmath40 and @xmath39 . if no constraints are required on @xmath41 and @xmath39 , the kronecker products @xmath42 and @xmath43 are still identifiable but the individual row and column parameter matrices are not . the kronecker covariance structure supposes a relational characteristic of the matrix - variate @xmath0 , as the covariances of the column vectors of @xmath35 are all proportional to @xmath40 and the covariances of the row vectors of @xmath35 are all proportional to @xmath44 . such relationships are usually desirable for matrix - valued variables , especially for multivariate repeated measures and multivariate longitudinal data because of the intrinsic relationships among elements . for instance , the eeg data contains measurements of each subject from different time ( row ) and different scalp locations ( column ) . it seems reasonable to model the data with similar variations among measurements over rows and measurements over columns . another advantage of formulating the kronecker covariance structure for such data is that the number of parameters in @xmath36 $ ] can be dramatically reduced when the matrix dimension is high . when the kronecker structure does not hold , a general covariance matrix @xmath36={\mathrm{\boldsymbol\sigma}}$ ] can be applied . hypothesis tests for the kronecker covariance structure can be found in shitan and brockwell ( 1995 ) , lu and zimmerman ( 2005 ) , and roy and khattree ( 2005 ) . let @xmath45 denote the @xmath46 indicator vector with a @xmath47 in the @xmath48-th element and zeros elsewhere , and imagine for the moment that @xmath34 is known . then the multivariate regression implied by ( [ eq : reg1 ] ) for the @xmath49-th column of @xmath0 , @xmath50 , can be seen as standard multivariate regression with response vector @xmath51 , coefficient matrix @xmath33 and predictor vector @xmath52 . the matrix model ( [ eq : reg1 ] ) can be interpreted in the same way , except the predictors for each column of @xmath0 are estimated as linear combinations of the columns of @xmath1 determined by the rows of @xmath34 . the matrix regression ( [ eq : reg1 ] ) is a new model formulation that as far as we know has not been discussed in the literature . it incorporates simple regression , multiple regression and multivariate multiple regression as special cases under different settings of @xmath1 and @xmath0 . for instance , when the response is univariate and the predictor is vector - valued , ( [ eq : reg1 ] ) reduces to a multiple regression . when both the response and the predictor are vector - valued , ( [ eq : reg1 ] ) coincides with the usual multivariate regression model . by capturing the row and column relationships , model ( [ eq : reg1 ] ) reduces the number of parameters by @xmath53 in comparison to ( [ eq : reg0 ] ) , where @xmath54 is the parameter reduction from the coefficients and the rest is the parameter reduction from the covariance matrices . the total number of the reduced parameters could be very large when the matrix dimensions of @xmath1 and @xmath0 are relatively high . model ( [ eq : reg1 ] ) can be seen as a new regression tool for matrix - variate analyses in chemometrics . a variant on partial least square ( pls ) regression is the primary method used to study matrix - variate problems in chemometrics ( smilde et al . , 2004 ) , which regresses both response and predictor on a common latent term with a bilinear coefficient structure . the relationship between the matrix - variate regression ( [ eq : reg1 ] ) and matrix pls is similar to the relationship between regression and pls in the conventional univariate and multivariate settings . * concomitant columns*. when the column dimension of @xmath1 is the same as the corresponding dimension of @xmath0 , and in particular when the columns of @xmath1 and the columns of @xmath0 represent repeated measures or similar characteristics , it might be appropriate to build ( [ eq : reg1 ] ) with @xmath55 , an @xmath56 identity matrix , since each column of @xmath0 is usually associated with the corresponding column of @xmath1 . in this case , the matrix regression model can be simplified to : @xmath57 so the regression of each column of @xmath0 on the corresponding column of @xmath1 has the same coefficient matrix @xmath33 . in application , one can apply a likelihood ratio test or other model selection methods to select between ( [ eq : reg1 ] ) and ( [ eq : reg2 ] ) . similar comments apply to the row dimensions of @xmath1 and @xmath0 . * univariate predictor . * when the predictor @xmath58 is univariate , a matrix regression might allow a different linear regression for each element of @xmath0 on @xmath59 : @xmath60 where @xmath61 . in this case , ( [ eq : reg1 ] ) still has a multiplicative coefficient structure but now the column dimensions of @xmath62 and @xmath41 need not be restricted to match the dimension of @xmath59 : @xmath63 where @xmath64 and @xmath65 , for some @xmath66 . it is easy to see that ( [ eq : reg3 ] ) is a reduced rank ( izenman , 1975 ; reinsel and velu , 1998 ) form of ( [ eq : reg4 ] ) . again , model selection tools can be applied to select the proper model between ( [ eq : reg4 ] ) and ( [ eq : reg3 ] ) . * multivariate pod model . * the matrix - variate regression ( [ eq : reg1 ] ) can be seen as an extension of the partial one - dimensional regression model ( pod ) studied by cook and weisberg ( 2004 ) . the pod was proposed to study data with one response variable , one grouping variable or factor , and several covariates . a typical pod model represents the response @xmath67 at factor level @xmath68 as @xmath69 , where @xmath70 and @xmath71 represent the @xmath68-th level factor effect , and @xmath72 represents the one - dimensional covariate effect ( @xmath73 ) . if each subject contains measurements over all @xmath74 categories , then a multivariate response @xmath75 will be observed for each subject . in this case , the pod model can be extended as @xmath76 where @xmath77 and @xmath78 are @xmath74 dimensional vectors . the mean function of this model has the exactly the same coefficient structure as shown in ( [ eq : reg1 ] ) except the response and the predictor are vector - valued . when @xmath79 is matrix - valued , for example , @xmath80 response variables measured at each factor level , ( [ eq : pod1 ] ) can be extended to the exact matrix - variate regression setting . in this section , we focus on the estimation of ( [ eq : reg1 ] ) . the estimation of the special formulations can be similarly derived . without a specific parametric distribution on the random error , one can estimate the coefficient parameters in ( [ eq : reg1 ] ) by using a loss function , like a squared loss function , and estimate the covariance matrices using moments . however , for statistical inference , as in conventional linear regressions , we assume that the random error @xmath35 follows a matrix normal distribution @xmath81 . background on the matrix normal distribution can be found in dawid ( 1981 ) and de waal ( 1985 ) . we next describe maximum likelihood estimation of the parameters in ( [ eq : reg1 ] ) assuming the matrix normal . assume that @xmath82 , @xmath83 , are sampled from the conditional distribution @xmath84 , and that they are independent . without loss of generality , suppose @xmath85 . then the mle of @xmath86 is @xmath87 . in matrix - variate analysis , explicit mles of @xmath62 , @xmath41 , @xmath40 and @xmath39 are generally unobtainable . we propose a two - step iterative algorithm to construct the estimators . we first hold @xmath41 and @xmath39 fixed and estimate the remaining parameters . we then use an iterative algorithm to obtain the full set of the mles . let @xmath88 and @xmath89 . according to the log - likelihood function ( in the supplement ) , given @xmath41 and @xmath39 fixed , the estimators @xmath62 and @xmath40 are @xmath90 similarly , let @xmath91 and @xmath92 . given @xmath62 and @xmath40 fixed , @xmath41 and @xmath39 can be estimated by @xmath93 let @xmath94 , @xmath95 , @xmath96 and @xmath97 be the mles of @xmath62 , @xmath41 , @xmath40 and @xmath39 respectively . therefore , by initializing @xmath98 with the required frobenius normalization and initializing @xmath99 , one can obtain @xmath94 , @xmath95 , @xmath96 and @xmath97 by iterating between ( [ eq : mrmle1 ] ) and ( [ eq : mrmle2 ] ) with updated values until the log - likelihood function of ( [ eq : reg1 ] ) meets a convergence criterion . then normalize @xmath95 as @xmath100 and rescale @xmath94 as @xmath101 , where @xmath102 with @xmath103 being the first element of @xmath95 and ` @xmath104 ' representing the frobenius norm ; and normalize @xmath97 as @xmath105 and rescale @xmath96 as @xmath106 , where @xmath107 and @xmath108 is the first diagonal element of @xmath97 . if unique identification is not required for the individual parameter matrices @xmath109s and @xmath110s , the last step of normalization and rescaling in the algorithm is not necessary . in this case , we obtain mles for @xmath42 and @xmath111 but not the individual @xmath109s and @xmath110s . in addition , the aforementioned estimation procedure provides the same results as estimating the likelihood directly under the normalization constraints on @xmath41 and @xmath39 , because the corresponding mles with and without constraints are proportional . the justification is similar to the matrix normal estimation shown in dutilleul ( 1999 ) and srivastava et al . ( 2008 ) . the existence of the estimators @xmath94 , @xmath95 , @xmath96 and @xmath97 depends on the existence of the inverse matrices @xmath112 and @xmath113 , which requires at least @xmath114 ( dutilleul , 1999 ) , and on the existence of @xmath115 and @xmath116 , which requires @xmath117 and @xmath118 respectively , when @xmath94 and @xmath95 are full rank matrices . in general , these conditions are fairly mild and require only the row dimensions of the response and the predictor less than their column dimensions multiplied by sample size and vice versa . in section [ sec:5 ] , we further investigate the asymptotic properties of the mles in ( [ eq : reg1 ] ) and show their asymptotic efficiency relative to estimators from conventional vector regression . to guage how well the matrix regression ( [ eq : reg1 ] ) fits the observed data , one can test the goodness of fit of the model compared to an alternative . for instance , to compare ( [ eq : reg1 ] ) with the vector - variate regression of @xmath119 on @xmath120 given in ( [ eq : reg0 ] ) assuming that @xmath121 , one can test the hypothesis : @xmath122 versus @xmath123 is not true . since ( [ eq : reg1 ] ) is nested within ( [ eq : reg0 ] ) , the likelihood ratio test can be applied . let @xmath124 and @xmath125 be the maximized likelihoods ( [ eq : reg1 ] ) and ( [ eq : reg0 ] ) . then the test statistic @xmath126 follows a chi - squared distribution with degree of freedom @xmath127 , which is equal to the number of reduced parameters from ( [ eq : reg0 ] ) to ( [ eq : reg1 ] ) . goodness of fit can also be evaluated by other model selection methods , such as aic , bic or cross validation . the goodness of fit of ( [ eq : reg1 ] ) compared to other models , such as to the special formulations in section [ sec:2.2 ] , can be similarly derived . since the envelope methodology introduced by cook et al . ( 2010 ) is still relatively new , we provide a brief review in this section , before turning to envelopes for matrix - variate regressions in section [ sec:3 ] . to facilitate the discussion , we introduce the following notations that will be used in the rest of the article . for an @xmath128 matrix @xmath129 , @xmath130 is the subspace of @xmath131 spanned by the columns of @xmath129 , @xmath132 is the projection onto @xmath130 , and @xmath133 is the orthogonal projection , where @xmath134 is the moore - penrose inverse . let @xmath135 . the projections @xmath136 and @xmath137 can be equivalently denoted as @xmath138 and @xmath139 . a subspace @xmath140 is said to be a reducing subspace of @xmath141 if @xmath142 decomposes @xmath143 as @xmath144 . if @xmath142 is a reducing subspace of @xmath143 , we say that @xmath142 reduces @xmath143 . envelope methodology was proposed originally to improve efficiency in the vector - variate linear model @xmath145 where the response vector @xmath146 , the predictor vector @xmath147 , @xmath148 , @xmath149 , and the random error @xmath150 is independent of @xmath1 . the motivation for enveloping in this context arose from asking if there are linear combinations of @xmath0 whose distribution is invariant to changes in the non - stochastic predictor @xmath1 . we refer to such linear combinations as @xmath1-invariants . to gain an operational version of this notion , suppose that the subspace @xmath151 satisfies the two conditions ( i ) the marginal distribution of @xmath152 does not depend on @xmath1 , and ( ii ) given the predictor @xmath1 , @xmath153 and @xmath152 are independent . then a change in @xmath1 can affect the distribution of @xmath0 only via @xmath153 ( cook et al . informally , we think of @xmath153 as the part of @xmath0 that is material to the regression , while @xmath152 is @xmath1-invariant and thus immaterial . let @xmath154 . cook et al . ( 2010 ) showed that the statistical conditions ( i ) and ( ii ) are equivalent to the algebraic conditions : ( a ) @xmath155 , and ( b ) @xmath156 . therefore , @xmath142 is a reducing subspace of @xmath157 that contains @xmath158 . the intersection of all reducing subspaces of @xmath157 that contain @xmath158 is called the @xmath157-envelope of @xmath158 , denoted as @xmath159 , or @xmath160 when used as a subscript . the envelope @xmath159 serves to distinguish @xmath161 and the maximal @xmath1-invariant @xmath162 in the estimation of @xmath163 and can result in substantial increases in efficiency , sometimes equivalent to taking thousands of additional observations . reparameterizing ( [ eq : m1 ] ) in terms of a semi - ortogonal basis matrix @xmath164 for @xmath159 , we have its envelope version @xmath165 with @xmath166 . the standard errors of elements of the maximum likelihood estimator of @xmath167 are often substantially smaller than the corresponding standard errors based on ( [ eq : m1 ] ) . the overarching premise of the envelope @xmath159 that there may be linear combinations of @xmath0 whose distribution is invariant to changes in @xmath1 can be seen as inducing a type of generalized sparsity in both @xmath163 and @xmath157 . since @xmath168 , @xmath169 and thus linear combinations of @xmath163 are 0 rather than individual elements . envelopes go a step further by simultaneously requiring a generalized form of sparsity in @xmath166 . we employ data on concho water snakes ( johnson and wichern , 2007 ) to illustrate the idea of enveloping intuitively . the data contain measurements on the gender , age , tail length ( mm ) , and snout to vent length ( mm ) for 37 female concho water snakes and 29 male concho water snakes . we want to investigate the gender effect on the tail length and snout to vent length for young snakes ( age @xmath170 years ) . the data can be modeled using ( [ eq : m1 ] ) , where @xmath171 is a bivariate response vector with @xmath172 and @xmath173 representing the tail length and snout to vent length , respectively , and @xmath58 is an indicator variable with zero indicating a male snake and with one indicating a female snake . thus , the intercept @xmath174 is the mean of the female population , and the coefficient vector @xmath175 is mean difference between the female population and the male population , which implies the gender effect . figure 1(a ) shows the data , where the circles represent the male snakes and the cross points represent the female snakes . under model ( [ eq : m1 ] ) , inference on the gender difference in tail lengths proceeds by first projecting the data onto the horizontal axis of figure 1(a ) and then using univariate inference methods with the projected data . the two dashed curves represent the marginal distributions of the male tail length and the female tail length . these curves overlap substantially , so it may take a large sample size to detect the gender effect . the estimated envelope and its orthogonal complement are represented by the solid and the dashed lines in figure 1(a ) . we can see that the two samples differ notably in the direction of the envelope , while differing little if at all in the direction of its orthogonal complement . accordingly , envelope estimation proceeds by first projecting the data onto the envelope to remove @xmath162 , the @xmath59-invariant part of @xmath0 , and then projecting onto the horizontal axis to focus on the gender difference in tail length . this process is depicted in figure 1(b ) , which shows the approximated marginal distributions of the male s and the female s tail lengths after enveloping , represented by the solid curves . the variances of two marginal distributions after enveloping are considerably reduced compared to the original ones ( the dashed curves ) , which reflects the efficiency gain . the efficiency gains can be massive when the immaterial variation @xmath176 is substantially larger than the material variation @xmath177 . although the matrix - variate regression ( [ eq : reg1 ] ) is parsimoniously parameterized relative to the naive model ( [ eq : reg0 ] ) , there still may be linear combinations of the rows or columns of @xmath178 whose distribution is invariant to changes in the predictors . to allow for the possibility of @xmath1-invariant linear combinations , we extend the vector - variate model described in section [ sec:3.1 ] to matrix - variate regression . the rationale for doing so is generally the same as that described for vector - variate regressions : we hope to achieve efficiency in estimation and prediction better than that for model ( [ eq : reg1 ] ) . we first establish the envelope structure for the matrix - variate regression model ( [ eq : reg1 ] ) . envelope models for the special cases are given in the supplement . to allow for the possibility that there are @xmath1-invariants in both the rows and columns of @xmath0 , we suppose that there exist subspaces @xmath179 and @xmath180 so that @xmath181 where @xmath182 and @xmath183 were defined in section [ sec:2.1 ] . these conditions are similar to conditions ( i ) and ( ii ) introduced in section [ sec:3.1 ] for defining @xmath1-invariants of @xmath0 in vector - variate regressions . let @xmath184 and @xmath185 . conditions ( [ eq : mel]a ) and ( [ eq : mer]a ) mean that the marginal distributions of @xmath186 and @xmath187 do not depend on @xmath1 , which is equivalent to requiring that @xmath188 and @xmath189 . conditions ( [ eq : mel]b ) and ( [ eq : mer]b ) indicate that @xmath186 does not respond to changes in @xmath1 through a linear association with @xmath190 and that @xmath191 does not respond to changes in @xmath1 through a linear association with @xmath192 . further , condition ( [ eq : mel]b ) holds if and only if @xmath193 , which is equivalent to requiring that @xmath194 reduce @xmath195 , so @xmath196 . similarly , @xmath197 must reduce @xmath198 , so @xmath199 . the intersection of all reducing subspaces @xmath194 of @xmath40 that contain @xmath200 is the @xmath40-envelope of @xmath200 , denoted as @xmath201 or @xmath202 when used as a subscript . similarly , the intersection of all reducing subspaces @xmath197 of @xmath39 that contain @xmath203 is the @xmath39-envelope of @xmath203 , denoted as @xmath204 or @xmath205 when used as a subscript . these subspaces @xmath201 and @xmath204 always exist , are uniquely defined and serve as the fundamental constructs that allow row and column reduction of @xmath0 . in concert , they imply that @xmath0 responds to changes in @xmath1 only via @xmath206 and they hold the promise of much better estimation of @xmath62 and @xmath41 . a theoretical justification regarding the variance reduction of enveloping is given by proposition [ prop4 ] in section [ sec:5 ] . because the column spaces @xmath207 and @xmath208 , and the orthogonal decompositions on @xmath40 and @xmath39 , are all invariant under a multiplicative constant , the column and row envelopes @xmath209 and @xmath210 are always uniquely defined , whether normalization is required or not . we next use @xmath201 and @xmath204 to reparameterize ( [ eq : reg1 ] ) and to establish the envelope model . let @xmath211 and @xmath212 be semi - orthogonal bases of @xmath201 and @xmath204 , respectively , where @xmath213 and @xmath214 are the dimensions of the corresponding row and column envelopes and are temporarily assumed to be known . the determination of @xmath213 and @xmath214 will be discussed in section [ sec:6 ] . by definition , we know that @xmath215 and @xmath216 , so there exist two coordinate matrices @xmath217 and @xmath218 such that @xmath219 and @xmath220 . let @xmath221 and @xmath222 be orthogonal matrices . then the matrix regression model ( [ eq : reg1 ] ) can be re - parameterized as the following envelope model : @xmath223 where @xmath224 , @xmath225 , and @xmath226 and @xmath227 are unknown , @xmath228 . as @xmath229 and @xmath230 are overparameterized , the matrices @xmath231 and @xmath232 themselves are not identifiable but their column spaces @xmath233 and @xmath234 are identifiable . the two parameter spaces are grassmannians of dimension @xmath213 and @xmath214 in @xmath131 and @xmath235 with the numbers of unknown real parameters @xmath236 and @xmath237 respectively . therefore , the total number of real parameters in ( [ eq : envreg1 ] ) is @xmath238 , which is equal to the sum of the numbers of parameters @xmath239 in @xmath86 , @xmath240 in @xmath241 , @xmath242 in @xmath243 , @xmath236 in @xmath231 , @xmath237 in @xmath232 , @xmath244 in @xmath245 , @xmath246 in @xmath247 , and @xmath248 in @xmath249 and @xmath250 , whereas ( [ eq : reg1 ] ) has @xmath251 parameters . here the normalization constraints on @xmath41 and @xmath39 eliminate two free parameters in each model . we see that the envelope model further reduce @xmath252 parameters from ( [ eq : reg1 ] ) . efficiency gain of envelope model ( [ eq : envreg1 ] ) over matrix model ( [ eq : reg1 ] ) can arise in two distinct ways . the first is through parameter reduction , particularly when the number of real parameters in ( [ eq : reg1 ] ) is large relative to that in ( [ eq : envreg1 ] ) . but the largest efficiency gains are often realized when the variation in the @xmath1-invariant part of @xmath0 is large relative to the material variation @xmath0 . letting @xmath253 denote the spectral norm of a matrix , this happens when @xmath254 or @xmath255 . for vector - variate regression , this is reflected in figure [ ch3_fig1 ] , since the variation along @xmath256 is notably larger than that along @xmath159 . in this section , we describe the mles for the unknown parameters in ( [ eq : envreg1 ] ) . as @xmath85 , the mle of @xmath86 is still @xmath87 but the remaining parameter estimates can not be expressed in closed form . consequently , we propose a two - step iteration algorithm to obtain the mles . let @xmath257 , @xmath258 , @xmath259 , @xmath260 , @xmath261 and @xmath262 be the envelope mles of the corresponding parameters . since closed - form expressions for these estimators are unavailable , we first hold @xmath34 and @xmath198 fixed and give the estimators of the remaining parameters . we later describe an iterative algorithm for computing the full set of mles . assuming then that @xmath34 and @xmath198 are given , let @xmath263 , and let @xmath264 and @xmath265 be defined in ( [ eq : mrmle1 ] ) . then @xmath231 can be estimated as @xmath266 where the objective function @xmath267 and @xmath268 is taken over all semi - orthogonal matrices @xmath269 . the estimators of @xmath33 and @xmath195 are then @xmath270 and @xmath271 , where @xmath272 is orthogognal . similarly , given @xmath62 and @xmath40 , let @xmath273 , and let @xmath274 and @xmath275 be defined in ( [ eq : mrmle2 ] ) . then @xmath276 where @xmath277 and @xmath278 is taken over all semi - orthogonal matrices @xmath279 . then the intermediate estimators of @xmath34 and @xmath198 are @xmath280 and @xmath281 , where @xmath282 is orthogognal . the full mles can now be obtained by alternating between @xmath283 and @xmath284 , and then rescaling them : 1 . initialize @xmath285 and @xmath286 as the mles from ( [ eq : reg1 ] ) . let @xmath287 and let @xmath288 . 2 . given @xmath260 and @xmath262 , estimate @xmath231 by ( [ eq : envlme1 ] ) , @xmath289 and @xmath290 . 3 . given @xmath259 and @xmath261 , estimate @xmath232 by ( [ eq : envlme2 ] ) , @xmath291 and @xmath292 . 4 . iterate 2 - 3 until the log - likelihood function of ( [ eq : envreg1 ] ) converges . then normalize @xmath260 as @xmath293 and rescale @xmath259 as @xmath294 , where @xmath295 and @xmath296 is the first element of @xmath260 , and normalize @xmath262 as @xmath297 and rescale @xmath261 as @xmath298 , where @xmath299 and @xmath300 is the first diagonal element of @xmath262 . similar to the discussion in section [ sec:2.3 ] , the proposed estimation procedure provides the same results as estimating the parameters directly under constrained @xmath41 and @xmath39 . if one is interested only in estimating the entire kronecker products @xmath42 and @xmath111 , but not individual @xmath109s and @xmath110s , the last rescaling step in the algorithm is unnecessary . moreover , the optimization of @xmath301 and @xmath302 are attained by a non - grassman algorithm ( cook et al . 2015 ) that is more efficient and faster than existing grassmannian optimization approaches . in particular , cook et al . ( 2015 ) demonstrated its superiority over the 1d algorithm used by li and zhang ( 2015 ) . the resulting envelope estimators @xmath303 and @xmath304 in steps 2 and 3 can be viewed as the projections of their corresponding estimators , @xmath264 and @xmath305 , from ( [ eq : reg1 ] ) onto the estimated row and column envelopes @xmath306 and @xmath307 . in addition , the envelope mles @xmath261 and @xmath262 are both partitioned into the estimated @xmath1-variant and @xmath1 invariant parts of @xmath0 . similar to the intuitive arguments in figure [ ch3_fig1 ] , with the envelope projections , the formulations here can lead to more efficient estimation of the parameters . so far , we have assumed that the dimensions @xmath213 and @xmath214 of the row and column envelopes are known . extending the methods described by cook et al . ( 2010 ) , these dimensions , which are essentially model selection parameters , can be determined by using an information criterion , say aic or bic , or using other model selection methods such as cross validation and likelihood ratio tests . for the information criteria , the envelope dimensions @xmath213 and @xmath214 can be selected by minimizing the objective function @xmath308 , where @xmath309 is the maximized log - likelihood function of an envelope matrix regression model , @xmath310 is equal to log@xmath311 for bic and is equal to 2 for aic , and @xmath312 is the total number of parameters in the target model . when the matrix regression model ( [ eq : reg1 ] ) holds , these procedures for selecting @xmath313 and @xmath314 are in effect methods for determining the appropriateness of a proper envelope model . if it is inferred that @xmath315 and @xmath316 , then the data do not support a proper envelope model . on the other hand , the inference that @xmath315 and @xmath316 may itself be useful depending on the application , because it indicates the absence of @xmath1-invariant linear combinations of the row and columns of @xmath0 . in this section , we investigate the asymptotic properties of the mles from ( [ eq : reg1 ] ) and ( [ eq : envreg1 ] ) . we show that the mles from ( [ eq : reg1 ] ) are asymptotically more efficient than the mles obtained from ( [ eq : reg0 ] ) and that the envelope estimators from ( [ eq : envreg1 ] ) can gain efficiency over the mles from ( [ eq : reg1 ] ) . we neglect @xmath86 in the theoretical development , since its mle is @xmath87 in all the three models , which is asymptotically independent of the mles of all other parameters in the models . to show the relative efficiency of ( [ eq : reg1 ] ) to ( [ eq : reg0 ] ) , we need to consider only the joint asymptotic distribution of the mles of @xmath317 and @xmath318 , whereas the individual parameter matrices @xmath109s and @xmath110s are not of interest . in this case , the matrix regression ( [ eq : reg1 ] ) can be considered as an over - parameterized structural model of ( [ eq : reg0 ] ) . hence we can apply proposition 4.1 in shapiro ( 1986 ) to derive the target property . for notation convenience , let @xmath319 and @xmath320 , where ` vech ' denotes the half - vectorization operator . then under ( [ eq : reg1 ] ) , we can rewrite @xmath164 as @xmath321 , where @xmath322 and @xmath323 . define @xmath324 to be the standard mle of @xmath164 from ( [ eq : reg0 ] ) and define @xmath325 to be the mle of @xmath164 from the matrix regression ( [ eq : reg1 ] ) . for a general statistic @xmath326 , denote the asymptotic variance of @xmath327 as @xmath328 . let @xmath329 be the fisher information function of @xmath164 and let @xmath330 be the gradient matrix . proposition [ prop1 ] establishes the asymptotic distribution of @xmath325 and its asymptotic efficiency relative to @xmath324 . under ( [ eq : reg1 ] ) with a normal error , @xmath331 converges in distribution to a normal random vector with mean zero and covariance matrix @xmath332 . moreover , @xmath333 . [ prop1 ] consequently , when the matrix regression structure holds , omitting row and column information in the coefficient and covariance formulation can result in loss of efficiency in parameter estimation . we next establish asymptotic properties for the mle from ( [ eq : envreg1 ] ) and show that enveloping can further improve efficiency in estimating ( [ eq : reg1 ] ) . let @xmath334 be the collection of the envelope parameters . under ( [ eq : envreg1 ] ) , @xmath335 is reformulated as @xmath336 , where @xmath337 , @xmath338 , @xmath339 , and @xmath340 . thus , @xmath341 . correspondingly , the envelope mle of @xmath164 is @xmath342 . let @xmath343 be a gradient matrix . under ( [ eq : envreg1 ] ) with a normal error , @xmath344 converges in distribution to a normal random vector with mean zero and covariance matrix @xmath345 . moreover , @xmath346 . [ prop2 ] so far we have derived the asymptotic distributions for the estimators of @xmath164 under different model settings and compared the relative efficiency among different estimators . we next investigate the asymptotic properties of the individual row and column parameter estimators of @xmath335 in ( [ eq : reg1 ] ) and ( [ eq : envreg1 ] ) , given that these parameters are uniquely defined by normalization as discussed in the model settings . let @xmath347 denote the mle of @xmath335 from ( [ eq : reg1 ] ) , and let @xmath348 denote the envelope mle of @xmath335 from ( [ eq : envreg1 ] ) . under ( [ eq : reg1 ] ) with a normal error and uniquely defined row and column parameter matrices , @xmath349 converges in distribution to a normal random vector with mean zero and covariance matrix @xmath350 , where @xmath351 and @xmath352 are matrix - valued functions of @xmath335 given in the supplement . [ prop3 ] under ( [ eq : envreg1 ] ) with a normal error and uniquely defined row and column parameter matrices , @xmath353 converges in distribution to a normal random vector with mean zero and covariance matrix @xmath354 . moreover , @xmath355 . [ prop4 ] propositions [ prop3 ] and [ prop4 ] demonstrate the asymptotic normality for both enveloped and non - enveloped estimators of @xmath335 . proposition [ prop4 ] further indicates that as long as there exists @xmath1-invariants in ( [ eq : reg1 ] ) , one can gain potential efficiency in estimation by enveloping the matrix regression model . since the row and column coefficient matrices , @xmath356 and @xmath357 , are often of interest in regression analysis , we further derived asymptotic variances for @xmath358 and @xmath359 , which are given in the supplement . when the random error in ( [ eq : reg1 ] ) is not normal , the estimator of @xmath335 under a type of generalized least square estimation ( as shown in the supplement ) is the same as the mle of ( [ eq : reg1 ] ) under normality . we thus use the same notation @xmath347 . under mild conditions , the estimator @xmath347 is still asymptotically normal , except having a more complex asymptotic covariance matrix . the terms @xmath360 and @xmath361 described in proposition [ prop3 ] will be used in the following propositions . under ( [ eq : reg1 ] ) with uniquely defined row and column parameter matrices , suppose that the random error @xmath35 follows a general matrix - valued distribution with zero mean and column and row covariance matrices @xmath40 and @xmath39 . if @xmath352 is invertible at the true value @xmath335 and @xmath362 is @xmath363 at some @xmath364 between the estimator @xmath347 and the true @xmath335 , where @xmath365/\partial { \boldsymbol\theta}_1^t\|_{{\mathrm{f}}}$ ] , then @xmath366 converges in distribution to @xmath367 , where @xmath368 is a matrix of second - order limiting estimating functions given in the supplement . [ prop5 ] for the envelope model ( [ eq : envreg1 ] ) with a non - normal error @xmath35 , a good way to achieve parameter estimation is to use the same objective functions ( [ eq : envlme1 ] ) and ( [ eq : envlme2 ] ) to estimate the row and column envelopes , and then estimate the remaining parameters as described in section [ sec:3.3 ] . the next proposition shows that without normality on the error , the envelope estimators @xmath348 obtained from this procedure still retain desired asymptotic properties . under ( [ eq : envreg1 ] ) and the conditions in proposition [ prop5 ] , @xmath353 converges in distribution to @xmath369 , where @xmath370 moreover , the envelope estimators are asymptotically more efficient than the estimators from ( [ eq : reg1 ] ) if @xmath371 is a reducing subspace of @xmath372 . in this section , we evaluate the performance of matrix - variate regression ( [ eq : reg1 ] ) and envelope matrix - variate regression ( [ eq : envreg1 ] ) numerically , and compare the two models along with the vector regression model ( [ eq : reg0 ] ) . we first generated data based on ( [ eq : envreg1 ] ) with @xmath373 , @xmath374 , @xmath375 , @xmath376 , @xmath377 , and @xmath378 . here @xmath379 and @xmath380 . the semi - orthogonal matrices @xmath231 and @xmath232 were generated by orthogonalizing matrices of independent uniform ( 0,1 ) random variables . the elements of @xmath86 , @xmath241 , @xmath243 and the predictors were selected as independent standard random normal variables . we then fitted three models : ( [ eq : reg0 ] ) , ( [ eq : reg1 ] ) , and ( [ eq : envreg1 ] ) to the data and evaluated their estimation accuracy of the coefficient parameters , according to the criteria : @xmath381 for the first two matrix regression models , and @xmath382 for the vector regression ( [ eq : reg0 ] ) . a corresponding evaluation of the covariance estimators showed results similar to the coefficient and is thus omitted . we used six different sample sizes , @xmath383 . within each sample size , 200 replicates were simulated . the average estimation errors were computed over the 200 random samples for each sample size , under each model . the envelope dimensions were set to the true dimensions @xmath374 . with this setup models ( [ eq : reg0 ] ) , ( [ eq : reg1 ] ) and ( [ eq : envreg1 ] ) have 975 , 105 and 75 real parameters to be estimated . figure [ fig:2 ] shows the comparison results from the three models . in particular , the left panel of figure [ fig:2 ] shows a comparison of the three models , and the right panel highlights the improvement of enveloping . over all selected sample sizes , the two matrix regression models provided much smaller estimation errors than did the vector regression . in addition , by effectively removing @xmath1-invariants from estimation , the envelope model further improved estimation accuracy in comparison to the matrix regression model without enveloping . figure [ fig:3 ] shows the efficiency gains by assessing the asymptotic , actual , and bootstrap standard errors of the first elements in @xmath384 ( or @xmath385 ) from the three models . similar results were obtained for other elements . the asymptotic standard errors were estimated according to the results presented in propositions [ prop1 ] and [ prop2 ] . the actual standard errors were computed using the sample standard errors of @xmath384 ( or @xmath385 ) over 200 samples for each selected sample size . the bootstrap standard errors were obtained by bootstrapping one sample for each sample size . again , the left panel in figure [ fig:3 ] provides a comparison among the three models and the right panel highlights the comparison between the two matrix regression models with and without enveloping . clearly , the asymptotic , actual and bootstrap standard errors are very close , indicating that the asymptotic covariance matrices proposed in section [ sec:5 ] are accurate . in addition , over all selected sample sizes , the matrix regression without enveloping was asymptotically more efficient in parameter estimation than the vector regression , and imposing additional envelope structure further improved the efficiency gains . the importance of enveloping can be seen clearly in the right panel of figure [ fig:3 ] . indeed , the ratios of the asymptotic standard errors between the estimators from ( [ eq : reg0 ] ) and ( [ eq : reg1 ] ) range from 23.7 to 54.9 , while the ratios of the asymptotic standard errors between the estimators from ( [ eq : reg0 ] ) and ( [ eq : envreg1 ] ) range from 309.0 and 523.9 . therefore , effectively utilizing matrix - variate information in model fitting , and at the same time , removing @xmath1-invariants from estimation , one can gain substantial efficiency in parameter estimation . we next assess the performance of the two matrix regression models ( [ eq : reg1 ] ) and ( [ eq : envreg1 ] ) in terms of the individual row and column parameter estimations , particularly the estimation of @xmath62 and @xmath41 . the data were generated as in the previous example but to make the row and column parameter matrices identifiable , we normalized @xmath41 and @xmath39 such that each of them has frobenius norm 1 and positive element in its first row and column . we then evaluated the estimation accuracy of @xmath62 and @xmath41 according to the criterion @xmath386 , @xmath387 for both models . the results are shown in figure [ fig:4 ] wherein the envelope model notably outperformed the matrix regression ( [ eq : reg1 ] ) for both @xmath62 and @xmath41 at all sample sizes . figure [ fig:5 ] demonstrates the substantial efficiency gains of the envelope model by comparing the standard errors of the first elements in @xmath259 and @xmath260 obtained from ( [ eq : envreg1 ] ) and ( [ eq : reg1 ] ) . the comparison results similarly hold for other elements . again , asymptotic , actual and bootstrap standard errors are reported for each model . the asymptotic standard errors were estimated based on the results from propositions [ prop3 ] and [ prop4 ] . the actual and bootstrap standard errors were computed as in the previous example . from figure [ fig:5 ] , we see that the asymptotic standard errors of the estimators of @xmath62 and @xmath41 are very close to their corresponding actual and bootstrap standard errors for both models . this again supports the accuracy of the asymptotic covariance matrices given in section [ sec:5 ] . in addition , the envelope model shows asymptotic efficiency in comparison to ( [ eq : reg1 ] ) over all selected sample sizes . the ratios of the standard errors between the estimators from ( [ eq : reg1 ] ) and ( [ eq : envreg1 ] ) over all elements in @xmath62 range from 2.7 to 11.8 . the ratios of the standard errors between the corresponding estimators of @xmath41 range from 2.5 to 27.2 . enveloping improved both accuracy and efficiency in parameter estimation . we applied the proposed matrix - variate regression models to a cross - over assay of insulin based on rabbit blood sugar concentration ( v@xmath3lund , 1980 ) . the design partitioned the animals into four groups of nine rabbits each , with a different treatment assigned to each group . hence @xmath388 . let @xmath389 and @xmath390 denote the low and high dose levels , 0.75 units and 1.5 units of the standard treatment , and let @xmath391 and @xmath392 denote the same two dose levels of the test treatment . the treatment assignment is shown in table [ ch4_tb1 ] . after injection of the insulin dose each day , the blood sugar concentration of each rabbit was measured at 0 , 1 , 2 , 3 , 4 , and 5 hours . we consider the percentage decreases of the blood sugar concentrations at 1 , 2 , 3 , 4 , and 5 hours relative to the initial concentration at 0 hours . the measurements on each rabbit then form a matrix @xmath395 , where the two rows represent the percentage decreases under two different treatments assigned on day 1 and day 2 , and the columns indicate the hourly percentage decreases each day . we modeled the insulin assay data with treatment effects and dose levels as covariates to explore how these factors affect the percentage decreases of the blood sugar concentrations . we defined the covariates of each rabbit as a matrix @xmath396 , where the two rows of @xmath1 represent standard and the test treatments , and the two columns of @xmath1 represent day 1 and day 2 . for example , for the rabbits in group 1 , the covariate matrix @xmath397 , a diagonal matrix with elements 0.75 and 1.5 . that is because the rabbits in group 1 received the standard treatment at the low dose level 0.75 in day 1 , and received the test treatment at the high dose level 1.5 in day 2 . similarly , for the rabbits in groups 2 , 3 , and 4 , the covariate matrices can be formed as @xmath398 , @xmath399 and @xmath400 . we then fitted the relationship between the response @xmath401 and the predictor @xmath396 based on ( [ eq : reg1 ] ) , ( [ eq : reg2 ] ) and ( [ eq : reg0 ] ) as : @xmath402 @xmath403 @xmath404 where @xmath405 is the intercept matrix , @xmath406 , @xmath407 , @xmath408 , and @xmath409 are the corresponding coefficient matrices in the three models , and @xmath35 is the random error that is assumed to be matrix normal in the matrix - variate regression models ( [ eq : rabbitreg1 ] ) and ( [ eq : rabbitreg3 ] ) , and @xmath410 is assumed to be multivariate normal under the multivariate regression ( [ eq : rabbitreg2 ] ) . since the blood sugar concentration levels observed in the first day can not be affected by the treatment received in the second day , we set the elements in the upper - right corner of @xmath41 in ( [ eq : rabbitreg1 ] ) and of @xmath411 in ( [ eq : rabbitreg2 ] ) to be zero , that is , @xmath412 where @xmath413 are scalars , and @xmath414 . in this setting , the percentage decreases of the blood sugar concentrations in day 2 can depend on the treatments received in both day 1 and day 2 , whereas the percentage decreases in day 1 can only depend on the treatment received in day 1 . since ( [ eq : rabbitreg2 ] ) , ( [ eq : rabbitreg1 ] ) and ( [ eq : rabbitreg3 ] ) are nested models , we applied likelihood ratio tests to select the best model among the three and concluded that ( [ eq : rabbitreg3 ] ) was sufficient . we then performed envelope fitting for ( [ eq : rabbitreg3 ] ) to achieve efficiency gains for the parameter estimation . using bic , the envelope dimension was selected to be @xmath415 . the estimates ( est ) and the asymptotic standard errors ( se ) for each element in @xmath416 are summarized in table [ tab3 ] , with subscripts ` e ' and ` s ' indicating the results from the envelope fit and the fit from ( [ eq : rabbitreg3 ] ) without enveloping . c|ccc|ccc|c + & & & ratio + & @xmath417 & @xmath418 & @xmath419 & @xmath420 & @xmath421 & @xmath422 & @xmath423 + & 7.74&1.78&4.35&6.34&2.21&2.87&1.24 + & 13.07&2.48&5.28&14.42&2.76&5.22&1.12 + & 18.98&3.22&5.9&21.12&3.36&6.28&1.04 + & 17.41&3.11&5.6&16.67&3.35&4.98&1.08 + & 9.63&2.17&4.43&6.03&2.65&2.28&1.22 + & 7.33&1.73&4.23&6.33&2.21&2.86&1.28 + & 12.37&2.44&5.06&12.45&2.76&4.51&1.13 + & 17.96&3.21&5.6&20.1&3.36&5.98&1.05 + & 16.47&3.09&5.34&16.42&3.35&4.9&1.09 + & 9.11&2.12&4.3&5.68&2.65&2.15&1.25 + in comparison to the fit from ( [ eq : rabbitreg3 ] ) , the envelope model shows smaller standard errors for all the elements in @xmath416 . all the parameters in @xmath424 are significant based on the two models . yet the magnitudes of the ratios @xmath425 from the envelope fitting are mostly higher than those from the fitting without enveloping . we further evaluated the sum of squared prediction error ( sspe ) by partition the data into training sets of seven subjects within each group and testing sets of two subjects within each group . the results based on the prediction error @xmath426 are summarized in figure [ ch4_fig7 ] , which shows that the envelope regression dominates for all choices of the envelope dimension @xmath213 . the smallest prediction error occurs at the optimal dimension @xmath415 . when @xmath427 , the envelope model coincides with the non - enveloped model and so the prediction errors of the two models are identical . in addition , we also modeled the data with group effects and compared estimates and standard errors with and without enveloping . the model and the results are given in the supplement . in this case , enveloping provides substantial efficient gains in estimation and can successfully identify the effects of different treatments , while the model without enveloping fails to detect such effects . the eeg data was briefly introduced in the introduction section . it contains two groups of subjects : 77 subjects from the alcoholic group and 45 subjects from the control group . each subject has measurements of electrical scalp activity , which form a @xmath5 matrix . to explore the influence of the alcoholism on the brain activity , let @xmath428 if a subject is alcoholic and @xmath429 if a subject is nonalcoholic . we applied ( [ eq : reg4 ] ) to model the data as @xmath430 where @xmath431 is the matrix - valued brain measurements , @xmath432 is the mean brain activity of the nonalcoholic population ( control group ) , @xmath433 is the difference between the mean brain activity of the alcoholic subjects , denoted as @xmath434 , and the mean brain activity of the nonalcoholic subjects , @xmath435 . therefore , the coefficient matrix , @xmath436 , compares the alcoholic subjects and the nonalcoholic subjects based on their brain signals . the random error @xmath35 is assumed to follow @xmath437 . a saturated model such as ( [ eq : reg0 ] ) is not estimable for the eeg data as the dimension of @xmath119 is much higher than the sample size , and the the covariance matrix of @xmath119 given @xmath59 is not likely to be sparse . matrix - variate regression provides one way to handle such data . in ( [ eq : eeg ] ) , it is easy to see that the mle of @xmath163 without enveloping is @xmath438 , the difference between the sample means of alcoholic brain activities and nonalcoholic brain activities . to achieve estimation efficiency , we considered fitting an envelope model on ( [ eq : eeg ] ) as @xmath439 where @xmath231 and @xmath232 are the bases of the @xmath40-envelope of @xmath440 and the @xmath39-envelope of @xmath441 respectively , @xmath442 is the the coordinate matrix , and the rest of the parameters are similarly defined as in ( [ eq : envreg1 ] ) . the bic chose the envelope dimensions to be @xmath443 and @xmath444 . similar to the results observed in the bioassay example , the envelope model can be more effective at identifying the differences between the alcoholic and nonalcoholic brain activities because it shows smaller standard errors for the estimators of @xmath163 . with @xmath443 and @xmath444 , we calculated the relative ratios of the standard errors of the estimators from ( [ eq : eeg ] ) to those from ( [ eq : eegenv ] ) , and 90% of them fall in the range of 1.2 and 5.5 . moreover , since brain regions associated with alcoholism could of interest to researchers , we further investigated the effect of alcoholism on different brain locations ( channels ) by averaging the two estimators of @xmath163 over time ( 256 rows ) . we then obtained location - based estimators from ( [ eq : eeg ] ) and ( [ eq : eegenv ] ) , each of dimension 64 . bootstrap standard errors were computed for each element of the estimators and corresponding t - tests were applied . because this is a multiple hypothesis testing problem , we employed the approach proposed by benjamini and yekutieli ( 2001 ) to control the false discovery rate ( fdr ) . the fdr adjusted p - values ( in @xmath445 scale ) for the estimators with and without enveloping are reported in the left and right panels of figure [ fig:7 ] . the dashed lines in both panels represent the significance levels @xmath446 and @xmath447 . some plot characteristics are due to the adjustment process . for instance , about 25 p - values were adjusted to 1 , which corresponds to @xmath448 in figure [ fig:7 ] . at the level @xmath446 , the envelope model suggests that 35 out of 64 brain regions are associated with alcoholism , while the model without enveloping was unable to identify relevant regions with only a few borderline significance . li and zhang ( 2015 ) also analyzed these data , but computationally their method of estimation had to first reduce the number of time points from 256 to 64 by averaging adjacent groups of four . no such reduction was necessary with our estimation method . this is an important distinction . by averaging adjacent groups of four time points , a tuning parameter was essentially introduced into their estimation method . different result would surely be obtained by varying the averaging group size . for instance , using bic their method estimated envelope dimensions of @xmath449 and @xmath450 , while using the full data we estimated @xmath443 and @xmath444 . by averaging adjacent groups , they could lose dimensions . zhao , q. , caiafa , c. f. , mandic , d. p. , chao , z. c. , nagasaka , y. , fujii , n. , zhang , l. , and cichocki , a. ( 2013 ) . higher order partial least squares ( hopls ) : a generalized multilinear regression method . _ ieee trans . pattern anal . machine intell . _ * 35 * , 16601673 .

in this paper , we give an explanation to the failure of two likelihood ratio procedures for testing about covariance matrices from gaussian populations when the dimension is large compared to the sample size . next , using recent central limit theorems for linear spectral statistics of sample covariance matrices and of random f - matrices , we propose necessary corrections for these lr tests to cope with high - dimensional effects . the asymptotic distributions of these corrected tests under the null are given . simulations demonstrate that the corrected lr tests yield a realized size close to nominal level for both moderate ( around 20 ) and high dimension , while the traditional lr tests with approximation fails . another contribution from the paper is that for testing the equality between two covariance matrices , the proposed correction applies equally for non - gaussian populations yielding a valid pseudo - likelihood ratio test .

0902.0552

the rapid development and wide application of computer techniques permits to collect and store a huge amount data , where the number of measured variables is usually large . such high dimensional data occur in many modern scientific fields , such as micro - array data in biology , stock market analysis in finance and wireless communication networks . traditional estimation or test tools are no more valid , or perform badly for such high - dimensional data , since they typically assume a large sample size @xmath2 with respect to the number of variables @xmath0 . a better approach in this high - dimensional data setting would be based on asymptotic theory which has both @xmath2 and @xmath0 approaching infinity . to illustrate this purpose , let us mention the case of hotelling s @xmath3-test . the failure of @xmath3-test for high - dimensional data has been mentioned as early as by @xcite . as a remedy , dempster proposed a so - called non - exact test . however , the theoretical justification of dempster s test arises much later in @xcite inspired by modern random matrix theory ( rmt ) . these authors have found necessary correction for the @xmath3-test to compensate effects due to high dimension . in this paper , we consider two lr tests concerning covariance matrices . we first give a theoretical explanation for the fail of these tests in high - dimensional data context . next , with the aid of random matrix theory , we provide necessary corrections to these lr tests to cope with the high dimensional effects . first , we consider the problem of one - sample covariance hypothesis test . suppose that @xmath4 follows a @xmath0-dimensional gaussian distribution @xmath5 and we want to test @xmath6 where @xmath7 denotes the @xmath0-dimensional identity matrix . note that testing @xmath8 with an arbitrary covariance matrix @xmath9 can always be reduced to the above null hypothesis by the transformation @xmath10 . let @xmath11 be a sample from @xmath4 , where we assume @xmath12 . the sample covariance matrix is @xmath13 and set @xmath14 the likelihood ratio test statistic is @xmath15 keeping @xmath0 fixed while letting @xmath16 , then the classical theory depicts that @xmath17 converges to the @xmath18 distribution under @xmath19 . however , as it will be shown , this classical approximation leads to a test size much higher than the nominal test level in the case of high - dimensional data , because @xmath17 approaches infinity for large @xmath0 . as seen from table 1 in [ sec : test1 ] , for dimension and sample sizes @xmath20 , the realized size of the test is 22.5% instead of the nominal 5% level . the result is even worse for the case @xmath21 , with a 100% test size . based on a recent clt for linear spectral statistics ( lss ) of large - dimensional sample covariance matrices @xcite , we construct a corrected version of @xmath17 in [ sec : test1 ] . as shown by the simulation results of [ sec : simul1 ] , the corrected test performs much better in case of high dimensions . moreover , it also performs correctly for moderate dimensions like @xmath22 or 20 . for dimension and sample sizes @xmath23 cited above , the sizes of the corrected test are 5.9% and 5.2% , respectively , both close to the 5% nominal level . the second test problem we consider is about the equality between two high - dimensional covariance matrices . let @xmath24 and @xmath25 @xmath26 be observations from two @xmath0-dimensional normal populations @xmath27 , respectively . we wish to test the null hypothesis @xmath28 the related sample covariance matrices are @xmath29 where @xmath30 , @xmath31 are the respective sample means . let @xmath32 where @xmath33 and @xmath34 denote @xmath35 the likelihood ratio test statistic is @xmath36 and when @xmath37 , we get @xmath38 under @xmath39 . of cause , in this limit scheme , the data dimension @xmath0 is held fixed . however , employing this @xmath1 limit distribution for dimensions like 30 or 40 , increases dramatically the size of the test . for instance , simulations in [ sec : simul2 ] show that , for dimension and sample sizes @xmath40 , the test size equals 21.2% instead of the nominal 5% level . the result is worse for the case of @xmath41 , leading to a 49.5% test size . the reason for this fail of classical lr test is the following . modern rmt indicates that when both dimension and sample size are large , the likelihood ratio statistic @xmath42 drifts to infinity almost surely . therefore , the classical @xmath1 approximation leads to many false rejections of @xmath19 in case of high - dimensional data . based on recent clt for linear spectral statistics of @xmath43-matrices from rmt , we propose a correction to this lr test in [ sec : test2 ] . although this corrected test is constructed under the asymptotic scheme @xmath44 , @xmath45 , @xmath46 , simulations demonstrate an overall correct behavior including small or moderate dimensions @xmath0 . for example , for the above cited dimension and sample sizes @xmath47 , the sizes of the corrected test equal 5.6% and 5.2% , respectively , both close to the nominal 5% level . related works include @xcite , @xcite and @xcite . these authors propose several procedures in the high - dimensional setting for testing that i ) a covariance matrix is an identity matrix , proportional to an identity matrix ( spherecity ) and is a diagonal matrix or ii ) several covariance matrices are equal . these procedures have the following common feature : their construction involves some well - chosen distance function between the null and the alternative hypotheses and rely on the first two spectral moments , namely the statistics tr@xmath48 and tr@xmath49 from sample covariance matrices @xmath48 . therefore , the procedures proposed by these authors are different from the likelihood - based procedures we consider here . another important difference concerns the gaussian assumption on the random variables used in all these references . actually , for testing the equality between two covariance matrices , the correction proposed in this paper applies equally for non - gaussian and high - dimensional data leading to a valid pseudo - likelihood test . the rest of the paper is organized as following . preliminary and useful rmt results are recalled in [ sec : useful ] . in [ sec : test1 ] and [ sec : test2 ] , we introduce our results for the two tests above . proofs and technical derivations are postponed to the last section . we first recall several results from rmt , which will be useful for our corrections to tests . for any @xmath50 square matrix @xmath51 with real eigenvalues @xmath52 , @xmath53 denotes the empirical spectral distribution ( esd ) of @xmath51 , that is , @xmath54 we will consider random matrix @xmath51 whose esd @xmath53 converges ( in a sense to be precised ) to a limiting spectral distribution ( lsd ) @xmath55 . to make statistical inference about a parameter @xmath56 , it is natural to use the estimator @xmath57 which is a so - called linear spectral statistic ( lss ) of the random matrix @xmath51 . let @xmath58 be a double array of @xmath59 complex variables with mean 0 and variance 1 . set @xmath60 , the vectors @xmath61 is considered as an @xmath62 sample from some @xmath0-dimensional distribution with mean @xmath63 and covariance matrix @xmath7 . therefore the sample covariance matrix is @xmath64 for @xmath65 , let @xmath66 and @xmath67 . the marenko - pastur distribution of index @xmath68 , denoted as @xmath69 , is the distribution on @xmath70 $ ] with the following density function @xmath71[x - a(\theta)]},\quad a(\theta)\le x\le b(\theta).\ ] ] assume that @xmath83 , and @xmath84 are @xmath59 random variables , such that @xmath85 moreover , @xmath86 as @xmath87 + then : + _ ( i ) _ real case . assume @xmath84 are real and @xmath88 then the random vector @xmath89 weakly converges to a @xmath90-dimensional gaussian vector with mean vector , @xmath91 and covariance function @xmath92 + where @xmath93 is the stieltjes transform of @xmath94 . the contours in ( [ 04var ] ) are non overlapping and both contain the support of @xmath95 . + _ ( ii ) _ complex case . assume @xmath84 are complex and @xmath96 , @xmath97 . then the conclusion of _ ( i ) _ also holds , except the mean vector is zero and the covariance function is half of the function given in ( [ 04var]).[t1.1 ] it is worth noticing that theorem 1.1 in @xcite covers more general sample covariance matrices of form @xmath98 where @xmath99 is a given sequence of positive - definite hermitian matrices . in the `` white '' case @xmath100 as considered here , in a recent preprint @xcite , the authors offer a new extension of the clt where the constraints @xmath101 or 2 , as stated above , are removed . let @xmath58 and @xmath102 are two independent double arrays of @xmath59 complex variables with mean 0 and variance 1 . write @xmath103 and @xmath104 . also , for any positive integers @xmath105 , the vectors @xmath106 and @xmath107 can be thought as independent samples of size @xmath108 and @xmath109 , respectively , from some @xmath0-dimensional distributions . let @xmath110 and @xmath111 be the associated sample covariance matrices , @xmath112 @xmath113 then , the following so - called f - matrix generalizes the classical fisher - statistics for the present @xmath0-dimensional case , @xmath114 where @xmath115 . here we use the notation @xmath116 . let @xmath117 under suitable moment conditions , the esd @xmath118 of @xmath119 has a lsd @xmath120 , which has a density [ see p72 of @xcite ] , given by @xmath121 & \displaystyle{0 , \quad\quad \quad\quad \mbox{otherwise}. } \end{array } \right.\label{lsdden}\ ] ] where @xmath122 recently , @xcite establishes a general clt for lss of large - dimensional f matrix . the following theorem is a simplified one quoted from it , which will play an important role . let @xmath132 , and assume : + for each p , @xmath133 and @xmath134 variables are @xmath135 , @xmath136 @xmath137 @xmath138 @xmath139 + then + _ ( i ) _ real case . assume @xmath140 and @xmath141 are real , @xmath142 , then the random vector @xmath143 weakly converges to a k - dimensional gaussian vector with the mean vector @xmath144\nonumber\\ & & \quad\frac{1}{4\pi i}\oint_{|\zeta|=1 } f_j(z(\zeta))\left[\frac{1}{\zeta-{1\over r}}+\frac{1}{\zeta+{1\over r}}-\frac{2}{\zeta+{y_2\over { hr}}}\right]d\zeta \label{e1}\\ & & \quad + \frac{\beta\cdot y_1(1-y_2)^2}{2\pi i \cdot h^2}\oint_{|\zeta|=1}f_j(z(\zeta))\frac{1}{(\zeta+\frac{y_2}{hr})^3}d\zeta\label{e1betax}\\ & & \quad + \frac{\beta\cdot y_2(1-y_2)}{2 \pi i \cdot h}\oint_{|\zeta|=1}f_j(z(\zeta))\frac{\zeta + \frac{1}{hr}}{(\zeta+\frac{y_2}{hr})^3}d\zeta , \label{e1betay } \quad\quad j=1 , \cdots , k , \end{aligned}\ ] ] where @xmath145 , \quad h = \sqrt{y_1+y_2-y_1y_2 } , $ ] @xmath146 and the covariance function as @xmath147 @xmath148}\nonumber\\ & & -\displaystyle\frac{1}{2\pi^2}\oint_{|\zeta_2|=1 } \oint_{|\zeta_1|=1}\frac{f_j(z(r_1\zeta_1))f_\ell(z(r_2\zeta_2))r_1r_2}{(r_2\zeta_2-r_1\zeta_1)^2 } d\zeta_1d\zeta_2,\label{cov1}\\ & & -\frac{\beta \cdot ( y_1+y_2)(1-y_2)^2}{4\pi^2h^2 } \oint_{|\zeta_1|=1}\frac{f_j\left ( z(\zeta_1 ) \right)}{(\zeta_1+\frac{y_2}{hr_1})^2}d\zeta_1 \oint_{|\zeta_2|=1}\frac{f_\ell\left ( z(\zeta_2 ) \right)}{(\zeta_2+\frac{y_2}{hr_2})^2}d\zeta_2 \label{cov1betax}\\ & & j , \ell \in \{1,\cdots , k\}.\nonumber \end{aligned}\ ] ] _ ( ii ) _ complex case . assume @xmath140 and @xmath141 are complex , @xmath149 then the conclusion of _ ( i ) _ also holds , except the means are @xmath150 $ ] and the covariance function is @xmath151 , $ ] where @xmath152 [ t2.1 ] we should point out that zheng s clt for @xmath43-matrices covers more general situations then those cited in theorem [ t2.1 ] . in particular , the fourth - moments @xmath153 and @xmath154 can be different . the following lemma will be used in [ sec : test2 ] for an application of theorem [ t2.1 ] to obtain the formula ( [ teste ] ) and ( [ testvar ] ) . for the function @xmath155 , let @xmath156 be the unique solution to the equations @xmath157 analogously , let @xmath158 be the constants similar to @xmath159 but for the function @xmath160 then , the mean and covariance functions in ( [ e1 ] ) and ( [ cov1 ] ) equal to @xmath161 [ lem1 ] to test the hypothesis @xmath162 , let be the sample covariance matrix * s * and likelihood ratio statistic @xmath17 as defined in ( [ s ] ) and ( [ singlvclassict ] ) , respectively . for @xmath163 the array @xmath164 contains @xmath0-dimensional standard normal variables under @xmath19 . let @xmath165 and @xmath166 assuming that the conditions of theorem [ t1.1 ] hold , @xmath167 is defined as ( [ l * ] ) and @xmath168 . then , under @xmath19 and when @xmath169 @xmath170 \rightarrow n \left ( 0 , 1\right ) , \label{singlsta}\ ] ] where @xmath171 is the marenko - pastur law of index @xmath172 . because the difference between @xmath173 and @xmath174 is a rank-1 matrix , @xmath173 and @xmath175 have the same lsd . so , @xmath167 and @xmath176 have the same asymptotic distribution . we also have @xmath177 so that @xmath178 by theorem [ t1.1 ] , @xmath179 weakly converges to a gaussian vector with the mean @xmath180 and variance @xmath181 for the real case , which are calculated in [ sec : proofs ] . for the complex case , the mean @xmath182 is zero and the variance is half of @xmath183 then , by ( [ singlesd - plsd ] ) we arrive at @xmath184 where @xmath185 can be calculated by the density of lsd of sample covariance matrix in [ sec : proofs ] . because @xmath176 and @xmath186 have the same asymptotic distribution and ( [ 1connec ] ) , finally we get @xmath187 \rightarrow n \left ( 0 , 1\right).\nonumber \end{aligned}\ ] ] for different values of @xmath23 , we compute the realized sizes of traditional likelihood ratio test ( lrt ) and the corrected likelihood ratio test ( clrt ) proposed previously . the nominal test level is set to be @xmath188 , and for each @xmath189 , we run 10,000 independent replications with real gaussian variables . results are given in table 1 and figure 1 below . .sizes and powers of the traditional lrt and the corrected lrt , based on 10,000 independent applications with real gaussian variables . powers are estimated under the alternative @xmath190 . [ cols="<,^,^,^,^,^ " , ] with real gaussian variables . 10 000 independent runs with 5% nominal level and sample size @xmath191.[1],title="fig:",width=340 ] + as seen from table 1 , the traditional lrt always rejects @xmath19 when @xmath0 is large , like @xmath192 or 300 , while the sizes produced by the corrected lrt perfectly matches the nominal level . for moderate dimensions like @xmath193 , the corrected lrt still performs correctly while the traditional lrt has a size much higher than 5% . let @xmath194 and @xmath195 be observations from two normal populations @xmath27 , respectively . we examine the test defined in ( [ a0 ] ) and ( [ l1ab ] ) . the aim is to find a good scaling of the lr statistic @xmath42 , such that the scaled statistic weakly converges to some limiting distribution . let @xmath196 where @xmath197 denotes the common covariance matrix under @xmath19 . note that in a strict sense , the vectors @xmath198 and the matrices @xmath199 depend on @xmath0 . however we do not signify this dependence in notations for ease of statements . due to gaussian assumption , the arrays @xmath200 and @xmath201 contain i.i.d . @xmath202 variables , for which we can apply theorem [ t2.1 ] . let @xmath203 where @xmath204 note that @xmath205 forms a random f - matrix and we have @xmath206 [ t4.1 ] assuming that the conditions of theorem [ t2.1 ] hold under @xmath19 , @xmath207 as defined in ( [ l1ab ] ) and @xmath208 then , under @xmath19 and as @xmath209 , @xmath210 \rightarrow n \left ( 0 , 1\right).\label{lst}\ ] ] as @xmath211 and @xmath212 are rank-1 random matrices , @xmath213 and @xmath214 have the same lsd . also by ( [ s1s2cd ] ) , @xmath215 and @xmath216 have the same asymptotic distribution . because @xmath217df_n^{v_n}(x ) . \end{aligned}\ ] ] define @xmath218 , by @xmath219 and @xmath220 , also it can be written as @xmath221 from @xmath222 we get @xmath223 by theorem [ t2.1 ] , @xmath224 weakly converges to a gaussian vector with mean @xmath225\label{teste}\ ] ] and variance @xmath226 for the real case , which are calculated by lemma [ lem1 ] in [ sec : proofs ] . for the complex case , the mean @xmath227 is zero and the variance is half of @xmath228 . in other words , @xmath229 where @xmath230 \quad & + & \displaystyle{\frac{(y_{n_1}+y_{n_2}-y_{n_1}y_{n_2})}{y_{n_1}y_{n_2}}\log{(y_{n_1}+y_{n_2 } ) } } + \displaystyle{\frac{y_{n_1}(1-y_{n_2})}{y_{n_2}(y_{n_1}+y_{n_2})}\log{(1-y_{n_2})}}\nonumber\\ \quad\quad\quad&+&\displaystyle{\frac{y_{n_2}(1-y_{n_1})}{y_{n_1}(y_{n_1}+y_{n_2})}\log{(1-y_{n_1})}},\nonumber \label{limit } \end{aligned}\ ] ] is derived by use of the density of @xmath129 in [ sec : proofs ] . because @xmath215and @xmath216 have the same asymptotic distribution and by ( [ 2connec ] ) , we get by letting @xmath231 , @xmath232 \rightarrow n \left ( 0 , 1\right).\ ] ] for different values of @xmath233 , we compute the realized sizes of the traditional lrt and the corrected lrt with 10,000 independent replications . the nominal test level is @xmath234 and we use real gaussian variables . results are summarized in table 2 and figure 2 . & & + ( p , @xmath235 , @xmath236 ) & size & difference with 5% & power & size & power + ( 5 , 100 , 100 ) & 0.0770&0.0270&1 & 0.0582&1 + ( 10 , 200 , 200 ) & 0.0680&0.0180&1 & 0.0684&1 + ( 20 , 400 , 400 ) & 0.0593&0.0093&1 & 0.0872&1 + ( 40 , 800 , 800 ) & 0.0526&0.0026&1 & 0.1339&1 + ( 80 , 1600 , 1600 ) & 0.0501 & 0.0001&1&0.2687&1 + ( 160 , 3200 , 3200 ) & 0.0491 & -0.0009&1&0.6488&1 + ( 320 , 6400 , 6400 ) & 0.0447 & -0.0053&0.9671&1&1 + + & & + ( p , @xmath235 , @xmath236 ) & size & difference with 5% & power & size & power + ( 5 , 100 , 50 ) & 0.0781&0.0281&0.9925 & 0.0640&0.9849 + ( 10 , 200 , 100 ) & 0.0617&0.0117&0.9847 & 0.0752&0.9904 + ( 20 , 400 , 200 ) & 0.0573&0.0073&0.9775 & 0.1104&0.9938 + ( 40 , 800 , 400 ) & 0.0561&0.0061&0.9765 & 0.2115&0.9975 + ( 80 , 1600 , 800)&0.0521 & 0.0021&0.9702 & 0.4954&0.9998 + ( 160 , 3200 , 1600 ) & 0.0520&0.0020&0.9702&0.9433&1 + ( 320 , 6400 , 3200 ) & 0.0510 & 0.0010&1&0.9939&1 + . right : @xmath237.[fig:2],title="fig:",width=264 ] . right : @xmath237.[fig:2],title="fig:",width=264 ] as we can see , when the dimension @xmath0 increases , the traditional lrt leads to a dramatically high test size while the corrected lrt remains accurate . furthermore , for moderate dimensions like @xmath238 or 40 , the sizes of the traditional lrt are much higher than 5% , whereas the ones of corrected lrt are very close . by a closer look at the column showing the difference with 5% , we note that this difference rapidly decrease as @xmath0 increases for the corrected test . figure 2 gives a vivid sight of these comparisons between the traditional lrt and the corrected lrt in term of test sizes . as said in introduction , previous related works as @xcite , @xcite or @xcite all assume gaussian variables . in contrast , theorem [ t4.1 ] applies for general distributions having a fourth moment . for these non gaussian data , we consider the corrected lrt as generalized pseudo - likelihood ratio test ( or gaussian lrt ) . moreover , the methods proposed by these authors all rely on an appropriate normalization of the trace of squared difference between two sample covariances following the idea of @xcite . we believe that their method would strongly depend on the normality assumption ( which was supported by simulation results below ) . on the other hand , based on general understanding , the lrt contains much higher information from data and its poor performance observed up to now is just caused by its large bias when dimension is large . thus , from the intuitive understanding , we are confined ourselves to modify the lrt . let us develop in more details an example . assume that * x * follows a normalized @xmath239-distribution with 5 degree of freedom , that is @xmath240 * x * and * y * are i.i.d . , hence @xmath241 @xmath242 and @xmath243 we still employ the result in theorem [ t4.1 ] for the test of equality between two covariance matrices , where @xmath244 & \quad\quad\quad\quad\quad\quad\quad\quad + \displaystyle{\frac{6y_1 ^ 2y_2}{(y_1+y_2)^2}+\frac{6y_1y_2 ^ 2}{(y_1+y_2)^2}\big]}\label{m1f}\end{aligned}\ ] ] and @xmath245 instead of @xmath227 and @xmath228 for real case , respectively . ( [ m1f ] ) and ( [ v1f ] ) are calculated in [ sec : proofs ] . the following table [ tab : t ] summarizes a simulation study where we compare this corrected pseudo - lrt with the test proposed in @xcite . we use 1,000 independent replications with the above @xmath239-distributed variables . again , the nominal test level is @xmath234 . as we can see , the corrected pseudo - lrt performs correctly while schott s test is no more valid here since the variables are not gaussian . & clrt size & schott s size + ( 10,100 , 200 ) & 0.067&0.517 + ( 20 , 200 , 400 ) & 0.065&0.603 + ( 40 , 400 , 800 ) & 0.054&0.703 + ( 80 , 800 , 1600)&0.048 & 0.764 + ( 160 , 1600 , 3200 ) & 0.045&0.826 + ( 320 , 3200 , 6400 ) & 0.051&0.854 + by theorem [ t1.1 ] , for @xmath168 , by using the variable change @xmath246 , we have @xmath247d\theta\\ & = & \frac{y-\log(1-y)}{2}-\frac{1}{4\pi}\int_{0}^{2\pi } \left[y-2\sqrt{y}\cos\theta-\log|1-\sqrt{y}e^{i \theta}|^2\right]d\theta\\ & = & -\frac{\log(1-y)}{2},\end{aligned}\ ] ] where @xmath248 is calculated in @xcite . for @xmath249 , by theorem [ t1.1 ] , we have @xmath250 and @xmath251 it is easy to see that @xmath252 , where * 1 * means constant function equals to 1 . for stieltjes transform of @xmath95 , the following equation is given in @xcite , for @xmath253 , @xmath254 + let @xmath255 . for fixed @xmath256 , we have on a contour enclosed 1 , @xmath257 and -1 , but not 0 , @xmath258 and @xmath259^{-1 } \cdot ( m_{2}+1)^{-2 } \cdot ( 1-\frac{m_{1}+1}{m_{2}+1})^{-2 } dm_{1}}\\ & = & \displaystyle { y\oint ( \frac{1}{1+m_{1}}+\frac{1-y}{y } ) \cdot \sum\limits^\infty_{j=0}(1+m_1)^j ( m_2 + 1)^{-2 } \sum\limits^\infty_{\ell=1}\ell(\frac{m_{1}+1}{m_{2}+1})^{\ell-1}dm_1}\\ & = & \displaystyle{2\pi i\cdot \frac{y}{(m_2 + 1)^{2}}}.\end{aligned}\ ] ] then we also get @xmath260 . similarly , @xmath261 . furthermore , @xmath262 and @xmath263^{-1 } dm_2}\\ & = & \displaystyle { \frac{y}{\pi i } \oint ( \frac{1}{m_2 + 1}- \frac { 1}{m_2 - 1/(y-1)})(\frac{1}{1+m_{2}}+\displaystyle\frac{1-y}{y } ) \sum\limits^\infty_{j=0}(1+m_2)^j dm_2 } \\ & = & 2y.\end{aligned}\ ] ] by a computation in @xcite , we know that @xmath264 . finally , we obtain @xmath265 since @xmath171 is the marenko - pastur law of index @xmath266 , by using the variable change @xmath267 we have @xmath2684y_n\sin^2\theta d\theta\\ & = & \frac{1}{2\pi}\int_0^{2\pi}\left[2\sin^2\theta-\frac{2\sin^2\theta}{1+y_n-2\sqrt{y_n}\cos\theta } \left(\log |1-\sqrt{y_n}e^{i\theta}|^2 - 1\right)\right]d \theta\\ & = & 1-\frac{y_n-1}{y_n}\log ( 1-y_n),\end{aligned}\ ] ] where @xmath269 is calculated in @xcite . we use the variable change @xmath270 , where @xmath271 . when @xmath272 satisfy @xmath273 we have @xmath274 similarly , @xmath275 let @xmath276 note that @xmath277 . by theorem [ t2.1 ] , we have @xmath278d\xi}\nonumber\\ & = & \displaystyle{\frac{1}{4\pi}\int^{2\pi}_{0 } f(z(e^{i\theta}))\left[\frac{1}{e^{i\theta}-{1\over r } } + \frac{1}{e^{i\theta}+{1\over r}}-\frac{2}{e^{i\theta}+{y_2\over { hr}}}\right]e^{i\theta}d\theta}\nonumber\\ & = & \displaystyle{\frac{1}{4\pi}\int^{2\pi}_{0 } f(z(e^{i\theta}))\left[\frac{1}{e^{-i\theta}-{1\over r } } + \frac{1}{e^{-i\theta}+{1\over r}}-\frac{2}{e^{-i\theta}+{y_2\over { hr}}}\right]e^{-i\theta}d\theta}\nonumber\\ & = & \displaystyle\frac{1}{8\pi}\int^{2\pi}_{0 } f(z(e^{i\theta}))\bigg\{\left[\frac{1}{e^{i\theta}-{1\over r } } + \frac{1}{e^{i\theta}+{1\over r}}-\frac{2}{e^{i\theta}+{y_2\over { hr}}}\right]e^{i\theta}+\nonumber\\ & & \left[\frac{1}{e^{-i\theta}-{1\over r}}+\frac{1}{e^{-i\theta}+{1\over r } } -\frac{2}{e^{-i\theta}+{y_2\over { hr}}}\right]e^{-i\theta}\bigg\}d\theta\nonumber\\ & = & \frac{1}{8\pi}\re\bigg\{\int^{2\pi}_{0 } \widetilde{f}(z(e^{i\theta}))\bigg[\left(\frac{1}{e^{i\theta}-{1\over r } } + \frac{1}{e^{i\theta}+{1\over r}}-\frac{2}{e^{i\theta}+{y_2\over { hr}}}\right)e^{i\theta}+\nonumber\\ & & \left(\frac{r}{r - e^{i\theta}}+\frac{r}{r+e^{i\theta } } -\frac{2hr}{y_2e^{i\theta}+hr}\right)\bigg]d\theta\bigg\}\nonumber\\ & = & \re\bigg\{\frac{1}{8\pi i}\oint_{|\xi|=1 } \widetilde{f}(z(\xi))\bigg[\left(\frac{1}{\xi-{1\over r}}+\frac{1}{\xi+{1\over r}}-\frac{2}{\xi+{y_2\over { hr } } } \right)\nonumber\\ & & + \left(\frac{r}{r-\xi}+\frac{r}{r+\xi } -\frac{2hr}{y_2\xi+hr}\right)\xi^{-1}\bigg]d\xi\bigg\}\nonumber\\ & = & \frac{1}{4}\left(\widetilde{f}(z(\frac{1}{r}))+\widetilde{f}(z(-\frac{1}{r } ) ) -2\widetilde{f}(z(-\frac{y_2}{hr}))\right)\nonumber\\ & \rightarrow & ^{r\downarrow1}\frac{1}{4}\left[\widetilde{f}(z(1))+\widetilde{f}(z(-1))-2\widetilde{f}(z(-\frac{y_2}{h}))\right]\nonumber\\ & = & \frac{1}{2}\log\frac{(c^2-d^2)h^2}{(ch - y_2d)^2}.\nonumber\end{aligned}\ ] ] let @xmath279 , where @xmath280 @xmath281 and @xmath282 . by theorem [ t2.1 ] , we have @xmath283 when @xmath284 are poles . we can then choose @xmath285 so that @xmath286 is a not a pole . then we get @xmath287 & = & \displaystyle{\oint_{|\xi_1|=1}\frac{\left(\log(a+bz(r_1\xi_1))\right)'}{r_1\xi_1-r_2\xi_2}\cdot r_2 d\xi_1}\\[3 mm ] & = & \displaystyle{\oint_{|\xi_1|=1}\bigg[\frac{bhr_1\xi_1}{(r_1\xi_1-r_2\xi_2)(c+dr_1\xi_1)c}\cdot \frac{1}{\xi_1+\frac{d}{cr_1}}}\\ & & \quad-\frac{bhr_1^{-1}}{(r_1\xi_1-r_2\xi_2)(c+dr_1\xi_1)c}\cdot \frac{1}{(\xi_1+\frac{d}{cr_1})\xi_1}\cdot r_2\bigg]d\xi_1\\[5 mm ] & = & \displaystyle{2\pi i}\left ( \frac{bhd^{-1}c^{-1}}{\xi_2}-\frac{bhd^{-1}r_2}{d+cr_2\xi_2}\right).\end{aligned}\ ] ] so , @xmath288 since the function @xmath289 is analytic , when @xmath290 but sufficiently close to 1 , we have @xmath291 for some constant @xmath292 . thus we have @xmath293 \left(\frac{bhd^{-1}c^{-1}}{\xi_2}-\frac{bhd^{-1}r_2}{d+cr_2\xi_2}\right)d\xi_2\right|}\\[4 mm ] \rightarrow & 0 \quad \mbox{as } \quad r_2 \downarrow 1 , \end{array}\ ] ] where the estimations are done according to@xmath294 or @xmath295 or not . thus , @xmath296 where @xmath297 because @xmath298 for @xmath158 satisfying @xmath299 and if @xmath300 we have @xmath301 . therefore , @xmath302d\theta\\ & = & \frac{1}{2\pi}\re\bigg\{\int_{0}^{2\pi}\widetilde{g}(z(e^{i\theta } ) ) \bigg[\left(\frac{bhd^{-1}c^{-1}}{e^{i\theta}}-\frac{bhd^{-1}r_2}{d+cr_2e^{i\theta}}\right ) e^{i\theta } + bhd^{-1}c^{-1}-\frac{bhd^{-1}r_2}{de^{i\theta}+cr_2}\bigg]d\theta\bigg\}\\ & = & \re\bigg\{\frac{1}{2\pi i}\oint_{|\xi|_2=1}\widetilde{g}(z(\xi_2))\bigg[\left(\frac{bhd^{-1}c^{-1}}{\xi_2}-\frac{bhd^{-1}r_2}{d+cr_2\xi_2}\right ) + \left(bhd^{-1}c^{-1}-\frac{bhd^{-1}r_2}{d\xi_2+cr_2}\right)\xi_2^{-1}\bigg]d\xi_2\bigg\}\\ & = & bhd^{-1}c^{-1}\left[\widetilde{g}(z(0))-\widetilde{g}(z(-\frac{d}{cr_2}))\right]\\ & \rightarrow & bhd^{-1}c^{-1}\left[\widetilde{g}(z(0))-\widetilde{g}(z(-\frac{d}{c}))\right]\\ & = & 2bhd^{-1}c^{-1}\log\frac{c\gamma}{c\gamma - d\eta}.\end{aligned}\ ] ] because * @xmath303 * and * @xmath304 * are gaussian variables , for real case , @xmath305**@xmath303**@xmath306 then ( [ e1betax ] ) , ( [ e1betay ] ) and ( [ cov1betax ] ) are all 0 . consider ( [ e1 ] ) and ( [ cov1 ] ) , as @xmath307 , by the computations done in the proof of lemma [ lem1 ] , we see that termes tending to zero could be neglected in the considered contour integrals . hence we can put @xmath308 and use @xmath309 instead of @xmath310 . consider the variable change @xmath270 , where @xmath311 , \quad h = \sqrt{y_1+y_2-y_1y_2}$ ] . as @xmath312 we have by lemma [ lem1 ] , @xmath313}\\ & = & \displaystyle{\frac{1}{2}\left[\log\left(\frac{y_1+y_2-y_1y_2}{y_1+y_2}\right ) -\frac{y_1}{y_1+y_2}\log(1-y_2)-\frac{y_2}{y_1+y_2}\log(1-y_1)\right]},\end{aligned}\ ] ] and @xmath314 by ( [ f(x ) ] ) and the density of @xmath315 ( the limiting distribution in ( [ lsdden ] ) but with @xmath130 in place of @xmath131 ) , where @xmath316 , @xmath317 and @xmath318 using the substitution @xmath319 we have @xmath320 therefore , @xmath321 & = & { \int_{a_n}^{b_n}f(x)\frac{(1-y_{n_2})\sqrt{(b_n - x)(x - a_n)}}{2\pi x(y_{n_1}+y_{n_2}x)}dx}\\[6 mm ] & = & { ( 1-y_{n_2})\int_{a_n}^{b_n } \left[\log\left(y_{n_1}+y_{n_2}x\right ) -\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\log x\right]\frac{\sqrt{(b_n - x)(x - a_n)}}{2\pi x(y_{n_1}+y_{n_2}x)}dx}\\ & & \quad\quad-{\log\left(y_{n_1}+y_{n_2}\right)}\\[6 mm ] & = & { \frac{2(1-y_{n_2})}{\pi}\int_{0}^{\pi}\left [ \log\frac{\left|h_n - y_{n_2}e^{i\theta}\right|^2}{(1-y_{n_2})^2 } -\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\log\frac{\left|1-h_ne^{i\theta}\right|^2 } { ( 1-y_{n_2})^2}\right]}\\[6 mm ] & & \quad\quad { \cdot\frac{h_n^2\sin^2\theta}{\left|1-h_ne^{i\theta}\right|^2 \left|h_n - y_{n_2}e^{i\theta}\right|^2}d\theta}- { \log\left(y_{n_1}+y_{n_2}\right)}\\[6 mm ] & = & { \frac{2(1-y_{n_2})}{\pi}\int_{0}^{\pi } \left [ \log\left|h_n - y_{n_2}e^{i\theta}\right|^2 -\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\log\left|1-h_ne^{i\theta}\right|^2 \right ] } \\[4 mm ] & & \cdot{\frac{h_n^2\sin^2\theta } { \left|1-h_ne^{i\theta}\right|^2\left|h_n - y_{n_2}e^{i\theta}\right|^2}d\theta } -2\left(1-\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\right ) \log(1-y_{n_2})-\log{(y_{n_1}+y_{n_2})}\\ & = & { \re\bigg\{\frac{2(1-y_{n_2})}{\pi}\int_{0}^{2\pi } \left [ \log(h_n - y_{n_2}e^{i\theta } ) -\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\log(1-h_ne^{i\theta})\right]}\\ & & { \frac{h_n^2\sin^2\theta}{\left|1-h_ne^{i\theta}\right|^2 \left|h_n - y_{n_2}e^{i\theta}\right|^2}d\theta\bigg\}}-\frac{2y_{n_1}}{y_{n_1}+y_{n_2 } } \log(1-y_{n_2})-\log{(y_{n_1}+y_{n_2})}\\[5 mm ] & = & { \re\bigg\{\frac{-(1-y_{n_2})}{2\pi i}\oint_{|z|=1 } \left[\log(h_n - y_{n_2}z)-\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\log(1-h_nz)\right]}\\[5 mm ] & & \cdot{\frac{h_n^2(z - z^{-1})^2}{z\left|1-h_nz\right|^2 \left|h_n - y_{n_2}z\right|^2}dz\bigg\}}-{\frac{2y_{n_1}}{y_{n_1}+y_{n_2 } } \log(1-y_{n_2})-\log{(y_{n_1}+y_{n_2})}}\\[5 mm ] & = & { \re\bigg\{\frac{y_{n_2}-1}{y_{n_2}}\cdot\frac{1}{2\pi i}\oint_{|z|=1}\left [ \log(h_n - y_{n_2}z)-\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\log(1-h_nz)\right]}\\[6 mm ] & & \cdot{\frac{(z^2 - 1)^2}{z(z - h_n)(z-\frac{1}{h_n } ) ( z-\frac{y_{n_2}}{h_n})(z-\frac{h_n}{y_{n_2}})}dz}\bigg\ } -{\frac{2y_{n_1}}{y_{n_1}+y_{n_2 } } \log(1-y_{n_2})-\log{(y_{n_1}+y_{n_2})}}.\end{aligned}\ ] ] there are three poles inside the unit circle : 0 , @xmath322 . their corresponding residues are @xmath323,\nonumber\\ r(\frac{y_{n_2}}{h_n})&= & \frac{(y_{n_2}^2-h_n^2)}{y_{n_2}(y_{n_2}-h_n^2 ) } \left[\log ( h^2_n - y_{n_2}^2)-\log ( h_n)-\frac{y_{n_2}}{y_{n_1}+y_{n_2}}\log ( 1-y_{n_2 } ) \right].\nonumber\end{aligned}\ ] ] therefore , @xmath324 & & \quad + \displaystyle{\frac{(y_{n_1}+y_{n_2}-y_{n_1}y_{n_2})}{y_{n_1}y_{n_2}}\log{(y_{n_1}+y_{n_2 } ) } } + \displaystyle{\frac{y_{n_1}(1-y_{n_2})}{y_{n_2}(y_{n_1}+y_{n_2})}\log{(1-y_{n_2})}}\nonumber\\ & & \quad\quad\quad+\displaystyle{\frac{y_{n_2}(1-y_{n_1})}{y_{n_1}(y_{n_1}+y_{n_2})}\log{(1-y_{n_1})}}.\nonumber\end{aligned}\ ] ] because * x * and * y * are random variables from normalized @xmath239-distribution with 5 degree of freedom , * x * and * y * are @xmath59 , @xmath241 @xmath242 and @xmath243 for real case , @xmath305**@xmath303**@xmath325 ( [ e1 ] ) and ( [ cov1 ] ) items are the same to the gaussian variables . consider the items ( [ e1betax ] ) , ( [ e1betay ] ) and ( [ cov1betax ] ) . as the same explanation in proof of ( [ teste ] ) and ( [ testvar ] ) , we use @xmath326 instead . for ( [ e1betax ] ) , we have @xmath327\\ & & \quad\quad\quad\quad\quad\quad \quad\quad\quad\quad\quad\quad \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad \quad\quad\quad\cdot \frac{1}{(\xi+\frac{y_2}{hr})^3 } d\xi\\ & = & \frac{\beta\cdot y_1(1-y_2)^2}{2\pi i \cdot h^2 } \oint_{|\xi|=1 } 2 \mathcal{r}\big\{\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\big\}\cdot \frac{1}{(\xi+\frac{y_2}{hr})^3 } d\xi\\ & = & \frac{\beta\cdot y_1(1-y_2)^2}{2\pi i \cdot h^2 } \oint_{|\xi|=1}\bigg\{\log(h+y_2\xi)+\log(h+y_2\overline{\xi})\\ & & \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad -\frac{y_2}{y_1+y_2}\big[\log(1+h\xi)+\log(1+h\overline{\xi})\big]\bigg\ } \cdot\frac{1}{(\xi+\frac{y_2}{hr})^3 } d\xi\\ & = & \frac{\beta\cdot y_1(1-y_2)^2}{2\pi i \cdot h^2 } \oint_{|\xi|=1}\left[\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\right ] \\ & & \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad \quad\quad\quad\quad\quad\quad\quad\quad\quad \left[\frac{1}{(\xi+\frac{y_2}{hr})^3 } + \frac{\left(\frac{hr}{y_2}\right)^3\xi}{(\frac{hr}{y_2}+\xi)^3 } \right ] d\xi\\ & = & \frac{\beta\cdot y_1(1-y_2)^2}{2\pi i \cdot h^2 } \cdot 2\pi \cdot \frac{1}{2 } \left[\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\right]''\bigg|_{\xi=-\frac{y_2}{hr}}\\ & = & \frac{\beta\cdot y_1(1-y_2)^2}{2 h^2 } \left[-\frac{y_2 ^ 2}{(h+y_2\xi)^2}+\frac{y_2}{y_1+y_2}\frac{h^2}{(1+h\xi)^2}\right]\bigg|_{\xi=-\frac{y_2}{hr}}\\ & = & \frac{\beta y_1 ^ 2y_2}{2(y_1+y_2)^2}.\end{aligned}\ ] ] for ( [ e1betay ] ) , we have @xmath328\\ & & \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\cdot \frac{\xi^2-\frac{y_2}{h^2r^2}}{(\xi+\frac{y_2}{hr})^2}\left [ \frac1{\xi-\frac{\sqrt{y_2}}{hr}}+\frac1{\xi+\frac{\sqrt{y_2}}{hr}}-\frac2{\xi+\frac{y_2}{hr}}\right]d\xi\\ & = & \frac{\beta\cdot ( 1-y_2)y_2}{2\pi i \cdot h } \oint_{|\xi|=1 } \left[\log\frac{|h+y_2\xi|^2}{(1-y_2)^2}-\frac{y_2}{y_1+y_2}\log\frac{|1+h\xi|^2}{(1-y_2)^2}-\log(y_1+y_2)\right]\\ & & \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\cdot \left[\frac{\xi+\frac{1}{hr}}{(\xi+\frac{y_2}{hr})^3}\right ] d\xi\\ & = & \frac{\beta\cdot ( 1-y_2)y_2}{2\pi i \cdot h } \oint_{|\xi|=1}2 \mathcal{r}\big\{\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\big\ } \cdot \left[\frac{\xi+\frac{1}{hr}}{(\xi+\frac{y_2}{hr})^3}\right ] d\xi\\ & = & \frac{\beta\cdot ( 1-y_2)y_2}{2\pi i \cdot h}\oint_{|\xi|=1}\big [ \log(h+y_2\xi)+\log(h+y_2\overline{\xi})\\ & & \quad\quad\quad\quad\quad\quad\quad\quad -\frac{y_2}{y_1+y_2}\left(\log(1+h\xi)+\log(1+h\overline{\xi})\right ) \big]\cdot \left[\frac{\xi+\frac{1}{hr}}{(\xi+\frac{y_2}{hr})^3}\right]d\xi\\ & = & \frac{\beta\cdot ( 1-y_2)y_2}{2\pi i \cdot h}\oint_{|\xi|=1}\left[\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\right]\\ & & \quad\quad\quad\quad\quad\quad\quad\cdot\big[\frac{\xi+\frac{1}{hr}}{(\xi+\frac{y_2}{hr})^3}+ \frac{\frac{h^2r^2}{y_2 ^ 3}(\xi+hr)}{(\xi+\frac{hr}{y_2})^3}\big]d\xi\\ & = & \frac{\beta\cdot ( 1-y_2)y_2}{2\pi i \cdot h}2\pi i \cdot \frac1{2}\left([\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi ) ] \cdot(\xi+\frac{1}{hr})\right)''\bigg|_{\xi=-\frac{y_2}{hr}}\\ & = & \frac{\beta y_1y_2 ^ 2}{2(y_1+y_2)^2}.\end{aligned}\ ] ] therefore , @xmath329.\nonumber\end{aligned}\ ] ] for covariance , we have @xmath330 \cdot \frac{1}{(\xi+\frac{y_2}{hr})^2 } d\xi\\ & = & \oint_{|\xi|=1 } 2 \mathcal{r}\big\{\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\big\}\cdot \frac{1}{(\xi+\frac{y_2}{hr})^2 } d\xi\\ & = & \oint_{|\xi|=1}\bigg\{\log(h+y_2\xi)+\log(h+y_2\overline{\xi})\\ & & \quad\quad -\frac{y_2}{y_1+y_2}\big[\log(1+h\xi)+\log(1+h\overline{\xi})\big]\bigg\ } \cdot\frac{1}{(\xi+\frac{y_2}{hr})^2 } d\xi\\ & = & \oint_{|\xi|=1}\left[\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\right ] \left[\frac{1}{(\xi+\frac{y_2}{hr})^2 } + \frac{\left(\frac{hr}{y_2}\right)^2}{(\xi+\frac{hr}{y_2})^2 } \right ] d\xi\\ & = & 2\pi i \cdot \left[\log(h+y_2\xi)-\frac{y_2}{y_1+y_2}\log(1+h\xi)\right]'\bigg|_{\xi=-\frac{y_2}{hr}}\\ & = & \pi i \cdot \left[\frac{y_2}{h+y_2\xi}+\frac{y_2}{y_1+y_2}\frac{h}{1+h\xi}\right]\bigg|_{\xi=-\frac{y_2}{hr}}\\ & = & 0.\end{aligned}\ ] ]

we present new results on the structure of the solar core , obtained with new sets of frequencies of solar low - degree p modes obtained from the bison network . we find that different methods used in extracting the different sets of frequencies cause shifts in frequencies , but the shifts are not large enough to affect solar structure results . we find that the bison frequencies show that the solar sound speed in the core is slightly larger than that inferred from data from mdi low - degree modes , and the uncertainties on the inversion results are smaller . density results also change by a larger amount , and we find that solar models now tend to show smaller differences in density compared to the sun . the result is seen at all radii , a result of the fact that conservation of mass implies that density differences in one region have to cancel out density differences in others , since our models are constructed to have the same mass as the sun . the uncertainties on the density results are much smaller too . we attribute the change in results to having more , and lower frequency , low - degree mode frequencies available . these modes provide greater sensitivity to conditions in the core .

0905.0651

helioseismology is now an extremely successful and proven technique , which can be used to infer the structure and dynamics of the interior sun ( see . eg . antia & basu 1994 ; gough et al . 1995 ; basu et al . 2000a , b ; basu 2003 ; couvidat et al . 2003 ; schou et al . 1998a ; pijpers 2006 ; howe 2008 , etc . ) . however , there is still considerable uncertainty about the structure of the solar core . different data sets can often give results that show discrepancies . basu et al . ( 2000b ) found that different solar sound - speed results were obtained depending on whether data from the michelson doppler imager ( mdi ) or the ground - based global oscillation network group ( gong ) were used . the differences in the inferred sound - speed in the solar core are illustrated in figure 2 of bahcall et al . the sensitivity of the inferred structure to the mode sets has been noted in several investigations ( see e.g. basu et al . 2000a , turck - chieze et al . 2001 , etc . ) the model fitted to mode peaks in the oscillation power spectrum , as part of the `` peak bagging '' procedures used to extract estimates of the mode frequencies , can also give rise to discrepant inferences . for example , toutain et al . ( 1998 ) found that sound - speed differences in the core were about 0.2% to 0.3% higher when they used frequencies obtained by fitting asymmetric profiles to the asymmetric peaks in the oscillation power spectrum , rather than what had been the usual practice of fitting symmetric ( lorentzian ) profiles . however , basu et al . ( 2000a ) claimed that the differences were caused by a few unreliable modes , and that the change in frequencies brought about by fitting the more accurate asymmetric profiles would not be expected to affect significantly the inversion results . so , although the structure of the sun , in particular the sound - speed profile of the outer 80% by radius , is known very well , the structure of the core is relatively uncertain . the basic reason for this uncertainty is that we only have frequency estimates of acoustic modes , or p modes . the p - modes that probe the inner 0.25@xmath0 of the sun are modes of low - degree @xmath1 , i.e. , with @xmath2 , 1 , 2 and 3 . all solar p - modes have their highest amplitudes close to the solar surface , and hence even modes that probe the core ( and there are relatively few of them ) are less sensitive to the conditions there than in the outer parts of the solar interior . frequencies of solar g - modes ( buoyancy modes ) would help reduce uncertainties , because they have their highest amplitudes in the core . however , g modes are extremely hard to detect , due to their small amplitudes at the solar surface ( see e.g. , appourchaux 2008 ; appourchaux et al . 2000 , 2006 ; gabriel et al . 2002 ; elsworth et al . there are recent reports of the possible detection of the signature of several g modes ( see e.g. , garca et al . 2008a , b ) . there are as yet no convincing candidates for individual g - mode frequencies . thus , for the moment , we still have to rely on p - modes to determine the structure of the solar core . the most reliable data on low - degree solar p - modes are obtained from specialized instruments that observe the sun as a star . the birmingham solar - oscillations network ( bison ; chaplin et al . 1996 ) is a network of such instruments . data from the network give precise estimates of the frequencies of the low - degree p modes . in this paper we use new sets of data from bison to re - examine the structure of the solar core . to extract information about the solar core properly through inversions of solar frequencies we also need the higher , intermediate - degree modes . inversions are possible using only low - degree modes , but the results have large errors and the resolution can be poor ( see e.g. , basu 2003 ) . while this may be acceptable for other stars , where for the foreseeable future the only modes we can hope to observe are low - degree modes , this is not good enough for the sun . unfortunately instruments that determine frequencies of low - degree modes precisely [ such as bison or golf ( gabriel et al . 1997 ) ] can not determine frequencies of intermediate - degree modes , and instruments that determine frequencies of intermediate - degree modes precisely ( such as mdi and gong ) do not obtain as robust frequencies at low degree as bison or golf . the standard way to overcome this problem is to combine precise low - degree data from one source with precise intermediate - degree data from another . combining data from different sources requires caution since solar frequencies are known to change with solar cycles ( elsworth et al . 1990 ; libbrecht & woodard 1990 ; howe et al . 1999 , etc ) . it has been demonstrated by basu et al . ( 1996 ; 1997 ) that inverting combined sets when there is a mismatch in the solar activity level of the low- and intermediate - degree data sets can lead to misleading results about the solar core . thus one either has to use contemporaneous low- and intermediate - degree data ; or use a robust method for correcting one of the sets of frequencies to the solar activity level of the other set . here , we compare results using both approaches . the intermediate - degree mode frequencies that we use are those estimated by schou et al . ( 1998b ) . this set was obtained from observations made by the mdi instrument on board soho during its first year of operation and covers a period from may 9 , 1996 to april 25 , 1997 . the activity level of the sun was quite low over the entire period that the data were collected , and hence frequency changes over the year in question were not a matter of much concern . the bison data sets we use in this paper are contemporaneous with the mdi set . in our 1997 paper ( basu et al . 1997 ) , we had used bison data that were nearly contemporaneous with the intermediate - degree data set obtained from the lowl instrument ( tomczyk et al . the bison data set used then was a combination of frequencies obtained from an 8-month time series , five 2-month time series , and low - frequency data obtained from a 32-month time series . this combined set is referred to in basu et al . ( 1997 ) as the `` best set '' and we use the same nomenclature here to refer to this older set . since 1997 there have of course been improvements in our understanding of the properties of the solar oscillation power spectrum and in the peak - bagging techniques used to extract estimates of the frequencies . as a result , we believe it is time to revisit the question of what the solar core looks like using the bison data . the rest of the paper is organized as follows : we describe our data sets in [ sec : data ] . the inversion technique and reference models are described in [ sec : inv ] . we present our results in [ sec : res ] , and discuss their implications in [ sec : disc ] . we use a number of different low - degree sets for this work , most of them obtained with the bison instruments . we have , however , also made use of low-@xmath1 frequencies extracted from sun - as - a - star observations made by golf ( bertello et al . 2000 , garca et al . 2001 ) and iris ( fossat et al . 2003 ) , and from analysis of disc - integrated mdi data ( toutain et al . 1998 ) . the golf , iris and mdi frequencies are in the literature . the bison frequencies come from our analysis of the bison data , which is described below . since the golf , iris and the disc - integrated mdi data are not contemporaneous with the 360-day mdi intermediate - degree set of schou et al . ( 1998b ) which hereafter we shall refer to as data set mdi-1 for these sets we only use modes with frequencies less than 1800@xmath3hz . solar - cycle related changes at these low frequencies are negligible and smaller in size than the uncertainties in the frequencies . the data sets we have used for this work , along with the names we use to identify the sets , are listed in table [ tab : data ] . all the bison sets are supplemented with @xmath4 data from mdi-1 . the golf , mdilow , and iris sets are supplemented with mdi-1 data with @xmath5hz data for the low - degree modes , and all the higher @xmath1 mdi-1 data . the majority of the bison frequencies came from analysis of data collected over two periods : a 360-day period commensurate with the mdi-1 observations ; and a much longer 4752-day period , beginning 1992 december 31 and ending 2006 january 3 ( which includes the shorter 360-day period ) . estimates of the frequencies of very low - frequency modes came from analysis of datasets prepared specifically to reduce low - frequency background noise ( see appourchaux et al . 2000 , and chaplin et al . 2002 , for more details ) . these ` low - frequency optimized ' datasets included one dataset prepared from observations made over the 4752-d period ; and two others prepared from shorter 2000-day and 3071-day periods within the 4752-day period . there are two main advantages to using frequencies from the 4752-day period and the other longer periods . first , the frequencies have superior internal precision , e.g. , frequency uncertainties in the 4752-day dataset are @xmath6-times smaller , on average , than those in the 360-day dataset . second , very low - frequency modes are either more prominent , or only detectable , in the longer datasets . the main downside to using bison frequencies from the longer periods is that they do not cover the same levels of solar activity as the 360-day mdi period . the average activity over the 4752-day period , as determined by various proxies of global activity , is about 60% higher than the average activity over the 360-day period . since an increase in surface activity brings with it an increase in mode frequencies , adjustments must be made to the bison frequencies to correct them _ down _ to values expected for the 360-day period . only then can the longer - period bison frequencies be usefully combined with the 360-day mdi frequencies for inversion . solar - cycle variability is much less of a cause for concern in the lower - frequency modes , since , as noted above , they show significantly smaller solar - cycle frequency shifts than their higher - frequency counterparts . we therefore augmented the 360-day bison frequency sets with frequencies of very low - frequency modes from the longer bison datasets . the golf , mdi and iris frequency sets that we used were also comprised only of low - frequency modes ( having @xmath7 ) , and so the impact of the different epochs over which the various data were collected should not have had a significant impact on the modes that we used . there is one other potential complication for combining the bison and mdi frequencies . this complication arises from the fact that bison observations are made of the `` sun as a star '' . the mdi observations are in contrast made at high spatial resolution . the two types of observation show a marked difference in sensitivity to modes having certain combinations of @xmath1 and the azimuthal order @xmath8 , and as a result the frequencies may show a different response to the azimuthally dependent surface activity . we now go on to provide more detail on the bison analysis , and on the important points raised above . we begin in [ sec : bagging ] with a brief summary of how frequencies were estimated from the data . in [ sec : corr ] we discuss how the long - dataset bison frequencies were corrected to the activity level of the 360-day mdi period . then in [ sec : mismatch ] we explain the inherent mismatch between the bison sun - as - a - star and mdi resolved - sun data , and the extent to which it is an issue for the frequency data used in our studies , and in [ subsec : comp ] we compare the different frequency sets . a set of bison sun - as - a - star observations gives a single time series whose power frequency spectrum contains many closely spaced mode peaks . parameter estimation must contend with the fact that within the non - radial ( @xmath9 ) mode multiplets the various @xmath8 lie in very close frequency proximity to one another . suitable models , which seek to describe the characteristics of the @xmath8 present , must therefore be fitted to the components simultaneously . furthermore , overlap between modes adjacent in frequency means the modes are usually fitted in pairs . we fitted multi - component models to modes in power frequency spectra of the various bison datasets to extract estimates of the mode frequencies . this fitting was accomplished by maximizing a likelihood function commensurate with the @xmath10 , 2-d.o.f . statistics of the power spectral density . we adopted the usual approach to fitting the sun - as - a - star spectra , and the low-@xmath1 modes were fitted in pairs ( @xmath2 with 2 , and @xmath11 with 3 ) . further background on details of the procedures may be found in chaplin et al . ( 1999 ) . our procedure for removing the solar - cycle frequency shifts rests on the assumption that variations in certain global solar activity indices may be used as a proxy for the low-@xmath1 frequency shifts , @xmath12 . we assume the correction can be parameterized as a linear function of the chosen activity measure , @xmath13 . when the 10.7-cm radio flux ( tapping & de tracey 1990 ) is chosen as the proxy , this assumption is found to be reasonably robust ( e.g. , chaplin et al . 2004a ) at the level of precision of the data . consider then the set of measured bison mode frequencies , @xmath14 , that we wish to ` correct ' . let us take the example where the data come from observations collected over the @xmath15 period , when the mean level of the 10.7-cm radio flux was @xmath16 . our intention it to correct these frequencies to the mean level of @xmath17 observed over the 360-day period covered by the mdi data . the mean activity in this 360-day period only just exceeds the canonical quiet - sun level of the radio flux , which , from historical observations of the index , is usually fixed at @xmath18 ( see tapping & detracey 1990 ) . the magnitude of the solar - cycle correction which must be subtracted from the raw frequencies will then be : @xmath19 \cdot [ \left < a(t ) \right > - \left < a(t ) \right>_{360 } ] . \label{eq : corr}\ ] ] the @xmath20 are @xmath1-dependent factors that calibrate the size of the shift . these factors are required because the sun - as - a - star shifts alter significantly with @xmath1 ( see [ sec : mismatch ] below ) . to determine the @xmath21 , we divided the 4752-day timeseries into 44 independent 108-day segments . the resulting ensemble was then analyzed , in the manner described by chaplin et al . ( 2004a ) , to uncover the dependence of the solar - cycle frequency shifts on the 10.7-cm radio flux . the @xmath22 $ ] in equation [ eq : corr ] is a function that allows for the dependence of the shift on mode frequency . here , we used the determination of @xmath22 $ ] to be found in chaplin et al . ( 2004a , b ) . uncertainty in the correction is dominated by the errors on the @xmath21 . these errors must be propagated , together with the formal uncertainties from the mode fitting procedure , to give uncertainties on the corrected frequencies , @xmath23 . the corrected uncertainties are , on average , about 10% larger than those in the raw , fitted frequencies . at frequencies below about @xmath24 , estimated corrections for the 4752-day dataset , and the other long datasets , were smaller in size than the fitted frequency uncertainties . we therefore felt justified in augmenting the 360-day bison frequency set with very low - frequency estimates from the longer bison datasets . extant sun - as - a - star observations , such as those of the bison , are made from a perspective in which the plane of the rotation axis of the sun is nearly perpendicular to the line - of - sight direction . this means that only components with @xmath25 even have non - negligible visibility . resolved - sun observations like mdi in contrast have good sensitivity to all components in detected modes . sun - as - a - star and resolved - sun frequency data must as a result be combined very carefully since there there can be an inherent , underlying mismatch between frequency determinations from the two types of instrument , as will be explained below . the resolved - sun data allow for a direct measurement of the frequency centroids of the non - radial modes , because all components are detectable . the frequency centroids carry information on the spherically symmetric component of the internal structure , and are the input data that are required for the hydrostatic structure inversions . in the case of the sun - as - a - star data , some components are missing , and the centroids must be estimated from the subset of visible components . in the complete absence of the near - surface activity , the @xmath25 odd components ` missing ' in the sun - a - as - a - star data would be an irrelevance . all mode components would be arranged symmetrically in frequency , meaning centroids could be estimated accurately from the subset of visible components . a near - symmetric arrangement is found at the epochs of the modern cycle minima . however , when the observations span a period having medium to high levels of activity as a long dataset by necessity does the arrangement of components is no longer symmetric . the frequencies given by fitting the sun - as - a - star data then differ from the true centroids by an amount that is sensitive to @xmath1 . the @xmath1 dependence arises because in the sun - as - a - star data modes of different @xmath1 comprise visible components having different combinations of @xmath1 and @xmath26 ; and these different combinations show different responses ( in amplitude and phase ) to the spatially non - homogeneous surface activity . is it possible to correct the sun - a - as - a - star frequencies to remove the mismatch ? the answer is yes , and the procedure requires knowledge of the strength and spatial distribution of the surface activity over the epoch in question . chaplin et al . ( 2004c ) and appourchaux & chaplin ( 2007 ) show how to make the correction , using the so - called even @xmath27 coefficients from fits for the resolved - sun frequencies . does the mismatch matter for the datasets used here ? we should not expect the mismatch to be a cause for concern between the 360-day bison frequencies and 360-day mdi frequencies . that is because the surface activity is low throughout this period . we have verified that application of the correction procedure outlined in appourchaux & chaplin ( 2007 ) does not alter significantly the 360-day bison frequencies , nor does it affect significantly the results of the structure inversions . there will be a more serious mismatch between the 4752-day bison frequencies and the 360-day mdi frequencies . however , the solar - cycle frequency correction procedure , outlined in [ sec : corr ] above , adjusts _ by definition _ the bison frequencies to values expected at low levels of activity ; as such this in principle removes the sun - as - a - star and resolved - sun mismatch without the need to apply a further correction . panel ( a ) of figure [ fig : freqdif ] shows frequency differences between the bison-1 set and the mdi-1 set , while panel ( b ) of the same figure shows differences between the bison-1 set and the bison-2 set . the differences in panels ( a ) and ( b ) both have the same basic structure . this structure comes from the fact that the bison-2 and mdi-1 sets were derived from fits to the oscillation power spectra that used a symmetric ( lorentzian ) profile for each mode peak , while the bison-1 set came from fits made using a more accurate asymmetric profile . there is also likely to be a contribution to the bison-1 minus mdi-1 residuals from differences in the peak - bagging methodologies applied to sun - as - a - star ( bison ) and resolved - sun ( mdi ) data . the mdi-1 mode set has fewer low - degree modes than the bison-1 set , and as a result we would expect to see differences between inversion results obtained by the mdi-1 set and those obtained with bison sets supplemented by @xmath28 modes from the mdi-1 set . in order to judge the effect of the number of low - degree modes in the set , we have also inverted data using only those modes of the bison-1 set that are also present in the mdi-1 set . we call this restricted set the bison-1 m set for ease of reference . differences between the bison-1 and 4752-day bison-13 set , shown in panel ( c ) of figure [ fig : freqdif ] , are on the whole very small because both sets came from asymmetric - profile fits . there are a few @xmath29 outliers , which probably reflect the impact on the fitting of different realization noise in the two sets of data . the observed differences do also increase in size to above @xmath30 at high frequencies . this is the part of the oscillation spectrum where the mode peaks are very wide ( the modes are heavily damped ) , meaning the peaks of mode components , and modes , adjacent in frequency begin to overlap . some modest bias in the estimation of the frequencies can result , and this will be more severe in the shorter bison-1 set than in the longer bison-13 set because of the inferior frequency resolution of the 360-day bison-1 data . panel ( a ) of figure [ fig : best ] shows frequency differences between the bison-1 and the 1997 `` best set '' . as can be seen in the figure , the new bison-1 set has filled up some of the gaps in the @xmath31 space . also , the new set extends to somewhat lower as well as higher frequencies . the frequency differences shown in panel ( b ) of the same figure have a definite structure as a function of frequency , which has a similar pattern to the structure in panels ( a ) and ( b ) of figure [ fig : freqdif ] . these differences are again largely due to the fact that the `` best set '' was derived from fits that used a symmetric profile for the mode peaks . it should be noted that some of the difference in the frequencies probably also comes from other improvements to the peak - bagging routines that have been implemented since the mid 1990s . since the overall differences are not random , we might again expect to find some differences in the inferred structure of the solar core inversion for solar structure is complicated because the problem is inherently non - linear . the inversion generally proceeds through a linearization of the equations of stellar oscillations , using their variational formulation , around a known reference model ( see e.g. , dziembowski et al . 1990 ; dppen et al . 1991 ; antia & basu 1994 ; dziembowski et al . 1994 , etc . ) . the differences between the structure of the sun and the reference model are then related to the differences in the frequencies of the sun and the model by kernels . nonadiabatic effects and other errors in modeling the surface layers give rise to frequency shifts which are not accounted for by the variational principle . in the absence of any reliable formulation , these effects have been taken into account by including an arbitrary function of frequency in the variational formulation ( e.g. , dziembowski et al . 1990 ) . the fractional change in frequency of a mode can be expressed in terms of fractional changes in the structure of model characteristics , for example , the adiabatic sound speed @xmath32 and density @xmath33 , and a surface term . the frequency differences can be written in the form ( e.g. , dziembowski et al . 1990 ) : @xmath34 here @xmath35 is the difference in the frequency @xmath36 of the @xmath37th mode between the data and the reference model , where @xmath37 represents the pair ( @xmath38 ) , @xmath39 being the radial order and @xmath1 the degree . the kernels @xmath40 and @xmath41 are known functions that relate the changes in frequency to the changes in the squared sound speed @xmath42 and density @xmath33 respectively , and @xmath43 is the mode inertia . the kernels for the @xmath44 combination can be easily converted to kernels for others pairs of variables like @xmath45 , with no extra assumptions ( gough 1993 ) . the term @xmath46 is the `` surface term '' , and takes into account the near - surface errors in modeling the structure . equation ( [ eq : inv ] ) constitutes the inverse problem that must be solved to infer the differences in structure between the sun and the reference solar model . we carried out the inversions using the subtractive optimally localized averages ( sola ) technique ( pijpers & thompson 1992 ; 1994 ) . the principle of the inversion technique is to form linear combinations of equation [ eq : inv ] with weights @xmath47 chosen such as to obtain an average of @xmath48 localized near @xmath49 ( the ` target radius ' ) while suppressing the contributions from @xmath50 and the near - surface errors . in addition , the statistical errors in the combination must be constrained . if successful , the result may be expressed as @xmath51 where the @xmath52 , the averaging kernel at @xmath53 , is defined as @xmath54 of unit integral , and determines the extent to which we have achieved a localized measure of @xmath48 . in particular , the width in @xmath55 of @xmath56 , here calculated as the distance between the first and third quartile point , provides a measure of the resolution . although we attempt to obtain averaging kernels at the target radius @xmath57 this is not always possible , and in the core very often the kernel is localized at a different point . this point , unless the averaging kernel is really non - localized , coincides with the second quartile point of the averaging kernel . as a result all our results are plotted as a function of the 2nd quartile points of the averaging kernels obtained at the end of the inversion process . details of how sola inversions are carried out and how various parameters of the inversion are selected were given by rabello - soares et al . ( 1999 ) . since the solar core is difficult to invert for , we also inverted the data using the regularized least squares ( rls ) technique . given the complementary nature of the rls and sola inversions ( see sekii 1997 for a discussion ) , we can be more confident of the results if the two inversions agree . details on rls inversions and parameter selections can be found in antia & basu ( 1994 ) and basu & thompson ( 1996 ) . in this work we invert for the relative difference of the squared sound speed , i.e. @xmath48 between the sun and the reference solar model using @xmath44 kernel pairs , and for the relative density difference @xmath58 between the sun and reference model using @xmath45 kernel pairs . our main reference model is model bp04 of bahcall et al . we also use two other models in the discussion of our results , model s of christensen - dalsgaard et al . ( 1996 ) , and model bsb(gs98 ) of bahcall et al . ( 2006 ) , which in this paper we refer to as simply model bsb . all the models are standard solar models , but have been constructed with somewhat difference input physics . the characteristics and physics inputs of the models are listed in table [ tab : models ] . most of the structure inversion results found in the literature have been obtained with model s as the reference model . however , the input physics in model s is now somewhat outdated , in particular the equation of state is known to have problems under the conditions found in the solar core . the opacities are also somewhat outdated . as a result we use bp04 as our reference model . model bp04 is very similar to model s , but it is constructed with an improved equation of state and newer opacities . model bsb is in contrast quite different from the other two models . in particular it uses op opacities ( badnell et al . 2005 ) instead of opal opacities ( iglesias & rogers 1996 ) . the op opacities are somewhat smaller than opal opacities for conditions expected in the solar core , but somewhat larger than opal opacities at the base of the convection zone . the core structure has also been affected by the new and lower @xmath59n(@xmath60,@xmath61)@xmath62o reaction rate ( formicola et al . 2004 ) . it should be noted that all the models that we use are high-@xmath63 models , with @xmath63 values adopted from either grevesse & noels ( 1993 ) or from grevesse & sauval ( 1998 ) . asplund et al . ( 2005 ) compiled a table of solar abundance determinations based on new techniques that showed that solar metallicity is about 30% lower than previous estimates . this led to large changes in the structure of standard solar models , and in particular , made the agreement between the sun and solar models much worse [ see basu & antia ( 2008 ) for a review of the problem ] . since the aim of this paper is to take a fresh look into the structure of the sun rather than study solar models , we restrict ourselves to models that are known to agree well with the sun . the relative differences in the squared sound speed between the sun and model bp04 , as obtained with the different bison sets , are plotted in figure [ fig : csqbison ] . also plotted is a close - up of the differences in the core . we can see from the figure that the 1-year bison sets combined with @xmath64 modes from the mdi 360 day set give very similar results , but that the results are different from those obtained with the mdi-1 set , which is the mdi 360-day set for all degrees . the differences between the mdi-1 and bison sets are seen below about @xmath65 , where the @xmath66 modes begin to influence the results . while we were able to get smaller errors in the inversion results , we were not able to push the inversions much deeper without increasing the errors . the differences between the mdi and bison results are not merely a result of having more modes . if that were so , the result of inverting set bison-1 m would have been similar to the mdi-1 result . while this is indeed true between about @xmath67 and @xmath68 , in deeper layers the bison-1 m set gives very similar results to the other bison sets , pointing to the fact that the frequencies themselves , as well as the lower errors of the bison frequencies , play a rle . the sound - speed obtained with the symmetric bison-2 set is marginally higher than that obtained with the asymmetric bison-1 set , however , the results agree within errors . since all the 1-year bison sets give similar results , and since peaks in the solar oscillation power spectrum are known to be asymmetric , in further discussion of the 360-d bison inversions we shall use the results from bison-1 , not bison-2 . as far as the bison-13 set is concerned , it lets us go marginally deeper into the core , with the innermost averaging kernel centered at 0.062@xmath0 . we get lower sound speeds than bison-1 but higher speeds than mdi-1 in the deepest regions , as is seen in figure [ fig : csq13 ] . the bison-13 results above 0.2 @xmath69 lie between the bison-1 and mdi results ; while between 0.1 and 0.2 @xmath69 they lie slightly above the bison-1 and mdi-1 results . figure [ fig : avkcsq ] shows the innermost averaging kernel , as well as two others obtained with the bison-13 and bison-1 sets . also shown are the `` cross - term '' kernels corresponding to the innermost averaging kernels . note that the peak of the bison-13 averaging kernel is closer to the centre . figure [ fig : csqbest ] shows the sound - speed difference obtained by the bison-1 set compared with that of the `` best set '' of basu et al . we see that the differences in structure are small . with the new set we can though push closer to the core , which we believe is a result of the lower - frequency modes in the new set . one point should be noted : in the basu et al . ( 1997 ) paper , we had plotted the results against the `` target '' radius , instead of the 2nd quartile point as we have done here . our experience with inversions in the intervening period leads us to believe that plotting the results against the 2nd quartile point is more correct , since they are the points where the averaging kernels are localized . the bison-1 and mdi-1 results are compared with the other data sets listed in table [ tab : data ] in figure [ fig : csq_comp ] . for the external data sets , we show two inversions , one marked `` all '' and the other marked `` weeded . '' the inversions marked `` all '' used all the @xmath70hz modes in the sets , while the ones marked `` weeded '' had a few modes removed , usually because they had large residuals in the rls inversions . the mdilow `` all '' results show a steeper rise in solar sound - speed near the core . however , rls inversion of this set show that two modes ( @xmath71 and @xmath72 ) had extremely large residuals ( 16 and 22@xmath73 respectively ) . this problem had also been noted earlier by basu et al . ( 2000a ) , who used different reference models . the weeded results are quite similar to the bison results . the irislow set was problematic in that many modes had high residuals . the worst offenders were the @xmath74 , @xmath75 and @xmath72 modes . with these modes in place , the irislow data implied that the solar sound - speed increased enormously compared to the reference model at radii below @xmath76 . with these modes removed the results are still higher than the bison results , but within 1@xmath73 . the results for the two golf sets are much more interesting . the unweeded results ( `` all '' ) show a marked departure from the bison and mdi results below 0.15@xmath69 . these two golf sets contain very low - frequency detections that are controversial and extremely uncertain , which have not been verified independently by other data . the golf1low set set has an @xmath77 mode at @xmath78hz and an @xmath79 mode at @xmath80hz . the golf2low set has an @xmath79 at @xmath81hz . the lowest bison-1 frequency , for comparison , is the @xmath82 mode at @xmath83hz . we note that this mode has been observed in bison and golf data . the @xmath79 mode changes the inversion results for the golf2low set completely . given that there are no other modes in the set at comparably low frequencies , this raises a serious question mark over whether this is a real effect . in the absence of this mode , the golf2low results ( `` weeded '' in figure [ fig : csq_comp ] ) match the bison-1 results almost exactly . the situation is the same with the golf1low results . note that the same inversion parameters have been used for both the `` weeded '' and `` all '' sets . the @xmath79 mode appears to exert a disproportionate influence , the @xmath77 less so , but it also pulls down the solution . without these two modes , the sound - speed difference obtained is again very similar to that of the bison-1 set . we discuss the issue of low - frequency modes further in [ subsec : den ] . thus , if for the moment we ignore the effect of the @xmath66 , @xmath84 modes , it appears that the solar sound - speed still shows a dip in the sound - speed difference with respect to the solar model around @xmath85 and rises to positive values at radii below @xmath76 . the effect of the two very low - frequency modes in lowering the estimated sound speed in the core has also been noted by turck - chieze et al . . the changes between the `` all '' and `` weeded '' results are caused by differences in the inversion coefficients . for sola inversions , the modes that were weeded out changed the inversion coefficients of the neighbouring low - degree modes by a large amount . the changes were much smaller in the case of rls inversions , and because the low - degree modes have large residuals , the changes did not affect the solution . the density differences between the sun and model bp04 , as obtained by the different bison sets , are shown in figure [ fig : bisondensity](a ) . also shown in the figure is the result of inverting the mdi-1 set . as can be seen from the figure , the bison sets give fairly similar results , within errors . however , the mdi results are different at all radii , not just close to the core . at first glance this is surprising since the only difference between the mdi and bison sets are the @xmath86 modes , and one would expect differences only in the core , as was found for the sound - speed difference . however , the result is not really surprising . density inversions are carried out by imposing the condition that the mass of the reference model is the same as that of the sun . thus the density differences must integrate out to zero , i.e. , @xmath87 the density @xmath33 in the core is large , and thus a small relative difference @xmath58 in the core will lead to a large difference in the regions where the density is small . the higher number of low - degree models in the bison sets allows us determine the core structure much better than with the mdi set : we find a smaller difference in the core , which implies a smaller difference at all radii . that said , the difference between the bison and mdi results is not merely a matter of the larger bison mode set . if we only use bison frequencies of those modes that are present in the mdi set ( i.e. , set bison-1 m ) , the result lies between the mdi and the other bison sets . this can be seen in figure [ fig : bisondensity](a ) . these results show us that when we attempt to infer the solar density it is not enough to have good intermediate and high - degree modes : it is more important to have reliable data on low - degree modes . given that the bison low - degree sets still have gaps and do not yet have very low - frequency modes , we can be sure that solar density results will change if we get better low - degree sets . in figure [ fig : bisondensity](b ) we show the relative differences in the inferred solar density obtained from the different sets . the differences are shown with respect to the results obtained with the bison-1 set . as can be seen , the differences are completely systematic , even when within errors as is the case for bison-2 . judging by the difference between the results of the mdi-1 set and the results of the bison sets , we estimate that despite the small statistical errors , solar density inversions are uncertain at a level of about 0.6% . figure [ fig : rho13 ] shows the density differences obtained with the bison-13 set compared to those obtained with the bison-1 and mdi sets . the bison-13 set seems to imply a low - density solar core . figure [ fig : rho_best ] compares the density differences obtained using the `` best set '' of basu et al . ( 1997 ) and the bison-1 set . note that the density differences obtained with the `` best set '' are larger in general , and in the core they imply that the density in the sun appears to be much greater than in the reference model . this result is in seeming contradiction to the result reported by basu et al . ( 1997 ) ( even though that was for a different reference model , the model s ) , which was that the solar density was found to be much lower than the density in the reference model . this discrepancy appears to be a result of the use of the @xmath88 kernel combination by basu et al . density difference results obtained for the `` best set '' using model s as the reference and with @xmath45 and @xmath89 kernels are shown in figure [ fig : rho_yker ] . with @xmath45 kernels we see that the inferred density difference between the sun and the model is positive in the core . transformation to kernels involving @xmath90 requires the assumption that the equation of state and the heavy element abundance are known exactly . basu & christensen - dalsgaard ( 1997 ) showed that the density results obtained with these @xmath89 kernels are sensitive to errors in the equation of state assumed for the reference model . model s was constructed with an equation of state that was found to be deficient for conditions found in the solar core ( elliott & kosovichev 1998 ) , and as a result we believe that the results obtained using the @xmath91 variable combination are more robust than those obtained with @xmath89 , even though it is at the expense of larger statistical errors . basu et al . ( 1997 ) had discussed the possibility of systematic errors in the density results caused by the particular choice of kernels , but had not tested the results obtained using other sets of kernels . figure [ fig : rhoall ] shows the density differences obtained for the four external sets mdilow , irislow , golf1low , and golf2low . we only show the results of the weeded sets . although there are differences in the results obtained by the various sets , only set irislow gives completely discrepant results , probably implying that some of the frequencies are not accurate . while all other differences lie within a few @xmath73 there are systematic trends as expected from the conservation of mass constraint ( see above ) . the two golf results are again interesting . although the results obtained from the weeded sets shown in figure [ fig : rhoall ] are very similar to those obtained using the bison sets , the unweeded results are troublesome . figure [ fig : golf ] shows both the weeded and unweeded results obtained with the golf sets . we show both sola and rls results . the striking feature of the unweeded results is that the sola results do not match the rls results at all . this again leads us to believe that there could be some problems with the very low - frequency @xmath2 modes that were weeded out from the two sets . one of the drawbacks of the sola method is the implicit assumption that the frequency differences and the errors associated with them are correct . however , sometimes the observational errors can be either under- or over - estimated , or the modes may have unusual characteristics ( e.g. , much smaller line widths than modes adjacent in frequency ) , in which case the inversion results can be misleading . rls inversions are less prone to be affected by such outliers , and hence we are more inclined to believe the rls results , which agree with the weeded sola results . we have used a number of bison low-@xmath1 data sets , combined with @xmath92 data from the mdi 360-day set , to obtain sound speed and density differences between our reference model bp04 and the sun . we find that we are able to go slightly deeper into the core than we had been able to do with earlier bison sets . while the sound - speed results are similar to what we had seen earlier ( e.g. , basu et al . 1997 ) , the density results show larger differences . since the new sets of bison frequencies have more modes , and smaller estimated uncertainties , we believe that our current results are more reliable . differences in inversion results obtained by using as input frequencies estimated from symmetric and asymmetric fits to the oscillation power spectra are within errors . this is what had also been found for golf data by basu et al . ( 2000a ) . similarly the inversion results obtained from frequencies with and without sun - as - a - star corrections are very similar . the differences in density results are larger , but although they are systematic , the differences are still within errors . the bison-13 set , which has low frequency data from a 4752-day long spectrum , allows us to get results with somewhat smaller errors . the solar sound - speed and density results obtained with this set are tabulated in table [ tab : seismic ] . the sound - speed results are very similar to what has been obtained before , except that we go slightly deeper in the core and the errors are slightly smaller . from figure [ fig : csqbison ] we can still see a discrepancy in the sound - speed difference around @xmath93 , a large positive difference just below the base of the convection zone , and a gentle fall in the convection zone itself . of these features , the discrepancy at the base of the convection zone is easiest to interpret . this feature is usually taken as evidence for mixing in the sun below the base of the convection zone , mixing that is absent in the models ( see e.g. , gough et al . 1996 ; basu et al . 1996 . 1997 ; etc . ) . this supposition is supported by inversions for the helium profile of the sun ( antia & chitre 1997 ) . support for this interpretation also comes from the fact that models that incorporate mixing below the base of the convection zone show a reduced discrepancy with respect to the sun ( see e.g. , gough et al . 1996 ; basu et al . there is also a mismatch in the position of the convection zone base in the sun ( @xmath94 ; christensen - dalsgaard et al . 1991 ; basu 1998 ) and our reference model ( @xmath95 ) , and this also contributes to the discrepancy . the discrepancy close to the core is more difficult to interpret , but is believed to be related to weak mixing at some time during the sun s evolution ( see e.g. gough et al . the third apparent discrepancy is for @xmath96 . basu et al . ( 2003 ) noted that for their reference model , the discrepancy decayed with depth from the surface at a rate roughly proportional to @xmath97 . since they could not find any physical reason for such a discrepancy in what is essentially an adiabatically stratified region , they believed that the feature was a systematic error in the inversions caused by imperfect suppression of the near - surface errors . before we try to interpret the features in the sound - speed difference between the sun and our models , it is instructive to look at differences between different standard solar models and the differences these models show with respect to the sun . figure [ fig : modelfig ] shows the relative sound - speed and density differences between model s and bp04 , and between model bsb and bp04 . these are exact differences , but convolved with the averaging kernels obtained from inverting set bison-1 . convolution with the averaging kernels makes it easier to compare these differences with those obtained from the inversions using the bison data . figures [ fig : csqsa ] and [ fig : csqsb ] show the sound speed differences , respectively for all radii and just for the core between the sun and models s and bsb ; while figure [ fig : rhos ] shows the density differences between these models and the sun . the first thing to note from figure [ fig : modelfig](a ) is that the sound - speed difference between bsb and bp04 has the same type of behavior at @xmath96 as does the sound - speed difference between the sun and model bp04 ( and model s ) , and in fact the sound - speed differences between the sun and model bsb are flatter in this region ( figure [ fig : csqsa ] ) . yet the model differences are exact and therefore should not suffer from imperfect suppression of near - surface errors . there is a small difference between the convolved and the unconvolved differences that points to effects of the finite resolution of the averaging kernels , but the effect of finite resolution is to reduce the sound - speed difference for @xmath98 . the difference between the sound - speed profiles of bsb and bp04 in the convection zone is a result of using different low - temperature opacities ( ferguson et al . 2005 in bsb and alexander & ferguson 2004 in bp04 ) . although , the bulk of the convection zone is stratified adiabatically , the structure closer to the surface is affected by differences in low - temperature opacities , which is manifested as the observed sound - speed differences in the convection zone of these models . this occurs predominantly because of differences in the mixing length parameter needed to model the present - day sun . in particular , we find that the mixing length parameter is 2.19 when opal opacities are used in conjunction with the low - temperature opacities of ferguson et al . ( 2005 ) ; is 2.21 when opal opacities are replaced by op opacities ; and is 2.11 when we use a combination of op opacities and the alexander & ferguson ( 1994 ) low - temperature opacities . therefore , we conclude that the differences in sound - speed between the sun and our models in the convection are most likely to be due to deficiencies of our models and not merely due to limitations of our inversion techniques . the next prominent feature in figure [ fig : modelfig](a ) is the increased difference just below the base of the convection zone . while at first glance this is reminiscent of the increase in figures [ fig : csqbison ] and [ fig : csqsa ] , a careful look shows that the difference actually occurs over a wider radius range , and is also smaller than the difference between the sun and bp04 . the differences between the models can basically be attributed to differences in opacity . in the case of model s and bp04 , both the convection - zone heavy element abundance and the opacity tables used are different . bp04 effectively has lower opacity ( and hence a shallower convection zone ) than model s. while bp04 and bsb have the same convection - zone metallicity , bsb used op rather than opal opacities . op opacities are marginally larger than opal opacities for conditions present at the base of the convection zone and that caused the small spike in the sound - speed difference between these two models . these differences do not explain the larger , and more localized , sound - speed difference between the sun and the three models . mixing below the solar convection zone remains the best explanation , and as mentioned earlier , there is other evidence for mixing in this region ( e.g. , antia & chitre 1997 ) . going back to the sound - speed differences between the sun and models bp04 , s and bsb at the core , we can see the solar sound - speed differences against all three models show a localized dip around @xmath93 , implying that the solar sound - speed is lower than that of the models . at even smaller radii the sound - speed in the sun appears to be larger than the sound - speed in the models , at least for the case of models s and bp04 . this feature , particularly in the case of model s , has been used as evidence for possible mixing at some time in the sun s past ( see gough et al . 1996 ) . while mixing in the past is a possibility , the evidence is somewhat less compelling when models with more updated physics are used . for example , if we look at the sound - speed differences between the sun and model bsb , the large , positive difference in the core is reduced in size . however , the deficit around @xmath93 remains , although it is shifted inwards and is closer to @xmath99 . basu et al . ( 2000a ) had claimed that the dip in the sound - speed difference is independent of the reference model and hence intrinsic to the sun , but it appears that their result was due to the fact their various reference models used very similar physics inputs . the difference in the reaction rates and opacities in bsb compared to those in models s and bp04 are evidently large enough to shift the position of the deficit . thus , we might expect models with newer inputs to show other changes . also notable is that the sound speed difference between bsb and bp04 ( see figure [ fig : modelfig ] ) shows a larger sound - speed in model bsb at low radii , and a lower sound speed around @xmath100 . these differences are purely due to changes in input microphysics , i.e. , opacities as well as nuclear reaction rates . of course , the conjecture that there may have been mixing in the solar core was not based on sound - speed inversions alone . gough et al . ( 1996 ) used the density inversion results as corroboration . one should note , however , that the density difference that gough et al . ( 1996 ) used was obtained with @xmath89 kernels , and like the @xmath89 result shown in figure [ fig : rho_yker ] , showed that density of the solar core is less than the density in model s in the core . gough et al . ( 1996 ) argued that the relatively steep positive gradient in @xmath58 in the core and immediately beneath the convection zone imply that the magnitude of the negative gradient of density is too high in the model . however , our more robust inversions show that the gradient of @xmath58 is not steep for any of the three models , and not positive for models bp04 and bsb . thus , we find that even with improved data , interpreting the sound - speed and density differences against different models is not completely straightforward . improvements in solar models need to go hand - in - hand with improvements in the observational data . to be able to invert closer to the solar core we require reliable frequency estimates of very low-@xmath39 , low - degree modes . since these modes have large amplitudes close to the core , they will help ( even in the absence of g - modes ) to get a better handle on the innermost layers if the sun . as we have seen in this paper , changes in the low - degree mode set can lead to large variations in estimated density differences between reference solar models and the sun . there is therefore a clear need for very low - frequency modes if we are to obtain robust , precise density estimates for the sun . this paper utilizes data collected by the birmingham solar - oscillations network ( bison ) , which is funded by the uk science technology and facilities council ( stfc ) . we thank the members of the bison team , colleagues at our host institutes , and all others , past and present , who have been associated with bison . this paper also utilizes from the solar oscillations investigation / michelson doppler imager ( soi / mdi ) on soho . sb would like to thank the hiros group at the university of birmingham , u.k . , for their hospitality during the period when this work was conceived and developed . she would also like to thank the institute for advanced study , princeton for its hospitality while this paper was being written . sb acknowledges partial support from nsf grants atm-0348837 and atm-0737770 . fossat e. , et al . , 2003 , in : proceedings of soho 12 / gong+ 2002 . local and global helioseismology : the present and future , 2002 , big bear lake , ca , usa . edited by h. sawaya - lacoste , esa sp-517 , noordwijk , netherlands : esa publications division , isbn 92 - 9092 - 827 - 1 , 2003 , p. 139 - 144 schou , j. , christensen - dalsgaard , j. , howe , r. , larsen , r. m. , thompson , m. j. , toomre , j. 1998b , in proc . soho 6/gong 98 , structure and dynamics of the interior of the sun and sun - like stars , esa sp-418 , 845 llll mdi-1 & @xmath101 & schou et al . ( 1998 ) + bison-1 & 1yr , asymmetric fitting & + bison-1 m & same as bison-1 , but restricted to the + & same low-@xmath1 modes as mdi-1 & + bison-2 & 1yr , symmetric fitting & + bison-13 & 13yr , asymmetric fitting , solar - cycle corrected & + golf1low & @xmath102 & bertello et al . ( 2000 ) + golf2low & @xmath102 & garcia et al . ( 2001 ) , table v + mdilow & @xmath102 & toutain et al . ( 1998 ) + irislow & @xmath102 & fossat et al . ( 2003 ) + [ tab : data ] llll & bp04 & s & bsb + + + mass & @xmath103 g & @xmath103 g & @xmath103 g + radius & @xmath104 cm & @xmath105 cm & @xmath104 cm + luminosity & @xmath106 erg / s & @xmath107 erg / s & @xmath106 erg / s + + ( z / x)@xmath108 & 0.0229 & 0.0245 & 0.0229 + & grevesse & sauval ( 1998 ) & grevesse & noels ( 1993 ) & grevesse & sauval ( 1998 ) + + eq . of state & opal & opal & opal + & rogers & nayfonov ( 2002 ) & rogers et al . ( 1996 ) & rogers & nayfonov ( 2002 ) + + opacity & opal & opal & op + & iglesias & rogers ( 1996 ) & rogers & iglesias ( 1992 ) & badnell et al . ( 2005 ) + & alexander & ferguson ( 1994 ) & kurucz ( 1991 ) & ferguson et al . ( 2005 ) + + reaction rates & bahcall & & bahcall & & bahcall & + & pinsonneault ( 2004 ) & pinsonneault ( 1995 ) & pinsonneault ( 2004 ) + & & & formicola et al . ( 2004 ) + + diffusion rates & thoul et al . ( 1994 ) & proffitt & & thoul et al . ( 1994 ) + & & michaud(1991 ) + + * characteristics : * + + @xmath109 & 0.243 & 0.247 & 0.243 + @xmath110 & 0.7146 & 0.7129 & 0.7138 + [ tab : models ] cccccc 6.1949e-02 & 5.1170e+07 & 9.2178e+03 & 5.8884e-02 & 1.2078e+02 & 1.1601e-01 + 7.4948e-02 & 5.1169e+07 & 4.3834e+03 & 7.7225e-02 & 1.0532e+02 & 7.7585e-02 + 1.0052e-01 & 5.0803e+07 & 4.9108e+03 & 1.0208e-01 & 8.5937e+01 & 5.1525e-02 + 1.2733e-01 & 4.9858e+07 & 4.4274e+03 & 1.2706e-01 & 6.9226e+01 & 3.1670e-02 + 1.5208e-01 & 4.8559e+07 & 4.4193e+03 & 1.5162e-01 & 5.5510e+01 & 2.0105e-02 + 1.7636e-01 & 4.7033e+07 & 3.5829e+03 & 1.7709e-01 & 4.3772e+01 & 1.2053e-02 + 2.0184e-01 & 4.5316e+07 & 3.4470e+03 & 2.0233e-01 & 3.4221e+01 & 6.9189e-03 + 2.2746e-01 & 4.3569e+07 & 3.0592e+03 & 2.2730e-01 & 2.6514e+01 & 4.6064e-03 + 2.5253e-01 & 4.1897e+07 & 2.8858e+03 & 2.5277e-01 & 2.0234e+01 & 4.3910e-03 + 2.7771e-01 & 4.0301e+07 & 2.5884e+03 & 2.7803e-01 & 1.5355e+01 & 4.5718e-03 + 3.0320e-01 & 3.8798e+07 & 2.4506e+03 & 3.0333e-01 & 1.1570e+01 & 3.9911e-03 + 3.2847e-01 & 3.7395e+07 & 2.3018e+03 & 3.2879e-01 & 8.6838e+00 & 3.2633e-03 + 3.5370e-01 & 3.6090e+07 & 2.1119e+03 & 3.5403e-01 & 6.5346e+00 & 2.5846e-03 + 3.7912e-01 & 3.4866e+07 & 1.9529e+03 & 3.7946e-01 & 4.9123e+00 & 1.9819e-03 + 4.0439e-01 & 3.3710e+07 & 1.8638e+03 & 4.0474e-01 & 3.7160e+00 & 1.5338e-03 + 4.2970e-01 & 3.2629e+07 & 1.7383e+03 & 4.3006e-01 & 2.8201e+00 & 1.1410e-03 + 4.5504e-01 & 3.1604e+07 & 1.6411e+03 & 4.5535e-01 & 2.1521e+00 & 8.9351e-04 + 4.8031e-01 & 3.0634e+07 & 1.5152e+03 & 4.8063e-01 & 1.6517e+00 & 6.7778e-04 + 5.0566e-01 & 2.9712e+07 & 1.4385e+03 & 5.0592e-01 & 1.2737e+00 & 5.4124e-04 + 5.3096e-01 & 2.8822e+07 & 1.3808e+03 & 5.3121e-01 & 9.8850e-01 & 4.1931e-04 + 5.5630e-01 & 2.7972e+07 & 1.3146e+03 & 5.5651e-01 & 7.7054e-01 & 3.3861e-04 + 5.8160e-01 & 2.7130e+07 & 1.2604e+03 & 5.8181e-01 & 6.0445e-01 & 2.6438e-04 + 6.0693e-01 & 2.6311e+07 & 1.1919e+03 & 6.0711e-01 & 4.7634e-01 & 2.1495e-04 + 6.3223e-01 & 2.5483e+07 & 1.1390e+03 & 6.3243e-01 & 3.7756e-01 & 1.6995e-04 + 6.5754e-01 & 2.4637e+07 & 1.0619e+03 & 6.5772e-01 & 3.0095e-01 & 1.4047e-04 + 6.8283e-01 & 2.3706e+07 & 1.0045e+03 & 6.8304e-01 & 2.4197e-01 & 1.1334e-04 + 7.0811e-01 & 2.2614e+07 & 9.3680e+02 & 7.0838e-01 & 1.9685e-01 & 9.3937e-05 + 7.3337e-01 & 2.1257e+07 & 8.9853e+02 & 7.3371e-01 & 1.6281e-01 & 7.8683e-05 + 7.5869e-01 & 1.9881e+07 & 8.7096e+02 & 7.5903e-01 & 1.3311e-01 & 6.5124e-05 + 7.8404e-01 & 1.8495e+07 & 7.7763e+02 & 7.8435e-01 & 1.0703e-01 & 5.4001e-05 + 8.0934e-01 & 1.7089e+07 & 7.3027e+02 & 8.0967e-01 & 8.4315e-02 & 4.3385e-05 + 8.3464e-01 & 1.5649e+07 & 6.8648e+02 & 8.3499e-01 & 6.4667e-02 & 3.4486e-05 + 8.5995e-01 & 1.4156e+07 & 6.1451e+02 & 8.6031e-01 & 4.7800e-02 & 2.6671e-05 + 8.8527e-01 & 1.2582e+07 & 5.7896e+02 & 8.8564e-01 & 3.3507e-02 & 1.9704e-05 + 9.1060e-01 & 1.0881e+07 & 5.9297e+02 & 9.1096e-01 & 2.1657e-02 & 1.4339e-05 + 9.3590e-01 & 8.9771e+06 & 5.5548e+02 & 9.3628e-01 & 1.2157e-02 & 9.5495e-06 + 9.5661e-01 & 7.1388e+06 & 4.9286e+02 & 9.5754e-01 & 5.9835e-03 & 6.2548e-06 + [ tab : seismic ]

complete set of cylindrical modes is constructed for the electromagnetic field inside and outside a cylindrical shell in the background of-dimensional ds spacetime . on the shell , the field obeys the generalized perfect conductor boundary condition . for the bunch - davies vacuum state , we evaluate the expectation values ( vevs ) of the electric field squared and of the energy - momentum tensor . the shell - induced contributions are explicitly extracted . in this way , for points away from the shell , the renormalization is reduced to the one for the vevs in the boundary - free ds bulk . as a special case , the vevs are obtained for a cylindrical shell in the-dimensional minkowski bulk . we show that the shell - induced contribution in the electric field squared is positive for both the interior and exterior regions . the corresponding casimir - polder forces are directed toward the shell . the vacuum energy - momentum tensor , in addition to the diagonal components , has a nonzero off - diagonal component corresponding to the energy flux along the direction normal to the shell . this flux is directed from the shell in both the exterior and interior regions . for points near the shell , the leading terms in the asymptotic expansions for the electric field squared and diagonal components of the energy - momentum tensor are obtained from the corresponding expressions in the minkowski bulk replacing the distance from the shell by the proper distance in the ds bulk . the influence of the gravitational field on the local characteristics of the vacuum is essential at distances from the shell larger than the ds curvature radius . the results are extended for confining boundary conditions of flux tube models in qcd . pacs numbers : 04.62.+v , 03.70.+k , 11.10.kk

1604.06244

the casimir effect ( for reviews see @xcite ) is among the most interesting boundary - induced effects in quantum field theory . nearly seventy years after its discovery , the effect continues to be area of active theoretical and experimental research today . this is due to of fundamental and practical importance of the casimir effect in various fields of modern physics and technology . the imposition of boundary conditions on a quantum field shifts the expectation values of physical observables . the particular features of the effect depend on the spin of the field , on the geometries of boundaries and background spacetime , and on the specific boundary conditions . an interesting topic in the investigations of the casimir effect is the dependence of the physical characteristics , such as the vacuum energy and forces acting on the boundaries , on the dimension of the background spacetime . in particular , this is motivated by the role of the casimir effect in higher - dimensional models . the dependence of the casimir energy on the size of the extra dimensions provides a mechanism for their stabilization . the latter is an important issue in both the kaluza - klein - type and braneworld models . in addition , the casimir energy related to extra dimensions can serve as a model of dark energy driving the accelerating expansion of the universe at recent epoch . in the present paper we investigate combined effects of boundaries and background geometry on the local characteristics of the electromagnetic vacuum in an arbitrary number of spatial dimensions . as a background spacetime we consider @xmath0-dimensional de sitter ( ds ) spacetime . this choice is motivated by several reasons . the ds spacetime has maximal number of symmetries and because of this a large number of physical problems can be solved exactly on its background . a better understanding of the influence of classical gravitational field on quantum matter in ds spacetime serves as a way to deal with less symmetric geometries . the quantum effects in ds bulk play an important role in inflationary scenarios ( see @xcite and mart13 for recent reviews ) . quantum fluctuations in the inflationary universe with ds geometry produce observable effects in the form of small density perturbations in the post - inflationary stage . these perturbations have nearly scale - invariant spectrum and form the base of the currently most popular mechanism for the generation of the seeds for galaxy formation . the corresponding predictions are in good agreement with the recent observational data on the temperature anisotropies of the cosmic microwave background radiation . the importance of ds spacetime as a gravitational background further increased after the discovery of the accelerating expansion of the universe at recent epoch @xcite . within the framework of general relativity , among the most popular cosmological models is the one with a positive cosmological constant as a driving source behind the accelerating expansion . in this model the ds spacetime is the future attractor for the geometry of our universe . an important ingredient in the problems of the casimir effect is the geometry of boundaries . here we consider a cylindrical shell in background of ds spacetime , described by inflationary coordinates . on the shell the electromagnetic field obeys the generalized perfect conductor boundary condition . along with the boundaries of planar and spherical symmetry , the cylindrical boundaries are among the most popular geometries in the investigations of the casimir effect ( see @xcite and references therein ) . the patch of ds spacetime covered by inflationary coordinates is conformally flat and the corresponding results for the casimir effect in the case of conformally invariant fields are obtained from the expressions for the minkowski bulk by using the standard conformal relation between two problems ( see , for instance , @xcite ) . in particular , this is the case for the electromagnetic field in 4-dimensional spacetime . the case of conformally coupled massless scalar field obeying dirichlet boundary condition on a spherical shell in 4-dimensional ds spacetime has been considered in @xcite . for conformally non - invariant fields the investigation of the casimir effect in ds spacetime requires a separate consideration . in this case , at distances from the boundaries larger than the curvature radius of the background spacetime , the influence of the gravitational field may lead to a behavior of the local characteristics of the vacuum essentially different from that in the minkowski bulk . the scalar casimir effect for a massive field with general curvature coupling and with robin boundary conditions on planar boundaries in the background of @xmath0-dimensional ds spacetime has been investigated in @xcite . the corresponding problem with a spherical boundary was discussed in @xcite . an important feature for conformally non - invariant fields is that the vacuum energy - momentum tensor , in addition to the diagonal components , acquires an off - diagonal component corresponding to the energy flux along the direction normal to the boundary . the casimir densities induced by a spherical ds bubble in the minkowski bulk has been studied in @xcite . for a scalar field , the geometry with a cylindrical boundary with robin boundary condition was discussed in saha15 . depending on the mass of the field , at distances from the boundaries larger than the ds curvature radius , two different regimes are realized with monotonic or oscillatory decay of the vacuum expectation values . the electromagnetic casimir effect for planar boundaries with generalized perfect conductor boundary conditions in ds bulk , having an arbitrary number of spatial dimensions , has recently been studied in saha14 ( for the propagators of vector fields on ds background in the absence of boundaries see @xcite ) . the background geometry of friedmann - robertson - walker cosmologies with power - law scale factors was discussed in @xcite . the organization of the paper is the following . in the next section we consider the complete set of cylindrical modes for the electromagnetic field in ds spacetime , obeying the generalized perfect conductor boundary condition on a cylindrical shell . by using these modes , in section sec : e2 , the vacuum expectation value ( vev ) of the electric field squared inside the shell is evaluated . various asymptotic regions of the parameters are discussed . the vev of the energy - momentum tensor is studied in section [ sec : emt ] . the vevs of the electric field squared and of the energy - momentum tensor in the exterior region are investigated in section [ sec : ext ] . section [ sec : conc ] summarizes the main results of the paper . in appendix [ sec : app1 ] the cylindrical modes in @xmath0-dimensional minkowski spacetime are discussed . the integrals , appearing in the expressions for the vevs for the special case @xmath1 , are evaluated in appendix [ sec : app2 ] . in the numerical evaluations of the vevs we have considered this special case . we consider a quantum electromagnetic field in background of a @xmath0-dimensional ds spacetime , in the presence of a perfectly conducting cylindrical shell having the radius @xmath2 . in accordance with the problem symmetry , we write the ds line element in cylindrical coordinates @xmath3 : @xmath4 , \label{ds2}\]]where @xmath5 . the parameter @xmath6 is expressed in terms of the cosmological constant @xmath7 through the relation @xmath8 . below , in addition to the synchronous time coordinate @xmath9 we will use the conformal time @xmath10 , defined as @xmath11 . in terms of this coordinate the metric tensor takes a conformally flat form @xmath12 , where @xmath13 is the minkowskian metric tensor . we are interested in the changes of the vevs for the electromagnetic field induced by a cylindrical boundary @xmath14 in the background of the geometry given by ( [ ds2 ] ) . in the canonical quantization procedure a complete orthonormal set of solutions to the classical field equations is required and , in this section , we present this set for the geometry at hand . for a free electromagnetic field the maxwell equations have the form @xmath15where @xmath16 is the electromagnetic field tensor , @xmath17 . we assume that on the surface @xmath14 the field obeys the boundary condition @xmath18where @xmath19 is the normal to the boundary , @xmath20 is the dual of the field tensor @xmath16 . for @xmath21 the condition ( [ bc1 ] ) reduces to the boundary condition on the surface of a perfect conductor . we want to find the complete set of solutions to the equation ( [ meq ] ) in the coordinates @xmath22 . in the coulomb gauge one has @xmath23 , @xmath24 , @xmath25 . for the geometry under consideration the latter equation is reduced to@xmath26which is the same as that in the minkowski spacetime . if we present the solution in the factorized form , @xmath27 , then it can be shown that the parts of the mode functions corresponding to @xmath28 are found in a way similar to that for the @xmath0-dimensional minkowski bulk . the corresponding modes are presented in appendix [ sec : app1 ] . from the field equations ( [ meq ] ) for the function @xmath29 one gets @xmath30 , where @xmath31 and @xmath32 is a cylinder function of the order @xmath33 . it can be taken as a linear combination of the hankel functions @xmath34 . the relative coefficient in the linear combination depends on the vacuum state under consideration . we assume that the field is prepared in the bunch - davies vacuum @xcite for which @xmath35 . among a one - parameter family of maximally symmetric vacuum states in ds spacetime the bunch - davies vacuum is the only one with the hadamard singularity structure . as a result , by taking into account the expressions for the minkwoskian modes ( [ a1 m ] ) and ( [ amum ] ) , for the modes in ds spacetime realizing the bunch - davies vacuum state one finds the expressions@xmath36for the polarization @xmath37 , and @xmath38for the remaining polarizations @xmath39 , with @xmath40 . here , @xmath41 , @xmath42 is a cylinder function , @xmath43 , @xmath44 , and@xmath45with @xmath46 . for the polarization vectors @xmath47 one has @xmath48 and the relations ( [ rel1 m ] ) , ( [ rel2 m ] ) . the set of quantum numbers specifying the modes is reduced to @xmath49 . it can be easily checked that the modes ( [ a1 ] ) and ( [ a2 ] ) obey the gauge condition ( [ gcond ] ) . the eigenvalues for the quantum number @xmath50 are determined by the boundary condition ( [ bc1 ] ) . first we consider the region inside the cylindrical shell , @xmath51 . from the regularity condition at @xmath52 it follows that@xmath53with @xmath54 being the bessel function . for the mode @xmath37 the allowed values for @xmath50 are roots of the equation@xmath55where the prime means the derivative with respect to the argument of the function . for the modes @xmath39 , the radial derivative enters in the expression for the component @xmath56 only . for these modes the boundary condition is reduced to@xmath57 in what follows we will denote the eigenmodes by @xmath58 , @xmath59 , where @xmath60 for @xmath37 and @xmath61 for @xmath62 . hence , one has @xmath63 , with @xmath64 and @xmath65 . the normalization coefficients @xmath66 in ( [ a1 ] ) and ( [ a2 ] ) are determined from the orthonormalization condition for the vector potential : @xmath67=4i\pi \delta _ { \beta \beta ^{\prime } } , \label{normcond}\]]where the integration over the radial coordinate goes over the region inside the cylinder and @xmath68 is understood as the kronecker symbol for the discrete components of the collective index @xmath69 and the dirac delta function for the continuous ones . by using the relation ( [ rel1 m ] ) it can be seen that the modes ( [ a2 ] ) are orthogonal . from ( [ normcond ] ) one finds @xmath70for @xmath71 , where we have introduced the notation@xmath72 ^{-1}. \label{tnu}\ ] ] having the complete set of mode functions ( [ a1 ] ) and ( [ a2 ] ) we can evaluate the vev of any physical quantity @xmath73 bilinear in the field . by expanding the operator of the vector potential in terms of the modes ( [ a1 ] ) and ( [ a2 ] ) and using the commutation relations for the annihilation and creation operators , the following mode - sum formula is obtained@xmath74where @xmath75 stands for the vacuum state , @xmath76 includes the summation over the discrete quantum numbers and the integration over the continuous ones . the expression in the right - hand side of ( [ vev ] ) is divergent and requires a regularization with the subsequent renormalization . the regularization can be done by introducing a cutoff function or by the point splitting . the consideration below does not depend on the specific regularization scheme and we will not specify it . as a local characteristic of the vacuum state we consider the vev of the squared electric field . this vev is obtained by making use of the mode - sum formula @xmath77note that the vev of the electric field squared determines the casimir - polder potential between the shell and a polarizable particle with a frequency - independent polarizability . substituting the eigenfunctions ( a1 ) and ( [ a2 ] ) , after the summation over @xmath78 with the help of ( [ rel2 m ] ) , for the vev inside the shell we find @xmath79l_{d/2 - 2}(\omega _ { m , n}^{(\lambda ) } \eta ) , \label{e2}\end{aligned}\]]where @xmath80 , the prime on the sign of the sum means that the term @xmath81 should be taken with the coefficient @xmath82,@xmath83and we have used the notation@xmath84here , instead of the hankel functions we have introduced the macdonald function @xmath85 . the function @xmath86 $ ] is defined by the relations@xmath87 = \left\ { \begin{array}{cc } ( k^{2}r^{2}/x^{2})\left [ f^{\prime 2}(x)+m^{2}f^{2}(x)/x^{2}\right ] + \left [ \left ( d-3\right ) \left ( 1+k^{2}r^{2}/x^{2}\right ) + 1\right ] f^{2}(x ) , & \lambda = 0 , \\ ( 1+k^{2}r^{2}/x^{2})\left [ f^{\prime 2}(x)+m^{2}f^{2}(x)/x^{2}\right ] , & \lambda = 1.% \end{array}% \right . \label{gm}\ ] ] the eigenvalues @xmath88 are given implicitly and the representation ( [ e2 ] ) is not convenient for the further investigation of the vev . for the further evaluation of the mode - sum in ( [ e2 ] ) , we apply to the series over @xmath89 the generalized abel - plana summation formula sahabook @xmath90 , \label{sumform}\end{aligned}\]]where @xmath91 is the neumann function and @xmath92 is the modified bessel function of the first kind . in ( sumform ) it is assumed that the function @xmath93 is analytic in the right half plane of the complex variable @xmath94 . for the series in ( e2 ) , the corresponding function @xmath93 is given by@xmath95 . \label{fz}\]]note that this function has branch points @xmath96 on the imaginary axis . for the function ( [ fz ] ) one has the relation@xmath97 \sum_{j=-1,1}jl_{d/2 - 2}(\eta \sqrt{z^{2}e^{-ji\pi } /a^{2}+k^{2 } } ) , \label{sumf}\]]with@xmath98here and in what follows we use the notation@xmath99 . \label{gnu}\ ] ] after the application of the summation formula ( [ sumform ] ) , the vev of the electric field squared is decomposed as @xmath100here , the first term in the right - hand side comes from the first integral in ( [ sumform ] ) . it does not depend on the shell radius @xmath2 and corresponds to the vev in ds spacetime in the absence of boundaries : @xmath101l_{d/2 - 2}(\eta \sqrt{k^{2}+x^{2 } } ) , \label{e20}\]]where@xmath102 & = & \left ( 1 + 2\frac{k^{2}r^{2}}{x^{2}}\right ) \left [ f^{\prime 2}(x)+\frac{m^{2}}{x^{2}}f^{2}(x)\right ] \notag \\ & & + \left [ \left ( d-3\right ) \left ( 1+\frac{k^{2}r^{2}}{x^{2}}\right ) + 1% \right ] f^{2}(x ) . \label{g0}\end{aligned}\]]the part of the vev @xmath103 is the contribution of the last integral in ( [ sumform ] ) . this part is induced by the presence of the cylindrical boundary and is given by the expression @xmath104 f_{d/2 - 2}\left ( xs\right ) , \label{e2b}\end{aligned}\]]with the notation@xmath105 = \left\ { \begin{array}{cc } \left ( 1-s^{2}\right ) \left [ f^{\prime 2}(x)+m^{2}f^{2}(x)/x^{2}\right ] + % \left [ s^{2}\left ( d-3\right ) + 1\right ] f^{2}(x ) , & \lambda = 0 , \\ -s^{2}\left [ f^{\prime 2}(x)+m^{2}f^{2}(x)/x^{2}\right ] , & \lambda = 1.% \end{array}% \right . \label{gb}\]]in deriving ( [ e2b ] ) , after using ( [ sumf ] ) and ( [ suml ] ) , we have introduced a new integration variable @xmath106 in accordance with @xmath107 and then passed to polar coordinates in the plane @xmath108 . the representation ( [ e2b ] ) is valid for all even values of @xmath109 and for @xmath110 in the case of odd @xmath109 . the shell - induced contribution ( [ e2b ] ) depends on the variables @xmath111 , @xmath2 , @xmath112 in the form of the ratios @xmath113 and @xmath114 . this feature is a consequence of the maximal symmetry of ds spacetime . note that the combination @xmath115 is the proper radius of the cylindrical shell and , hence , @xmath116 is the proper radius measured in units of the ds curvature scale @xmath6 . similarly , the ratio @xmath114 is the proper distance from the cylinder axis measured in units of @xmath6 . as is seen from ( [ e20 ] ) and ( [ e2b ] ) , in the new representation of the vev the explicit knowledge of the eigenvalues @xmath117 is not required . another advantage is that we have explicitly separated the part corresponding to the boundary - free ds spacetime . the presence of the boundary does not change the local geometry for points outside the shell . this means that at those points the divergences are the same in both the problems , in the absence and in the presence of the cylindrical shell . from here it follows that the divergences in ( [ e2dec ] ) are contained in the part @xmath118 only , whereas the boundary - induced contribution @xmath103 is finite for points away from the boundary and the regularization , implicitly assumed in the discussion above , can be safely removed in that part . hence , the renormalization is required for the boundary - free part only . note that the expression for the latter can be further simplified after the summation over @xmath119 by using the standard result for the series involving the square of the bessel function . let us consider the properties of the boundary - induced contribution in the vev of the field squared . first of all , by taking into account that for @xmath120 one has @xmath121 > 0 $ ] , @xmath122 < 0 $ ] , and the function @xmath123 is positive for the values of @xmath109 for which the representation ( [ e2b ] ) is valid , from ( [ e2b ] ) it follows that @xmath103 is always positive . the vev of the electric field squared inside a cylindrical shell in minkowski spacetime is obtained by the limiting transition @xmath124 for a fixed value of @xmath9 . in this limit one has @xmath125 and , hence , @xmath111 is large . passing to a new integration variable @xmath126 , we see that @xmath111 appears in the argument of the function @xmath127 . by taking into account that for large arguments one has @xmath128 , after the integration over @xmath129 , we get @xmath130 , where@xmath131 , \label{e2mb}\]]with@xmath132 = \left\ { \begin{array}{cc } \left ( d-2\right ) \left [ f^{\prime 2}(x)+(m^{2}/x^{2}+2)f^{2}(x)\right ] , & \lambda = 0 , \\ -f^{\prime 2}(x)-m^{2}f^{2}(x)/x^{2 } , & \lambda = 1,% \end{array}% \right . \label{gmmlam}\]]is the vev of the electric field squared inside a cylindrical shell in the minkowski bulk . in the special case of 4-dimensional ds spacetime one has @xmath21 and @xmath133 . after the integration over @xmath129 in ( [ e2b ] ) , we find@xmath134where @xmath135 is given by ( e2 mb ) with @xmath21 . in this case , the vev of the field squared is related to the corresponding result in minkowski spacetime by standard conformal transformation with the conformal factor @xmath136 . this is a direct consequence of the conformal invariance of the electromagnetic field in @xmath21 spatial dimensions and of the conformal flatness of the background geometry . the vev of the electric field squared for a conducting cylindrical shell coaxial with a cosmic string in 4-dimensional spacetime ( @xmath137 ) is investigated in @xcite . in the absence of planar angle deficit , the corresponding expression is reduced to ( [ e2 mb ] ) with @xmath21 . the corresponding casimir - polder forces were discussed in @xcite . on the axis of the shell , @xmath52 , the only nonzero contribution to the boundary - induced vev comes from the terms in ( [ e2b ] ) with @xmath138 : @xmath139 . \label{e2br0}\end{aligned}\]]the shell - induced vev diverges on the cylindrical boundary . the surface divergences in the vevs of local physical observables are well - known in the theory of the casimir effect @xcite ( see also @xcite for recent discussions ) . near the shell the dominant contribution to the vevcomes from large values of @xmath119 . for @xmath140 , introducing in ( [ e2b ] ) a new integration variable @xmath141 , we use the uniform asymptotic expansions for the modified bessel functions with the order @xmath119 and the asymptotic for the function @xmath142 for large arguments ( see , for instance , @xcite ) . to the leading order one finds@xmath143^{d+1}}. \label{e2bnear}\]]note that the combination @xmath144 is the proper distance from the boundary . the leading term ( [ e2bnear ] ) is obtained from that for the cylindrical shell with the radius @xmath2 in minkowski spacetime by the replacement @xmath145 . for points near the boundary the contribution of the modes with small wavelengths dominate and at distances from the shell smaller than the curvature radius of the ds spacetime the influence of the gravitational field is small . for the numerical evaluations we have taken the model with @xmath1 . in this case for the function @xmath146 in ( [ e2b ] ) one has@xmath147and the integrals over @xmath129 are of the form@xmath148with @xmath149 . these integrals are evaluated in appendix [ sec : app2 ] . for the shell - induced part one gets @xmath150 \delta _ { 0\lambda } \mathcal{i}_{1,0}(x)\right . \notag \\ & & -\left . \left [ i_{m}^{\prime 2}(y)+\left ( m^{2}/y^{2}-\delta _ { 0\lambda } \right ) i_{m}^{2}(y)\right ] \mathcal{i}_{3,0}(x)\right\ } _ { y = xr/\eta } , \label{e2bd4}\end{aligned}\]]with @xmath151 and @xmath152 given by ( [ int0 ] ) , ( [ int4 ] ) . in figure [ fig1 ] , this contribution is plotted versus the proper distance from the shell axis , measured in units of @xmath6 . for the proper radius of the shell , in the same units , we have taken @xmath153 . the corresponding casimir - polder force is expressed in terms of the derivative @xmath154 . this force is directed toward the cylindrical shell . another important local characteristic of the vacuum state is the vev of the energy - momentum tensor . this vev is evaluated by using the mode - sum formula@xmath155where @xmath156 is the field strength corresponding to the mode functions ( [ a1 ] ) and ( [ a2 ] ) . by using the expressions for the mode functions , the mode - sums for the diagonal components of the vev are presented in the form ( no summation over @xmath157)@xmath158l_{d/2 - 1-l}(\omega _ { m , n}^{(\lambda ) } \eta ) , \label{tii}\end{aligned}\]]where @xmath159here we have used the notations@xmath160=(a_{\lambda l}^{(i)}k^{2}+b_{\lambda l}^{(i)}\gamma ^{2})z_{m}^{(i)}[f(y)]+(\left ( d-3\right ) c_{\lambda l}^{(i)}k^{2}+d_{\lambda l}^{(i)}\gamma ^{2})f^{2}(y ) , \label{til}\]]and@xmath161 & = & f^{\prime 2}(y)+m^{2}f^{2}(y)/y^{2},\;i=0,3,\ldots , d , \notag \\ z_{m}^{(i)}[f(y ) ] & = & f^{\prime 2}(y)-m^{2}f^{2}(y)/y^{2},\;i=1,2 . \label{zi}\end{aligned}\]]for the energy density the coefficients in ( [ til ] ) are given by the expressions@xmath162where the rows and columns are numbered by @xmath163 and @xmath164 , respectively . for the coefficients in the radial and azimuthal components one has@xmath165and @xmath166 , @xmath167 , @xmath168 , @xmath169 . for the axial components , @xmath170 , we get@xmath171 in addition to the diagonal component , the vev of the energy - momentum tensor has a nonzero off - diagonal component@xmath172where @xmath173 , @xmath174 . this component describes energy flux along the radial direction ( along the direction normal to the boundary ) . after the application of the summation formula ( [ sumform ] ) , with the function@xmath175 , \label{fxemt}\]]to the series over @xmath89 in ( [ tii ] ) , the vev of the energy - momentum tensor is presented in the form@xmath176where @xmath177 is the corresponding vev in the boundary - free ds spacetime and the contribution @xmath178 is induced by the presence of the cylindrical shell . the boundary - free contribution corresponds to the first integral in the right - hand side of the formula ( [ sumform ] ) and the boundary - induced contribution comes from the second integral . for points outside the cylindrical shell the boundary - induced contribution in ( [ tdec ] ) is finite and the renormalization is reduced to the one for the boundary - free part . from the maximal symmetry of the bunch - davies vacuum state it follows that the latter does not depend on the spacetime point and has the form @xmath179 . with the help of the transformations similar to those we have used for the vev of the field squared , for the boundary - induced parts in the diagonal components we get ( no summation over @xmath157)@xmath180f_{d/2 - 1-l}(xs ) , \label{tiib}\end{aligned}\]]where we have defined the functions@xmath181=(a_{\lambda l}^{(i)}+b_{\lambda l}^{(i)}s^{2})z_{m}^{(i)}[f(y)]+(c_{\lambda l}^{(i)}+\left ( d-3\right ) d_{\lambda l}^{(i)}s^{2})f^{2}(y ) , \label{pi}\]]with @xmath182 $ ] given by ( [ zi ] ) . for the energy density the coefficients in ( [ pi ] ) are given by the expressions@xmath183for the radial and azimuthal stresses one has@xmath184with @xmath185 . and finally , for the axial stresses ( @xmath170 ) we get:@xmath186note that , unlike to the case of the minkowski bulk ( see below ) , for the ds bulk the axial stresses do not coincide with the energy density . for the off - diagonal component ( [ t10 ] ) , the contribution of the first integral in ( [ sumform ] ) to the vev vanishes . this directly follows from the relations @xmath187 and @xmath188 . the nonzero part is induced by the presence of the shell and is given by the expression@xmath189 i_{m}(xr/\eta ) i_{m}^{\prime } ( xr/\eta ) \notag \\ & & \times \int_{0}^{1}ds\,s^{2}\left ( 1-s^{2}\right ) ^{d/2 - 2}\left [ k_{d/2 - 1}\left ( y\right ) i_{2-d/2}\left ( y\right ) -k_{d/2 - 2}\left ( y\right ) i_{d/2 - 1}\left ( y\right ) \right ] _ { y = xs}. \label{t01b}\end{aligned}\]]in the special case @xmath21 the off - diagonal component vanishes . for other values of @xmath109 , for which the representation ( [ t01b ] ) is valid both the functions in the square brackets are positive and , hence , @xmath190 for @xmath191 . on the axis , the energy flux vanishes , @xmath192 . similar to the case of the field squared , the boundary - induced vevs ( [ tiib ] ) and ( t01b ) depend on @xmath111 , @xmath2 , @xmath112 in the form of the ratios @xmath116 and @xmath193 . with the expressions ( [ tiib ] ) and ( [ t01b ] ) , we can see that the boundary - induced contributions obey the covariant continuity equation @xmath194 . for the geometry at hand , this equation is reduced to the following relations between the vevs:@xmath195let us denote by @xmath196 the shell - induced contribution in the vacuum energy in the region @xmath197 , per unit coordinate lengths along the directions @xmath198 : @xmath199by taking into account the first equation in ( [ conteq ] ) , the corresponding derivative with respect to the synchronous time coordinate @xmath9 is expressed as ( see also @xcite)@xmath200this relation shows that @xmath201 is the energy flux per unit proper surface area . the quantity @xmath202 is the shell - induced contribution to the vacuum pressure along the @xmath203-th direction and the first term in the right - hand side of ( [ eder ] ) is the work done by the surrounding on the selected volume . the last term in ( eder ) is the energy flux through the surface @xmath204 . inside the shell one has @xmath205 and the flux is directed from the shell to the axis @xmath52 . let us discuss special cases of the general expressions for the vevs of the energy - momentum tensor components . first we consider the minkowskian limit , corresponding to @xmath124 . introducing in ( [ tiib ] ) a new integration variable @xmath126 , we see that the argument of the functions @xmath206 is large and we can use the asymptotic expression @xmath207 . after the integration over @xmath129 , to the leading order we get @xmath208 , where for the vevs on the minkowski bulk one has ( no summation over @xmath157)@xmath209+b_{\lambda } ^{(i)}i_{m}^{2}(rx)\right\ } , \label{tiim}\]]with the coefficients@xmath210for @xmath211 . for the stresses along the directions @xmath170 we have @xmath212 . in the special case @xmath21 , from ( [ tiim ] ) we obtain the results previously derived in @xcite . as is seen , for the minkowski bulk the axial stresses are equal to the energy density . this result could be directly obtained on the base of the invariance of the problem with respect to the lorentz boosts along the directions of the axis @xmath213 , @xmath214 . the off - diagonal component vanishes in the minkowskian limit . for the leading term in the corresponding asymptotic expansion from ( [ t01b ] ) we find@xmath215 i_{m}(xr)i_{m}^{\prime } ( xr ) . \label{t01bm}\ ] ] in the special case @xmath21 the off - diagonal component of the vacuum energy - momentum tensor vanishes , @xmath216 , and the diagonal components are connected to the corresponding quantities for a cylindrical shell in the minkowski bulk by the relation ( no summation over @xmath157)@xmath217 in this special case @xmath218 for @xmath219 , and @xmath220 for all @xmath157 . note that the casimir self - stress for an infinite perfectly conducting cylindrical shell in background of 4-dimensional minkowski spacetime has been evaluated in @xcite on the base of a green s function technique . the corresponding casimir energy was investigated by using the zeta function technique in @xcite and the mode - by - mode summation method in @xcite . the geometry of a cylindrical shell coaxial with a cosmic string was considered in beze07,brev95 . near the cylindrical shell , the asymptotic expressions for the components of the energy - momentum tensor are found in the way similar to that for the vev of the electric field squared , by using the uniform asymptotic expansions for the modified bessel functions . the leading terms are given by the expressions ( no summation over @xmath157)@xmath221^{d+1 } } , \label{tiinear}\]]for @xmath222 , and@xmath223the leading terms for the diagonal components coincide with those for a cylindrical shell in minkowski bulk with the distance from the shell replaced by the proper distance @xmath144 . in the special case @xmath21 the leading terms vanish . the latter feature is related to the conformal invariance of the electromagnetic field in @xmath21 . in the special case @xmath1 , the integrals over @xmath129 in ( [ tiib ] ) are evaluated in appendix [ sec : app2 ] . the expression for the diagonal components takes the form ( no summation over @xmath157)@xmath224\right . \notag \\ & & \left . + ( c_{\lambda l}^{(i)}\mathcal{i}_{1,1-l}(x)+d_{\lambda l}^{(i)}% \mathcal{i}_{3,1-l}(x))i_{m}^{2}(xr/\eta ) \right\ } , \label{tiid4}\end{aligned}\]]where the coefficients are given by ( [ a0 ] ) , ( [ ai1 ] ) and ( [ ai2 ] ) with @xmath1 . in the expression ( [ t01b ] ) for the energy flux , with @xmath1 , the integral over @xmath129 is evaluated by using the formulas ( [ int11 ] ) and ( [ int12 ] ) . this leads to the following result : @xmath225 \notag \\ & & \times i_{m}(xr/\eta ) i_{m}^{\prime } ( xr/\eta ) i_{1}\left ( x\right ) k_{1}\left ( x\right ) . \label{t01d4}\end{aligned}\ ] ] in figure [ fig2 ] we have plotted the boundary - induced contribution in the energy density and the energy flux as functions of the proper distance from the shell axis , measured in units of the ds curvature scale @xmath6 . for the corresponding value of the shell radius we have taken @xmath153 . as is seen , in the interior region the energy density is negative near the shell and positive near the axis of the shell . the energy flux is negative inside the shell . this means that the energy flux is directed from the shell . the corresponding energy density in the minkowski bulk is negative everywhere . [ cols="^,^ " , ] in the region outside the shell , @xmath226 , the radial functions @xmath42 in ( [ a1 ] ) and ( [ a2 ] ) are linear combinations of the bessel and neumann functions . the relative coefficients in the linear combinations are determined from the boundary condition ( [ bc1 ] ) on the cylindrical surface @xmath14 . in this way we can see that @xmath227for the modes @xmath163 . now , in the normalization condition ( normcond ) the integration over the radial coordinate goes over the region @xmath228 and in the right - hand side the delta symbol for the quantum number @xmath50 is understood as the dirac delta function @xmath229 . as the normalization integral diverges for @xmath230 , the main contribution to the integral comes from large values of @xmath112 . by making use of the asymptotic formulas for the bessel and neumann functions with large arguments , for the normalization coefficients in ( [ a1 ] ) and ( [ a2 ] ) one gets @xmath231^{-1}}{4\left ( 2\pi \alpha \right ) ^{d-3}\gamma } . \label{cout}\]]similar to the case of the interior region , we consider the vevs of the field squared and energy - momentum tensor separately . for the vev of the electric field squared , by using the mode - sum formula ( [ e2mode ] ) , one finds @xmath232}{j_{m}^{(\lambda ) 2}(\gamma a)+y_{m}^{(\lambda ) 2}(\gamma a)}l_{d/2 - 2}(\omega \eta ) , \label{e2out}\end{aligned}\]]with the same notations as in ( [ e2 ] ) . with the help of the identity@xmath233}{% j_{m}^{(\lambda ) 2}(\gamma a)+y_{m}^{(\lambda ) 2}(\gamma a)}=g_{m}^{(\lambda ) } [ j_{m}(\gamma r)]-\frac{1}{2}\sum\limits_{j=0,1}\frac{j_{m}^{\left ( \lambda \right ) } \left ( \gamma a\right ) } { h_{m}^{\left ( j\right ) \left ( \lambda \right ) } \left ( \gamma a\right ) } g_{m}^{(\lambda ) } [ k , h_{m}^{\left ( j\right ) } ( \gamma r ) ] , \label{ident}\]]the vev is presented in the decomposed form ( [ e2dec ] ) where the shell - induced contribution is given by the expression@xmath234 for the further transformation , in ( [ e2bout ] ) , we rotate the contour of integration in the complex plane @xmath50 by the angle @xmath235 for the term with @xmath236 and by the angle @xmath237 for @xmath238 . introducing the modified bessel functions , the expression ( [ e2bout ] ) takes the form @xmath239 f_{d/2 - 2}\left ( xs\right ) , \label{e2bout2}\end{aligned}\]]where the functions @xmath240 $ ] are defined by the formulae ( [ gb ] ) . comparing this result with ( [ e2b ] ) , we see that the expressions for the interior and exterior regions are related by the interchange @xmath241 . in particular , the result in the minkowskian limit is obtained from ( [ e2 mb ] ) by the same replacements . in the special case @xmath21 the electromagnetic field is conformally invariant and we have the relation ( [ e2conf ] ) . similar to the case of the interior region , the shell - induced contribution ( [ e2bout ] ) in the vev of the electric field squared is positive . for points near the shell , the leading term in the asymptotic expansion over the distance from the boundary is obtained from ( [ e2bnear ] ) by the replacement @xmath242 . at large proper distances from the shell compared with the ds curvature radius , we have @xmath243 . introducing in ( [ e2bout ] ) a new integration variable @xmath244 and assuming that @xmath245 , we use the expansions of the functions @xmath246 and @xmath247 for small values of the arguments . the leading contribution comes from the term with @xmath248 and @xmath81 . for even @xmath249 one gets @xmath250 in the case @xmath1 the leading term is given by @xmath251for @xmath21 and @xmath252 we find@xmath253at large distances , the total vev is dominated by the boundary - free part @xmath254 . for a cylindrical shell in the minkowski bulk , at large distances , @xmath245 , one has the following asymptotic behavior:@xmath255for all values @xmath256 . note that for @xmath257 the leading term ( e2bmfar ) is obtained from ( [ e2bfard35 ] ) by the replacement of the proper distance from the axis , @xmath258 . for the special case @xmath1 , the shell - induced contribution in the vev of the squared electric field , for the exterior region , is presented in figure fig1 as a function of the ratio @xmath114 . similar to the region inside the shell , the corresponding casimir - polder force is directed toward the shell . now let us consider the vev of the energy - momentum tensor outside the cylindrical shell . by using the mode - sum formula ( [ tmunu ] ) with the exterior modes , for the vev of the diagonal components we get the following representation ( no summation over @xmath157)@xmath259}{j_{m}^{(\lambda ) 2}(\gamma a)+y_{m}^{(\lambda ) 2}(\gamma a)}l_{d/2 - 1-l}(\omega \eta ) , \label{tiiext}\end{aligned}\]]with the notation ( [ til ] ) . the expression for the off - diagonal component has the form@xmath260where @xmath261 . further transformation of the vevs is similar to that we have used for the field squared . in the case of the diagonal components we employ the relation@xmath262}{% j_{m}^{(\lambda ) 2}(\gamma a)+y_{m}^{(\lambda ) 2}(\gamma a)}=t_{\lambda l}^{(i)}[k,\gamma , j_{m}(\gamma r)]-\frac{1}{2}\sum\limits_{j=1,2}\frac{% j_{m}^{\left ( \lambda \right ) } \left ( \gamma a\right ) } { h_{m}^{\left ( j\right ) \left ( \lambda \right ) } \left ( \gamma a\right ) } t_{\lambda l}^{(i)}[k,\gamma , h_{m}^{\left ( j\right ) } ( \gamma r ) ] , \label{ident2}\]]the part with the first term in the right - hand side coincides with the vev in the boundary - free ds spacetime . in the part corresponding to the last term in ( [ ident2 ] ) we rotate the integration contours by the angle @xmath263 for the term with @xmath236 and by the angle @xmath237 for @xmath238 . in this way for the boundary - induced contributions in the vevs of the diagonal components we find ( no summation over @xmath157)@xmath264f_{d/2 - 1-l}(xs ) . \label{tiibout}\end{aligned}\]]in a similar way , the expression for the vev of the off - diagonal component is presented as @xmath265 k_{m}(xr/\eta ) k_{m}^{\prime } ( xr/\eta ) \notag \\ & & \times \int_{0}^{1}ds\,s^{2}\left ( 1-s^{2}\right ) ^{d/2 - 2}\left [ k_{d/2 - 1}\left ( y\right ) i_{2-d/2}\left ( y\right ) -k_{d/2 - 2}\left ( y\right ) i_{d/2 - 1}\left ( y\right ) \right ] _ { y = xs}. \label{t10bout}\end{aligned}\]]for @xmath266 , in the range of validity of this representation one has @xmath267 . the corresponding energy flux is directed from the cylindrical shell to the infinity . again , we can see that the shell - induced contributions obey the relations ( [ conteq ] ) . the vevs in the minkowskian limit , @xmath124 , are obtained in the way similar to that for the interior region . in this limit the energy flux vanishes and the corresponding expressions for the diagonal components are obtained from ( [ tiim ] ) by the replacements @xmath268 . in the special case @xmath21 , the vevs in the ds and minkowski bulks are connected by the conformal relation ( [ tiid3 ] ) . for points near the shell , the leading terms in the asymptotic expansions over the distance from the boundary for the components @xmath269 with @xmath270 are given by ( [ tiinear ] ) with the replacement @xmath242 . for the normal stress and the energy flux we have the relations ( [ t10near ] ) . hence , near the shell and for @xmath271 , the boundary - induced contribution in the energy density has the same sign in the exterior and interior regions , whereas the normal stress and the energy flux have opposite signs . let us consider the asymptotics of the boundary - induced contributions at large distances from the shell . introducing in ( [ tiibout ] ) and ( t10bout ) a new integration variable @xmath272 , we see that for @xmath273 the arguments of the functions @xmath274 , @xmath275 and @xmath276 are small . by using the corresponding asymptotic expressions one can show that the dominant contribution comes from the term @xmath81 and @xmath248 . for even values of @xmath277 , to the leading order one gets the following expressions ( no summation over @xmath157)@xmath278with the coefficients@xmath279@xmath280 . the corresponding energy density is positive . by taking into account that near the shell the energy density is negative for @xmath281 , we conclude that at some intermediate value of the radial coordinate the boundary - induced contribution in the energy density vanishes . the asymptotic ( [ tiilarge ] ) for the energy flux is valid in the case @xmath282 as well . for @xmath1 , the leading term in the expansion of the diagonal components is given by ( no summation over @xmath157)@xmath283where @xmath284 , and @xmath285 for@xmath286 . for the minkowski bulk , the large distance asymptotic is given by ( no summation over @xmath157 ) @xmath287where@xmath288the corresponding energy density is positive for @xmath21 and negative for @xmath281 . the shell - induced contribution to the energy density and the energy flux in the exterior region are plotted in figure [ fig2 ] for the @xmath1 model . the large distance asymptotic is given by ( [ tiilarged4 ] ) and the energy density is positive . for points near the shell we have the asymptotic behavior ( [ tiinear ] ) , with the replacement @xmath242 , and the energy density is negative . note that for the @xmath1 minkowski bulk the corresponding energy density is negative everywhere . for the @xmath21 model , the energy density is positive in the exterior region and negative in the interior region . in the discussion above we have considered the boundary condition ( [ bc1 ] ) that generalizes the condition at the surface of a conductor for arbitrary number of spatial dimensions . another type of boundary conditions for a gauge field is used in bag models of hadrons and in flux tube models of confinement in quantum chromodynamics ( see , for instance , @xcite ) . this boundary condition has the form@xmath289on the boundary of a volume inside of which the gluons are confined . in flux tube models the gauge field is confined inside a cylinder . the corresponding casimir densities inside and outside a cylindrical shell are investigated in a way similar to that we have described above . the mode functions still have the form ( [ a1 ] ) and ( [ a2 ] ) . imposing the boundary condition ( bc2 ) , we can see that , in the interior region , eigenvalues for @xmath50 are roots of the equation ( [ bcb ] ) for the mode @xmath37 and the roots of ( [ bca ] ) for @xmath39 . the final expressions for the vevs are obtained from those given above by the replacement@xmath290 in both the interior and exterior regions . in particular , the vev of the squared electric field is negative in these regions . in the investigations of the casimir effect the cylindrically symmetric boundaries are among the most popular geometries . in the present paper we have investigated the local casimir densities for the electromagnetic field inside and outside a cylindrical shell in background of @xmath0-dimensional ds spacetime . on the shell , the field tensor obeys the boundary condition ( [ bc1 ] ) . in the special case @xmath21 this corresponds to the perfect conductor boundary condition . the procedure , we employed for the evaluation of the vevs bilinear in the field , is based on the mode - sum formula ( vev ) . in this procedure the complete set of cylindrical mode functions for the electromagnetic field , obeying the boundary condition , is required . in the problem under consideration one has a single mode of the te type and @xmath291 modes of the tm type . for the bunch - davies vacuum state the corresponding vector potentials are given by the expressions ( [ a1 ] ) and ( [ a2 ] ) with the radial functions ( [ radin ] ) and ( [ radout ] ) for the exterior and interior regions , respectively . we have investigated the combined effects of a cylindrical boundary and background gravitational field on the vevs of the electric field squared and of the energy - momentum tensor . in the interior region the eigenvalues of the quantum number @xmath50 are expressed in terms of the zeros of the bessel function @xmath54 for the tm modes and in terms of the zeros of the derivative @xmath292 in the case of the te mode . for the summation of the series over these zeros we have used the generalized abel - plana summation formula ( [ sumform ] ) . this allowed us to extract from the vevs the parts corresponding to the boundary - free ds spacetime and to present the shell - induced contributions in terms of strongly convergent integrals , for points away from the boundary . with this separation , the renormalization of the vevs is reduced to the one for the boundary - free geometry . as a result , inside the shell , the vevs are decomposed as ( e2dec ) and ( [ tdec ] ) with the shell - induced parts given by ( [ e2b ] ) for the electric field squared and by ( [ tiib ] ) for the diagonal components of the energy - momentum tensor . a similar decomposition is provided for the exterior region . the expressions for the shell - induced parts in this region differ from the ones inside the shell by the replacements @xmath241 of the modified bessel functions ( see ( [ e2bout2 ] ) and ( [ tiibout ] ) ) . for both the interior and exterior regions the shell - induced contributions to the vev of the electric field squared are positive . in addition to the diagonal components , the vev of the energy - momentum tensor has nonzero off - diagonal component @xmath293 . it corresponds to the energy flux along the radial direction and is given by the expressions ( [ t01b ] ) and ( [ t10bout ] ) for the exterior and interior regions . the off - diagonal component is negative inside the shell and positive in the exterior region . this means that the energy flux is directed from the shell in both the regions . on the axis of the shell the flux vanishes . we have considered various special cases of general formulas . in the limit @xmath294 , the vevs inside and outside a cylindrical shell in the background of @xmath0-dimensional minkowski spacetime are obtained . the corresponding expressions generalize the results previously known for @xmath137 to an arbitrary number of spatial dimensions . note that for @xmath21 the electromagnetic field is conformally invariant and the shell - induced vevs in the ds bulk are obtained from those in minkowski spacetime by the standard conformal transformation . for points near the cylindrical boundary the contribution of small wavelengths dominates in the shell - induced vevs . the leading terms in the corresponding asymptotic expansions for the field squared and diagonal components of the energy - momentum tensor coincide with those for a cylindrical shell in the minkowski bulk with the distance from the shell replaced by the proper distance in ds bulk . the leading term in the energy flux is given by the relation ( [ t10near ] ) . the effects of the background gravitational field on the shell - induced vevs are essential at distances from the boundary larger than the curvature radius of the ds spacetime . in particular , for the numerical example considered by us in the case @xmath1 , at large distances the shell - induced contribution to the vacuum energy density is negative for the minkowski bulk and positive for ds background . near the shell , the energy density is negative in both these cases . as a consequence , for the ds bulk it vanishes for some intermediate value of the radial coordinate . another boundary condition , used for the confinement of gauge fields in bag models of hadrons and in flux tube models of qcd , is the one given by ( bc2 ) . the corresponding expressions for the vevs of the field squared and energy - momentum tensor are obtained from those for generalized perfect conductor boundary condition by the replacement ( [ replqcd ] ) . in this case , the boundary - induced contribution on the vev of the squared electric field is neagtive in both the interior and exterior regions . a. a. s. and n. a. s. were supported by the state committee of science ministry of education and science ra , within the frame of grant no . scs 15t-1c110 , and by the armenian national science and education fund ( ansef ) grant no . in this section we consider the cylindrical modes for the electromagnetic field in @xmath0-dimensional minkowski spacetime . for the electromagnetic field one has @xmath295 polarization states specified by @xmath296 . the vector potential for the polarization @xmath78 will be denoted by @xmath297 , @xmath298 . we will impose the gauge condition @xmath299 , where @xmath300 is the covariant derivative operator associated with the minkowskian metric tensor @xmath301 . from the field equation @xmath302 one gets@xmath303for @xmath304 , and@xmath305for the radial and azimuthal components . here@xmath306 for the polarization @xmath37 we take @xmath307it can be easily checked that this function obeys the gauge condition . the field equations ( [ aeqm2 ] ) are satisfied if the function @xmath308 obeys the equation@xmath309 for the polarizations @xmath39 we present the vector potential in the form@xmath310with scalar functions @xmath308 and @xmath311 . from the field equations ( [ aeqm ] ) it follows that the functions @xmath308 obey the wave equation ( [ psieqm ] ) . the gauge condition is reduced to@xmath312 the solutions for the scalar functions @xmath308 , @xmath313 , have the form@xmath314where @xmath42 is a cylinder function of the order @xmath315 , @xmath316 , and @xmath317 is given by ( [ om ] ) . we will normalize the polarization vectors in accordance with the relation@xmath318from the gauge condition one has @xmath319 , where @xmath320 . combining with ( [ epsnorm ] ) the following relations are obtained:@xmath321and@xmath322for @xmath323 . an alternative form for the cylindrical modes ( [ a2 m ] ) with the polarizations @xmath39 is obtained by making the gauge transformation@xmath324with the function@xmath325 in the new gauge the scalar potential vanishes , @xmath326 , and one has@xmath327 hence , for @xmath0-dimensional minkowski spacetime the cylindrical modes for the electromagnetic field in the gauge @xmath328 , @xmath329 are given by ( [ a1 m ] ) and ( [ amum ] ) , where for the scalar function @xmath308 one has the expression ( [ psi ] ) . the radial function @xmath330 is a linear combination of the bessel and neumann functions . the relative coefficient in this linear combination depends on the specific problem . for example , inside a cylindrical shell one has @xmath331 . in the special case @xmath21 , the modes ( [ a1 m ] ) and ( [ amum ] ) are reduced to the well known te and tm modes in cylindrical waveguides ( see , for instance , @xcite ) . in the @xmath332-dimensional case we have a single mode of the te type and @xmath333 modes of the tm type . the corresponding boundary conditions are discussed in section [ sec : modes ] . in this section we evaluate the integrals of the form ( [ ical ] ) appearing in the expressions for the vevs in the special case @xmath1 . first of all , by using the integration formula from @xcite we can see that@xmath334 , \notag \\ \mathcal{i}_{1,1}(x ) & = & \mathcal{i}_{1,0}(x)-\frac{2}{x}i_{1}(x)k_{0}(x ) . \label{int0}\end{aligned}\]]our starting point in the evaluation of the remaining integrals is the formula ( see , for instance , @xcite)@xmath335applying the operator @xmath336 on the left and right hand sides of this formula we find@xmath337next , combining with the relation @xmath338 one gets@xmath339 for the evaluation of the integral @xmath340 we apply the operator @xmath341 on the left- and right - hand sides of ( [ int1 ] ) . this gives@xmath342 + \frac{1}{6}\left [ f_{0}\left ( x\right ) + \left ( 1 - 4/x^{2}\right ) f_{1}\left ( x\right ) \right ] . \label{int2}\]]integrating by parts the integral @xmath343 we can see that@xmath344by taking into account the relations ( [ int11 ] ) and ( [ int2 ] ) we obtain@xmath345 + \frac{1}{6}\left [ f_{0}\left ( x\right ) + \left ( 1 + 2/x^{2}\right ) f_{1}\left ( x\right ) \right ] \ , . \label{int4}\ ] ] 99 g. plunien , b. mller , w. greiner , phys . rep . * 134 * , 87 ( 1986 ) ; v.m . mostepanenko , n.n . trunov , _ the casimir effect and its applications _ ( oxford university press , oxford , 1997 ) ; k.a . milton , _ the casimir effect : physical manifestation of zero point energy _ ( world scientific , singapore , 2002 ) ; m. bordag , g.l . klimchitskaya , u. mohideen , v.m . mostepanenko , _ advances in the casimir effect _ ( oxford university press , oxford , 2009 ) ; _ casimir physics _ , edited by d. dalvit , p. milonni , d. roberts , f. da rosa , lecture notes in physics vol . 834 ( springer - verlag , berlin , 2011 ) . a.d . linde , _ particle physics and inflationary cosmology _ ( harwood academic publishers , chur , switzerland 1990 ) ; d.h . lyth , a. riotto , phys . rep . * 314 * , 1 ( 1999 ) ; b.a . bassett , s. tsujikawa , d. wands , rev . 78 , 537 ( 2007 ) . riess , et al . j. , * 116 * , 1009 ( 1998 ) ; s. perlmutter , et al . , astrophys . j. * 517 * , 565 ( 1999 ) ; a.g . riess et al . , astrophys . j. 659 , 98 ( 2007 ) ; d.n . spergel et al . , astrophys . 170 , 377 ( 2007 ) ; e. komatsu et al . , * 180 * , 330 ( 2009 ) ; p.a.r . ade et al . , a&a * 571 * , a16 ( 2014 ) . a.a . saharian , t.a . vardanyan , classical quantum gravity * 26 * , 195004 ( 2009 ) ; e. elizalde , a.a . saharian , t.a . vardanyan , phys . d * 81 * , 124003 ( 2010 ) ; a.a . saharian , int . j. mod a * 26 * , 3833 ( 2011 ) ; p. burda , jetp lett . * 93 * , 632 ( 2011 ) . b. allen , t. jacobson , commun . math . phys . * 103 * , 669 ( 1986 ) ; n.c . tsamis , r.p . woodard , j. math . * 48 * , 052306 ( 2007 ) ; t. garidi , j.p . gazeau , s. rouhani , m.v . takook , j. math . phys . * 49 * , 032501 ( 2008 ) ; a. youssef , phys lett . * 107 * , 021101 ( 2011 ) ; m.b . frb , a. higuchi , j. math . phys . * 55 * , 062301 ( 2014 ) . philbin , c. xiong , u. leonhardt , ann . 325 , 579 ( 2010 ) ; k.a . milton , phys . d * 84 * , 065028 ( 2011 ) ; f.d . mazzitelli , j.p . nery , a. satz , phys . d * 84 * , 125008 ( 2011 ) ; n. bartolo , r. passante , phys . a * 86 * , 012122 ( 2012 ) ; w.m.r . simpson , s.a.r . horsley , u. leonhardt , phys . a * 87 * , 043806 ( 2013 ) ; k.a . milton , s.a . fulling , p. parashar , p. kalauni , t. murphy , arxiv:1602.00916 . p. candelas , ann . phys . * 167 * , 257 ( 1986 ) ; p.m. fishbane , s.g . gasiorowich , p. kauss , phys . d * 37 * , 2623 ( 1988 ) ; b.m . barbashov , v.v . nesterenko , _ introduction to the relativistic string theory _ ( world scientific , singapore , 1990 ) .

we present maps of 7 young massive molecular clumps within 5 target regions in the (=10 ) line emission , using the nobeyama 45 m telescope . these clumps , which are not associated with clusters , lie at distances between 0.7 to 2.1kpc . we find clumps with radii of 0.51.7pc , masses of 4704200 , and velocity widths of 1.43.3 . all of the clumps are massive and approximately in virial equilibrium , suggesting they will potentially form clusters . three of our target regions are associated with h ii regions ( cwhrs " from clump with h ii regions ) , and the other two are without h ii regions ( cwohrs ) . the clumps can be classified into two morphological types : cwhrs shape a filamentary or shell - like structure , cwohrs are spherical . the two cwohrs have systematic velocity gradients . using the publicly released database , class i and class ii protostellar candidates were identified within the clumps . the fraction of the class i candidates among all yso candidates ( class i+class ii ) is 50 in cwhrs , and 50 in cwohrs . we conclude that effects from the h ii regions can be seen in ( 1 ) spatial distributions of the clumps : filamentary or shell - like structure running along the h ii regions , ( 2 ) velocity structures of the clumps : large velocity dispersion along shells , and ( 3 ) small age spreads of ysos . the small spread in age of the ysos show that the presence of h ii regions tend to trigger coeval cluster formation .

1301.3199

most stars , in particular high - mass stars ( @xmath9@xmath28@xmath3 ) , form as stellar clusters @xcite . stellar clusters form in dense and massive molecular clumps ( size @xmath10@xmath21@xmath2pc , mass @xmath10@xmath21001000@xmath2@xmath3 , density @xmath10 10@xmath11@xmath2@xmath12 ) with large velocity widths ( 24@xmath2km@xmath2s@xmath13 ) @xcite . typical median ages of stellar members in these clusters range between 1 and 3 myr @xcite , although the reason of this spread is unclear . since dense gas in the cluster forming clumps is significantly dispersed by the feedback from newly formed stars ( e.g. , * ? ? ? * ; * ? ? ? * ) , the initial conditions and details of cluster formation are not understood yet . therefore , it is necessary to study molecular clumps in the early stages of cluster formation . recently , infrared dark clouds ( irdcs ) have been identified as dark extinction features against the bright emission using @xmath14 @xcite , and glimpse 8@xmath15 data @xcite . irdcs were catalogued by @xcite and @xcite . their physical conditions are similar to those of cluster - forming clumps @xcite , which are the dense regions in molecular clouds . most irdcs are found to be associated with yso candidates in @xmath16 and @xmath17 observations @xcite . thus , irdcs are suggested to be precursors of cluster - forming clumps which still preserve initial conditions for the formation of stellar clusters @xcite because the dense gas component is not yet dispersed by the stellar feedback . some irdcs are detected near h ii regions ( e.g. , * ? ? ? * ; * ? ? ? in such cases , cluster formation may be triggered by h ii regions . recently , bubbles " , which are surrounding expanding h ii regions have been detected from @xmath16 and @xmath17 @xcite . the uv radiation from massive stars excite polycyclic aromatic hydrocarbons ( pahs ) at the bubble rims and also heat dust grains . the former can be readily observed at 8@xmath15 by irac @xcite and the latter at longer wavelengths using mips on @xmath16 @xcite . cold dense material has been observed around the borders of the h ii regions ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? how the cold dense material can be accumulated by the expanding shells from the h ii region has been discussed ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? most of the relevant observations have been conducted with mm dust continuum emission . considering dynamical effects such as shock waves and expanding h ii region shells should strongly influence surrounding molecular gas and change the kinematic structures significantly . it is a subject of great interest to investigate velocity structures in molecular gas near the boundary of h ii regions in a systematic way . @xcite presented maps of @xmath0(@xmath1=21 ) toward 48 massive molecular clumps and found that about half of their sample sources have velocity gradients of @xmath10 2 km@xmath2s@xmath13@xmath2pc@xmath13 . @xcite found similar velocity gradients in nh@xmath18 lines toward some irdcs . these studies do not focus on the external environments around the clumps . understanding the nature and origin of velocity gradients may be crucial in understanding the initial condition of cluster formation . if a clump is associated with an h ii region , the velocity structure of the clump will be influenced by shock waves and an expanding shell . understanding how clumps are influenced by h ii regions require comparison of velocity structures among clumps associated with and without h ii regions . in this paper , we present high - resolution observations ( effective beam size of @xmath19 ) with the nobeyama 45 m telescope toward seven nearby ( @xmath8 2.1@xmath2kpc ) massive clumps ( e.g. , infrared dark ) within five target regions . three of the objects are associated with well - known h ii regions ( named cwhr " , i.e. , clump with h ii region " ) , the others are without h ii region ( cwohr " , i.e. , clump without h ii region " ) . we used the @xmath0(@xmath1=10 ) molecular line emission which traces dense clumps with density around 10@xmath20@xmath2@xmath12 regions @xcite . @xmath0 is an excellent tracer of the h@xmath21 column density because in most of cases they are optically thin , and the relative abundance of co and its isotopes are in most cases unaffected by the prevailing changing in the gas studied ( see the discussion in @xcite ) . we analyzed @xmath5 archival data available for all target regions , in order to identify yso candidates which may be associated with clumps , and estimate their relative ages through its class i fraction in each region . here , we investigate the differences of spatial distributions , physical parameters , velocity structures , and activity of star formation ( e.g. , yso distributions , relative ages ) between cwhrs and cwohrs . our targets are selected from clumps that are expected to be in early stages of cluster formation @xcite . our criteria are that the clumps are massive ( @xmath10 100@xmath2@xmath3 ) , which estimated from dust continuum emission , and nearby within 2.1@xmath2kpc , in order to resolve the cloud structures corresponding to the jeans length ( assuming gas temperature of 15 k ) or finer with the 45 m telescope at 110@xmath2ghz . we finally selected seven young massive clumps ( e.g. , infrared dark ) within 5 target regions , as listed in table [ para ] . the association with h ii region or lack of such association is also indicated in table [ para ] . we carried out @xmath0(@xmath1=10 ; 109.782182 ghz ) observations using the 45 m telescope at the nobeyama radio observatory ( nro ) between december 2007 and may 2008 . we used the 25-beam array receiver system ( bears ) for the front end receiver , which has 5@xmath225 beams separated by [email protected] with respect to each other on the plane of the sky @xcite . at 110 ghz , the fwhm main beam size was 15@xmath24 and the main beam efficiency was 0.45 . for the back end , we used 25 sets of 1024 channel auto - correlators ( ac ) , each with 32 mhz bandwidth and 37.8 khz frequency resolution @xcite . the frequency resolution corresponds to a velocity resolution of 0.10@xmath2 km @xmath25 at 110 ghz . we applied on - the - fly mapping technique @xcite and covered areas of 6@xmath26@xmath226@xmath26 to 10@xmath26@xmath2210@xmath26 which corresponds to [email protected] to [email protected]@xmath2pc at a distance of 2.1@xmath2kpc . during the observations , the system temperatures in the double sideband mode ranged from 400 to 500@xmath2k . we used emission - free regions near the observed sources as the off positions . the standard chopper wheel method was used to calibrate the intensity scale into @xmath27 , the antenna temperature corrected for the atmospheric attenuation . the accuracy of telescope pointing was checked every 1.5 hr by observing sio maser sources near ( within @xmath10 45@xmath28 ) the target regions , and was confirmed to be better than 3@xmath23 . we correct the difference in intensity scales among the 25 beams of the bears by using reference data of w51 , which was obtained in observations with a single sideband filtered sis receiver s100 , and the acousto - optical spectrometer ( aos ) as the backend . we used a convolution scheme with a spheroidal function @xcite to calculate the intensity at each grid point of the final map - cube ( r.a . , decl . , and @xmath29 ) data with a spatial grid size of [email protected] , which is half of the beamsize ; the final effective resolution is 22@xmath23 and the effective integration time at each grid point is about 15 minutes . as a result , the typical rms noise level in channel maps for all of the target regions was 0.14 @xmath30 0.03 k in 0.1 km s@xmath13 . we also carried out @xmath31(1,1 ; 23.694495 ghz ) and @xmath31(2,2 ; 23.722633 ghz ) observations as the backup for bad weather . at 24 ghz , the fwhm main beam size and the main beam efficiency were 80@xmath24 and 0.8 , respectively . for the frontend , we used the 20 ghz hemt receiver ( h20 ) . the standard chopper wheel method was used to convert the received intensity into @xmath32 . the system noise temperature was 100 to 500 k. for the backend , we used the aoss , which have a velocity resolution of 0.5 km s@xmath13 at 24 ghz . we observed the @xmath31 ( 1,1 ) and ( 2,2 ) lines simultaneously by using two aoss each with a bandwidth of 32 mhz . the integration time of each point was typically 180 seconds , resulting in a rms noise level in the channel maps of 0.05 k. the telescope pointing was checked every 3 hours by observing the sio maser sources . the pointing accuracy was better than 15@xmath24 . data reduction and analysis were made with the newstar package developed at the nobeyama radio observatory ( nro ) . figures [ map1 ] to [ map5 ] show the integrated intensity maps of the @xmath0(@xmath1=10 ) emission ( contours ) superposed on infrared color images in which we used the @xmath5 3.4 , 4.6 , and 12 @xmath15 composite color images , and @xmath5 22 @xmath15 color images . we defined the @xmath0 emission enclosed the 3@xmath2@xmath33 contours of the individual map as @xmath0 clump . for irdc 20081 + 2720 , afgl 333 , and g 79.3 + 0.3 , the clumps are associated with well - known h ii regions ( see description of individual sources ) , while for irdc 19410 + 2336 and msxdc g 053.11 + 00.05 , there is no evidence of the existence of h ii region associated with the clumps . ] hereafter , we named the clumps associated with h ii regions as cwhr " ( clump with h ii region ) , and cwohr " ( clump without h ii region ) . from the morphological structure of the @xmath0 clumps , we can classify @xmath0 clumps into two types : a ) filamentary or shell - like structures ( irdc 20081 + 2720 , afgl 333 , and g 79.3 + 0.3 ) , b ) spherical ( irdc 19410 + 2336 , msxdc g 053.11 + 00.05 ) . although the number of samples is small , the cwhrs tend to have the shell - like structures , while the cwohrs have spherical structures from our targets . in order to investigate the detailed velocity structures of the clumps , i.e. , presence or absence of velocity gradients , areas of large velocity dispersion , we produced 1st and 2nd moment maps ( figures [ mm1 ] to [ mm5 ] ) . from the 1st moment images , we identify the objects which have a velocity gradient ( @xmath10 2 km@xmath2s@xmath13 pc@xmath13 ) within the cwohrs : irdc 19410 + 2336 , and msxdc g 053.11 + 00.05 . for the cwhrs , the velocity gradient within the clumps can not be well identified . furthermore , from the 2nd moment images , we can find that large velocity dispersion ( @xmath33 @xmath10 0.8 km@xmath2s@xmath13 ) appears along the filamentary structure for afgl 333 and irdc 20081 + 2720 , which may indicate the interactions between molecular gas and the expanding shell of the h ii region . for g 79.3 + 0.3 , there are two h ii regions ( g79.307 + 0.277 : south - east , g79.29 + 0.46 : north - west ) , we can see shell - like structures in the channel maps ( 0.251.2 km s@xmath13 in figure [ mapch3 ] ) of the whole of the g 79.3 + 0.3 clump . in this section , we describe the derivation of the physical properties of the @xmath0 clumps ( i.e. , the area within the 3 @xmath33 contours ) . a summary of the physical properties is given in table [ para ] . to define the clump radius , we estimated the projected area of the clump on the sky plane , @xmath34 and the observed radius was derived as @xmath35=@xmath36 . then we corrected for the spatial resolution as , @xmath37^{1/2 } , \end{aligned}\ ] ] where @xmath38 of @xmath19 the full width half maximum ( fwhm ) effective beam size . the velocity width of the clump ( one - dimensional fwhm velocity widths ) , @xmath39 was derived as follows . we used the fwhm of the gaussian fit of spectra averaged over the clumps to derive the virial masses of the clumps ( see below ) . we derived the @xmath0(@xmath1=10 ) line widths , @xmath40 by fitting the @xmath0(@xmath1=10 ) spectra averaged over the clumps with a gaussian function . we estimated intrinsic velocity widths of the clumps , @xmath39 , by correcting for the velocity resolution of the instruments , @xmath41 , as @xmath39 = @xmath42 . the velocity widths , @xmath39 range from 1.4 to 3.3@xmath2@xmath4 , suggesting considerable non - thermal contributions ; the thermal velocity widths ( one - dimensional velocity widths in fwhm ) of @xmath0 molecules are 0.120.24@xmath2@xmath4 with @xmath43 = 1030@xmath2k . the h@xmath21 column density of the clumps , @xmath44 was calculated as , @xmath45 , \end{aligned}\ ] ] where @xmath46 is the fractional abundance of @xmath0 relative to @xmath47 , @xmath48 is the excitation temperature of the transition , @xmath49 is the optical depth , and @xmath50 is the total integrated intensity of the @xmath0 line emission . for @xmath46 , we used 1.7 @xmath22 @xmath51 by frerking et al . we applied @xmath52 , and @xmath53 of 1 . hofner et al . ( 2000 ) revealed that most of massive star forming regions are optically thin in @xmath0 emission , so that we consider that the assumption is reasonable . the total mass under the local thermodynamic equilibrium ( lte ) condition , @xmath54 is calculated as , @xmath55 , \end{aligned}\ ] ] where @xmath56 is the distance to the object and @xmath57 is the radius of the clump . the rotation temperatures , @xmath58(2,2 ; 1,1 ) were derived from the nh@xmath18 data using the same method presented by ho @xmath59 townes ( 1983 ) . we applied the relation between the kinetic temperature and the rotation temperature , @xmath43 @xmath10 @xmath60 as proposed by danby et al . ( 1988 ) . under the lte condition , we consider that the excitation temperature is comparable to the kinetic temperature ( @xmath61 @xmath10 @xmath43 ) . we used the kinetic temperature derived from nh@xmath18 data data ( irdc 19410 + 2336 and g 79.3 + 0.3 ) , we adopted 15@xmath62 which is the typical temperature in irdcs regions . ] as the excitation temperature of the @xmath0 gas . we also calculated the virial masses of the clumps using ( @xmath63@xcite . we obtained radii of 0.51.7@xmath2pc , masses of 4704200@xmath2@xmath3 , and velocity widths in fwhm of 1.43.3@xmath2@xmath4 ( see table [ para ] ) . we also calculated virial masses , and found that all observed @xmath0 clumps are in virial equilibrium within the error in estimate including the uncertainty of abundance ratio and kinetic temperatures . in order to discuss the kinematics of the clumps , we calculated effective jeans lengths of 0.120.34@xmath2pc . the effective jeans length , @xmath64 was calculated as @xmath65 , where @xmath43 is the kinetic temperature of the clump and @xmath66 is the density of the clump @xcite . the density of the clump , @xmath66 was calculated as @xmath67 , where @xmath54 is lte mass of the clump , @xmath68 is mean mass of molecule , and @xmath57 is the radius of the clump . we also calculated the effective crossing time , 2@xmath57/@xmath69 ( @xmath69 = @xmath70 ) of ( 0.93)@xmath2210@xmath71@xmath2yr , and the internal kinetic energy , @xmath72=(1/2)@xmath54@xmath73 of ( 0.176.6)@xmath2210@xmath74@xmath2erg ( see table [ energy ] ) . comparing the physical parameters between cwhrs and cwohrs , there are no significant differences of the physical parameters ( radii of 0.71.7@xmath2pc , velocity widths in fwhm of 1.93.3@xmath2@xmath4 , masses of 7403200@xmath2@xmath3 in cwohrs , and 0.51.4@xmath2pc , 1.43.0@xmath2@xmath4 , 4704200@xmath2@xmath3 in cwhrs ) and kinetic energies ( ( 0.66)@xmath2210@xmath74@xmath2erg in cwohrs , ( 0.27)@xmath2210@xmath74@xmath2erg in cwhrs ) we calculated the virial ratio using our results , and found that they are gravitationally bound . @xcite suggest that the star formation efficiencies [ sfe=@xmath75 of the cluster - forming clumps range from approximately 10 to 30@xmath2@xmath7 . the sfes of a cluster increases to a maximum value of @xmath10 3040@xmath2@xmath7 due to gas dispersal by the stellar feedback @xcite . if we assume maximum sfes of 30@xmath2@xmath7 , cluster masses are expected from 100 to 1000 @xmath3 . @xcite suggest an empirical mass - size threshold for cluster formation including massive stars . using the mass - size relation ; @xmath76 @xmath9 870@xmath2@xmath3@xmath2@xmath77@xmath78 presented in @xcite , is the effective radius of the clump , @xmath76 is the mass of the clump ] all of the @xmath0 clumps exceed the limit shown in figure [ plot ] , suggesting that they have a potential to form a cluster including massive stars . in order to investigate the existence of yso candidates in our target regions , we used the information on stellar members in the point source catalogs of the wide - field infrared survey explorer ( @xmath5 ; wright et al . 2010 ; jarrett et al . 2011 ) for 10@xmath26@xmath2210@xmath26 areas . we adopted the source classification scheme in @xcite . they applied the established source classification scheme of @xmath16 irac data @xcite to the @xmath5 four bands ( 3.4 , 4.6 , 12 , and 22 @xmath15 ) . we selected all sources with photometric uncertainty @xmath79 0.2 mag in four bands . to reduce the risk that differential extinction will confuse our classification of ysos , we have adopted color - color diagrams of [ 3.4][4.6 ] vs. [ 4.6][12 ] and [ 4.6][22 ] to distinguish between class i and class ii candidates . figure [ ccd_1 ] shows source color - color diagrams of [ 3.4][4.6 ] and [ 4.6][12 ] made using @xmath5 3.4 , 4.6 , and 12 @xmath15 data . class i candidates are identified if their colors match : [ 3.4][4.6 ] @xmath9 1.0 and [ 4.6][12 ] @xmath9 2.0 . class ii ( t tauri star candidates ) are slightly reddened objects , and were selected by the condition of [ 3.4][4.6 ] @xmath9 0.25 and [ 4.6][12 ] @xmath9 1.0 . figure [ ccd_2 ] shows the @xmath5 3.4 , 4.6 , and 22 @xmath15 color - color diagrams . class ii sources are re - classified as [ 4.6][22 ] @xmath9 4.0 @xcite . the rest are considered to be diskless sources as in @xcite . we selected class i candidates " , which are selected from both [ 3.4][4.6 ] vs. [ 4.6][12 ] diagram in figure [ ccd_1 ] , and [ 3.4][4.6 ] vs. [ 4.6][22 ] diagram in figure [ ccd_2 ] . we detected 52 class i and 28 class ii candidates superposed on enclosed 3 @xmath33 contours of the @xmath0 clumps . we expect that contamination from non - yso candidates , such as galaxies ( stern et al . 2005 ) is small ( less than 10 non - ysos per square degree ) from the analysis in allen et al . ( 2007 ) . furthermore , comparing the spatial distributions of yso candidates with @xmath0 emissions in channel maps , most of yso candidates are associated with @xmath0 emission . thus we presume yso candidates superposed on @xmath0 clumps as a result of star formation within the clumps . stellar candidates with intermediate to high masses were identified as the class i sources with the magnitudes lower than 5 in 12 @xmath15 ( e.g. , * ? ? ? @xcite identified the brightest sources with @xmath16 8.0 @xmath15 @xmath8 6 mag , which corresponds to star that can produce photo dissociation regions ( pdrs ) . most of extended nebulae are probably local pdrs created by the radiation of the massive stars . in addition , these are not massive enough to form h ii regions , they can heat the surrounding dust and create local pdrs . although the wavelength is different from @xcite , the positions of the bright sources with 12 @xmath15 @xmath8 5 mag corresponds to h ii regions or pdrs in our objects . from these empirical results , 30 @xmath7 of detected class i candidates can be candidates of intermediate to massive stars . the number density of yso candidates does not satisfy the definition of cluster " in @xcite ( e.g. , 35 or more stars whose space densities exceed 1@xmath2@xmath3@xmath2pc@xmath80 ) . consequently , we call these yso distributions as association " in this study . note that the agn and pah candidates have been removed as in @xcite , although there is still uncertainty because individual candidates have not been identified by the sed . furthermore , in order to check the existence of more evolved yso candidates , we use the point source catalogs of the 2mass . we identified yso candidates from the 2mass @xmath1 , @xmath81 , and @xmath82 data which cover the mapping areas of our dense clump data , using the color - color diagrams and methods as in @xcite , @xcite , and @xcite . figure [ ccd ] shows @xmath1@xmath81 versus @xmath81@xmath82 color - color diagrams of point sources for the all target regions with 10@xmath26@xmath2210@xmath26 regions . the spatial distributions of the yso candidates are shown in figures [ mm1 ] to [ mm5 ] . there is a difference in the spatial distributions of class i and class ii candidates between cwhrs and cwohrs . for cwohrs ( irdc 19410 + 2336 and msxdc g 053.11 + 00.05 ) , the class i candidates are distributed around 69 @xmath33 contours of @xmath0 emission ( within @xmath10 0.5@xmath2pc ) , although it depends on observations . while , the class ii candidates are distributed @xmath10 2@xmath2pc apart from the positions of strong @xmath0 emission , which is close to edge of 3 @xmath33 contours . for cwhrs ( afgl 333 , irdc 20081 + 2720 ) , class i candidates are distributed in the shell - like structure within the clump , i.e. , they seem to align along the expanding shell from the associated h ii region . for the whole clump of g 79.3 + 0.3 , there seem to be scattering of the distributions of class ii candidates compared with the distributions of class i candidates . the fraction of class i candidates among yso candidates , @xmath83=@xmath84/(@xmath84+@xmath85 ) was calculated in table [ sep ] . @xmath83 was found to be higher than 50@xmath2@xmath7 ( 5060@xmath2@xmath7 ) in cwhrs , and smaller than 50@xmath2@xmath7 ( 1040@xmath2@xmath7 ) in cwohrs ( see figure [ histo ] ) . if we add yso candidates identified by the 2mass into calculation of the fraction , this tendency does not change . in particular , the fraction of class i candidates among yso candidates is proportional to the cluster age @xcite . comparing the results of 2mass with _ wise _ , cwohrs tend to have yso candidates identified by 2mass . this result is consistent with small fraction of class i candidates among yso candidates in cwohrs . the higher fraction of class i sources indicates that cwohrs are relatively younger than cwohrs . furthermore , this trend implies that the age spread of ysos in cwhrs is relatively smaller than those of cwohrs . there are three clear signatures that distinguish cwhrs from cwohrs , ( 1 ) spatial distributions : shell - like or spherical , ( 2 ) velocity structures : absence or presence of the distinct velocity gradient , large velocity dispersion along shells , and ( 3 ) age spreads of ysos : small or large ( see table [ score ] ) . our results imply that h ii regions strongly impact on the spatial distributions , velocity structures , and age spreads of ysos . however , such an impact is not evident in the other physical proprieties of clumps ( size , velocity width , mass , and kinetic energies ) . first , the shell - like structure in cwhrs indicates an interaction between the clumps and expansion shell from the h ii region . comparing the images of dust continuum ( also for our targets ) and molecular lines ( e.g. , co(10 ) , @xmath86co(10 ) ) , the molecular clumps associated with the h ii region tend to have shell - like or filamentary structures @xcite . these morphological structures may have been shaped by an h ii region . filaments will be unstable due to the interactions and will be easy to fragment ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite presented that there are two factors that can trigger star formation by an h ii region ; dynamical compression due to the h ii region and change in the self - gravity . in a numerical study of star formation , @xcite suggest that an expanding h ii region is able to trigger star formation within the ambient molecular clouds traced by co(10 ) , @xmath86co(10 ) lines . we discuss this in more detail in sec.[4 - 2 - 1 ] . second , regarding the difference of velocity structures , shock waves and the expanding shell of an h ii region should strongly influence surrounding molecular gas and change significantly the kinematic structure of the clumps . @xcite present maps with various molecular lines around the h ii region , sh 2 - 217 . they see no obvious velocity gradients . velocity structures in cwhrs has not been discussed previously because most of previous surveys of dense material around h ii regions have been conducted with mm - continuum . this is discussed in more detail in sec.[4 - 2 - 2 ] . third , the value of @xmath83 is considered to be a clear indication of the relative evolutionary status of the yso association @xcite . @xcite present a plot of the ratio , @xmath83 for the cluster - forming regions as a function of the central cluster age . using their correlation , the ages of our stellar associations obtained by _ wise _ are less than 1@xmath2myr . in particular , the ages of the associations are estimated to be less than 0.5@xmath2myr if the value of @xmath83 is higher than 50@xmath2@xmath7 . the higher fraction of class i sources indicates that cwohrs are relatively younger than cwohrs . furthermore , this trend implies that the age spread of ysos in cwhrs is relatively smaller than those of cwohrs . comparing the spatial distributions of yso candidates with the @xmath0 emissions , class i candidates are distributed uniformly in the shell - like structure in cwhrs . on the other hand , the class i candidates are distributed with the peak emission in @xmath0 , and class ii candidates are distributed outward of the class i candidates in cwohrs . @xcite discussed the triggering mechanism of an h ii region , rcw 120 ( d @xmath10 1.3@xmath2kpc ) , and identified yso candidates using _ spitzer _ data . the spatial distribution of yso candidates within condensation 1 " around rcw 120 , whose distance are @xmath10 0.8@xmath2pc from the exciting stars , are similar to those of g 79.3 + 0.3 , irdc 20081 + 2720 , and afgl 333 . class i candidates are uniformly distributed and aligned on the shell structure @xcite . from the spatial distribution of the yso candidates toward rcw 120 ( figure 5 in @xcite ) , we calculated the value of @xmath87 and found it to be @xmath10 50@xmath2@xmath7 , which is consistent with our results in cwhrs . star formation may be triggered by the expanding shell motion of an h ii region for afgl 333 , irdc 20081 + 2720 , and the whole clump of g 79.3 + 0.3 . dynamical effects of h ii regions can be seen in the spatial distributions , velocity structures , and small age spreads of ysos . here , we discuss a possible triggering mechanism , which leads these different signatures . there are mainly two models proposed for triggered star formation by h ii regions , radiation - driven implosion @xcite and collect - and - collapse model @xcite . in the collect - and - collapse model , swept - up molecular gas is accumulated to form dense gas in the shock front by the supersonic expansion of an h ii region . the shell is mainly occupied by dense molecular gas whose mass increases with evolution . @xcite discussed the dynamical expansion of rcw 120 using the model of @xcite . figure 14 shows the gas shell swept up by the shock front , with densities in the range of 10@xmath88@xmath2cm@xmath80 . this layer may become unstable and fragment , which make dense core ( size @xmath10 0.1@xmath2pc , density @xmath10 10@xmath88@xmath2cm@xmath80 ; @xcite ) . this process tends to sweep up amounts of gas and allows to form massive dense clumps @xcite , which is considered to form clusters including massive stars . on the other hand , radiation - driven implosion models are predicted to form only low- to intermediate stars @xcite . our clumps are massive enough to form a cluster including massive stars @xcite . in addition , cwhrs have a few candidates of intermediate to massive stars are distributed around the h ii region in g 79.3 + 0.3 and afgl 333 . from those results , we suggest that the collect - and - collapse model may be more adequate than the radiation - driven implosion . let us assume that there is one 20@xmath2@xmath3 star , which is located at 0.5@xmath2pc ( the separations between the positions of h ii region and the strong @xmath0 emission peak are 0.3 to 1@xmath2pc ) apart from the strong @xmath0 emission peak of the clumps . we estimate the internal kinetic energy , @xmath90 of the individual molecular clumps to be @xmath10 10@xmath91@xmath2erg ( see table [ energy ] ) . we consider the kinetic energy transferred to the clump through the stellar wind . using the values of @xcite , we estimated the kinetic energy supply rate by the stellar wind from a 20@xmath2@xmath3 star to be @xmath10 10@xmath92 erg@xmath2s@xmath13 . here , we assume that the stellar wind is isotropically driven to the molecular clump . the kinetic energy of stellar wind during the crossing time ( @xmath93 in niwa et al . 2009 ) is estimated to be @xmath10 10@xmath94 erg , which is larger than internal kinetic energies of the clumps . therefore our estimation suggests that h ii regions have potential to change the kinematics of the clumps located on the border of the h ii region . we can present that the energy input from the h ii regions through the shocks of stellar wind are sufficient to drive the internal kinetic energy of the clump . triggered star formation by an h ii region can explain small age spreads of ysos . according to @xcite , the dynamical expansion of an h ii region ( excited by a star of @xmath10 20 @xmath3 ) reaches a size of 1@xmath2pc in 0.2@xmath2myr . an h ii region will accumulate dense gas and derive the formation of filaments on the border of the h ii region . dynamical interactions between an h ii region and a filament will occur . this filament will become gravitationally unstable due to the interaction . once dense filament forms , fragmentation will occur along the filament to form dense cores , whose typical separation are the order of the jeans length @xcite . this mechanism will coevally form dense cores along the filament . these cores will immediately start collapsing in the free - fall time @xcite , which explains the small age spread of ysos in cwhrs . in fact , fragment components can be seen in the channel maps around the h ii region in figure [ mapch5 ] to [ mapch4 ] . yso candidates with the separations of the order of jeans length are distributed along the filament in observed cwhrs . for cwohrs , the clumps have spherical structure . they are more stable than filamentary one ( e.g. , * ? ? ? nevertheless , a spherical clump will eventually fragment , which initiates cluster formation ( e.g. , * ? ? ? * ; * ? ? ? after fragmentation , the local condensations will become distinct peaks , i.e. , dense cores , where stellar clusters will be eventually born . @xcite classified 14 cluster - forming clumps into three evolutionary stages and found that spherical clumps in the early stage tend to have a single peak . this result may imply that the fragmentation rate in a spherical clump can be lower than in a filament . during the cluster formation , the cores within the clumps will form stars in a wide mass range and dense gas is dispersed by their stellar activities @xcite . once clusters are formed , young stars , particularly most massive ones will disperse the surrounding gas with evolution making cavities around them @xcite . stellar feedback also forms dense cores ( e.g. , * ? ? ? * ) , the structures of the evolved clumps are expected to be less condensed than those of the younger clumps . the differences of the fragmentation rate between cwhrs and cwohrs will explain the difference of the age spreads of ysos . @xcite observed swept up dense gas surrounding h ii region , rcw 82 , and several class i candidates along the expanding shell . @xcite investigated the spread of ages in sh-284 with @xmath16/irac , and presented that despite the different scales ( @xmath10 20@xmath2pc ) observed in its multiple - bubble morphology , the spread of ages among the powering high - mass clusters is @xmath10 3@xmath2myr . @xcite presented the evidence of an age spread of at least 23@xmath2myr in w3 main . those scales of h ii regions are a magnitude larger than those of our objects . if we use the age spread - linear size relationship described by @xcite , the age spread is roughly estimated to @xmath10 0.5@xmath2myr ( i.e. , consistent with free - fall time ) in our observed scales . this value is consistent with the ages of associations derived from relation between the ratio of class i candidate and cluster age in @xcite . lifetimes of the low to intermediate stars ( class i and class ii candidates ) are considered to be @xmath10 10@xmath2myr @xcite . thus , an age spread of 0.5@xmath2myr is significantly small to speak of coeval " cluster formation . we propose that there are two modes of cluster formation ; one is coeval cluster formation , the other is non - coeval one . this difference is related to their environments ( existence or non - existence of the h ii region ) of the associated clumps . h ii regions play three roles : ( 1 ) they accumulate the dense gas by an expanding shell , ( 2 ) they shape the dense component into a filament , and ( 3 ) they inject energy . our study can reveal differences between the mode of cluster formation as we focused on clumps in the earliest stages of cluster formation . interestingly , both cwhrs and cwohrs have potential to form a cluster containing massive stars . in particular , cwohrs contain candidates of intermediate to massive stars , although their fraction of class ii candidates is relatively higher . if the differences of cluster - forming modes depend on the differences of the environments of the clumps , some unknown issues of cluster formation ( e.g. , imf , star formation efficiency , age spread of the cluster members ) can be clarified by carrying out such an investigation toward the clumps in the early stages of the cluster formation ( e.g.,@xmath2irdcs ) . although more statistical studies should be needed , these studies will lead the understanding the fundamental topic of the cluster formation . we have mapped 7 young massive molecular clumps within 5 target regions in the @xmath0 line emission with the nobeyama 45 m radio telescope . our aim of this study is to understand the dynamical impact of h ii regions on the clumps . we found different signatures of the clumps with h ii region as compared to clumps or without one . our results and conclusions are summarized as follows : 1 . we made the @xmath0 maps with a size of @xmath10 6@xmath26@xmath226@xmath26 to 10@xmath26@xmath2210@xmath26 areas of 7 nearby ( @xmath56 @xmath8 2.1@xmath2kpc ) young massive clumps within 5 targets . as a result , all the clumps are detected in @xmath0 lines , whose radii , masses , and velocity widths are 0.51.7@xmath2pc , 4704200@xmath2@xmath3 , and 1.43.3@xmath2 km s@xmath13 , respectively . they are similar to those of the cluster - forming clumps . 2 . all the clumps are likely to be in virial equilibrium , and exceed the massive star formation limit ; mass - size relation presented in @xcite , suggesting that they have a potential to form a cluster including massive stars . 3 . in order to investigate the existence of yso candidates in our target regions , we identified stellar members using the point source catalogs of the _ wise_. there are 52 class i and 28 class ii candidates within the all @xmath0 clumps . 30 @xmath7 of the detected class i candidates may be intermediate to massive stars . the fraction of class i to yso candidates : @xmath83=@xmath84/(@xmath84+@xmath85 ) is presented in table [ sep ] . it is @xmath6 50@xmath7 ( 5060 @xmath7 ) in cwhrs , and @xmath8 50@xmath7 ( 1040 @xmath7 ) in cwohrs . 4 . there are three clear signatures that distinguish cwhr and cwohr , ( 1 ) spatial distributions : shell - like or spherical , ( 2 ) velocity structures : absence or presence of the distinct velocity gradient , large velocity dispersion along shells , and ( 3 ) the age spread of ysos : small or large an h ii region can play three roles : ( 1 ) accumulate the dense gas by expanding shell , ( 2 ) shape the dense component into a filament , and ( 3 ) inject energy . the expanding shell can accumulate dense gas around the border of h ii region and make the dense gas filament . interaction between h ii regions and filaments will tend to form dense core coevally . once dense cores form , these cores will immediately start collapsing in a free - fall time , which can explain the small age spread of ysos . we thank the anonymous referee for constructive comments that helped to improve this manuscript . we are grateful to the staff of the nobeyama radio observatory ( nro ) for both operating the nobeyama 45 m telescope and helping us with the data reduction . we also thank s.komugi for useful discussion . from the @xmath5 composite image ( figure [ map1 ] ) , we can see two main yso associations in msxdc g 053.11 + 00.05 region located on north and south ( also see figure [ mm1 ] ) . the channel maps in figure [ mapch1 ] clearly show that strong @xmath0 line emission is associated with these yso candidates in different velocity ranges , for the south one which is brighter is in a range of 20.8 km s@xmath13 to 22.8 km s@xmath13 , for the other in north is in a range of 20.8 km s@xmath13 to 24.2 km s@xmath13 . the @xmath0 clump seems to have a velocity gradient along the north - south direction . we consider that these emission components correspond to parental clumps of these yso associations ( see figure [ mm1 ] ) , which are labeled as c1 for the south one and c2 ( corresponds to msxdc g 053.25 + 00.04 in @xcite ) for the north one . for clump c1 , compact dust condensation is detected by @xcite in 1.2 mm continuum observations . msxdc g 053.11 + 00.05 c1 have region of large velocity dispersion where yso candidates are distributed , while msxdc g 053.11 + 00.05 c2 have the large velocity dispersion where yso candidates are not distributed . from the images of figure [ map2 ] , we can see a yso association in the @xmath5 composite image around irdc 19410 + 2336 central region , corresponding to iras 19410 + 2336 @xcite . the channel maps in figure [ mapch2 ] clearly show that strong @xmath0 line emission is associated with yso candidates in a velocity range from 21.2 km@xmath2s@xmath13 to 23.2 km@xmath2s@xmath13 , and the @xmath0 clump seems to have a velocity gradient in the range from 20.8 km@xmath2s@xmath13 to 23.8 km@xmath2s@xmath13 in the direction from north - east to south - west . the multiple co(@xmath1=21 ) outflows ( blue ; 518 km@xmath2s@xmath13 , red ; 2647 km@xmath2s@xmath13 ) have been detected in the direction from north - east to south - west by @xcite , which implies that irdc 19410 + 2336 is a young active star forming region . there is bright infrared emission in the @xmath5 composite image ( figure [ map3 ] ) , which corresponds to an h ii region @xcite . the h ii region is located @xmath10 0.3@xmath2pc away from the strong @xmath0 line emission . in a velocity range from 5.2 km@xmath2s@xmath13 to 5.8 km@xmath2s@xmath13 , we can see the @xmath0 emission have shell - like structure , which may be shaped by the h ii region , i.e. , from the interaction between dense gas and expanding shell . the yso candidates are distributed within the shell - like structures of @xmath0 emission around the h ii region . there is a bright h ii region ( figure [ map4 ] ) , dr 15 ( g79.307 + 0.277 ) , which lies behind and slightly south ( @xmath10 0.6@xmath2pc ) of the g 79.3 + 0.3 region @xcite . dr 15 is formed by one or two b0 zero - age main - sequence stars , observed as the far - infrared source ( odenwald et al . 1990 ) , having a total luminosity of @xmath10 3 @xmath22 10@xmath71@xmath2@xmath95 ( oka et al . 2001 ) . a ring shaped nebula in 22 @xmath15 emission corresponds to a photo - dissociation region , g79.29 + 0.46 , which is the outskirts of an evolved massive star @xcite . from the morphological distributions between @xmath0 and 22 @xmath15 emission , in a velocity range from @xmath960.25 km@xmath2s@xmath13 to 1.2 km@xmath2s@xmath13 , we can see the @xmath0 emission have an shell - like structure in the north area , which is an possibly compressed by g79.29 + 0.46 . the yso candidates are distributed in a shell - like fashion around the h ii region ( dr 15 ) . the channel maps in figure [ mapch3 ] clearly show that strong @xmath0 line emission is associated with yso candidates for the east one in a range of 0.25 km@xmath2s@xmath13 to 2.8 km@xmath2s@xmath13 . for the other condensations in west , it is still dark in the 22 @xmath15 image . in g 79.3 + 0.3 region , the global trend of velocity gradients can not be well defined . the clump of the association for the east one are labeled as c1 ( correspond to p1 in @xcite ) . for the west one labeled as c2 ( correspond to p3 in @xcite ) , the @xmath0 clump is associated with an infrared dark area . we can see several yso candidates in the @xmath0 emission ( figure [ map5 ] ) , whose separations are comparable to a jeans length ( @xmath10 0.2@xmath2pc ) . in addition , we also see the bright infrared emission in the east of the @xmath0 emission , which corresponds to an h ii region generated by a b0.5 type star @xcite . it is located @xmath10 1@xmath2pc away from the strong @xmath0 line emission . in the north area of the @xmath0 emission , we can see the relatively bright and diffuse 22 @xmath15 emission , which corresponds to a herbig ae / be star @xcite . the channel maps in figure [ mapch4 ] clearly show that strong @xmath0 line emission is associated with these ysos candidates in a range of @xmath9650.2 km@xmath2s@xmath13 to @xmath9647.2 km@xmath2s@xmath13 . in a velocity range from @xmath9648.2 km@xmath2s@xmath13 to @xmath9649.2 km@xmath2s@xmath13 , the @xmath0 emission have an shell - like structure . the yso candidates are distributed in a shell - like fashion around the h ii regions . in a velocity range from @xmath9650.8 km@xmath2s@xmath13 to @xmath9648.2 km@xmath2s@xmath13 , we can see the @xmath0 emission in northern edge have also an shell - like structure , which seem be compressed by the radiation or stellar wind of the herbig ae / be star . bessell , m. s. , & brett , j. m. 1988 , , 100 , 1134 beuther , h. , schilke , p. , & stanke , t. 2003 , , 408 , 601 beuther , h. , schilke , p. , menten , k. m. , motte , f. , sridharan , t. k. , & wyrowski , f. 2002 , , 566 , 945 bik , a. , henning , t. , stolte , a. , et al . 2012 , , 744 , 87 bonnell , i. a. , bate , m. r. , & vine , s. g. 2003 , , 343 , 413 brand , j. , massi , f. , zavagno , a. , deharveng , l. , & lefloch , b. 2011 , , 527 , a62 carey , s. j. , clark , f. o. , egan , m. p. , price , s. d. , shipman , r. f. , & kuchar , t. a. 1998 , , 508 , 721 carey , s. j. , feldman , p. a. , redman , r. o. , egan , m. p. , macleod , j. m. , & price , s. d. 2000 , , 543 , l157 carpenter , j. m. 2000 , , 120 , 3139 churchwell , e. , watson , d. f. , povich , m. s. , et al . 2007 , , 670 , 428 churchwell , e. , et al . 2006 , , 649 , 759 cohen , j. g. , persson , s. e. , elias , j. h. , & frogel , j. a. 1981 , , 249 , 481 deharveng , l. , et al . 2010 , , 523 , a6 deharveng , l. , zavagno , a. , schuller , f. , caplan , j. , pomars , m. , & de breuck , c. 2009 , , 496 , 177 deharveng , l. , zavagno , a. , & caplan , j. 2005 , , 433 , 565 egan , m. p. , shipman , r. f. , price , s. d. , carey , s. j. , clark , f. o. , & cohen , m. 1998 , , 494 , l199 elmegreen , b. g. 2000 , , 530 , 277 elmegreen , b. g. , & lada , c. j. 1977 , , 214 , 725 evans , n. j. , ii , dunham , m. m. , jrgensen , j. k. , et al . 2009 , , 181 , 321 frerking , m. a. , langer , w. d. , & wilson , r. w. 1982 , , 262 , 590 fuente , a. , martin - pintado , j. , bachiller , r. , neri , r. , & palla , f. 1998 , , 334 , 253 fuente , a. , martin - pintado , j. , bachiller , r. , rodriguez - franco , a. , & palla , f. 2002 , , 387 , 977 fukuda , n. , & hanawa , t. 2000 , , 533 , 911 higuchi , a. e. , kurono , y. , saito , m. , & kawabe , r. 2009 , , 705 , 468 higuchi , a. e. , kurono , y. , saito , m. , & kawabe , r. 2010 , , 719 , 1813 hosokawa , t. , & inutsuka , s .- i . 2006 , , 646 , 240 ho , p. t. p. , & townes , c. h. 1983 , , 21 , 239 hofner , p. , wyrowski , f. , walmsley , c. m. , & churchwell , e. 2000 , , 536 , 393 hughes , v. a. , & viner , m. r. 1982 , , 87 , 685 ikeda , n. , sunada , k. , & kitamura , y. 2007 , , 665 , 1194 inutsuka , s .- miyama , s. m. 1992 , , 388 , 392 jarrett , t. h. , cohen , m. , masci , f. , et al . 2011 , , 735 , 112 kauffmann , j. , & pillai , t. 2010 , , 723 , l7 kessel - deynet , o. , & burkert , a. 2003 , , 338 , 545 knee , l. b. g. , & sandell , g. 2000 , , 361 , 671 koenig , x. p. , leisawitz , d. t. , benford , d. j. , et al . 2012 , , 744 , 130 lada , c. j. , & lada , e. a. 2003 , , 41 , 57 lada , c. j. 2010 , royal society of london philosophical transactions series a , 368 , 713 lefloch , b. , & lazareff , b. 1994 , , 289 , 559 lpez - sepulcre , a. , cesaroni , r. , & walmsley , c. m. 2010 , , 517 , a66 martn - hernndez , n. l. , bik , a. , puga , e. , nrnberger , d. e. a. , & bronfman , l. 2008 , , 489 , 229 mauersberger , r. , & henkel , c. 1993 , reviews in modern astronomy , 6 , 69 niwa , t. , tachihara , k. , itoh , y. , oasa , y. , sunada , k. , sugitani , k. , & mukai , t. 2009 , , 500 , 1119 nakamura , f. , sugitani , k. , shimajiri , y. , et al . 2011 , , 737 , 56 oka , t. , yamamoto , s. , iwata , m. , et al . 2001 , , 558 , 176 odenwald , s. f. , campbell , m. f. , shivanandan , k. , et al . 1990 , , 99 , 288 peretto , n. , & fuller , g. a. 2009 , , 505 , 405 pomars , m. , zavagno , a. , deharveng , l. , et al . 2009 , , 494 , 987 puga , e. , hony , s. , neiner , c. , et al . 2009 , , 503 , 107 plume , r. , jaffe , d. t. , evans , n. j. , ii , martin - pintado , j. , & gomez - gonzalez , j. 1997 , , 476 , 730 rathborne , j. m. , jackson , j. m. , chambers , e. t. , stojimirovic , i. , simon , r. , shipman , r. , & frieswijk , w. 2010 , , 715 , 310 rathborne , j. m. , jackson , j. m. , & simon , r. 2006 , , 641 , 389 ragan , s. e. , heitsch , f. , bergin , e. a. , & wilner , d. 2012 , , 746 , 174 redman , r. o. , feldman , p. a. , wyrowski , f. , ct , s. , carey , s. j. , & egan , m. p. 2003 , , 586 , 1127 rizzo , j. r. , jimnez - esteban , f. m. , & ortiz , e. 2008 , , 681 , 355 ridge , n. a. , wilson , t. l. , megeath , s. t. , allen , l. e. , & myers , p. c. 2003 , , 126 , 286 sakai , t. , oka , t. , & yamamoto , s. 2007 , , 662 , 1043 sawada , t. , et al . 2008 , , 60 , 445 simon , r. , jackson , j. m. , rathborne , j. m. , & chambers , e. t. 2006 , , 639 , 227 sorai , k. , sunada , k. , okumura , s. k. , tetsuro , i. , tanaka , a. , natori , k. , & onuki , h. 2000 , , 4015 , 86 sridharan , t. k. , beuther , h. , saito , m. , wyrowski , f. , & schilke , p. 2005 , , 634 , l57 watson , c. , povich , m. s. , churchwell , e. b. , et al . 2008 , , 681 , 1341 whitworth , a. p. , bhattal , a. s. , chapman , s. j. , disney , m. j. , & turner , j. a. 1994 , , 268 , 291 wright , e. l. , eisenhardt , p. r. m. , mainzer , a. k. , et al . 2010 , , 140 , 1868 yamaguchi , c. , sunada , k. , iizuka , y. , iwashita , h. , & noguchi , t. 2000 , , 4015 , 614 zavagno , a. , deharveng , l. , comern , f. , brand , j. , massi , f. , caplan , j. , & russeil , d. 2006 , , 446 , 171 zavagno , a. , pomars , m. , deharveng , l. , et al . 2007 , , 472 , 835 zavagno , a. , russeil , d. , motte , f. , et al . 2010 , , 518 , l81 zinnecker , h. , & yorke , h. w. 2007 , , 45 , 481 l l l l l l l l l l l c c c c c c c c c c c c msxdc g53.11 + 00.05 c1 & 19h29m18s & 17d56m33s & 1.8 & [email protected] & [email protected] & 920@xmath30560 & [email protected] & 3200@xmath301400 & 18 & no + msxdc g53.11 + 00.05 c2 & 19h29m37s & 18d02m00s & 1.9 & [email protected] & [email protected] & 610@xmath30360 & [email protected] & 740@xmath30370 & 15 & no + irdc 19410 + 2336 & 19h43m10s & 23d45m04s & 2.1 & [email protected] & [email protected] & 4000@xmath302000 & [email protected] & 2900@xmath301200 & 15 & no + irdc 20081 + 2720 & 20h10m13s & 27d28m18s & 0.7 & [email protected] & [email protected] & 230@xmath30140 & [email protected] & 470@xmath30150 & 15@xmath97 & yes + g79.3 + 0.3 c1 & 20h32m23s & 40d20m00s & 0.8 & [email protected] & [email protected] & 650@xmath30310 & [email protected] & 1000@xmath30340 & 15@xmath97 & yes + g79.3 + 0.3 c2 & 20h32m06s & 40d18m00s & 0.8 & [email protected] & [email protected] & 860@xmath30360 & [email protected] & 630@xmath30230 & 15@xmath97 & yes + afgl 333 & 02h28m19s & 61d29m40s & 2.0 & [email protected] & [email protected] & 2500@xmath301300 & [email protected] & 4200@xmath301800 & 15 & yes + l l l l l l c c c c c msxdc g53.11 + 00.05 c1 & 0.21 & 3@xmath2210@xmath71 & 2.1@xmath2210@xmath74 & 2.2@xmath2210@xmath98 + msxdc g53.11 + 00.05 c2 & 0.17 & 2@xmath2210@xmath71 & 5.6@xmath2210@xmath99 & 1.1@xmath2210@xmath98 + irdc 19410 + 2336 & 0.34 & 2@xmath2210@xmath71 & 5.9@xmath2210@xmath74 & 7.9@xmath2210@xmath98 + irdc 20081 + 2720 & 0.16 & 2@xmath2210@xmath71 & 1.7@xmath2210@xmath99 & 2.9@xmath2210@xmath100 + g79.3 + 0.3 c1 & 0.13 & 1@xmath2210@xmath71 & 8.6@xmath2210@xmath99 & 2.0@xmath2210@xmath98 + g79.3 + 0.3 c2 & 0.12 & 9@xmath2210@xmath101 & 8.7@xmath2210@xmath99 & 3.1@xmath2210@xmath98 + afgl 333 & 0.20 & 2@xmath2210@xmath71 & 6.6@xmath2210@xmath74 & 9.8@xmath2210@xmath98 + l l l l l c c c c msxdc g53.11 + 00.05 c1 & 4 ( 6 ) & 3 ( 7 ) & 3 & 57 + msxdc g53.11 + 00.05 c2 & 3 ( 4 ) & 4 ( 7 ) & 0 & 43 + msxdc g53.11 + 00.05 ( total ) & 7 ( 10 ) & 7 ( 14 ) & 3 & 50 + irdc 19410 + 2336 & 6 & 9 ( 11 ) & 3 & 40 + irdc 20081 + 2720 & 7 & 3 & 0 & 70 + g79.3 + 0.3 c1 & 11 & 9 & 0 & 55 + g79.3 + 0.3 c2 & 3 & 0 & 0 & 100 + g79.3 + 0.3 ( total ) & 14 & 9 & 0 & 61 + afgl 333 & 18 ( 21 ) & 2 & 0 & 90 + l l l c c source name & irdc 20081 + 2720 , g79.3 + 0.3 , afgl 333 & msxdc g53.11 + 00.05 , irdc 19410 + 2336 + spatial distribution & filamentary , shell - like & spherical + velocity gradient & present ( @xmath10 2 km@xmath2s@xmath13pc@xmath13 ) & absent + velocity dispersion & large ( @xmath10 1@xmath2km@xmath2s@xmath13)@xmath97along shell - like structure & large ( @xmath10 1@xmath2km@xmath2s@xmath13)@xmath97at central region + age of yso association & @xmath8 0.5@xmath2myr & @xmath10 1@xmath2myr +

in this paper we propose a general spectral theory for tensors . our proposed factorization decomposes a tensor into a product of orthogonal and scaling tensors . at the same time , our factorization yields an expansion of a tensor as a summation of outer products of lower order tensors . our proposed factorization shows the relationship between the eigen - objects and the generalised characteristic polynomials . our framework is based on a consistent multilinear algebra which explains how to generalise the notion of matrix hermicity , matrix transpose , and most importantly the notion of orthogonality . our proposed factorization for a tensor in terms of lower order tensors can be recursively applied so as to naturally induces a spectral hierarchy for tensors .

1008.2923

in 1762 joseph louis lagrange formulated what is now known as the eigenvalue - eigenvector problem , which turns out to be of significant importance in the understanding several phenomena in applied mathematics as well as in optimization theory . the spectral theory for matrices is widely used in many scientific and engineering domains . in many scientific domains , data are presented in the form of tuples or groups , which naturally give rise to tensors . therefore , the generalization of the eigenvalue - eigenvector problem for tensors is a fundamental question with broad potential applications . many researchers suggested different forms of tensor decompositions to generalize the concepts of eigenvalue - eigenvector and singular value decomposition . in this paper we propose a mathematical framework for high - order tensors algebra based on a high - order product operator . this algebra allows us to generalize familiar notions and operations from linear algebra including dot product , matrix adjoints , hermicity , permutation matrices , and most importantly the notion of orthogonality . our principal result is to establish a rigorous formulation of tensor spectral decomposition through the general spectral theorem . we prove the spectral theorem for hermitian finite order tensors with norm different from @xmath0 . finally we point out that one of the fundamental consequence of the spectral theorem is the existence of a spectral hierarchy which determines a given hermitian tensor of finite order . there are certain properties that a general spectral theory is expected to satisfy . the most fundamental property one should expect from a general formulation of the spectral theorem for tensors is a factorization of a cubic tensor into a certain number of cubic tensors of the same dimensions . our proposed factorization decomposes a hermitian tensor into a product of orthogonal and scaling tensors . our proposed factorization also extends to handle non - hermitian tensors . furthermore our proposed factorization offers an expansion of a tensor as a summation of lower order tensors that are obtained through outer products . our proposed factorization makes an explicit connection between the eigen - objects and the reduced set of characteristic polynomials . the proposed framework describes the spectral hierarchy associated with a tensor . finally the framework aims to extend linear algebraic problems found in many domains to higher degree algebraic formulations of corresponding problems . the organization of this paper is as follows ; section [ 2 ] reviews the state of the art in tensor decomposition and its relation to the proposed formulation . section [ 3 ] introduces our proposed tensor algebra for order three tensors . section [ 4 ] introduces and proves our proposed spectral theorem for order three tensors . section [ 5 ] discusses some important properties following from the proposed spectral decomposition . section [ 6 ] proposes a computational framework for describing the characteristic polynomials of a tensor . section [ 7 ] generalizes the introduced concepts to higher order tensors and introduces the notion of the spectral hierarchy . section [ 8 ] discusses in details the relation between the proposed framework and some existing tensor decomposition frameworks . section [ 9 ] concludes the paper with a discussion on the open directions . in this section we recall the commonly used notation by the multilinear algebra community where a @xmath1-tensor denotes a multi - way array with @xmath1 indices @xcite . therefore , a vector is a @xmath0-tensor and a matrix is a @xmath2-tensor . a @xmath3-tensor @xmath4 of dimensions @xmath5 denotes a rectangular cuboid array of numbers . the array consists of @xmath6 rows , @xmath7 columns , and @xmath8 depths with the entry @xmath9 occupying the position where the @xmath10 row , the @xmath11 column , and the @xmath12depth meet . for many purposes it will suffice to write @xmath13 we now introduce generalizations of complex conjugate and inner product operators . + the _ order @xmath8 conjugates _ of a scalar complex number @xmath14 are defined by : @xmath15 where @xmath16 and @xmath17 respectively refer to the imaginary and real part of the complex number @xmath14 , equivalently rewritten as @xmath18 from which it follows that @xmath19 the particular _ inner product _ operator that we introduce relates the inner product of a @xmath8-tuple of vectors in @xmath20 to a particular @xmath21 norm operator @xmath20 in a way quite similar to the way the inner product of pairs of vectors relate to the usual @xmath22 vector norm . we refer to the norm operator @xmath23 ( for every integer @xmath24 ) as the @xmath21 norm defined for an arbitrary vector @xmath25 by @xmath26^{\frac{1}{p}},\ ] ] the inner product operator for a @xmath8-tuple of vectors in @xmath20 denoted @xmath27 is defined by @xmath28 some of the usual properties of inner products follow from the definition @xmath29 and most importantly the fact that @xmath30 and @xmath31 we point out that the definitions of inner products is extended naturally to tensors as illustrated bellow @xmath32 @xmath33 more generally for arbitrarly finite order tensor the inner product for the family of tensors @xmath34 is defined by : @xmath35 note that the addition in the indices are performed modulo @xmath7 . + generalization of other concepts arising from linear algebra have been investigated quite extensively in the literature . cayley in @xcite instigated investigations on hyperdeterminants as a generalization of determinants . gelfand , kapranov and zelevinsky followed up on cayley s work on the subject of hyperdeterminants by relating hyperdeterminants to @xmath36-discriminants in their book @xcite . + a recent approach for generalizing the concept of eigenvalue and eigenvector has been proposed by liqun qi in @xcite and followed up on by lek - heng lim@xcite , cartwright and sturmfels @xcite . the starting point for their approach will be briefly summarized using the notation introduced in the book @xcite . assuming a choice of a coordinate system @xmath37 associated with each one of the vector space @xmath38 . we consider a multilinear function @xmath39 expressed by : @xmath40 equivalently the expression above can be rewritten as @xmath41 which of course is a natural generalization of bilinear forms associated with a matrix representation of a linear map for some choice of coordinate system @xmath42 it follows from the definition of the multilinear function @xmath43 that the function induces @xmath44 not necessarily distinct multilinear projective maps denoted by @xmath45 expressed as : @xmath46 the various formulations of eigenvalue eigenvector problems as proposed and studied in @xcite arise from investigating solutions to equations of the form : @xmath47 applying symmetry arguments to the tensor @xmath4 greatly reduces the number of map @xmath48 induced by @xmath4 . for instance if @xmath4 is supersymmetric ( that is @xmath4 is invariant under any permutation of it s indices ) then @xmath4 induces a single map . furthermore , different constraints on the solution eigenvectors @xmath49 distinguishes the @xmath50-eigenvectors from the @xmath51-eigenvectors and the @xmath52-eigenvectors as introduced and discussed in @xcite . our treatment considerably differs from the approaches described above in the fact that our aim is to find a decomposition for a given tensor @xmath4 that provides a natural generalization for the concepts of hermitian and orthogonal matrices . furthermore our approach is not limited to supersymmetric tensors . in connection with our investigations in the current work , we point out another concepts from linear algebra for which the generalization to tensor plays a significant role in complexity theory , that is the notion of matrix rank . indeed one may also find an extensive discussions on the topic of tensor rank in @xcite . the tensor rank problem is perhaps best described by the following optimization problem . given an @xmath44-tensor @xmath53 we seek to solve the following problem which attempts to find an approximation of @xmath4 as a linear combination of rank one tensors . @xmath54 our proposed tensor decomposition into lower order tensors relates to the tensor rank problem but differs in the fact that the lower order tensors arising from the spectral decomposition of @xmath3-tensors , named eigen - matrices are not necessarily rank @xmath0 matrices . several approaches have been introduced for decomposing @xmath1-tensors for @xmath55 in a way inspired by matrix svd . svd decomposes a matrix @xmath4 into @xmath56 and can be viewed as a decomposition of the matrix @xmath4 into a summation of rank-1 matrices that can be written as @xmath57 where @xmath44 is the rank of @xmath4 , @xmath58 are the @xmath59-th columns of the orthogonal matrices @xmath60 and @xmath61 , and @xmath62 s are the diagonal elements of @xmath63 , i.e. , the singular values . here @xmath64 denotes the outer product . the canonical and parallel factor decomposition ( canecomp - parafac , also caller the cp model ) , independently introduced by @xcite , generalize the svd by factorizing a tensor into a linear combination of rank-1 tensors . that is given @xmath65 , the goal is to find matrices @xmath66 , @xmath67 and @xmath68 such that @xmath69 where the expansion is in terms of the outer product of vectors @xmath70 are the i - th columns of @xmath60 , @xmath61 , and @xmath71 , which yields rank-1 tensors . the rank of @xmath4 is defined as the minimum @xmath44 required for such an expansion . here there are no assumption about the orthogonality of the column vectors of @xmath60 , @xmath61 , and @xmath71 . the cp decomposition have been show to be useful in several applications where such orthogonality is not required . there are no known closed - form solution to determine the rank @xmath44 , or to find a lower rank approximation as given directly by matrix svd . tucker decomposition , introduced in@xcite , generalizes over eq [ eq : rank1decomp ] , where an @xmath72 tensor @xmath4 is decomposed into rank-1 tensor expansion in the form @xmath73 where @xmath74 , @xmath75 , and @xmath76 . the coefficients @xmath77 form a tensor that is called the core tensor @xmath78 . it can be easily seen that if such core tensor is diagonal , i.e. , @xmath79 unless @xmath80 , tucker decomposition reduces to the cp decomposition in eq [ eq : rank1decomp ] . orthogonality is not assumed in tucker decomposition . orthogonality constraints can be added by requiring @xmath81 to be columns of orthogonal matrices @xmath60,@xmath61 , and @xmath71 . such decomposition was introduced in@xcite and was denoted by high order singular value decomposition ( hosvd ) . tucker decomposition can be written using the mode-@xmath7 tensor - matrix multiplication defined in @xcite as @xmath82 where @xmath83 is the mode-@xmath7 tensor - matrix multiplication . similar to tucker decomposition , the core tensor of hosvd is a dense tensor . however , such a core tensor satisfies an _ all - orthogonality _ property between its slices across different dimensions as defined in @xcite . hosvd of a tensor can be computed by flattening the tensor into matrices across different dimensions and using svd on each matrix . truncated version of the expansion yields a lower rank approximation of a tensor @xcite . several approaches have been introduced for obtaining lower rank approximation by solving a least square problem , _ e.g. _ @xcite . recently an extension to tucker decomposition with non - negativity constraint was introduced with many successful applications @xcite . all the above mentioned decompositions factorizes a high order tensor as a summation of rank-1 tensors of the same dimension , which is inspired by such an interpretation of matrix svd as in eq [ eq : svd ] . however , none of these decomposition approaches can describe a tensor as a product of tensors as would be expected from an svd generalization . the only known approach to us for decomposing a tensor to a product of tensors was introduced in a technical report @xcite . this approach is based on the idea that a diagonalization of a circulant matrix can be obtained by discrete fourier transform ( dft ) . given a tensor , it is flattened then a block diagonal matrix is constructed by dft of the circulant matrix formed from the flattened tensor . matrix svd is then used on each of the diagonal blocks . the inverse process is then used to put back the resulting decompositions into tensors . this approach results in a decomposition in the form @xmath84 where the product is defined as @xcite @xmath85 however , such decomposition does not admit a representation of the decomposition into an expansion in terms of rank-1 tensors . the product is mainly defined by folding and unfolding the tensor into matrices . from the above discussion we can highlight some fundamental limitations of the known tensor decomposition frameworks . existing tensor decomposition frameworks are mainly expansions of a tensor as a linear combination of rank-1 tensors , which are the outer products of vectors under certain constraints ( orthogonality , etc . ) and do not provide a factorization into product of tensors of the same dimensions . tucker decomposition , although a generalization of svd , falls short of generalizing the notion of the spectrum for high - order tensors . there is no connection between the singular values and the spectrum of the corresponding cubic hermitian tensors . unfortunately , no such relation is proposed by the tucker factorization . the tucker decomposition does not suggest at all how to generalize such objects as the trace and the determinant of higher order tensors . in the appendix of this paper we show that tucker decomposition and hosvd uses notion of matrix orthogonality . the most widely used formulation for tensor decomposition is the orthogonal version of tucker decomposition ( hosvd ) @xcite . hosvd is a multilinear rank revealing procedure @xcite and therefore , it has been widely used recently in many domains for dimensionality reduction and to estimate signal subspaces of tensorial data@xcite . in computer vision , hosvd has been used in@xcite for analysis of face images with different sources of variability , _ e.g. _ different people , illumination , head poses , expressions , etc . it has been also used in texture analysis , compression , motion analysis @xcite , posture estimation , gait biometric analysis , facial expression analysis and synthesis , _ e.g. _ @xcite , and other useful applications @xcite . hosvd decomposition gives a natural way for dealing with images as matrices @xcite . the relation between hosvd and independent component analysis ica was also demonstrated in @xcite with applications in communication , image processing , and others . beyond vision and image processing , hosvd has also been used in data mining , web search , _ @xcite , and in dna microarray analysis @xcite . we propose a formulation for a general spectral theory for tensors coined with consistent definitions from multilinear algebra . at the core of the formulation is our proposed _ spectral theory for tensors _ . in this section , the theory focuses on @xmath3-tensors algebra . we shall discuss in the subsequent section the formulations of our theory for @xmath7-tensor where @xmath7 is positive integer greater or equal to @xmath2 . a @xmath86 _ 3-tensor _ @xmath4 denotes a rectangular cuboid array of numbers having @xmath6 rows , @xmath7 columns , and @xmath8 depths . the entry @xmath9 occupies the position where the @xmath10 row , the @xmath11 column , and the @xmath12 depth meet . for many purposes it will suffice to write @xmath87 we use the notation introduced above for matrices and vectors since they will be considered special cases of @xmath3-tensors . thereby , allowing us to indicate matrices and vectors respectively as oriented slice and fiber tensors . therefore , @xmath88 , @xmath89 , and @xmath90 tensors indicate vectors that are respectively oriented vertically , horizontally and along the depth direction furthermore they will be respectively denoted by @xmath91 , @xmath92 , @xmath93 . similarly @xmath94 , @xmath95 , and @xmath96 tensors indicate that the respective martrices of dimensions @xmath97 , @xmath98 and @xmath99 can be respectively thought of as a vertical , horizontal , or depth slice denoted respectively @xmath100 , @xmath101 , and @xmath102 . + + there are other definitions quite analogous to their matrix ( @xmath2-tensors ) counterparts such as the definition of addition , kronecker binary product , and product of a tensor with a scalar , we shall skip such definitions here . + _ ternary product of tensors : _ at the center of our proposed formulation is the definition of the ternary product operation for @xmath3-tensors . this definition , to the best of our knowledge has been first proposed by p. bhattacharya in @xcite as a generalization of matrix multiplication . let @xmath103 be a tensor of dimensions @xmath104 , @xmath105 a tensor of dimensions @xmath106 , and @xmath107 a tensor of dimensions @xmath108 ; the ternary product of @xmath4 , @xmath109 and @xmath78 results in a tensor @xmath110 of dimensions @xmath111 denoted @xmath112 and the product is expressed by : @xmath113 the specified dimensions of the tensors @xmath4 , @xmath109 and @xmath78 provide constraints for triplet of @xmath3-tensors that can be multiplied using the preceding product definition . the dimensions constraints are best illustrated by fig.[2 ] . there are several ways to generalize matrix product . we chose the previous definition because the entries of the resulting tensor @xmath114 relate to the general _ inner product _ operator as depicted by fig.[1 ] . therefore , the tensor product in eq [ eq : ternaryproduct ] expresses the entries of @xmath115 as inner products of the triplet of horizontal , depth , and vertical vectors of @xmath4 , @xmath109 and @xmath78 respectively as can be visualized in fig.[1 ] . we note that matrix product is a special instance of a tensor product and we shall discuss subsequently products of @xmath7-tensor where @xmath7 is positive integer greater or equal to @xmath2 . furthermore the proposed definition of the tensor multiplication suggests a generalization of the binary vector outer product operator to a ternary operator of slices . the ternary outer product is defined such that given tensors @xmath4 of dimensions @xmath116 , @xmath109 of dimensions @xmath117 , and @xmath78 of dimensions @xmath118 , their ternary outer product @xmath115 , noted @xmath119 , is an @xmath111 tensor defined by : @xmath120 note that @xmath4 , @xmath121 and @xmath78 here are slices arising from oriented matrices . the above definition generalizes the binary vector outer product operation to a ternary matrix outer product operation defined by @xmath122 similarly to matrix multiplication , where the operation of multiplying appropriate sized matrices can be viewed as a summation of outer product of vectors , the product of appropriate sized triplet of tensors in eq [ eq : ternaryproduct ] can be viewed as a summation of ternary outer product of slices @xmath123 _ _ + _ ternary dot product with a background tensor : _ the ternary dot product above can be further generalized by introducing the notion of a background tensor as follows for @xmath124 , @xmath125 and @xmath126 @xmath127 the preceding will be referred to as the _ triplet dot product _ operator with _ background tensor _ @xmath128 . background tensors plays a role analogous to that of the metric tensor . the triplet dot product with non trivial background tensor corresponds to a pure trilinear form . furthermore the outer product of @xmath2-tensors can be generalized using the notion of background tensors to produce a @xmath3-tensor @xmath115 which result from a product of three @xmath2-tensors namely @xmath129 , @xmath130 and @xmath131 as follows , @xmath132 the preceding product expression is the one most commonly used as a basis for tensor algebra in the literature as discussed in @xcite . + we may note that the original definition of the dot product for a triplets of vectors corresponds to a setting where the background tensor is the kronecker delta @xmath133 that is @xmath134 where @xmath135 denotes hereafter the kronecker tensor and can be expressed in terms of the kronecker @xmath2-tensors as follows @xmath136 equivalently @xmath133 can be expressed in terms of the canonical basis @xmath137 in @xmath138-dimensional euclidean space described by : @xmath139 tensor . ] hence @xmath140 in general it follows from the algebra described in the previous section for @xmath3-tensors that : @xmath141 in some sense the preceding illustrates the fact that the product operator is non associative over the set of tensors . however tensor product is weakly distributive over tensor addition that is to say @xmath142,\,\boldsymbol{c},\,\boldsymbol{d}\right))=\circ\left(\boldsymbol{a},\boldsymbol{c},\boldsymbol{d}\right)+\circ\left(\boldsymbol{b},\boldsymbol{c},\boldsymbol{d}\right),\ ] ] however in general @xmath143 _ _ + _ transpose of a tensor : _ given a tensor @xmath144 we define it s _ transpose _ @xmath145 and it s _ double transpose _ @xmath146as follows : @xmath147 @xmath148 it immediately follows from the definition of the transpose that for any tensor @xmath4 , @xmath149 incidentally the transpose operator corresponds to a cyclic permutation of the indices of the entries of @xmath4 . therefore we can defined a inverse transpose @xmath150 , generally we have @xmath151 furthermore , a tensor @xmath4 is said to be symmetrical if : @xmath152 as a result for a given arbitrary @xmath3-tensor @xmath4 , the products @xmath153 , @xmath154 and @xmath155 all result in symmetric tensors . it also follows from the definitions of the transpose operation and the definition of ternary product operation that : @xmath156^{t}=\circ\left(\boldsymbol{b}^{t},\boldsymbol{c}^{t},\boldsymbol{a}^{t}\right)\ ] ] and @xmath156^{t^{2}}=\left[\circ\left(\boldsymbol{b}^{t},\boldsymbol{c}^{t},\boldsymbol{a}^{t}\right)\right]^{t}=\circ\left(\boldsymbol{c}^{t^{2}},\boldsymbol{a}^{t^{2}},\boldsymbol{b}^{t^{2}}\right).\ ] ] + _ adjoint operator : _ for @xmath157 we introduce the analog of the _ adjoint _ operator for @xmath3-tensors in two steps . the first step consists in writing all the entries of @xmath4 in their complex polar form . @xmath158 the final step expresses the adjoint of the tensor @xmath4 noted @xmath159 as follows @xmath160 the adjoint operator introduced here allows us to generalize the notion of hermitian matrices or self adjoint matrices to tensors . a tensor is hermitian if the following identity holds @xmath161 incidentally the products @xmath162 , @xmath163 and @xmath164 result in self adjoint tensors or hermitian tensors . + _ identity tensor : _ let @xmath165 denotes the tensor having all it s entries equal to one and of dimensions @xmath111 . recalling that @xmath133 denotes the kronecker 3-tensor , we define the _ identity tensors _ @xmath166 to be : @xmath167 @xmath168 furthermore we have : @xmath169 @xmath170 @xmath171 @xmath172 for all positive integer @xmath173 . the identity tensor plays a role quite analogous to the role of the identity matrix since @xmath174 we have @xmath175 * proposition 1 : * _ @xmath176 _ + we prove the preceding assertion in two steps , the first step consists of showing that the @xmath166 is indeed a solution to the equation @xmath177 let @xmath178 be the result of the product @xmath179 @xmath180 @xmath181 we note that @xmath182 hence @xmath183 the last step consists in proving by contradiction that @xmath166 is the unique solution with positive entries to the equation @xmath177 suppose there were some other solution @xmath184 with positive entry to the above equation , this would imply that @xmath185 @xmath186 @xmath187\ ] ] since this expression must be true for any choice of the values of @xmath188 we deduce that it must be the case that @xmath189 @xmath190 @xmath191 the requirement that @xmath192 which results in the sought after contradiction @xmath193 . , @xmath194and @xmath195,title="fig : " ] , @xmath194and @xmath195,title="fig:"],@xmath194and @xmath195,title="fig : " ] _ inverse : _ by analogy to matrix inverse @xmath196 we recall that for a matrix @xmath4 , @xmath196 is its inverse if @xmath197 , for any non zero matrix @xmath198 . we introduce here the notion of inverse pairs for tensors . the ordered pair @xmath199 and @xmath200 are related by inverse relationship if for any non zero @xmath3-tensor @xmath198 with appropriated dimensions the following identity holds @xmath201 _ _ + _ permutation tensors : _ incidentally one may also discuss the notion of _ permutation tensors _ associated with any element @xmath202 of the permutation group @xmath203 . @xmath204 @xmath205 the @xmath3-tensor @xmath206 perform the permutation @xmath202 on the depth slices of a @xmath3-tensor @xmath4 through the product @xmath207 , consequently the products @xmath208 and @xmath209 perform the same permutation respectively on the row slices and the column slices of @xmath4 . + + * proposition 2 : * any permutation of the depth slices of @xmath4 can be obtained by finite sequence of product of transposition , and the sequence is of the form @xmath210 the preceding is easily verified using the definition above and the permutation decomposition theorem @xcite . furthermore permutation tensors suggest a generalization of bi - stochastic matrices to bi - stochastic tensors through the birkhoff - von neumann bi - stochastic matrix theorem . from linear algebra we know that permutation matrices belong to both the set of bi - stochastic matrices and to the set of orthogonal matrices . we described above a approach for defining bi - stochastic @xmath3-tensors , we shall address in this section the notion of orthogonality for @xmath3-tensors . we recall from linear algebra that a matrix @xmath211 is said to be orthogonal if @xmath212 when we consider the corresponding equation for 3-tensors two distinct interpretations arise . the first interpretation related to orthonormal basis induced by the row or column vectors of the orthogonal matrix @xmath211 that is : @xmath213 the corresponding equation for a @xmath3-tensor @xmath214 of dimensions @xmath215 is given by : @xmath216 or explicitly we can write : @xmath217 the second interpretation arises from the kronecker invariance equation expressed by : @xmath218 the corresponding kronecker invariance equation for @xmath3-tensor is given by : @xmath219 @xmath220^{\dagger}=\left[\circ\left(\circ\left(\boldsymbol{q},\circ\left(\boldsymbol{q}^{\dagger},\boldsymbol{q}^{\dagger^{2}},\boldsymbol{\delta}\right),\boldsymbol{q}^{\dagger^{2}}\right),\boldsymbol{q},\boldsymbol{q}^{\dagger}\right)\right]^{\dagger^{2}}.\label{eq : kronecker invariance}\ ] ] while kronecker invariance properly expresses a generalization of the conjugation operation and the @xmath3-uniform hypergraph isomorphism equation it does not follow from the first interpretation of orthogonality , that is to say @xmath221 we now discuss _ scaling tensors . _ the scaling _ _ tensor play a role analogous to diagonal matrices in the fact that tensor multiplication with scalling tensor results in a tensor whose vectors are scalled . first we observe that the identity pairs of tensors should corespond to special scaling tensors . the general family of diagonal tensors are expressed by pairs of tensors @xmath222 , @xmath223 such that @xmath224 @xmath225 the product @xmath114 yields @xmath226 @xmath227 the expression above illustrates the fact that @xmath228 and @xmath229 scale the entry @xmath230 of the tensor @xmath4 , or equivalently one may view the expression above as describing the non - uniform scaling of the following vector @xmath231 . the vector scaling transform is expressed by @xmath232 furthermore the scaling factors for a given vector may be viewed as coming from the same vector of the scaling matrix @xmath233 if the matrix @xmath71 is symmetric . finally we may emphasize the analogy with diagonal matrices , which satisfy the following equation independently of the value assigned to their non zero entries . for a given @xmath115 , we solve for @xmath78 such that @xmath234 we recall from matrix algebra that : @xmath235 and furthermore @xmath236 @xmath237 by analogy we may define scaling tensors to be tensors satisfying the following equation independently of the value of the nonzero tensors . @xmath238 a possible solution is given by @xmath239 @xmath240 @xmath241 this is easily verified by computing the product @xmath242 @xmath243 @xmath244 @xmath245 fig[4 ] provides an example of diagonal tensors . it so happens that @xmath4 , @xmath109 , @xmath78 discussed above are related by transpose relation for third order tensors . this fact considerably simplifies the formulation of the to diagonality property common to both matrices and @xmath3-tensors . by analogy to matrices we say for @xmath3-tensors that a tensor @xmath246 is diagonal if independently of the value of the non zero entries of @xmath115 we have : @xmath247 * proposition 3 : * if a 3-tensor @xmath115 can be expressed in terms of a symmetric matrix @xmath248 in the form @xmath249 then @xmath115 is diagonal . + the proof of the proposition follows from the fact that : @xmath250 @xmath251 from which it follows that @xmath252 [ [ observations - from - the - eigen - valuevector - equations . ] ] observations from the eigen - value / vector equations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we briefly review well established properties of matrices and their spectral decomposition , in order to emphasize how these properties carry over to spectral decomposition of tensors . from the definition of eigen - value / vector equation , we know that for a square hermitian matrix @xmath4 , there must exist pairs of matrices @xmath211 , @xmath178 and pairs of diagonal matrices @xmath115 , @xmath253 such that @xmath254 where the columns of @xmath255 corresponds to the left eigenvectors of @xmath4 , the rows of @xmath178 corresponds to the right eigenvectors of @xmath4 and the entries of the diagonal matrix @xmath256 correspond to eigenvalues of @xmath4 . @xmath257 let @xmath258 , i.e. , the entries of the matrix resulting from the outer product of the @xmath1-th left eigenvector with the @xmath1-th right eigenvector , incidentally the spectral decomposition yields the following expansion which is crucial to the principal component analysis scheme . @xmath259 the preceding amounts to assert that the spectral decomposition offers for every entry of the @xmath2-tensor @xmath4 a positional encoding in a basis formed by the eigenvalues of the matrix . assuming that the eigenvalues are sorted in decreasing order , the preceding expression suggest an approximation scheme for the entries of @xmath4 and , therefore , an approximation scheme for the @xmath2-tensor @xmath4 itself . [ [ definition ] ] definition + + + + + + + + + + the spectrum of an @xmath7-tensor corresponds to the collection of lower order tensors the entry of which are solutions to the characteristic system of equations . [ [ spectrum - of - hermitian - tensors ] ] spectrum of hermitian tensors + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the aim of this section is to rigorously characterize the spectrum of a symmetric tensor of dimensions @xmath260 . fig.[5 ] depicts the product and the slice that will subsequently also be referred to as eigen - matrices . we may state the _ spectral theorem _ as follows + * theorem 1 : * _ ( spectral theorem for @xmath3-tensors ) : _ for an arbitrary hermitian non zero @xmath3-tensor @xmath4 with @xmath261 there exist a factorization of the form : @xmath262^{\dagger^{2}},\left[\circ\left(\boldsymbol{s},\boldsymbol{f},\boldsymbol{f}^{t}\right)\right]^{\dagger}\right)\\ \boldsymbol{\delta } & = & \circ\left(\boldsymbol{q},\:\boldsymbol{r}^{\dagger^{2}},\:\boldsymbol{s}^{\dagger}\right ) \end{array}\end{cases}\label{eq:3tensor_factorization}\ ] ] where @xmath115 , @xmath253 , @xmath263 denote scaling tensors . for convenience we introduce the following notation for scaled tensors @xmath264 and simply expresses the tensor decomposition of @xmath4 as : @xmath265 in what follows the polynomial ideal generated by the set of polynomials @xmath266 is noted @xmath267 . we first emphasize the similarity between the spectral theorem for tensors and matrices , by providing an alternative proof of a weaker form of the spectral theorem for hermitian matrices with forbenius norm different from @xmath0 . finally we extend the proof technic to @xmath3-tensors and subsequently to @xmath7-tensors . our aim is to prove that the spectral decomposition exists for an arbitrary matrix @xmath4 with forbenius norm different @xmath0 . for this we consider the ideals induced by the characteristic system of equations for matrices . the spectral decomposition of @xmath4 refers to the decomposition : @xmath268 the spectral decomposition equation above provides us with polynomial system of equations in the form @xmath269 conveniently rewritten as @xmath270 the ideal being considered is : @xmath271.\ ] ] where the variables are the entries of the pairs of matrices @xmath211 , @xmath178 and @xmath272 * weak spectral theorem * _ _ ( for @xmath2-tensors ) _ _ * * : * * for an arbitrary non zero hermitian @xmath2-tensor @xmath4 with @xmath273 the spectral system of polynomial equations : @xmath274 admits a solution . we prove this theorem by exhibiting a polynomial @xmath275 which does not belong to the following ideal @xmath276 consider the polynomial @xmath277 we claim that @xmath278 since @xmath279 which contradicts to the assumption that @xmath280 . hence we conclude that @xmath281 which completes the proof . @xmath193 + in the proof above hermicity played a crucial role in that it ensures that the eigenvalues are not all zeros since for non zero hermitian @xmath2-tensor @xmath4 @xmath282 we procede to derive the existence of spectral decomposition for @xmath3-tensors using the proof thechnic discussed above @xmath262^{\dagger^{2}},\left[\circ\left(\boldsymbol{s},\boldsymbol{f},\boldsymbol{f}^{t}\right)\right]^{\dagger}\right)\\ \boldsymbol{\delta } & = & \circ\left(\boldsymbol{q},\:\boldsymbol{r}^{\dagger^{2}},\:\boldsymbol{s}^{\dagger}\right ) \end{array}\end{cases}\ ] ] equivalently written as @xmath283 the variables in the polynomial system of equations are the entries of the @xmath3-tensor @xmath211 , @xmath178 , @xmath284 and the entries of the scaling tensors @xmath115 , @xmath253 , @xmath263 . + it is somewhat insightfull to express the system of equations in a similar form to that of matrix spectral system of equations using inner product moperators : @xmath285 where @xmath286 is a diagonal matrix whose entries are specified by @xmath287 the characteristic system of equations yields the ideal @xmath288 defined by @xmath289 where @xmath290 . which corresponds to a subset of the polynomial ring over the indicated set of variables . the following theorem is equivalent to theorem 1 . + * theorem : * _ ( for @xmath3-tensors ) _ if @xmath4 is a non zero hermitian and @xmath261 then the spectral system of equations expressed as @xmath285 admits a solution . similarly to the @xmath2-tensor case , we exhibit a polynomial @xmath8 which does not belong to the ideal @xmath288 defined bellow . @xmath291 such a polynomial @xmath8 is expressed by @xmath292 @xmath293 since @xmath294 which contradicts our assumption that @xmath261 , this completes the proof . @xmath193 + hermiticity also ensure that the solution to the spectral decomposition is not the trivial all zero solution since for non zero @xmath3-tensor @xmath4 @xmath295 similarly to the formulation for the spectral theorem for matrices , we can also discuss the notion of eigen - objects for tensors . in order to point out the analogy let us consider the matrix decomposition equations in eq [ eq : matrix_decomp1 ] and eq [ eq : matrix_decomp2 ] , one is therefore led to consider the matrices @xmath296 as the scaled matrix of eigenvectors . according to our proposed decomposition , the corresponding equations for @xmath3-tensors is given by @xmath297 recall that the tensor @xmath298 collects as slices what we refer to as the _ scaled eigen - matrices_. the analogy with eigenvectors is based on the following outerproduct expansion . @xmath299 the equation emphasizes the fact that a hermitian matrices can be viewed as a sum of exterior products of scaled eigenvectors and the scaling factor associated to the rank one matrix resulting from the outerproduct corresponds to the eigenvalue . similarly , a symmetric @xmath3-tensor may also be viewed as a sum outer products of slices or matrices and therefore we refer to the corresponding slices as scaled _ eigen - matrices_. the outerproduct sum follows from the identity @xmath300 expressed as : @xmath301 which can be equivalently written as @xmath302 @xmath303 where @xmath304 denote the @xmath1-th component expressed @xmath305 _ we may summarize by simply saying that : as one had eigenvalues and eigenvectors for matrices one has eigenvectors and eigen - matrices for @xmath3-tensors . _ we shall first provide an algorithmic description of the characteristic polynomial of matrix without assuming the definition of the determinant of matrices and furthermore show how the description allows us to define characteristic polynomials for tensors . we recall for a matrix that the characteristic system of equations is determined by the algebraic system of equations @xmath306 as discussed above induces the following polynomial ideal @xmath307.\ ] ] let @xmath308 be the reduced gr@xmath309bner basis of @xmath288 using the ordering on the monomials induced by the following lexicographic ordering of the variables . @xmath310 in the case of matrices it has been established that there is a polynomial relationship between the eigenvalues ; more specifically the eigenvalues are roots to the algebraic equation @xmath311 by the elimination theorem @xcite we may computationaly derive the characteristic polynomials as follows @xmath312=\det\left(\boldsymbol{a}-\lambda_{l}\boldsymbol{i}\right)\ ] ] it therefore follows from this observation that the reduced gr@xmath309bner basis of @xmath288 determines the characteristic polynomial of @xmath4 . let @xmath308 denote the reduced gr@xmath309bner basis of the ideal @xmath288 using the the lexicographic order on the monimials induced by the following lexicographic order of the variables . @xmath313 where @xmath314 the reduced characteristic set of polynomials @xmath315 associated with the hermitian @xmath3-tensor @xmath4 is a subset of the reduced groebner basis @xmath308 such that @xmath316\ ] ] where @xmath317 $ ] denotes the polynomial ring in the entries of the sacaling tensor with complex coefficients . the reduced should here be thougth of as generalization of the characteristic polynomial associated with matrices . an @xmath318 @xmath7-tensor @xmath4 is a set of elements of a field indexed by the set resulting from the cartesian product @xmath319 the dimensions of @xmath4 is specified by @xmath318 where @xmath320 @xmath321 , @xmath322 specifies the dimensions of the tensor . we may also introduce a dimension operator defined by @xmath323 finally , we shall simply use the notation convention @xmath324 for describing @xmath4 once the dimensions have been specified . + in what follows we will discuss general tensor products for @xmath7-tensors where @xmath7 is a positive integer greater or equal to 2 . let us start by recalling the definition of matrix multiplication @xmath325 the preceding matrix product generalizes to the proposed @xmath3-tensor product as follows @xmath326 by closely inspecting the expression of the product we note that if * @xmath327 * is a @xmath328 tensor , and @xmath329 is a @xmath330 tensor then the resulting tensor @xmath109 expressed by @xmath331 will be of dimensions @xmath94 . the product above expresses the action of @xmath3-tensor @xmath332 of dimension @xmath333 on the pair of matrices arising from @xmath327 and @xmath329 . furthermore for @xmath332 having entries such that @xmath334 the result of the action of @xmath332 on the pair of matrices arising from the tensors@xmath327 and @xmath329 simply corresponds to a matrix multiplication . for @xmath335-tensor the product operator is expressed as : @xmath336 similarly the tensor @xmath329 can be chosen to be all - one tensor which reduces the product above to the product operation for @xmath3-tensors . this nested relationship will also apply to higher order tensors . + we may now write the expression for the product of @xmath7-tensor . let @xmath34 denotes a set of @xmath7-tensors . the product operator has therefore @xmath7 operands and is noted : @xmath337 defined by @xmath338 @xmath339 it follows from the definition that the dimensions of the tensors in the set @xmath34must be chosen so that : @xmath340 which describes the constraints on the dimension relating all the @xmath7 tensors in the product . the constraints accross the @xmath341 other dimensions for each tensor are described by the following relation . @xmath342 the tensor @xmath109 resulting from the product is a @xmath7-tensor of dimensions . @xmath343 _ note that the product of tensors of lower order all arise as special cases of the general product formula describe above . _ _ tensor action : _ + the action of @xmath344 order tensor @xmath324 on @xmath345-tuple of order @xmath345 tensors @xmath346is defined as @xmath347 the equation above generalizes the notion of matrices action on a vector . _ tensor outerproduct : _ the outer - product of @xmath7-tuple @xmath345-tensors is denoted by : @xmath348 and defined such that : @xmath349 the kronecker @xmath7-tensor is defined as @xmath350 _ order @xmath7 tensor transpose / adjoint : _ + given a tensor @xmath351 the transpose @xmath145 is defined such that @xmath352 for a complex valued tensor where the entries are expressed in their polar form as follows : @xmath353 the generalized adjoint is given by @xmath354 @xmath355 where @xmath356 denotes the composition of @xmath1 cyclic permutation of the indices from which it follows that @xmath357 in order to formulate the spectral theorem for @xmath358 we will briefly discussed notion of orthogonal and scaling @xmath7-tensors , which can be expressed as @xmath359 that is @xmath360 where @xmath361 denotes the transpose operation , which still corresponds to a cyclic permutation of the indices . we first provide the formula for the scaling tensor whose product with @xmath4 leaves the tensor unchanged . @xmath362 @xmath363 @xmath364 the above family of tensors play the role of identity operator and are related to one another by transposition of the indices . the more general expression for the scaling tensors is therefore given by @xmath365 where @xmath233 is a symmetric matrix . the expression for the scaled orthogonal tensor is therefore expressed by @xmath366 we therefore obtain that the scaled tensor which will be of the form : @xmath367 * theorem 2 : * _ ( spectral theorem for @xmath7-tensors ) : _ for any non zero hermitian tensor @xmath358 such that @xmath368 , there exist a factorization in the form @xmath369 the expression above generalizes eq [ eq : tensor factorization scaled ] the spectral decompostion yields the following system of equations @xmath369 more insightfully rewritten as @xmath370 where @xmath371 is a diagonal matrix whose entries are specified by @xmath372 we had already pointed out earlier in the proof for the spectral theorem for @xmath3-tensors that the proof technique would apply to @xmath7-tensors with norm @xmath373 , where @xmath7 is a positive integer greater or equal to @xmath2 . similarly we consider the polynomial expression @xmath374 @xmath375 @xmath376 and observe that @xmath293 where @xmath288 defines the ideal iduced by the spectral system of equation since @xmath377 which contradicts our assumption that @xmath368 , hence we conclude that @xmath378 this completes the proof . @xmath193 + the @xmath138 `` slices '' of the scaled tensor @xmath379 constitutes what we call the _ scaled eigen - tensors _ of @xmath4 which are @xmath345-tensors . we recursively define the spectral hierarchy for a tensor @xmath358 . the base case for the recursion is the case of matrices . the spectrum of an @xmath380 matrix is characterized by a set of @xmath138 scaled eigen - vectors . the existence of the spectral hierarchy relies on the observation that the spectrum of an order @xmath7-tensor @xmath358 is determined by a collection of @xmath138-tuple @xmath345-tensors not necessarily distinct . each one of these @xmath138-tuples corresponding to a scaled orthogonal eigen - tensor . by recursively computing the spectrum of the resulting scaled orthogonal @xmath345-tensors , one determines a tree structure which completely characterizes the spectral hierarchy associated with the @xmath7-tensor @xmath4 . the leaves of the tree will be made of scaled eigenvectors when the spectral decomposition exists for all the resulting lower order tensors . it therefore follows that the tensor @xmath4 can be expressed as a nested sequence of sums of outer products . we illustrate the general principle with @xmath3-tensors . let @xmath4 denotes a third order tensor which admits a spectral decomposition in the form described by eq [ eq : outer product expansion ] . we recall that the spectral decomposition for @xmath3-tensors is expressed by @xmath381 @xmath382 by computing the spectrum of the scaled eigen - matrices we have : @xmath383 @xmath384 @xmath385 where @xmath386 , @xmath387,@xmath388,@xmath389 and @xmath390,@xmath391 denote the eigenvalues and corresponding eigenvectors respectively for the matrices @xmath392 , @xmath393 , @xmath394 . it therefore follows that @xmath4 can be expressed by the following nested sum of outer product expressions @xmath395 @xmath396,\right.\ ] ] @xmath397,\left[\sum_{1\le j_{3}\le l}\left(\sqrt{\beta_{j_{3}}(k)}\cdot\vec{\boldsymbol{y}}_{j_{3}}(k)\right)\otimes\left(\sqrt{\beta_{j_{3}}(k)}\cdot\vec{\boldsymbol{z}}_{j_{3}}(k)\right)\right]\right)\ ] ] we shall present in this section a brief overview of the relationship between our framework and earlier proposed tensor decompositions let us show in this section how the tucker decomposition in fact uses matrix algebra more specifically orthogonality of matrices to express the singular value decomposition for @xmath3-tensors . we used for this section the notation and convention we introduced through this work . the tucker factorization scheme finds for an arbitrary @xmath3-tensor @xmath115 the following decomposition @xmath398 where @xmath128 denotes a @xmath3-tensor and @xmath399 denote matrices . the product expression used for the decomposition written above corresponds to our proposed definition for triplet dot product with non trivial background as described in eq [ eq : matrices to tensor ] . using our notation we can express the decomposition of @xmath115 as follows : @xmath400 our starting point is the following invariance relation , which arises from the matrix products with the identity matrix . @xmath401 where @xmath402 , @xmath403 and @xmath404 which correspond to transposes of the identity matrix . for any orthogonal matrices @xmath211 , @xmath284 and @xmath60 we know that @xmath405 incidentally the expression in eq [ eq : invariance tucker ] can be written as : @xmath406 by interchanging the order of the sums we get : @xmath407 we now separate out the products in the expressions to yield the general form of the tucker decomposition . @xmath408 @xmath409 the preceding emphasizes that the tucker decomposition reuses matrix orthogonality and does not provide a generalization of the notion of orthogonality for @xmath7-tensors . finally to determine the orthogonal matrices @xmath211 , @xmath284 and @xmath60 to use we specify the following constraints @xmath410 @xmath411 @xmath412 which is referred to as the total orthogonality condition . the rank 1 decomposition of tensor @xcite corresponds to solving the following optimization problem . given an @xmath44-tensor @xmath53 we seek to find : @xmath413 since johan h@xmath414stad in @xciteestablished the intractability of the tensor rank problem for @xmath3-tensors we briefly discuss the relationship to our framework . it follows from the definition of the outer product of matrices to form a @xmath3-tensor that @xmath415 we point out that for the very special matrices essentially made up of the same vector as depicted bellow : @xmath416 @xmath417 @xmath418 the outer product of the matrices @xmath419 this allows us to formulate the tensor rank problem in eq [ eq : tensor rank ] in terms of the outer product operator for slices as follows @xmath420 @xmath421 where @xmath422 are @xmath3-tensors arising from the collection of matrices associated with the collection of vectors . the preceding naturally related the tensor rank problem to our proposed tensor product . furthermore the generalized framework allows us to formulate the tensor rank problem for @xmath7-tensor where @xmath7 is a positive integer greater or equal to 2 as follows @xmath423 one may point out that the spectral decomposition associated with a hermitian tensor comes quite close to the sought after decomposition at the cost of the trading of the requirement that the matrices should be rank one to the fact the matrices should arise from scaled eigen - tensors . in this paper we introduced a generalization of the spectral theory for @xmath7-tensors where @xmath7 is a positive integer greater or equal to @xmath2 . we propose a mathematical framework for @xmath3-tensors algebra based on a ternary product operator , which generalizes to @xmath7-tensors . this algebra allows us to generalize notions and operators we are familiar with from linear algebra including dot product , tensor adjoints , tensor hermicity , diagonal tensor , permutation tensors and characteristic polynomials . we proved the spectral theorem for tensors having forbenius norm different from @xmath0 . finally we discussed the spectral hierarchy which confirms the intractability of determining the orthogonal vector components whose exterior product result in a given @xmath7-tensor . starting from the recently proposed product formula in eq [ eq : ternaryproduct ] for order @xmath3-tensors proposed by p. bhattacharya in @xcite we were able to formulate a general algebra for finite order tensors . the order @xmath3-tensor product formula suggests a definition for outer product of matrices as discussed in eq [ eq : outer - product ] , it also suggests how to express the action of a tensor on lower order tensors . most importantly with eq [ eq : background - tensors ] we propose a natural generalization for the dot product operator and a generalization for the riemann metric tensor ideas . furthermore the tensor algebra that we discuss sketches possible approaches to investigate generalizations of inner product space theory . one important characteristic of the product operator for tensor of order strictly greater than 2 is the fact that the product is not associative . incidentally by analogy to matrix theory where the lost of commutativity for matrix product results into a commutator theory and lie algeras which plays an important role in quantum mechanics , the lost of associativity as expressed in eq [ eq : non_associative ] could potentially give rise to an associator theory or generalizations of lie algebras . furthermore the transpose operator described in eq [ eq : transpose ] emphasizes the importance of the roots of unity in generalizing herminian and unitary tensors . the @xmath3-tensor permutation tensors provided a suprising representation for the permutation group @xmath203 which provide a glimpse at a tensor approach to a representation theory as well as a tensor approach to markov tensor models . at the heart of our work lies the concept of orthogonal tensors . we emphasize the fact the orthogonal tensors discussed here are generalizations of orthogonal matrices and are significantly different from orthogonal matrices . one significant difference lie in the two distinct interpretation of the orthogonality property for tensor . the first interpretation expressed by eq [ eq : orthonormaility ] is analogous to orthonormal for a set of vectors . the second interpretation relates to the invariance of the kronecker delta tensor under conjugation as expressed in eq[eq : kronecker invariance ] . furthermore we have through this work provided a natural generalization for the familiar characteristic polynomial using the important tool set of grobner basis . spectral analysis plays an important role in the theory and investigations of graphs . graph spectra have proved to be a relatively useful graph invariant for determining isomorphism class of graphs . it seem of interest to note that the symmetries of a graph described by it s corresponding automorphism group can also be viewed as depicting a 3-uniform hypergraph which can in turn be investigated by through it spectral properties . determining the relationship between spectral properties of a graph and the spectral properties of it corresponding automorphism seems worthy of attention in the context of determining isomorphism classes of graphs . the general framework which address the algebra for arbitrarily finite order tensor allowed us to derive the spectral hierarchy . the spectral hierarchy induces a bottom up construction for finite order tensor from vectors . this explicit construction may in fact prove useful in the context investigations on tensor rank problems which also validate as illustrated in eq [ eq : tensor rank problem ] our product operator . we are grateful to emilie hogan , professor doron zeilberger and professor henry cohn for helpful discussion regarding properties of ideals . the first author was partially supported by the national science foundation grant nsf - dge-0549115 . a. elgammal and c .- s . lee . separating style and content on a nonlinear manifold . in _ proceedings of the ieee computer society conference on computer vision and pattern recognition ( cvpr ) _ , volume 1 , pages 478485 , 2004 . m.e . kilmer , c.d . martin , and l. perrone . a third - order generalization of the matrix svd as a product of third - order tensors . technical report technical report number tr-2008 - 4 , tufts university department of computer science , medford , ma , october 2008 . tamara g. kolda , brett w. bader , and joseph p. kenny . higher - order web link analysis using multilinear algebra . in _ icdm 05 : proceedings of the fifth ieee international conference on data mining _ , pages 242249 , washington , dc , usa , 2005 . ieee computer society . tamara g. kolda and jimeng sun . scalable tensor decompositions for multi - aspect data mining . in _ icdm 2008 : proceedings of the 8th ieee international conference on data mining _ , pages 363372 , december 2008 . chan - su lee and ahmed elgammal . facial expression analysis using nonlinear decomposable generative models . in _ proceedings of ieee workshop on on analysis and modeling of faces and gestures ( amfg ) _ , pages 1731 , 2005 . chan - su lee and ahmed m. elgammal . towards scalable view - invariant gait recognition : multilinear analysis for gait . in _ proceedings of ieee conference on audio , video biometric people authentication ( avbpa ) _ , pages 395405 , 2005 . amnon shashua and tamir hazan . non - negative tensor factorization with applications to statistics and computer vision . in _ icml 05 : proceedings of the 22nd international conference on machine learning _ , pages 792799 , new york , ny , usa , 2005 . jian tao sun , hua - jun zeng , huan liu , and yuchang lu . cubesvd : a novel approach to personalized web search . in _ in proc . of the 14 th international world wide web conference ( www _ , pages 382390 . press , 2005 .

we calculate using perturbative calculations and ward identities the basic parameters of the fermi liquid : the scattering vertex , the landau interaction function , the effective mass , specific heat , and physical susceptibilities for a model of two - dimensional ( 2d ) fermions with a short ranged interaction at non - zero temperature . the leading temperature dependence of the spin components of the scattering vertex , the landau function , and the spin susceptibility is found to be linear . we find that the standard relationships for a galilean - invariant fermi liquid are violated by finite - temperature terms . the coefficients in the temperature corrections to these relationships involve a subtle interplay between contributions from small and large ( ) momentum processes . a connection with previous studies of the 2d fermi - liquid parameters is discussed . we conclude that the linear leading temperature dependence of the parameters is a generic feature of the 2d fermi liquid . = = amssym.def amssym

cond-mat0103155

for more than a decade two - dimensional ( 2d ) fermion systems have been a subject of great interest . one important motivation has been the non - fermi - liquid behavior observed in high-@xmath2 superconductors above @xmath2.@xcite in this context the existence and stability of the fermi liquid ( fl ) in @xmath3 has been extensively investigated within various modern approaches.@xcite it seems fair to say that no compelling theoretical evidence has been found for non - fermi - liquid behavior in systems with short ranged interactions and in the absence of coupling to gauge fields . the fl is stable , provided that standard conditions@xcite are satisfied . given this a natural question arises : suppose 2d fermions are in the fl phase in the sense of the landau s fermi liquid theory ( flt).@xcite we understand `` zero order '' fl quantities as those defined in the limit of zero temperature and zero momentum - energy transfers . then what is the low - energy behavior of such quantities ? in this paper we study the leading temperature corrections to the zero - order behavior of the the components of the landau function and the scattering vertex , the quasiparticle s effective mass , and the uniform response functions ( compressibility , spin susceptibility ) . in renormalization group ( rg ) language we are studying the corrections coming from the leading irrelevant terms.@xcite interest in the issue grown in recent years also because these leading corrections provide the `` bare '' temperature dependence of the parameters in theories describing quantum critical phenomena in metals.@xcite in particular , puzzling data in several materials @xcite led to explanations based on an unusual underlying temperature , momentum or frequency dependence of electronic susceptibilities.@xcite the leading temperature corrections to the parameters of a _ stable _ fl been studied for many years . rather surprisingly , the issue remains a subject of controversy . for example , it was found@xcite that the leading temperature correction to the specific heat coefficient @xmath4 was @xmath5 in d=3 spatial dimensions and @xmath6 in d=2.@xcite . recently these results were also rederived by multidimensional bosonization ( see , e.g. , ref . [ ] ) . whether the spin and charge susceptibilities display similarly anomalous ( i.e. , non-@xmath7 ) temperature dependence is the subject of a lengthy controversy in the literature : see , e.g. , ref . [ ] , discussion and references therein . for a most recent reassesment of such results see ref . [ ] . the prevailing conclusion was that of carneiro and pethick@xcite who found no leading @xmath8 correction to the spin susceptibility of the 3d fl . their analysis implies that terms @xmath9 are absent in 2d . the heuristic argument which runs commonly through the literature to account _ a posteriori _ for the absence of anomalous terms ( in @xmath6 or in @xmath10 ) in response functions is that although these terms are known to occur in the individual self - energy and vertex diagrams , they cancel in final results for symmetry reasons mathematically expressed via ward identities . as we now discuss , this argument is misleading . although exact symmetries of a model ( e.g. , global gauge and/or rotational invariance ) result in relationships between the self - energy and the vertex through the ward identities ( cf . [ wisec ] below ) , these are usually not enough to close the system of equations and demonstrate explicitly the cancellation of the self - energy and vertex corrections . additional symmetries like , e.g. , the chiral symmetry of the 1d tomonaga - luttinger model , are needed in order to do it.@xcite in @xmath11 analogous extra _ asymptotic _ symmetries first noted by haldane@xcite occur in the fermionic wilsonian _ low - energy effective action _ under some model s restrictions . in the recent series of papers by metzner , castellani and di castro ( reviewed in ref . [ ] ) the relationship between the extra asymptotic symmetry of the effective action and the rpa ( flt ) results for the response functions was clearly established . they also clarified a close connection between this approach and multidimensional bosonization.@xcite the additional ward identities are properties of a model in which all processes involving momentum transfers greater than a cutoff @xmath12 are discarded . these restrictions appear naturally in the wilsonian low - energy effective action with the uv cutoff ( [ lamea ] ) obtained from an rg approach such as that of shankar,@xcite in which interacting fermions are treated by progressive elimination of modes towards the fermi surface . only asymptotically ( i.e. , in the limit @xmath13 , @xmath0 ) this forward - scattering action possesses extra symmetries @xmath14 ( and @xmath15 for rotationally invariant case).@xcite ( it is assumed that the initial _ microscopic _ action ( hamiltonian ) has only ordinary @xmath16 [ and @xmath17 symmetry . ) however , these extra identities do not constrain possible @xmath6- or @xmath10- terms in physical quantities , coming from irrelevant terms . even in 1d , e.g. , the @xmath18-breaking marginally irrelevant term coming from backscattering results in @xmath19-leading correction to the uniform spin susceptibility of the luttinger liquid.@xcite in higher dimensions there are always [ even in the limit ( [ lamea ] ) ] special configurations of momenta near the fermi surface which give rise to irrelevant terms in the effective action which strongly interfere with the forward scattering channel@xcite and invalidate asymptotic symmetries . in the framework of the low - energy effective action approach there is also another potential source of temperature corrections , i.e. , the possibility of @xmath6-dependence of the wilsonian action s vertices ( couplings ) developing on the previous stages of modes elimination before one reaches the effective action scale ( [ lamea ] ) . apparently , this question has not been carefully studied . we note that in the most of existing literature studying the leading fl corrections it is assumed that the crucial coupling is between quasiparticles and long - wavelength collective modes , i.e. , only small momentum scales are taken into account . however the possibility of `` @xmath20 singularities '' , i.e. , anomalous temperature terms coming from processes involving large ( @xmath21 ) momentum transfers , has been pointed out by misawa for 3d fl already in early 70s.@xcite apparently due to the lack of experimental evidence of a @xmath8 term in the susceptibility of a generic 3d fl and also because misawa s results rely on the analysis of selected diagrams [ cf . previous paragraph ] , they were widely disregarded in favor of those of carneiro and pethick . in the context of semiconductor physics stern was the first to note@xcite that in a 2d electron gas the electron scattering rate was proportional to @xmath6 due to @xmath20 effects . the consequences of the @xmath20 effects for the leading @xmath6-dependence of 2d fl quantities have not been considered in the literature until recently . the issue of the leading correction to flt has recently been revived by several papers . belitz , kirkpatrick and vojta@xcite presented perturbative calculations , mode - coupling arguments and power counting estimates which showed that the leading @xmath10 dependence of the spin susceptibility ( but not the charge susceptibility ) was @xmath22 in 2d ( @xmath23 in 3d ) . they did not find the analogous @xmath6-correction explicitly , but concluded that one should generally expect a linear @xmath6-term in the 2d fl susceptibility ( @xmath24 in 3d ) . this dependence has important implications for the theory of the quantum critical metallic ferromagnet.@xcite snchal and one of us@xcite predicted the occurence of the linear @xmath6-corrections to the fl vertices from one - loop rg calculations based on a 2d effective action . those rg calculations of the irrelevant corrections rely more on the low - energy effective action s phase space constrains in their way to sort out the effective interactions and the scales involved , rather than on doing it according to the strength of the latter , as a perturbation theory does . that makes desirable to easily obtain a perturbative signal of the rg predictions . also , an analogous rg calculation of other fl quantities was not done . hirashima and takahashi@xcite performed numerical analyses of perturbative expressions which appeared to confirm the prediction of belitz _ et al_@xcite for 2d susceptibility . however due to numerical difficulties in handling divergences in some terms they were unable even to determine the sign of the coefficient in the leading @xmath10-term . also contrary to belitz _ et al _ who focused on the long - wavelength contributions , the authors of ref . [ ] emphasized the crucial role of @xmath20 contributions in their findings . following this misawa conjectured a phenomenological form for the free energy@xcite which results in the linear @xmath6-term in the 2d spin susceptibility and in the coefficient @xmath25 , and agrees well with the numerical calculations in lowest order.@xcite this previous work has left a number of important questions unresolved , including analytic treatment of anomalous terms ( important for verifying cancellations between different contributions ) , the relationship to flt and to ward identities , and the connection to the extensive literature from the 1960 - 70s . to elucidate the issues in the most transparent way , we apply the perturbation theory for 2d contact - interacting spin-@xmath26 fermions , starting from a _ microscopic _ action . although the landau flt is not a perturbative theory , for sufficiently weak interactions ( assuming the interaction being repulsive and we are above the kohn - luttinger temperature ) one should be able to find the parameters of the fl in terms of the coupling series . we present what is apparently the first analytic calculation of the leading @xmath6-dependence of the effective mass , landau parameters and response functions of a 2d electron gas , to second order in the interaction strength , _ including all channels and all momentum processes ( scales)_. we take into account the ward identities explicitly . as we show in this paper , the processes involving large ( @xmath27 ) momentum transfers are crucial to the anomalous temperature dependence of fl quantities . the rest of this paper is organized as follows . section [ model ] defines the model . section [ vertex ] defines the four - point vertex to be calculated and its relationship to the fermi liquid parameters . we also give there the basic equation for that vertex in the one - loop approximation . in section [ scatlan ] we present and discuss our results for the scattering amplitude and the landau function . in section [ wisec ] we give the ward identities which are used for the following calculations . in section [ efmass ] we calculate the effective mass . sections [ compr ] , [ susc ] present the results for the compressibility and the spin susceptibility , respectively . the results are recapitulated in the concluding section [ concl ] . appendices a and b contain detailed presentation of the calculations of the one - loop contributions entering the equation for the vertex . one of the goals of presenting these technicalities was to give the idea of how more involved two- and three - loop calculations of secs . [ efmass]-[susc ] were carried out . some of the results presented here were announced in a previous short communication.@xcite we treat interacting fermions at finite temperature in the standard path integral grassmannian formalism.@xcite the partition function is @xmath28 where the free part of the action is @xmath29\psi_{\alpha}({\bf 1})\ ] ] and [ notation ] @xmath30 where @xmath31 is the inverse temperature , @xmath32 the chemical potential , @xmath33 the fermion matsubara frequencies and @xmath34 a @xmath35-component grassmann field with a spin ( flavor , if @xmath36 ) index @xmath37 . summation over repeated indices is implicit throughout this paper . we set @xmath38 and @xmath39 . the @xmath40-invariant quartic interaction is @xmath41 here the conservation of energy and momentum is enforced by the symbolic delta function @xmath42 the @xmath40 extension of the physical @xmath43 symmetry is useful for different applications,@xcite and the case of the spin-@xmath26 electrons is recovered by setting @xmath44 . we can decompose the interaction by factorizing its symmetric and antisymmetric frequency - momentum- and spin- dependent parts in the way which explicitly manifests the antisymmetry of the function @xmath45 under exchange:@xcite @xmath46 where the functions @xmath47 and @xmath48 have the symmetry properties [ usym ] @xmath49 while two operators @xmath50 and @xmath51 , which are respectively symmetric and antisymmetric in the spin space , are defined as follows : [ base ] @xmath52 if we assume an instantaneous point interaction between electrons , then @xmath53 and @xmath54 . the sign of the interaction is chosen such that @xmath55 corresponds to repulsion . then @xmath56 in this paper we consider mainly 2d electrons with the bare spectrum of a free gas @xmath57 and the circular fermi surface , interacting via ( weak ) contact repulsion ( [ sinthub ] ) , but discuss the consequences of generic spectra and a non - circular fermi surface . we define the two - particle green s function as : @xmath58 the 1pi vertex @xmath59 is related to @xmath60 as @xmath61 where due to the symmetries of the model we can write the one - particle green s function as @xmath62 perturbatively , the vertex is constructed by considering only connected one - particle - irreducible ( 1pi ) diagrams with amputated external legs . in lowest order @xmath63 . to shorten notations we use the hat meaning that it is an operator in the spin space , and it also comprises two components . along with the representation ( [ pots ] ) with the components ( @xmath64 , @xmath65 ) for the vertex , which explicitly shows its antisymmetry , we will also use another representation more convenient for some applications , namely separating the vertex into charge and spin components via @xmath66 where @xmath67 ( @xmath68 ) are hermitian traceless generators of the @xmath40 group , coinciding with the pauli matrices for the case @xmath44 , and normalized such that @xmath69 the components of the vertex in different representations are related as [ abdef ] @xmath70 taking into account momentum and energy conservation , we write the vertex as : @xmath71 with the transfer vector @xmath72 where @xmath73 is a bosonic matsubara frequency . in what follows we will frequently use the property of the components @xmath74 and @xmath75 which is a consequence of the vertex s antisymmetry @xmath76 for the calculation of physical quantities of the fl we need @xmath77 at @xmath78 . as is well known@xcite the limit @xmath79 is not unique , since the vertex is non - analytic function at @xmath80 . two vertices @xmath81 @xmath82 can be defined unambiguously at @xmath83 and then continued to @xmath84 . ( see , e.g. , refs . [ ] on this subtlety ) . in the calculation of the flt vertices ( real electrons , @xmath44 ) the external momenta are chosen to lie on the fermi surface , and the external fermionic frequencies are put equal to the minimal matsubara frequency @xmath85 . in the case of a rotationally invariant fermi surface the vertex dependence on the external momenta @xmath86 and @xmath87 lying on the fermi surface can be parameterized by the relative angle @xmath88 between those momenta and then @xmath89 , @xmath90 can be identified with the scattering vertex and the landau function , respectively . namely , for the components of the scattering vertex ( @xmath91 we have ( cf . , e.g. , ref . [ ] ) @xmath92 where @xmath93 is the @xmath94-dimensional renormalized electron density of states per spin on the fermi level , @xmath95 is the @xmath94-dimensional area of the unit sphere ( for @xmath96 : @xmath97 ) , and @xmath98 is the field renormalization constant . so @xmath99 @xmath100 two components ( @xmath101 of the landau interaction function are defined by analogous equations , with the substitution @xmath102 , @xmath103 , @xmath104 . the one - loop approximation @xmath105 of the straightforward perturbation theory for @xmath106 in diagrammatic form is given in fig . 1 . for the model given by ( [ s0],[sinthub ] ) we obtain after some spin summations @xmath107 u_0 ^ 2 \bigg\ } \hat t\ ] ] with @xmath108 we denote @xmath109 . the functions @xmath110 and @xmath111 coming from the contributions of bubbles ( zs , zs ) and ( bcs ) , respectively , are defined and calculated in appendices [ apa ] and [ apb ] . from the one - loop approximation ( [ gammapertexp ] ) for the 1pi vertex @xmath112 valid for generic momenta , energies and transfer , we can obtain now the fermi liquid parameters ( vertices ) using the definitions ( [ two - limits1],[two - limits2],[a],[b ] ) . taking the appropriate zero transfer limit we obtain from eqs . ( [ gammapertexp],[a],[b ] ) the components of the scattering vertex @xmath113 @xmath114 with @xmath115 and @xmath116 . we absorbed the factor @xmath117 into the definition of coupling via @xmath118 definitions and detailed calculations of the functions involved in the equations above are given in appendices [ apa ] and [ apb ] . for the components of the landau function we obtain in the same way @xmath119 @xmath120 note that while calculating the fl vertices ( [ a],[b ] ) at the one - loop level we can put @xmath121 ( i.e. , @xmath122 ) , @xmath123 , and since @xmath124 , the zs contribution to @xmath125 , @xmath126 is zero in the limit ( [ two - limits2 ] ) . using the results ( [ chi0app],[chi1app],[c0f],[c1f ] ) of the appendices we obtain for the fourier components of the two vertices : @xmath127 @xmath128 @xmath129 @xmath130 @xmath131 @xmath132 the terms omitted in the above results are @xmath133 at most . within one - loop accuracy the standard rpa - type relationships between the fl parameters @xmath134 , @xmath135 hold note that taken separately , each contribution of the zs- or bcs bubble has a leading linear temperature - dependent term in the first two fourier components of the vertices [ cf . eqs.([chi0app],[chi1app ] ) and eqs.([c0f],[c1f ] ) , resp . ] . however , a cancellation of such terms coming from two graphs occurs in the `` charge sector '' ( i.e. , in @xmath136 , @xmath125 components ) , while the linear t - terms survive in the `` spin sector '' ( @xmath137 , @xmath126 components ) . to understand where the temperature dependence of the vertices comes from , note that @xmath138 as a function of @xmath139 has some `` temperature smearing '' near @xmath140 , otherwise it is virtually indistinguishable from its zero - temperature limit ( [ lt0 ] ) , provided the condition ( [ t0 ] ) applies . so the temperature dependence of the zs ( exchange ) contribution to the fourier components of the vertices comes from integrating of the function @xmath141 around the `` effective transfer '' @xmath140 in the zs graph , i.e. , when incoming momenta @xmath142 , while @xmath143 . notice that the transfer @xmath10 in the proper sense ( [ gammatr],[transfer ] ) is zero . this source of temperature dependence was not considered by pethick and carneiro , for example . in the same vein , one can see from eq.([cqf ] ) that the temperature dependence of the bcs contribution comes from regions of small @xmath144 , i.e. , again when @xmath142 . the coefficients in the temperature expansions of the fourier components of @xmath141 and @xmath145 are such [ cf . eqs.([chi0app],[chi1app],[c0f],[c1f ] ) ] that the cancellation occurs only for the linear terms in the charge components of the vertices , while the higher order terms in the temperature survive . since to the best of our knowledge there is no particular reason for this cancellation of the linear terms to happen , we expect that linear @xmath6-term of the charge component(s ) would come out of a more realistic calculations of the fl parameters . let us elaborate on the last point a bit more . first notice that in the perturbative approximation ( [ gammapertexp ] ) for the vertex in the model with a constant couplings @xmath146 ( bare vertex ) , all the three one - loop terms have the same factor @xmath147 in front of the bubble contributions @xmath148 . had these coupling factors been different for each of the three graph , the cancellation of the temperature corrections would not have happened . on the other hand , even the simplest approximation ( [ gammapertexp ] ) shows that fermionic interactions drive any bare _ coupling constant _ towards a _ coupling function _ of , generally speaking , three momenta [ cf . eq.([gammatr ] ) ] . more advanced calculations of the fl parameters , like , e.g. , rg , would reproduce such development of the momentum dependence in couplings in the regime of effective action ( [ lamea ] ) . this would prevent the cancellations between different terms , e.g. , between the one - loop zs and bcs contributions , which we have obtained in the simple perturbational framework.@xcite another artifact of the naive perturbative calculation of the vertex is its ultraviolet divergence coming from the bcs bubble . the @xmath149-term makes our perturbation theory applicable only if the interaction is small enough that the arbitrary , generally speaking , ultaviolet cutoff @xmath150 can be chosen to satisfy @xmath151 . however , we should not worry too much about this issue , since more sophisticated calculations based , e.g. , on summations of the diagrammatic series or an rg would result in an effective screening of the repulsive interaction in the bcs channel . see , e.g. , refs . ( although providing the possibility of the _ infrared _ instability ( divergence ) at exponentially low temperatures ( much lower than , say , we are working at ) through the kohn - luttinger mechanism.@xcite ) the linear dependence of the leading temperature corrections to the fl parameters seems not to be sensitive to the actual form of the one - particle spectrum . the same dependence ( apart from model - sensitive prefactors ) was obtained in the previous rg analysis of the _ effective action _ model for 2d spinless fermions with a linearized spectrum.@xcite in that study the integration of the `` effective '' transfer in the exchange ( zs ) graph contribution to the rg flow was also resulting in the linear @xmath6-term in the vertex corrections . according to ref . [ ] such temperature behavior can be understood from dimensional arguments . in order to derive the ward identities , we followed closely the methodology of ref . [ ] adapted to the path - integral technique according to standard approaches.@xcite the partition function is given by the path integral ( [ z ] ) . in coordinate - imaginary - time space the free part of the action is @xmath152\psi_{\alpha}({\it x})\ ] ] we consider pair interactions which preserve the total spin and number of particles @xmath153 here @xmath154 . notice that interaction ( [ sinthub ] ) is a special case of ( [ pint ] ) . then the following _ charge ward identity _ @xmath155 and the _ spin ward identity _ @xmath156 can be derived . from ( [ vrep],[lamnor ] ) one can find [ wivcs ] @xmath157 these equations are helpful in order to unclutter the r.h.s of the ward identities . a useful consequence of the ward identities ( [ wics],[wisuns ] ) is the the identity : the above relationships are non - trivial , since @xmath159 @xmath160 $ ] . there is also another ward identity@xcite @xmath161 which follows from the galilean invariance . in this subsection we calculate the effective mass using the ward identities . the self - energy is defined such that @xmath162 from now on we assume @xmath163 in the above equation to be not the exact bare green s function of the non - interacting system , but that with the shifted chemical potential . we gauge the shift to be such that it removes all tadpole insertions to the green s functions in diagrammatics . the effective mass @xmath164 is given by the following equation @xmath165 since we are working with the finite - temperature matsubara green s functions and @xmath166 is a fermionic matsubara frequency , we understand @xmath167 in the sense that the zero - frequency limit will be taken after the appropriate analytical continuation of the final result for @xmath168 . in the second order of the perturbation theory only the `` sunrise '' self - energy diagram contributes to the mass renormalization . in this approximation @xmath169 \bigg\vert^{i \omega_1 \leadsto 0}_{{\bf k}_1 \in s_f } + { \cal o}(u^3)\ ] ] the charge ward identity can be written as @xmath170 while the galilean - invariance ward identity ( [ wigal ] ) reads @xmath171 combining the three above equations together , we get @xmath172 \big\vert^{i \omega_1 \leadsto 0}_{{\bf k}_1 \in s_f}\ ] ] since the above equation has the accuracy @xmath173 , we can use the one - loop approximation for the vertex , while the green s functions can be approximated by @xmath174 . for the the contact interaction ( [ sinthub ] ) ( @xmath44 ) the vertex s one - loop approximation ( [ gammapertexp ] ) gives @xmath175 it is easy to verify that only the third and the fourth terms in the above expression ( coming from zs and bcs graphs , resp . ) give non - zero contributions to @xmath164 . plugging in the formulas for the zs and bcs graph contributions and carrying out some manipulations , we obtain ( recall that @xmath176 ) @xmath177 note that without applying the ward identities , the above equation can be derived directly from eq . ( [ mstarappr ] ) where the self - energy is given by the sunrise diagram . however , it takes much more tedious and lengthy calculations to verify the cancellation of other terms appearing at intermediate steps of the analysis . before proceeding further with the calculations , let us first comment on result ( [ mswiexpl ] ) . in the the zero - temperature limit @xmath178 enforces @xmath87 to lie on the fermi surface , while the integration over the angle @xmath179 is equivalent to the calculation of the first fourier harmonic of the expression in the curly brackets . the latter we have calculated already , and it is the sum of the zs and bcs contributions to the fl vertex . ( recall that the first order term and the zs graph do not contribute to the first fourier component of the vertex . ) then eq . ( [ mswiexpl ] ) in the the zero - temperature limit readily recovers the standard result of the flt:@xcite @xmath180 a straightforward extension of eq . ( [ msflt ] ) to finite temperatures like @xmath181 is not valid , since according to eq . ( [ mswiexpl ] ) @xmath182 contains an extra contribution from the `` off - shell '' integration over @xmath183 normal to the fermi surface , albeit the factor @xmath184 makes this contribution well - localized near the fermi surface . in other words the vertex entering the r.h.s . ( [ mswiexpl ] ) is not exactly the fl vertex @xmath185 as it is defined in the flt ( cf . definitions in sec . [ vertex ] ) , since one of its momenta ( namely , @xmath87 ) is not confined to the fermi surface . the calculation of the temperature expansion of the effective mass @xmath164 is quite involved.@xcite we did not go beyond the linear - temperature terms . evaluation of the contributions of two graphs to @xmath164 denoted as @xmath186 gives [ mterms ] @xmath187 thus we get @xmath188 ( recall that dimensionless coupling and temperature are introduced by ( [ usm ] ) and ( [ t0 ] ) , resp . ) in a close analogy with the cancellation of the linear - temperature terms in the fourier components of the fl vertices @xmath136 and @xmath125 , here the cancellation occurs between additive linear-@xmath6 corrections coming from both `` on - shell '' ( i.e. , linear @xmath6-term coming from the @xmath189-contribution to the vertex ) and `` off - shell '' ( i.e. , small - momentum contribution ) integrations in two diagrams . moreover , the `` on - shell '' ( `` off - shell '' ) @xmath6-term of the zs graph cancels the `` on - shell '' ( `` off - shell '' ) @xmath6-term of the bcs graph , correspondingly . in this subection we present an alternative direct evaluation of the leading temperature correction . the advantage of this evaluation is that it applies also to non - galilean invariant situations and helps to more clearly establish the relation to previous studies . it also indicates possible generalizations . we begin from the explicit expression for the self - energy correction given by the `` sunrise '' diagram @xmath190 this choice of variables is convenient because the @xmath191linear contributions to the effective mass will be seen to arise either from regions where @xmath192 is small or from regions where it has magnitude @xmath193 . the bare electron green function @xmath126 is given by ( [ g0def ] ) . evaluation of the matusbara sums leads to @xmath194[b(\xi _ { q}-\xi _ { k})+n_{\circ}(\xi_{\bf p+q+k})]}{i\omega + \xi _ { q}-\xi _ { k}-\xi_{\bf p+k+q}}\ ] ] and by using @xmath195[b(\xi _ { q}-\xi _ { k})]=n_{\circ}(\xi _ { q})[1-n_{\circ}(\xi _ { k})]$ ] ( @xmath196 and @xmath197 are the fermi - dirac and bose - einstein distribution functions , resp . ) , and relabelling variables we obtain @xmath198 with @xmath199 now the first term in the expression for @xmath200 can not give rise to a term of order @xmath6 because the integral of @xmath201 is not confined to the region near the fermi surface . indeed , differentiating this term on temperature leads to a numerator containing the factor @xmath202 . because the @xmath201 integral is not confined to the region near the fermi surface , obtaining a nonzero value for the @xmath202 integral would require an additional factor of @xmath202 divided by a scale of the order of @xmath203 , meaning that the integral would be of order @xmath204 as in the usual sommerfeld expansion . we therefore focus on possible t - linear contributions from the second term . to obtain these we will make the assumption , to be justified a posteriori , that all energies @xmath205 are near zero . we parameterize the @xmath206 , @xmath207 integrals by @xmath208 and angles . we choose @xmath209 as the angle between @xmath210 and @xmath192 and @xmath211 as the angle between @xmath206 and @xmath207 . for a spherical fermi surface we have @xmath212 with , ( up to first order in deviations from the fermi surface ) @xmath213\cos ^{2}(\theta _ { 2}/2)+\xi _ { p } \\ b & = & [ 4e_{f}+2\xi _ { p}+\xi _ { q}+\xi _ { k}]\cos ( \theta _ { 2}/2)\end{aligned}\ ] ] we may now do the integral over @xmath209 obtaining @xmath214 with @xmath215}+\frac{sgn(c_{2})}{% % tcimacro{\func{re}}% % beginexpansion \mathop{\rm re}% % endexpansion [ \sqrt{c_{2}^{2}-b^{2}}]}\ ] ] and @xmath216\cos ^{2}(\theta _ { 2}/2 ) \\ c_{2 } & = & -\omega + \xi _ { q}+\xi _ { k}-\xi _ { p}-[4e_{f}+\xi _ { q}+\xi _ { k}]\cos ^{2}(\theta _ { 2}/2)\end{aligned}\ ] ] the crucial point is now that ( after analytic continuation in @xmath217 ) the region of phase space in which the square root is real is very small , and in particular corresponds to @xmath218 or @xmath219 , as well as to small values of the energies , thus justifying the assumption made above . let us consider these two regions separately . @xmath220 @xmath221 : these angles correspond to processes in which a small total momentum is transferred to the electron , i.e. to the processes which are usually considered@xcite to give rise to anomalous terms in @xmath222 , etc . in this regime we write @xmath223 ; to acount for @xmath224 , y may have either sign . to leading order we have @xmath225 note that if all the @xmath226 are small , then @xmath227 is confined to small angles , where the approximation is accurate . performing the angle integral gives @xmath228 with @xmath229 the remainder of the evaluation is simple : to obtain the quasiparticle velocity renormalization we must take the sum of the @xmath217 and @xmath230 derivatives ; thus the first term does not contribute at all while the second term gives @xmath231 . for the @xmath217 derivative and the same for the @xmath232 derivative . substituting this into the integrals and evaluating gives @xmath233 @xmath220 @xmath234 : these angles correspond to processes in which a momentum near @xmath193 is tranferred to the electrons . in this regime we have @xmath235 further , as both @xmath236 and @xmath137 are of order @xmath237 we write @xmath238 . we have @xmath239 again the @xmath227 integral may be done , yielding @xmath240 with @xmath241 as above , one of the two terms give no contribution ; the other has a factor of two but when differentiated gives only @xmath242 not @xmath243 . @xmath244the sign is opposite . thus this term cancels the previous term , leading to no t - linear term in the effective mass enhancement . an almost identical calculation of the order @xmath245 term in the free energy shows that there is no t - linear term in the specific heat coefficient . as we see from these calculations , the coefficients of the linear @xmath6-contributions to the effective mass coming from small and nearly @xmath246 processes are model - dependent . any model modifications , e.g. , momentum dependent interaction or non - circular fermi surface , changes these coefficients , eliminating the cancellation and resulting in apperance of the linear @xmath6-term in the effective mass . the same conclusion applies to the specific heat coefficient . this agrees with the recent experimental data on liquid @xmath247 films.@xcite in this section we consider the density - density correlation function @xmath248 , defined as follows : @xmath249 where the density operator @xmath250 is @xmath251 and @xmath252 stands for the density fluctuation . using the definition for the 1pi four - point vertex ( [ 2pv ] ) , one obtains @xmath253\ ] ] the correlation function has the properties [ kapprop ] @xmath254 to derive ( [ kapprop]a ) it is enough to re - write the charge ward identity as @xmath255 using the above expression for the first term in the brackets on the r.h.s . of eq . ( [ kapc ] ) , one can easily establish eq . ( [ kapprop]a ) . as to the second eq . ( [ kapprop]b ) , it can be proved with the use of eqs.([versym ] ) . in order to find the compressibility of the system we need to calculate the density function in the non - trivial ( static ) zero transfer limit , which we denote @xmath256 . to do it in the second order over interaction by a naive perturbation theory , one needs to take into account the sunrise diagram for the self - energy in the dyson equation ( [ sigma ] ) and to use such approximation for the total green s function entering the first term in the brackets on the r.h.s . ( [ kapc ] ) . for the second term in that equation one can use the `` bare '' green s functions [ cf . comment after eq . ( [ sigma ] ) ] and the one - loop approximation for the vertex . in diagrammatic language this amounts to the calculation of a set of three - loop diagrams , which is very difficult in practice . the first problem is a proliferation of terms after doing each intermidiate step of integration ( summation ) for a given three - loop diagram . the second is that some of these terms exhibit ir divergencies.@xcite it takes additional efforts to assure the final cancellation of those divergencies between different terms . as in the effective mass calculation , the use of the ward identities drastically simplifies the problem . comment after eq . ( [ mswiexpl ] ) ] . in applying the ward identities for our calculations we will use the approaches developed in the classic derivations of flt.@xcite we denote [ gsq ] @xmath257 let us also introduce two quantities @xmath258 and @xmath259 as @xmath260 @xmath261 then from eqs.([kapc],[daqo ] ) we have @xmath262 where @xmath263 eqs . ( [ kapc],[kapprop]a ) give @xmath264 subtracting this from eq . ( [ kapq1 ] ) and using ( [ dr ] ) , we obtain @xmath265 + \int_{({\bf 1},{\bf 2 } ) } \delta_r({\bf 1 } ) \gamma_{\rm ch}^{\omega}({\bf 1},{\bf 2 } ) \delta_r({\bf 2 } ) -\varkappa_2\ ] ] note that from the charge ward identity @xmath266 so far we did not use any approximations , and eq . ( [ kapq2 ] ) is exact . now let us simplify it at the level @xmath267 for the model with interaction ( [ sinthub ] ) @xmath268 . from the one - loop approximation for the vertex ( [ cala ] ) we see that @xmath269 also the green s functions in eq . ( [ kap2 ] ) can be approximated by their bare forms , resulting in @xmath270 after the standard analytical continuation to real frequencies , one can find for the parameter @xmath271 , @xmath272 @xmath273 where @xmath274 eq . ( [ deltarap ] ) is the extension to the finite - temperature of the well - known result @xmath275 ( cf . , e.g. , ref . [ ] ) , combined with expansion of @xmath276 to order @xmath277 . up to the terms @xmath278 we have from eqs . ( [ kapq2],[wisig],[kap2ap],[deltarap ] ) @xmath279 as we have shown before for the calculation of the effective mass , the first integral on the r.h.s . of eq . ( [ kapq3 ] ) can be written via the ward identities as the second term on the r.h.s . ( [ mstarwiap ] ) ( with the only difference that in eq . ( [ kapq3 ] ) @xmath280 is arbitrary . ) since two limits of the vertex differ only by a constant term ( [ k1l ] ) which disappears after the integration over the angle @xmath179 , we can finally write @xmath281 eqs . ( [ kapq4],[cala ] ) for the response function are manifestly free of spurious ir divergences and readily recover the standard flt results at zero temperature . we recall [ cf . ( [ a ] ) ] that the components of the scattering vertex and the landau function are just the appropriate limits of the vertex @xmath74 times the factor @xmath282 @xmath283 . the latter reduces simply to @xmath284 in our approximation . then at @xmath0 [ cf . [ efmass ] for the explanations of @xmath285 limit , and note also that from eq . ( [ f0f ] ) @xmath286 we read off from eq . ( [ kapq4 ] ) @xmath287 which is nothing but the expansion to @xmath278 of 2d flt result ( cf . , e.g. , ref . [ ] ) @xmath288 using eq . ( [ cala ] ) for the vertex entering the first integral term in eq . ( [ kapq4 ] ) , we obtain : @xmath289 where we indicated explicitly two contributions to the response function coming from the zs and bcs one - loop corrections of the vertex . for the second integral on the r.h.s . ( [ kapq4 ] ) we have @xmath290 where the first term is the bare vertex contribution to @xmath291 , while the two other one - loop vertex corrections are given by the same formulas as in eq . ( [ f1 ] ) with an obvious replacement @xmath292 . we were able to analytically calculate the above terms in the leading order of their temperature dependence.@xcite the results are [ kapcomp ] @xmath293 up to the order @xmath294 . note that the bcs contribution to the term @xmath295 is ultraviolet divergent and we regularized it in the same way as we did it for that contribution to the vertex , i.e. , by introducing an uv cutoff @xmath296 . [ scatlan ] . ] applying eq . ( [ wisiggal ] ) for the term @xmath297 while passing from eq . ( [ kapq3 ] ) to eq . ( [ kapq4 ] ) may raise some questions from a cautious reader , since the ultraviolet regularization we used to calculate the bcs contribution would break the galilean invariance , which is indispensable for the derivation of ( [ wisiggal ] ) . however @xmath297 is uv - convergent , thus can be calculated in the limit of an infinite cutoff when the galilean invariance is preserved . combining the above results together , we obtain @xmath298 once again , the leading linear @xmath6-corrections which can be traced back to the zs- and bcs - loop contributions to the vertex cancel in each of the terms @xmath295 and @xmath297 separately . it is useful to keep in mind that albeit we expressed @xmath297 and @xmath295 in terms of the vertex only , these contributions entangle both `` purely vertex '' corrections and self - energy corrections to the response function . the latter are just expressed in terms of the vertex via the ward identities . let us finally remind@xcite that the compressibility @xmath299 , where @xmath300 is electron density . to find the spin susceptibility in the @xmath40 formalism , we consider the spin ( flavor , if @xmath36 ) correlation function @xmath301 , defined as follows : @xmath302 where the flavor density operators @xmath303 and @xmath304 is the gyromagnetic ratio . using eqs . ( [ 2pv],[wivcs ] ) one can find @xmath305\ ] ] note that in a paramagnetic state the response is the same along all the @xmath306 directions @xmath307 . then we follow the same steps as in the previous section , with the only difference that instead of the charge ward identity we use the spin identity ( [ wisuns ] ) . the latter can be casted in the form of eq . ( [ wi ] ) where @xmath308 . function @xmath309 has the same properties ( [ kapprop ] ) as those of the density function . the uniform susceptibility is given by the static limit @xmath310 ( no sum over @xmath307 ) . the equation we obtain for @xmath311 is the same as the r.h.s of eq . ( [ kapq2 ] ) , but @xmath312 . then we use the consequence of the both spin and charge ward identities [ cf . ( [ wiscr ] ) ] : @xmath313 and afterwards proceed as in the previous section . from the one - loop approximation ( [ gammapertexp ] ) @xmath314 thus up to the terms @xmath278 we obtain @xmath315 recalling once again what we said after eq . ( [ kapq4 ] ) and noting that from eq . ( [ g0f ] ) @xmath316 , we easily see that in the zero - temperature limit the above equation gives @xmath317 reproducing thus the second - order expansion of the 2d flt result@xcite @xmath318 luckily we do not have to do new calculations since @xmath319 so we find the spin susceptibility to have a linear leading temperature correction @xmath320 in this paper we have systematically examined the leading temperature corrections to flt in two spatial dimensions . we find for the model of a 2d electron gas with a contact interaction that to order @xmath277 the leading @xmath6-dependence of the fl parameters in the spin sector is linear in temperature , while for the parameters in the charge sector and for the effective mass a cancellation of the leading @xmath6-corrections occurs , and their expansions start from the terms quadratic in temperature . the particularly interesting result we found is the leading linear temperature dependence of the spin susceptibility @xmath321 according to the perturbative calculations of belitz _ et al_,@xcite the 2d fl susceptibility has a leading linear correction in @xmath322 at @xmath0 with a positive coefficient which is of the second order in interaction , i.e. , their result has a structure of eq . ( [ sus ] ) . this also agrees with the phenomenology of misawa@xcite and the numerical results.@xcite our results reveal the crucial importance of @xmath189 processes contributions into the low - energy parameters . we remind that recovering of the angular dependence of the fl vertices in the whole region where the angle varies , involves taking into account transfers varying from zero to @xmath189 . we have seen it from the direct bubbles evaluation , but the argument is non - perturbative . to appreciate this let us write the antisymmetry condition for the vertex as [ cross ] @xmath323 \mbox{\tiny a } \end{array } $ } \!\ ! \ } } ( { \bf k}_1,{\bf k}_2;{\bf q})= \pm \gamma^{\{\!\ ! \mbox{$ \begin{array}{c } \mbox{\tiny s } \\[-0.2 cm ] \mbox{\tiny a } \end{array } $ } \!\ ! \ } } ( { \bf k}_1,{\bf k}_2;{\bf k}_2-{\bf k}_1-{\bf q } ) \nonumber\end{aligned}\ ] ] so , small- and large - transfer scales are intrinsically related due to the pauli principle , and any calculation of the low - energy corrections should take this into account . eventually the integration over whole range of the transfer values through a given loop results in the temperature corrections not only to the vertex , but to the response functions as well , when that loop is a part of a more complex diagram . the big transfer ( @xmath1 ) contributions essentially gave rise to the spin susceptibility result ( [ sus ] ) in the three - loop approximation . we find that the relationships known from the classical flt derivations at @xmath0 for the parameters of galilean - invariant fl ( e.g , the effective mass , response functions vz components of the landau function ) are violated by finite - temperature terms . the coefficients in the temperature corrections to these relationships subtly involve contrubutions from small and large ( @xmath1 ) momentum processes . concerning the argument for the cancellation of anomalous terms in the response functions due to ward identities : we have calculated the vertices at the one - loop level @xmath324 . through the ward identities the self - energy corrections were taken into account with the same accuracy . there are no more terms of the order @xmath324 to cancel the temperature dependence ( [ sus ] ) . thus , the linear temperature dependence of susceptibility ( or weaker temperature dependence of the compressibility ) does not contradict the _ ward identities known to us , moreover in our results for the response functions both vertex and self - energy corrections are included on the same footing _ by using the ward identities_. for more realistic models of electrons in ( quasi)-2d crystals , i.e. , for various tight - binding spectra and fillings , the free - gas - like square - root @xmath189 singularities ( with @xmath325 depending on a chosen direction in @xmath206-space ) are known to exist in the lindhard functions.@xcite we think this is enough to result in linear @xmath6-terms in physical quantities analogous to what we found in this study . we argue that the cancellation of the @xmath6-terms in some fl parameters is special to second order perturbation theory and the model considered , while the leading linear temperature corrections are a generic feature of the 2d fl . we hope our results may be experimentally tested in real 2d fl systems . for example , a very naive fit of the temperature dependence of the spin susceptibility in @xmath326 system@xcite when it is in the 2d metallic regime ( above 3d crossover temperature ) shows that the data are compatible with the form ( [ sus ] ) . we expect our results stimulate a more detailed examination of the leading temperature dependences of response functions in 2d systems . stimulating conversations with s. lukyanov are gratefully acknowledged . we thank v. oudovenko for his help with the numerical tests of our results . this work is supported by nsf dmr0081075 . we define the lindhard function as @xmath327 it corresponds to the contribution of the particle - hole loop in graph zs or zs , depending on the actual value of @xmath328 we put . _ for the purpose of notational economy in the appendices we will reserve capital letters for momentum vectors , while small letters correspond to their dimensionless counterparts defined below . _ the free electron green s function entering the loop is given by @xmath329 with its spectrum @xmath330 a straightforward summation over matsubara frequencies results in @xmath331 we denote by @xmath196 the fermi - dirac distribution function @xmath332 we choose 2d polar axis along the transfer vector such that @xmath333 . then the lindhard function can be transformed to the expression @xmath334 the angular integration above is understood in the sense of the principal value . we define the dimensionless lindhard function @xmath141 at zero frequency transfer as @xmath335 where @xmath336 is the free 2d electron density of states per spin . let us also introduce the new ( dimensionless ) variables @xmath337 in order to proceed we use the following result for the principal value of the integral @xmath338 at @xmath339 : @xmath340 where @xmath341 is the event ( heaviside ) function . after the angular integration we obtain @xmath342 the above formula can be easily casted into the final form @xmath343 to make zero - temperature check of our result the following formula is useful @xmath344 then from eq . ( [ chi8 ] ) we easily recover the result known for @xmath0 , which was first found by stern:@xcite [ lt0 ] @xmath345 let us consider now the lindhard function at the `` effective '' transfer @xmath346 which occurs in the zs loop when the `` real '' transfer in the vertex is @xmath347 . parameterizing vectors @xmath348 and @xmath349 by their angles ( @xmath350 , @xmath351 ) relatively to the direction of vector @xmath347 , we get @xmath352 in terms of new angles @xmath353 and of the dimensionless variables ( [ dmlsvar ] ) one reads @xmath354 let us consider now @xmath138 at zero vertex transfer @xmath355 , i.e. , @xmath356 . for the fourier transform @xmath357 of @xmath358 given by eq.([chi8 ] ) , we have @xmath359 where @xmath360 then @xmath361 using the result @xmath362 we get @xmath363 thus @xmath364 the dimensionless temperature @xmath365 is @xmath366 taylor - expanding the square root under the integral , we can then extend the upper limit to the infinity with exponential accuracy . thus , @xmath367 after simple integration we arrive to the sought result : @xmath368 for the next fourier component using @xmath369 we obtain @xmath370 thus , @xmath371 and combining the last formula with the previous results , we have @xmath372 we define the contribution of the particle - particle loop in the bcs graph as @xmath373 where @xmath374 after summation of the matzubara frequencies we get @xmath375 for the case we are interested in ( i.e. , zero sum frequency @xmath376 and vectors @xmath377 , @xmath378 lying on the fermi surface , cf . also notations ( [ qprime ] , [ angles ] ) ) we have @xmath379 to regularize the momentum integral in ( [ cooper2 ] ) we will introduce an ultraviolet cutoff @xmath380 which we assume to be the largest momentum scale of the problem . under this condition we shift the momentum integration in the second term of the sum on the r.h.s . ( [ cooper2 ] ) , obtaining the expression @xmath381 two - dimensional integration in ( [ coopsh ] ) reads as : @xmath382 using the previously defined dimensionless parameters [ cf . notations ( [ dmlsvar ] ) ] along with new variables @xmath383 and the dimensionless function @xmath145 [ cf . notations ( [ lchi ] ) ] @xmath384 one gets @xmath385 or @xmath386 we denote @xmath338 according to eq.([ia ] ) with @xmath387 note that @xmath388 after some simple manipulations we get @xmath389 the square root is chosen to be positive on the real axis of integration @xmath390 . recalling that @xmath391 and , consequently , @xmath392 we see that the sign- and event functions play their role on the patch @xmath393 $ ] only . due to the symmetry properties of function @xmath394 it suffices to consider the angle range @xmath395 . after some analysis one can obtain @xmath396 the function @xmath397 is continuous . it can be easily calculated at zero temperature , giving the known result ( see . e.g. , ref . [ ] ) : @xmath398 as well as for zero total incoming momentum ( @xmath399 ) at low ( @xmath400 ) temperature , giving with exponential precision @xmath401 where @xmath402 is euler s constant . for the latter calculation cf . , e.g. , section 33.3 of ref . [ ] . in the above formulas and in what follows we track @xmath150 up to terms @xmath403 . from eq . ( [ cqf ] ) it is easy to calculate fourier components of function @xmath397 given by @xmath404 for the zeroth component after some manipulations we have @xmath405 the first term can be written as @xmath406 where @xmath407 is the complete elliptic integral of the first kind , which has the following taylor expansion : @xmath408 to proceed further we notice that @xmath409 \equiv { \bbb f}(z)\ ] ] where the taylor expansion of the hypergeometric function can be readily read from the term by term integration of the expansion ( [ elltaylor ] ) , i.e. , @xmath410 integrating by parts the r.h.s . ( [ c01 ] ) we obtain @xmath411 using the same ( exponentially accurate at low temperature ) approximations which lead us to eq . ( [ chi0tay ] ) , we end up @xmath412 in order to calculate @xmath413 it is convenient to come back to the integral representation of the elliptic integral , and then to evaluate it via changing the order of integration . @xmath414 here @xmath415 is catalan s constant . finally @xmath416 for the second contribution @xmath417 one integration by parts gives @xmath418 it is easy to see that the second term on the r.h.s of eq . ( [ c02bp ] ) is exponentially small , while the first one can be readily evaluated resulting in @xmath419 combining our results together , we have @xmath420 calculation of the first fourier component @xmath421 is very simple , since the first term on the r.h.s . of eq . ( [ cqf ] ) gives zero after the angular integration , while the second one can be evaluated exactly , resulting in @xmath422 for a recent comprehensive review and references see , e.g. , t. timusk and b. statt , rep . . phys . * 62 * , 61 ( 1999 ) . r. shankar , rev . phys . * 66 * , 129 ( 1994 ) . chitov and d. snchal , phys . b * 52 * , 13487 ( 1995 ) . a. houghton , h .- j . kwon , and j.b . marston , adv . phys * 49 * , 141 ( 2000 ) . w. metzner , c. castellani , and c. di castro , adv . phys * 47 * , 317 ( 1998 ) . m. disertori and v. rivasseau , phys . lett . * 85 * , 361 ( 2000 ) . landau , sov . jetp * 3 * , 920 ( 1957 ) ; * 5 * , 101 ( 1957 ) ; * 8 * , 70 ( 1959 ) . a.a . abrikosov , l.p . gorkov , and i.e. dzyaloshinski , _ methods of quantum field theory in statistical physics _ ( dover , new york , 1963 ) . e.m . lifshitz , l.p . pitayevskii , _ statistical physics ii _ ( pergamon press , oxford , 1980 ) . g.y . chitov and d. snchal , phys . b * 57 * , 1444 ( 1998 ) . hertz , phys . b * 14 * , 1165 ( 1976 ) ; a.j . millis , phys . b * 48 * , 7183 ( 1993 ) . h. von lhneysen , _ et al _ , phys . lett . * 72 * , 3262 ( 1994 ) . et al _ , jmmm * 177 - 181 * , 265 ( 1998 ) . a. schrder , g. aeppli , e. bucher , r. ramazashvili , and p. coleman , phys . * 80 * , 5623 ( 1998 ) . q. si , s. rabello , k. ingersent , and j.l . smith , cond - mat/0011477 . s. doniach and s. engelsberg , phys . * 17 * , 750 ( 1966 ) ; d.j . amit , j.w . kane , and h. wagner , phys . * 19 * , 425 ( 1967 ) . d. coffey and k.s . bedell , phys . lett . * 71 * , 1043 ( 1993 ) . s. misawa , j. phys . . jpn . * 68 * , 2172 ( 1999 ) ; preprint , to appear in physica b. g.m . carneiro and c.j . pethick , phys . b @xmath423 , 1933 ( 1977 ) . d. belitz , t.r . kirkpatrick , and t. vojta , phys . b @xmath424 , 9452 ( 1997 ) . i.e. dzyaloshinskii and a.i . larkin , sov . jetp * 38 * , 202 ( 1974 ) . haldane , in _ proceedings of the international school of physics `` enrico fermi '' , 1992 _ , ed . r.a . broglia and j.r . schrieffer ( north holland , amsterdam , 1994 ) . i.e. dzyaloshinskii and a.i . larkin , sov . jetp * 34 * , 422 ( 1972 ) ; recently the temperature dependence of the luttinger liquid due to marinally irrelevant terms was studied by h. nlisse , c. bourbonnais , h. touchette , y.m . vilk , and a .- m.s . tremblay , eur . j. b @xmath425 , 351 ( 1999 ) . analogous effects occur in spin-@xmath26 heisenberg chain : see s. lukyanov , nucl . b * 522 * [ fs ] , 533 ( 1998 ) . a more detailed presentation of those results was given in : s. misawa , physica b * 149 * , 162 ( 1988 ) ; s. misawa , t. tanaka and k. tsuru , europhys . * 14 * , 377 ( 1991 ) . see references on the previous work therein . f. stern , phys . * 44 * , 1469 ( 1980 ) . d. belitz , t.r . kirkpatrick , a.j . millis , and t. vojta , phys . b * 58 * , 14155 ( 1998 ) . d.s . hirashima and h. takahashi , j. phys . soc . jpn . * 67 * , 3816 ( 1998 ) . d.s . hirashima and k. kubo , j. phys . . jpn . * 68 * , 2174 ( 1999 ) . chitov and a.j . millis , cond - mat/0101399 . j.w . negele and h. orland , _ quantum many - particle systems _ ( addison - wesley , new york , 1988 ) . it should pointed out that we are not saying that in the case when the running vertices are not constants , but functions of the momenta , cancellations between different loop contributions could not take place . for example , as one can easily check from fig.1 , when @xmath426 momentum - antisymmetric parts of the direct ( zs ) and the exchange ( zs ) graphs cancel each other for any running coupling function which has the antisymmetric properties ( [ pots],[usym ] ) , thus preserving at the one - loop level the property @xmath427 following from the pauli principle . ( for more on this issue see ref . what we are saying is that even at special configurations of the vertex momenta , ( i.e. , @xmath428 , @xmath429 ) whence the loops temperature dependence comes mostly , we can not see analogous `` algebraic '' cancellations between zs abd bcs contributions for generic coupling function satisfying ( [ pots],[usym ] ) . w. kohn and j.m . luttinger , phys . 15 * , 524 ( 1965 ) . j. zinn - justin , _ quantum field theory and critical phenomena _ , third edition , ( clarendon press , oxford , 1999 ) . we will not present details of these calculations , since they are too technical . the reader may work them out along the lines of the simpler analytical procedure described in the appendices for the one - loop vertex corrections . the appearance of the ir divergencies in the multiple loops while calculating , e.g. , response functions of 2d fermions is a known fact . this problem was addressed recently by a. neumayr and w. metzner , phys . b , * 58 * , 15 449 ( 1998 ) . see also more references therein . p. bnard , l. cheng , and a .- m . s. tremblay , phys . b * 47 * , 15217 ( 1993 ) . y. maeno , _ et al _ , j. phys . 66 * , 1405 ( 1997 ) . f. stern , phys . lett . * 18 * , 546 ( 1967 ) . h. fukuyama , y. hasegawa , and o. narikiyo , j. phys . . jpn . * 60 * , 2013 ( 1991 ) .

the problem is to decide , given a set of real numbers , whether any three sum to zero . it is widely conjectured that a trivial-time algorithm is optimal and over the years the consequences of this conjecture have been revealed . this conjecture implies lower bounds on numerous problems in computational geometry and a variant of the conjecture implies strong lower bounds on triangle enumeration , dynamic graph algorithms , and string matching data structures . in this paper we refute the conjecture . we prove that the decision tree complexity of is and give two subquadratic algorithms , a deterministic one running in time and a randomized one running in time with high probability . our results lead directly to improved bounds for-variate linear degeneracy testing for all odd . the problem is to decide , given a linear function and a set , whether . we show the decision tree complexity of this problem is . finally , we give a subcubic algorithm for a generalization of the-product over real - valued matrices and apply it to the problem of finding zero - weight triangles in weighted graphs . we give a depth- decision tree for this problem , as well as an algorithm running in time .

1404.0799

the time hierarchy theorem @xcite implies that there exist problems in @xmath15 with complexity @xmath16 for every fixed @xmath6 . however , it is consistent with current knowledge that all problems of practical interest can be solved in @xmath17 time in a reasonable model of computation . efforts to build a useful complexity theory inside @xmath15 have been based on the conjectured hardness of certain archetypal problems , such as , @xmath12-matrix product , and cnf - sat . see , for example , the conditional lower bounds in @xcite . in this paper we study the complexity of and related problems such as linear degeneracy testing ( ) and finding zero - weight triangles . let us define the problems formally . * :* : : given a set @xmath9 , determine if there exists @xmath18 such that @xmath19 . * :* : : given a set @xmath20 , determine if there exists @xmath18 such that @xmath19 . * @xmath6- and @xmath6- : * : : fix a @xmath6-variate linear function @xmath21 , where @xmath22 . given a set @xmath9 , determine if @xmath23 for any @xmath24 . when @xmath25 is @xmath26 the problem is called @xmath6- . * :* : : given a weighted undirected graph @xmath27 , where @xmath28 , determine if there exists a triangle @xmath29 for which @xmath30 . ( from the definition of _ love : a score of zero _ , one could also call this the problem . ) these problems are often defined with further constraints that do not change the problem in any substantive way @xcite . for example , the input to can be three sets @xmath31 and the problem is to determine if there exists @xmath32 such that @xmath19 . even if there is only one set , there is sometimes an additional constraint that @xmath33 and @xmath34 be _ distinct _ elements . as a problem in its own right , has no compelling practical applications . however , lower bounds on imply lower bounds on dozens of other problems that _ are _ of practical interest . before reviewing existing algorithms we give a brief survey of conditional lower bounds that depend on the hardness of . it is often conjectured that requires @xmath2 time and that requires @xmath35 time @xcite . these conjectures have been shown to imply strong lower bounds on numerous problems in computational geometry @xcite dynamic graph algorithms @xcite , and pattern matching @xcite . for example , the conjecture implies that the following problems require at least @xmath2 time . 1 . given an @xmath0-point set in @xmath36 , determine whether it contains three collinear points @xcite . 2 . given two @xmath0-edge convex polygons , determine whether one can be placed inside the other via rotation and translation @xcite . 3 . given @xmath0 triangles in @xmath36 , determine whether their union contains a hole , or determine the area of their union @xcite . through a series of reductions , ptracu @xcite proved that the conjecture implies lower bounds on triangle enumeration and various problems in dynamic data structures , even when all updates and queries are presented in advance . some lower bounds implied by the conjecture include the following . 1 . given an undirected @xmath37-edge graph , enumerating up to @xmath37 triangles ( 3-cycles ) requires at least @xmath38 time @xcite . is optimal if the matrix multiplication exponent @xmath39 is 2 . ] 2 . given a sequence of @xmath37 updates to a directed graph ( edge insertions and edge deletions ) and two specified vertices @xmath40 , determining whether @xmath41 is reachable from @xmath42 after each update requires at least @xmath38 time @xcite . 3 . given an edge - weighted undirected graph , deciding whether there exists a zero - weight triangle requires at least @xmath43 time @xcite . in recent years conditional lower bounds have been obtained from two other plausible conjectures : that computing the @xmath12-product of two @xmath44 matrices takes @xmath43 time and that cnf - sat takes @xmath45 time . the latter is sometimes called the strong exponential time hypothesis ( strong eth ) . we now know that if the strong eth holds , no @xmath46 algorithm exists for @xmath6- @xcite and no @xmath47 algorithm exists for @xmath48-approximating the diameter of an @xmath37-edge unweighted graph @xcite . williams and williams @xcite proved that numerous problems are equivalent to @xmath12-matrix multiplication , inasmuch as a truly subcubic ( @xmath49 ) algorithm for one would imply truly subcubic algorithms for all the others . the evidence in favor of the and conjectures is rather thin . erickson @xcite and ailon and chazelle @xcite proved that any @xmath6-linear decision tree for solving @xmath6- must have depth @xmath50 when @xmath6 is even and @xmath51 when @xmath6 is odd . in particular , any 3-linear decision tree for has depth @xmath2 . ( an @xmath42-linear decision tree is one where each internal node asks for the sign of a linear expression in @xmath42 elements . ) the problem is obviously not harder than , but no other relationship between these two problems is known . indeed , the assumption that elements are integers opens the door to a variety of algorithmic techniques that can not be modeled as decision trees . using the fast fourier transform it is possible to solve in @xmath52 time , which is subquadratic even for a rather large universe size @xmath53 . baran , demaine , and ptracu @xcite showed that can be solved in @xmath54 time ( with high probability ) on the word ram , where @xmath55 and @xmath56 is the machine word size . the algorithm uses a mixture of randomized universe reduction ( via hashing ) , word packing , and table lookups . it is straightforward to reduce @xmath6- to a problem or unbalanced problem , depending on whether @xmath6 is even or odd . when @xmath6 is odd one forms certain sets @xmath57 where @xmath58 and @xmath59 , then sorts them in @xmath60 time . the standard three - set algorithm on @xmath57 takes @xmath61 time . when @xmath62 is even there is no set @xmath63 . using lambert s algorithm @xcite , @xmath64 and @xmath65 can be sorted is @xmath66 time while performing only @xmath67 comparisons . these algorithms can be modeled as @xmath6-linear decision trees , and are therefore optimal in this model by the lower bounds of @xcite . however , it was known that all @xmath6- problems can be solved by @xmath0-linear decision trees with depth @xmath68 @xcite , or with depth @xmath69 if the coefficients of the linear function are integers with absolute value at most @xmath70 @xcite . unfortunately these decision trees are not efficiently constructible . the time required to determine _ which _ comparisons to make is exponential in @xmath0 . the problem was highlighted in a recent article by williams and williams @xcite . they did not give any subcubic algorithm , but did show that a subcubic algorithm would have implications for via an intermediate problem called . * :* : : given a vector @xmath71 , determine if there exist @xmath72 for which @xmath73 . * :* : : the same as , except that @xmath74 and @xmath75 , where @xmath76 is the machine word size . ptracu @xcite defined the problem and gave a randomized reduction from to . williams and williams @xcite gave a reduction from to . neither of these reductions is frictionless . define @xmath77 and @xmath78 to be the complexities of the various problems on inputs of length @xmath0 , or graphs with @xmath0 vertices . clearly @xmath79 . the reductions show that for any @xmath6 , @xmath80 and @xmath81 . note that even if had an @xmath1-time algorithm ( optimal on dense graphs ) , this would only give an @xmath82 bound for . we give the first subquadratic bounds on both the decision tree complexity of and the algorithmic complexity of , which also gives the first deterministic subquadratic algorithm for .-bit word ram with the standard repertoire of unit - time @xmath83 operations : bitwise boolean operations , left and right shifts , addition , and comparison . ] our method leads to similar improvements to the decision tree complexity of @xmath6-when @xmath7 is odd . refer to figure [ fig : results ] for a summary of prior work and our results . [ thm:3sum ] there is a 4-linear decision tree for with depth @xmath3 . furthermore , can be solved deterministically in @xmath4 time and , using randomization , in @xmath84 time with high probability . [ thm : kldt ] when @xmath7 is odd , there is a @xmath85-linear decision tree for @xmath6- with depth @xmath11 , theorem [ thm:3sum ] refutes the conjecture and casts serious doubts on the optimality of many @xmath1 algorithms in computational geometry . theorem [ thm:3sum ] also answers a question of erickson @xcite and ailon and chazelle @xcite about whether @xmath86-linear decision trees are more powerful than @xmath6-linear decision trees in solving @xmath6- problems . in the case of @xmath87 , they are . we define a new product of three real - valued matrices called _ target - min - plus _ , which is trivially computable in @xmath88 time . we observe that is reducible to a target - min - plus product , then give subcubic bounds on the decision tree and algorithmic complexity of target - min - plus . theorem [ thm : zerotriangle ] is an immediate consequence . [ thm : zerotriangle ] the decision tree complexity of is @xmath13 on @xmath0-vertex graphs and its randomized decision tree complexity is @xmath89 with high probability . there is a deterministic algorithm running in @xmath90 time and a randomized algorithm running in @xmath91 time with high probability . any @xmath37-edge graph contains @xmath92 triangles which can be enumerated in @xmath92 time , so can clearly be solved in @xmath92 time as well . we improve this bound for all @xmath37 . [ thm : sparse - zerotriangle ] the decision tree complexity of on @xmath37-edge graphs is @xmath93 and , using randomization , @xmath94 with high probability . the problem can be solved in @xmath95 time deterministically or @xmath96 with high probability . by invoking the williams - williams reduction @xcite , our algorithms give subquadratic bounds on the complexity of . by designing algorithms from scratch we can obtain speedups comparable to those of theorem [ thm : zerotriangle ] . [ thm : convolutionthreesum ] the decision tree complexity of is @xmath3 and its randomized decision tree complexity is @xmath97 with high probability . the problem can be solved in @xmath5 time deterministically , or in @xmath98 time with high probability . all of our algorithms borrow liberally from fredman s 1976 articles on the decision tree complexity of @xmath12-matrix multiplication @xcite and the complexity of sorting @xmath99 @xcite . throughout the paper we shall refer to the ingenious observation that @xmath100 iff @xmath101 as _ fredman s trick_. in order to shave off @xmath102 factors in runtime we apply the geometric domination technique invented by chan @xcite and developed further by bremner , chan , demaine , erickson , hurtado , iacono , langerman , ptracu , and taslakian @xcite . in section [ sect : usefullemmas ] we review a number of useful lemmas due to fredman @xcite , buck @xcite , and chan @xcite about sorting with partial information , the complexity of hyperplane arrangements , and the complexity of dominance reporting in @xmath103 . in section [ sect : quadratic3sum ] we review a standard @xmath1-time algorithm and in section [ sect : nonuniform ] we present an @xmath104-depth decision tree for . subquadratic algorithms for are presented in section [ sect : uniform ] . section [ sect : lineardegeneracy ] presents new bounds on the decision tree complexity of @xmath6- for odd @xmath7 . section [ sect : zerotriangle ] presents new bounds on the decision tree and algorithmic complexity of and . section [ sect : conclusion ] concludes with some open problems . fredman @xcite considered the problem of sorting a list of @xmath0 numbers known to be arranged in one of @xmath105 permutations . when @xmath106 is sufficiently small the list can be sorted using a linear number of comparisons . [ lem : fredman ] ( fredman 1976 @xcite ) a list of @xmath0 numbers whose sorted order is one of @xmath106 permutations can be sorted with @xmath107 pairwise comparisons . throughout the paper @xmath108 $ ] denotes the first @xmath109 natural numbers @xmath110 , where @xmath111 may or may not be an integer . we apply lemma [ lem : fredman ] to the problem of sorting _ cartesian sums_. given lists @xmath112}$ ] and @xmath113}$ ] of distinct numbers , define @xmath114\}$ ] . we often regard @xmath115 as an @xmath116 matrix ( which may contain multiple copies of the same number ) or as a point in the @xmath117-dimensional space @xmath118 , whose coordinates are named @xmath119 . the points in @xmath118 that agree with a fixed permutation of @xmath115 form a convex cone bounded by the @xmath120 hyperplanes @xmath121\}$ ] . the sorted order of @xmath115 is encoded as a sign vector @xmath122 depending on whether @xmath123 lies on , above , or below a particular hyperplane in @xmath124 . therefore the number of possible sorted orders of @xmath115 is exactly the number of regions ( of all dimensions ) defined by the arrangement @xmath124 . ( regions of dimension less than @xmath117 correspond to instances in which some numbers appear multiple times . ) [ lem : buck ] ( buck 1943 @xcite ) consider the partition of space defined by an arrangement of @xmath37 hyperplanes in @xmath103 . the number of regions of dimension @xmath125 is at most @xmath126 and the number of regions of all dimensions is @xmath127 . in one of our algorithms we will construct the hyperplane arrangement explicitly . edelsbrunner , orourke , and seidel @xcite proved that the natural incremental algorithm takes @xmath127 time ( linear in the size of the arrangement ) , but any trivial @xmath128-time algorithm suffices in our application . the hyperplane arrangements we use correspond to fragments of the cartesian sum @xmath115 . lemma [ lem : sortx+y ] is a direct consequence of lemmas [ lem : fredman ] and [ lem : buck ] . [ lem : sortx+y ] let @xmath129}$ ] and @xmath113}$ ] be two lists of numbers and let @xmath130 ^ 2 $ ] be a set of positions in the @xmath44 grid . the number of realizable orders of @xmath131 is @xmath132 and therefore @xmath133 can be sorted with at most @xmath134 comparisons . it is sometimes convenient to assume that the elements of a cartesian sum are distinct ( and therefore have exactly one sorted order ) , even though numbers may appear multiple times . lemma [ lem : tiebreaking ] illustrates one way to break ties consistently . the proof is straightforward . [ lem : tiebreaking ] let @xmath135 and @xmath136 be two lists of numbers . define @xmath137 and @xmath138 . the cartesian sum @xmath139 is totally ordered , and is a linear extension of the partially ordered @xmath115 . ( addition over tuples is pointwise addition ; tuples are ordered lexicographically . the tuple @xmath140 can be regarded as a representation of a real number @xmath141 where @xmath142 are sufficiently small so as not to invert strictly ordered elements of @xmath115 . ) given a set @xmath143 of red and blue points in @xmath103 , the _ bichromatic dominating pairs _ problem is to enumerate every pair @xmath144 such that @xmath145 is red , @xmath146 is blue , and @xmath145 is greater than @xmath146 at each of the @xmath147 coordinates . a natural divide and conquer algorithm @xcite runs in time linear in the output size and @xmath148 . chan @xcite provided an improved analysis when @xmath147 is logarithmic in @xmath0 . for the sake of completeness we give a short proof of lemma [ lem : redblue ] in appendix [ sect : redblue ] . [ lem : redblue ] * ( bichromatic dominance reporting @xcite ) * given a set @xmath149 of red and blue points , it is possible to return all bichromatic dominating pairs @xmath144 in time linear in the output size and @xmath150 . here @xmath151 is arbitrary and @xmath152 . we typically invoke lemma [ lem : redblue ] with @xmath153 and @xmath154 , where @xmath155 is sufficiently small to make the running time subquadratic , excluding the time allotted to reporting the output . we shall review a standard @xmath1 algorithm for the three - set version of and introduce some terminology used in sections [ sect : nonuniform ] and [ sect : uniform ] . we are given sets @xmath156 and must determine if there exists @xmath32 such that @xmath19 . for each @xmath157 the algorithm searches for @xmath158 in the cartesian sum @xmath115 . each search takes @xmath159 time , for a total of @xmath160 . we view @xmath115 as being a matrix whose rows correspond to @xmath64 and columns correspond to @xmath65 , both listed in increasing order . 1.@xmath161 : : sort @xmath64 and @xmath65 in increasing order as @xmath162 and @xmath163 . 2.@xmath161 : : for each @xmath157 , 2.1.@xmath164 : : initialize @xmath165 and @xmath166 . 2.2.@xmath164 : : repeat : 2.2.1.@xmath167 : : if @xmath168 , report witness `` @xmath169 '' 2.2.2.@xmath167 : : if @xmath170 then decrement @xmath171 , otherwise increment @xmath172 . 2.3.@xmath173 : : until @xmath174 or @xmath175 . 3.@xmath161 : : if no witnesses were found report `` no witness . '' note that when a witness is discovered in step 2.2.1 the algorithm continues to search for more witnesses involving @xmath34 . since the elements in each row and each column of the @xmath115 matrix are distinct , it does not matter whether we increment @xmath172 or decrement @xmath171 after finding a witness . we choose to increment @xmath172 in such situations ; this choice is reflected in lemma [ lem : contour ] and its applications in sections [ sect : rand - param ] and [ sect : det - param ] . define the _ contour _ of @xmath176 , @xmath177 , to be the sequence of positions @xmath178 encountered while searching for @xmath176 in @xmath115 . when @xmath115 is understood from context we will write it as @xmath179 . if @xmath115 is viewed as a topographic map , with the lowest point in the nw corner and highest point in the se corner , @xmath179 represents the path taken by an agent attempting to stay as close to altitude @xmath176 as possible , starting in the ne corner ( at position @xmath180 ) and ending when it falls off the western or southern side of the map . lemma [ lem : contour ] is straightforward . [ lem : contour ] every occurrence of @xmath176 in @xmath115 lies on @xmath179 . let @xmath181 be any element of @xmath115 . then @xmath182 iff either @xmath183 lies strictly below @xmath179 or both @xmath183 and @xmath184 lie on @xmath179 . similarly , @xmath185 iff either @xmath183 lies strictly above @xmath179 or both @xmath183 and @xmath186 lie on @xmath179 . recall that we are given a set @xmath187 of reals and must determine if there exist @xmath18 summing to zero . we first state the algorithm , then establish its correctness and efficiency . 1.@xmath161 : : sort @xmath64 in increasing order as @xmath188 . partition @xmath64 into @xmath189 groups @xmath190 of size at most @xmath191 , where @xmath192 and @xmath193 may be smaller . the first and last elements of @xmath194 are @xmath195 and @xmath196 . 2.@xmath161 : : sort @xmath197 } \left(a_i - a_i\right ) = \{a - a ' \;|\ ; a , a ' \in a_i \mbox { for some } i\}$ ] . 3.@xmath161 : : for all @xmath198 $ ] , sort @xmath199 . 4.@xmath161 : : for @xmath6 from 1 to @xmath0 , 4.1.@xmath164 : : initialize @xmath165 and @xmath200 to be the group index of @xmath201 . 4.2.@xmath164 : : repeat : 4.2.1.@xmath167 : : if @xmath202 , report `` solution found '' and halt . 4.2.2.@xmath167 : : if @xmath203 then decrement @xmath171 , otherwise increment @xmath172 . 4.3.@xmath173 : : until @xmath204 . 5.@xmath161 : : report `` no solution '' and halt . with appropriate modifications this algorithm also solves the three - set version of , where the input is @xmath205 . [ [ efficiency - of - the - algorithm . ] ] efficiency of the algorithm . + + + + + + + + + + + + + + + + + + + + + + + + + + + + step 1 requires @xmath206 comparisons . by lemmas [ lem : fredman ] and [ lem : buck ] , step 2 requires @xmath207 comparisons to sort @xmath208 . using fredman s trick , step 3 requires no comparisons at all , given the sorted order on @xmath208 . ( if @xmath209 and @xmath210 , @xmath211 holds iff @xmath212 . ) for each iteration of the outer loop ( step 4 ) there are at most @xmath189 iterations of the inner loop ( step 4.2 ) since each iteration ends by either incrementing @xmath172 or decrementing @xmath171 . in step 4.2.1 we can determine whether @xmath213 is in @xmath214 with a binary search , in @xmath215 comparisons . in total the number of comparisons is on the order of @xmath216 , which is @xmath3 when @xmath217 . [ [ correctness - of - the - algorithm . ] ] correctness of the algorithm . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the purpose of the outer loop ( step 4 ) is to find @xmath218 , for which @xmath219 and @xmath220 . this is tantamount to finding indices @xmath221 for which @xmath222 , and @xmath202 . we maintain the loop invariant that if there exist @xmath223 for which @xmath224 , then both of @xmath225 and @xmath226 lie in @xmath227 . suppose the algorithm has not halted in step 4.2.1 , that is , there are no solutions with @xmath222 . if @xmath203 then there can clearly be no solutions with @xmath228 since @xmath229 , so decrementing @xmath171 preserves the invariant . similarly , if @xmath230 then there can be no solutions with @xmath231 since @xmath232 , so incrementing @xmath172 preserves the invariant . if it is ever the case that @xmath204 then , by the invariant , no solutions exist . [ [ algorithmic - implementation . ] ] algorithmic implementation . + + + + + + + + + + + + + + + + + + + + + + + + + + + this algorithm can be implemented to run in @xmath233 time while performing only @xmath234 comparisons . using any optimal sorting algorithm steps 13 can be executed in @xmath235 time while using @xmath236 comparisons . now the boxes @xmath237 have been explicitly sorted , so the binary searches in step 4.2.1 can be executed in @xmath238 time per search . the total running time is @xmath233 and the number of comparisons is now minimized when @xmath239 , for a total of @xmath234 comparisons . we do not know of any polynomial time algorithm that performs @xmath3 comparisons . in our decision tree , sorting @xmath208 ( step 2 ) is a comparison - efficient way to accomplish step 3 , but it only lets us _ deduce _ the sorted order of the boxes @xmath237 . it does not give us a useful representation of these sorted orders , namely one that lets us implement each comparison of the binary search in step 4.2.1 in @xmath240 time . in this section we present several methods for sorting the boxes based on bichromatic dominating pairs , as in chan @xcite and bremner et al . @xcite ; see lemma [ lem : redblue ] . the total time spent performing binary searches in step 4.2.1 will be @xmath241 , so our goal is to make @xmath191 as large as possible , provided that the cost of sorting the boxes is of a lesser order . [ [ overview . ] ] overview . + + + + + + + + + as a warmup we give , in section [ sect : simple - subquad - alg ] , a relatively simple subquadratic algorithm running in @xmath242 time . in section [ sssect : faster3sum ] we present a more sophisticated algorithm , some of whose parameters can be selected either deterministically or randomly . sections [ sect : rand - param ] and [ sect : det - param ] give two parameterizations of the algorithm , which lead to an @xmath243 time deterministic algorithm and @xmath244-time randomized algorithm . choose the group size to be @xmath245 . the algorithm enumerates _ every _ permutation @xmath246\rightarrow [ g]^2 $ ] , where @xmath247 is decomposed into row and column functions @xmath248 \rightarrow [ g]$ ] . by definition @xmath249 is the correct sorting permutation iff @xmath250 for all @xmath251 $ ] . is totally ordered . see lemma [ lem : tiebreaking ] . ] since @xmath252 this inequality can also be written @xmath253 . by fredman s trick this is equivalent to saying that the ( red ) point @xmath254 dominates the ( blue ) point @xmath255 , where @xmath256 we find all such dominating pairs . by lemma [ lem : redblue ] the time to report red / blue dominating pairs , over all @xmath257 invocations of the procedure , is @xmath258 , the last term being the total size of the outputs . for @xmath259 and @xmath260 the first term is negligible . the total running time is therefore @xmath261 for dominance reporting and @xmath262 for the binary searches in step 4.2.1 . since there are at most @xmath263 realizable permutations of @xmath264 , not @xmath257 ( see lemma [ lem : sortx+y ] and fredman @xcite ) , we could possibly shave off another @xmath265 factor by setting @xmath266 . however , with a bit more work it is possible to save @xmath102 factors , as we now show . to improve the running time of the simple algorithm we must sort larger boxes . our approach is to partition the blocks into layers and sort each layer separately . so long as each layer has size @xmath267 , the cost of red / blue dominance reporting will be negligible . the main difficulty is that the natural boundaries between layers are unknown and different for each of the blocks in @xmath237 . let @xmath268 ^ 2 $ ] be a set of @xmath145 positions in the @xmath269 grid that includes positions @xmath270 and @xmath271 . how we select the remaining @xmath272 positions in @xmath143 will be addressed later . for each @xmath273 , consider @xmath274 , that is , the path in @xmath275 ^ 2 $ ] taken by the standard algorithm of section [ sect : quadratic3sum ] when searching for @xmath276 inside @xmath264 . clearly @xmath277 goes through position @xmath278 . for any two @xmath279 , @xmath277 and @xmath280 may intersect in several places ( see figure [ fig : searchpath ] ) though they never cross . according to lemma [ lem : contour ] , the two contours define a _ @xmath281 of the positions of @xmath275 ^ 2 $ ] into three regions , where @xmath282\\ a_{i , j}(s ) & \subset ( a_{i , j}(l , m),\ , a_{i , j}(l',m'))\\ a_{i , j}(t ) & \subset [ a_{i , j}(l',m'),\ , \infty)\end{aligned}\ ] ] here @xmath283 for a subset @xmath284 ^ 2 $ ] . note that @xmath281 is fully determined by the shapes of the contours , not the specific contents of @xmath264 . totally ordered . refer to lemma [ lem : tiebreaking ] . ] see figure [ fig : tripartition](a ) for an illustration . [ cols="^,^,^,^,^,^,^,^,^,^",options="header " , ] throughout let @xmath285 be a sufficiently small constant . in the randomized implementation of our algorithm we choose @xmath286 and @xmath287 . the points @xmath270 and @xmath271 must be in @xmath143 and the remaining @xmath288 points are chosen uniformly at random . with these parameters the probability of a box being _ bad _ is sufficiently low to keep the expected cost per search @xmath238 . [ lem : badprob ] the probability a particular box is bad is at most @xmath289 . let @xmath249 be the sorted order for some box @xmath264 . the probability @xmath264 is bad is precisely the probability that there are @xmath290 consecutive elements ( according to @xmath249 ) that are not included in @xmath143 . the probability that this occurs for a particular set of @xmath290 elements is less than @xmath291 . by a union bound over all @xmath292 sets of @xmath290 consecutive elements , the probability @xmath264 is bad is at most @xmath289 . the expected time per search is therefore @xmath293 . by linearity of expectation the expected total running time is @xmath294 . we could have set the parameters differently and achieved the same running time . for example , setting @xmath295 and @xmath296 would also work . the advantage of keeping @xmath297 is simplicity : we can afford to enumerate all @xmath298 permutations of @xmath299 ^ 2 $ ] rather than explicitly construct a hyperplane arrangement in order to enumerate only those _ realizable _ permutations of @xmath300 . [ [ high - probability - bounds . ] ] high probability bounds . + + + + + + + + + + + + + + + + + + + + + + + + the running time of the algorithm may deviate from its expectation with non - negligible probability since the _ badness _ events for the boxes @xmath237 can be strongly positively correlated . the easiest way to obtain high probability bounds is simply to choose @xmath301 random point sets @xmath302}$ ] , estimate the cost of the algorithm under each point set , then execute the algorithm under the point set with the best estimated cost . the first step is to run a truncated version of the algorithm in order to determine which queries will be asked in step 4.2.1 . rather than answer the query @xmath303 we simply record the triple @xmath304 in a list @xmath305 to be answered later . the running time of the algorithm under @xmath306 is @xmath307 plus @xmath191 times the number of bad triples in @xmath305 , that is , those @xmath304 for which @xmath264 is bad according to @xmath306 . let @xmath308 be the true fraction of bad triples in @xmath305 according to @xmath306 and @xmath309 be the estimate of @xmath308 obtained by the following procedure . sample @xmath310 elements of @xmath305 uniformly at random and for each , test whether the given block is bad according to @xmath306 by sorting its elements , in @xmath311 time . if @xmath312 is the number of blocks discovered to be bad , report the estimate @xmath313 . by a standard version of the chernoff bound is the number of successes in @xmath0 independent bernoulli trials , @xmath314 and @xmath315 are both upper bounded by @xmath316 . see ( * ? ? ? ] we have @xmath317 by lemma [ lem : badprob ] , @xmath318 for each @xmath319 , so by markov s inequality @xmath320 . with high probability , each @xmath309 deviates from @xmath308 by at most @xmath289 , so the running time of the algorithm will be within a constant factor of its expectation with probability @xmath321 . the time to pick the best point set @xmath322 , @xmath323 being @xmath324 } \{\hat\epsilon_l\}$ ] , is @xmath325 . we could set @xmath326 and @xmath327 as high as @xmath328 , making the probability that the algorithm deviates from its expectation exponentially small , @xmath329 . we achieve a subquadratic worst - case algorithm by choosing @xmath330 and @xmath143 such that no block can be bad . fix @xmath331 and @xmath332 for an integer @xmath146 to be determined . aside from the two obligatory points , @xmath143 contains an evenly spaced @xmath333 grid in @xmath275\times [ g]$ ] . setting @xmath334 , @xmath143 is defined as @xmath335 see figure [ fig : tripartition](b ) . we now argue that no box can be bad if @xmath336 . for any legal pair of contours @xmath337 , in the corresponding tripartition @xmath281 no element of @xmath143 can be contained in @xmath300 , that is , in any row ( or column ) containing elements of @xmath143 , the width ( or height ) of the band @xmath300 is at most @xmath338 . since both @xmath339 and @xmath340 are monotone paths in @xmath275 ^ 2 $ ] ( non - decreasing by row , non - increasing by column ) , we always have @xmath341 . see figure [ fig : tripartition](b ) for a worst - case example . for @xmath342 sufficiently small the overhead for reporting dominating pairs will be negligible . the overall running time is therefore @xmath343 . recall that we are given a set @xmath344 and a function @xmath345 , for some real coefficients @xmath346 . the problem is to determine if there is a point @xmath347 where @xmath25 is zero . as we show below , our decision tree can be generalized in a straightforward way to solve @xmath6- with @xmath11 comparisons , when @xmath7 is odd . unfortunately , we do not see how to generalize our _ algorithms _ to solve @xmath6- in @xmath348 time , for any odd @xmath349 . ( of theorem [ thm : kldt ] ) define @xmath350 to be the set @xmath351 , where @xmath352 . begin by sorting the sets @xmath353 we have effectively reduced @xmath6- to an unbalanced problem . letting @xmath63 be the set @xmath354 , the problem is to determine if there exist @xmath32 such that @xmath19 . note that @xmath58 whereas @xmath59 . the standard algorithm of section [ sect : quadratic3sum ] performs @xmath355 comparisons . generalizing the decision tree of section [ sect : nonuniform ] ( from one list to three ) we can solve unbalanced using @xmath356 comparisons , which is @xmath11 when @xmath357 . our subquadratic algorithms do not extend naturally to unbalanced instances . when @xmath358 we can no longer afford to _ explicitly _ sort all @xmath269 boxes in @xmath115 as this would require at least @xmath359 time . boxes of interest . the dominating pairs approach does not let us sort an arbitrary selection of boxes in constant time per box , but it is possible to accomplish this task in a more powerful model of computation . on a souped - up word ram with @xmath360-bit words and a couple non - standard unit - time operations , any @xmath269 box can be sorted in @xmath240 time . simulating such a unit - time operation on the traditional word ram is a challenging problem . ] we consider a matrix product called _ target - min - plus _ that subsumes the @xmath12-product ( aka distance product ) and the problem of @xcite . given real matrices @xmath361 and a target matrix @xmath362 , the goal is to compute @xmath363 , where @xmath364 \mbox { and } a(i , k)+b(k , j ) \ge t(i , j)\,\right\}\ ] ] as well as the matrix of witnesses , that is , the @xmath6 ( if any ) for which @xmath365 . this operation reverts to the @xmath12-product when @xmath366 . it can also solve on a weighted graph @xmath367 by setting @xmath368 and @xmath369 as follows . let @xmath370 , where @xmath371 if @xmath372 , and let @xmath373 if @xmath374 and @xmath375 otherwise . if @xmath376 then there is a zero weight triangle containing @xmath183 and the witness matrix gives the third corner of the triangle . the trivial target - min - plus algorithm runs in @xmath377 time and performs the same number of comparisons . we can compute the target - min - plus product using fewer comparisons using fredman s trick . [ thm : target - min - plus ] the decision - tree complexity of the target - min - plus product of three @xmath44 matrices is @xmath13 . this product can be computed in @xmath14 time . we first show that the target - min - plus product @xmath378 can be determined with @xmath379 comparisons , where @xmath368 and @xmath369 are @xmath380 and @xmath381 matrices , respectively . begin by sorting the set @xmath382 , j\in[t ] , \mbox { and } k , k'\in [ s]\}.\ ] ] by lemma [ lem : sortx+y ] the number of comparisons required to sort @xmath208 is @xmath383 . we can now deduce the sorted order on @xmath384\},\ ] ] for any pair @xmath385\times [ t]$ ] , and can therefore find @xmath386 with a binary search over @xmath387 using @xmath388 additional comparisons . the total number of comparisons is @xmath379 . note that this provides no improvement when @xmath64 and @xmath65 are square , that is , when @xmath389 . following fredman @xcite we partition @xmath64 and @xmath65 into rectangular matrices and compute their target - min - plus products separately . choose a parameter @xmath191 and partition @xmath64 into @xmath390 and @xmath65 into @xmath391 where @xmath392 contains columns @xmath393 of @xmath64 and @xmath394 contains the corresponding rows of @xmath65 . for each @xmath395 $ ] , compute the target - min - plus product @xmath396 and set @xmath397 } ( c_{\ell}(i , j))$ ] . this algorithm performs @xmath398 comparisons to compute @xmath399}$ ] and @xmath400 comparisons to compute @xmath63 . when @xmath357 the number of comparisons is @xmath13 . to compute the product efficiently we use the geometric dominance approach of chan @xcite and bremner et al . choose a parameter @xmath401 and partition @xmath64 into @xmath402 matrices @xmath403 and @xmath65 into @xmath404 matrices @xmath405 . for each @xmath406 $ ] and permutation @xmath407 \rightarrow [ g]$ ] we will find those pairs @xmath408 ^ 2 $ ] for which @xmath249 is the sorted order on @xmath409\}$ ] . such a triple satisfies the inequality @xmath410 , for all @xmath411 $ ] . by fredman s trick this is equivalent to saying that the ( red ) point @xmath412 in each of the @xmath413 coordinates . by lemma [ lem : redblue ] the total time for all @xmath414 invocations of the dominance algorithm is @xmath415 plus the output size , which is precisely @xmath416 . for @xmath417 and @xmath401 the running time is @xmath418 . we can now compute the target - min - plus product @xmath396 in @xmath419 time by iterating over all @xmath420 ^ 2 $ ] and performing a binary search to find the minimum element in @xmath409\ } \cap [ t(i , j),\infty)$ ] . since @xmath421 contains the pointwise minima of @xmath422 , the total time to compute the target - min - plus product is @xmath423 . the @xmath424 factor in the decision tree complexity of target - min - plus arises comes from the binary searches , @xmath425 searches per pair @xmath420 ^ 2 $ ] . if the searches were sufficiently correlated ( either for fixed @xmath183 or fixed @xmath426 ) then there would be some hope that we could evade the information theoretic lower bound of @xmath427 per search . using random sampling we form a hierarchy of rectangular target - min - plus products such that the solutions at one level gives a hint for the solutions at the next lower level . the cost of finding the solution , given the hint from the previous level , is @xmath240 in expectation . the same approach lets us shave off another @xmath428 factor off the algorithmic complexity of target - min - plus . [ thm : target - min - plus - rand ] the randomized decision tree complexity of the target - min - plus product of three @xmath44 matrices is @xmath89 . it can be computed in @xmath91 time with high probability . as usual let @xmath368 and @xmath369 be @xmath44 matrices and @xmath191 be a parameter . we will eventually set @xmath429 . we partition the indices @xmath430 $ ] at @xmath428 levels . define @xmath431 to be the @xmath145th interval at level @xmath319 . in other words , level-@xmath319 intervals have width @xmath432 and a level-@xmath433 interval is the union of two level-@xmath319 intervals . form a series of nested index sets @xmath430 = j_0 \supset j_1 \supset \cdots \supset j_{\log\log n-1}$ ] , such that @xmath434 is a uniformly random subset of @xmath435 of size @xmath191 . in other words , each element of @xmath436 is promoted to @xmath437 with probability 1/2 , but in such a way that @xmath438 is precisely its expectation @xmath191 . after generating the sets @xmath439 the algorithm sorts @xmath208 with @xmath440 comparisons ( see lemma [ lem : sortx+y ] ) , where @xmath441 $ and $ k , k ' \in j_l \cap i_{l , p}$,}\\ \mbox{for some level $ l$ and index $ p$ } \end{array } \right . \right\}\ ] ] fix @xmath442 $ ] . we proceed to compute @xmath443 with @xmath444 comparisons with high probability . if @xmath445 $ ] is a set of indices , define @xmath446 to be the witness of the target - min - plus product restricted to @xmath70 , that is , @xmath447{0mm}{1 mm}\rule{0mm}{2.5 mm}}}\\ a(i , k ) + b(k , j ) \ge t(i , j ) } } ( a(i , k ) + b(k , j)).\ ] ] there may , in fact , be no such witness , in which case @xmath448 . let @xmath449 be short for @xmath450 . notice that by fredman s trick we can deduce the sorted order on @xmath451 without additional comparisons , for any @xmath319 and @xmath145 . we can therefore compute the top - level witnesses @xmath452}$ ] with @xmath453 comparisons via binary search . our goal now is to compute the witnesses at all lower level intervals with @xmath454 comparisons . suppose we have computed the level-@xmath433 witness @xmath455 and wish to compute the level-@xmath319 witnesses of the constituent sequences , namely @xmath456 and @xmath457 . define @xmath458 . note that @xmath459 is determined by @xmath455 and the sorted order on @xmath460 . the distance between @xmath456 and @xmath459 ( according to the sorted order on @xmath461 ) is stochastically dominated by a geometric random variable with mean 1 . @xmath462 ; with probability less than 1/4 @xmath456 is one less than @xmath459 according to the sorted order on @xmath461 , and so on . ] the expected number of comparisons needed to determine @xmath449 using linear search is therefore @xmath240 . these geometric random variables are independent due to the independence of the samples , so we can apply standard chernoff - type concentration bounds @xcite . the probability that the sum of these independent geometric random variables exceeds twice its expectation @xmath463 is @xmath464 . once we have computed all the witnesses for level-0 , @xmath465}$ ] , we simply have to choose the best among them , so @xmath466 \mbox { and } \kappa_{0,p } \neq { \perp}\}$ ] . the total number of witnesses computed for fixed @xmath72 is @xmath467 . the total number of comparisons is therefore @xmath468 to sort @xmath208 and @xmath418 to compute all the witnesses and @xmath63 , which is @xmath89 when @xmath239 . to improve the @xmath14 algorithm we apply the ideas above with different parameters . let @xmath469 . we consider the same partitions @xmath470 and nested index sets @xmath471 , but only use the first @xmath472 levels , not @xmath428 as before . for each level @xmath473 $ ] , index @xmath474 $ ] , and permutation @xmath475\rightarrow [ g]$ ] , we compute those pairs @xmath183 for which @xmath249 is the sorting permutation on the elements of @xmath476 . this can be done in time linear in the output size , at most @xmath477 , and @xmath478 } \frac{n}{g2^l } g!c_\epsilon n^{1+\epsilon}$ ] . when @xmath259 and @xmath191 is sufficiently small the time spent computing dominating pairs is @xmath418 . since @xmath401 we can encode the sorting permutation of each @xmath479 in one word and can answer a variety of queries about these permutations in @xmath240 time using @xmath480-size precomputed tables . fix a pair @xmath481 ^ 2 $ ] . when finding the top - level witnesses @xmath482}$ ] we can implement each step of the binary searches in @xmath240 time using table lookups , for a total of @xmath444 time . we can also implement each step of the linear searches for witnesses @xmath456 and @xmath457 in @xmath240 time using table lookups . ( in addition to encoding the sorting permutations on @xmath483 , @xmath484 , and @xmath485 , we also need to encode the positions of @xmath483 within @xmath486 as a length-@xmath487 bit vector . this is needed in order to find @xmath459 and @xmath488 in @xmath240 time , given @xmath455 and the sorted order on @xmath460 . ) over all @xmath408 ^ 2 $ ] the total number of comparisons is @xmath489 with high probability . the trivial time to solve on sparse @xmath37-edge graphs is @xmath92 . such graphs contain at most @xmath92 triangles , which can be enumerated in @xmath92 time . we now restate and prove [ thm : sparse - zerotriangle ] . + * theorem [ thm : sparse - zerotriangle ] . * _ the decision tree complexity of on @xmath37-edge graphs is @xmath93 and , using randomization , @xmath94 with high probability . the problem can be solved in @xmath95 time deterministically or @xmath96 with high probability . _ + we begin by greedily finding an acyclic orientation of the graph @xmath27 . iteratively choose the vertex @xmath490 with the fewest number of still unoriented edges and direct them all away from @xmath490 . since every @xmath37-edge graph contains a vertex of degree less than @xmath491 , the maximum outdegree in this orientation is less than @xmath492 . we now use @xmath493 instead of @xmath494 to emphasize that the set is oriented . select a random mapping @xmath495 $ ] , where @xmath70 will be fixed soon . the expected number of pairs of oriented edges @xmath496 having @xmath497 is less than @xmath498 . any coloring that does not exceed this expected value suffices ; we do not need to choose @xmath499 at random . we now sort the set @xmath208 with @xmath500 comparisons @xcite , where @xmath501 call a triangle on @xmath502 _ type-@xmath503 _ if the orientation of the edges is @xmath504 and @xmath505 . clearly every triangle is of one type and there are @xmath506 types . a type-@xmath503 zero - weight triangle exists iff @xmath507 appears in the set @xmath508 . by fredman s trick the sorted order of this set is determined by the sorted order of @xmath208 , since @xmath509 iff @xmath510 . see figure [ fig : triangle ] . we can therefore determine if there exists a zero - weight triangle of a particular type with @xmath511 comparisons via binary search . the total number of comparisons is @xmath512 , which is @xmath93 when @xmath513 . the @xmath514 factor can be shaved off using randomization , exactly as in theorem [ thm : target - min - plus - rand ] . we form @xmath428 levels of colorings , where color class @xmath145 at the @xmath433th level is the union of classes @xmath515 and @xmath516 at the @xmath319th level . after the searches are conducted at level @xmath517 , the expected cost per search at level @xmath319 is @xmath240 . to solve efficiently we greedily orient the graph as before , stopping when all remaining vertices have degree at least @xmath492 , where @xmath492 is a parameter to be fixed shortly . ( the unoriented subgraph remaining is called the _ @xmath492-core_. ) for each vertex @xmath518 and each pair of outgoing edges @xmath519 , we check whether @xmath520 is a triangle and , if so , whether it has zero weight . ( note that the edge @xmath521 , if it exists , may be in the @xmath492-core and therefore not have an orientation . ) this takes @xmath522 time . it remains to check triangles contained entirely in the @xmath492-core . since the @xmath492-core has at most @xmath523 vertices we can solve on it in @xmath524 time or @xmath525 time with high probability . the total cost is balanced when @xmath526 or @xmath527 depending on whether uses the randomized or deterministic algorithm . the problem is easily reducible to , so our @xmath3 and @xmath528 bounds for extend directly to . however , has additional structure , which makes it amenable to the same random sampling techniques used in theorem [ thm : target - min - plus - rand ] . we give only a sketch of the proof of theorem [ thm : convolutionthreesum ] as the analysis is essentially the same as that found in theorem [ thm : target - min - plus - rand ] . + * theorem [ thm : convolutionthreesum ] . * _ the decision tree complexity of is @xmath3 and its randomized decision tree complexity is @xmath97 with high probability . the problem can be solved in @xmath5 time deterministically , or in @xmath98 time with high probability . _ + ( sketch ) in the problem we must determine if there is a @xmath529 $ ] such that @xmath201 occurs on the @xmath6th antidiagonal of the matrix @xmath530 . in contrast to , the rows and columns of @xmath530 are not sorted . on the other hand , we do not need to look for @xmath201 in the whole matrix , just those locations along an antidiagonal . the @xmath3 decision tree bound is proved as in section [ sect : nonuniform ] , by partitioning the matrix into @xmath269 blocks and for each @xmath6 , conducting binary searches for @xmath201 in the appropriate antidiagonals of at most @xmath531 boxes . in order to shave off the @xmath424 factor we use the same random sampling approach of theorem [ thm : target - min - plus - rand ] . we partition @xmath530 at @xmath428 levels , where level-@xmath319 boxes have size @xmath532 and are the union of four level-@xmath533 boxes . the rows and columns are sampled at @xmath428 levels , where a row or column at level @xmath534 is promoted to level @xmath319 with probability 1/2 . note that an element of @xmath530 appears at level-@xmath319 if and only if both its row and column are in the level-@xmath319 sample . since elements along any antidiagonal share no rows or columns , the events that they appear at level-@xmath319 are entirely independent . this independence property allows us to search for @xmath201 in level-@xmath319 sampled boxes in @xmath240 expected time , given the predecessors of @xmath201 in the level-@xmath433 sampled boxes . algorithms running in @xmath535 ( deterministically ) or @xmath98 ( with high probability ) are obtained using the methods applied in theorems [ thm : target - min - plus ] and [ thm : target - min - plus - rand ] . alternatively , we could apply the williams - williams reduction @xcite from to and then invoke the algorithms of theorems [ thm : target - min - plus ] and [ thm : target - min - plus - rand ] as black boxes . since the introduction of fredman s @xcite @xmath12-product algorithm in 1976 , many have become comfortable with the idea that some numerical problems _ naturally _ have a large gap ( @xmath536 ) between their ( nonuniform ) decision - tree complexity and ( uniform ) algorithmic complexity .- convolution , @xmath537-convolution , polyhedral 3sum ( see bremner et al . @xcite ) , and erds - szekeres partitioning , that is , decomposing a sequence into @xmath454 monotonic subsequences . see bar - yehuda and fogel @xcite , dijkstra @xcite , and fredman @xcite . ] from this perspective , our decision trees for and ( with depth @xmath538 and @xmath539 ) do not constitute convincing evidence that and have truly subquadratic and subcubic algorithms . however , williams s @xcite recent breakthrough on the algorithmic complexity of @xmath12-product should shake one s confidence that these @xmath540 gaps are natural . to close them one may simply need to develop more sophisticated algorithmic machinery . the exponent @xmath541 has a special significance in ptracu s program @xcite of conditional lower bounds based on hardness of . his superlinear lower bounds on triangle enumeration and polynomial lower bounds on dynamic data structures depend on the complexity of being @xmath542 , for some @xmath543 . in most other -hardness proofs there is nothing sacred about the 3/2 threshold ( or any other exponent ) . for example , if requires @xmath544 time then finding three collinear points in a set @xmath545 also requires @xmath546 time @xcite . 10 o. weimann a. abboud , v. vassilevska williams . consequences of faster sequence alignment . in _ proceedings 41st intl colloquium on automata , languages , and programming ( icalp ) _ , 2014 . a. abboud and k. lewi . exact weight subgraphs and the @xmath6-sum conjecture . in _ proceedings of the 40th intl colloquium on automata , languages , and programming ( icalp ) _ , pages 112 , 2013 . a. abboud and v. vassilevska williams . popular conjectures imply strong lower bounds for dynamic problems . , abs/1402.0054 , 2014 . o. aichholzer , f. aurenhammer , e. d. demaine , f. hurtado , p. ramos , and j. urrutia . on @xmath6-convex polygons . , 45(3):7387 , 2012 . n. ailon and b. chazelle . lower bounds for linear degeneracy testing . , 52(2):157171 , 2005 . a. amir , t. m. chan , m. lewenstein , and n. lewenstein . consequences of faster sequence alignment . in _ proceedings 41st intl colloquium on automata , languages , and programming ( icalp ) _ , 2014 . r. bar - yehuda and s. fogel . partitioning a sequence into few monotone subsequences . , 35:421440 , 1998 . i. baran , e. d. demaine , and m. ptracu . subquadratic algorithms for 3sum . , 50(4):584596 , 2008 . g. barequet and s. har - peled . polygon containment and translational min - hausdorff - distance between segment sets are 3sum - hard . , 11(4):465474 , 2001 . a. bjorklund , r. pagh , v. vassilevska williams , and u. zwick . listing triangles . in _ proceedings 41st intl colloquium on automata , languages , and programming ( icalp ) _ , 2014 . m. blum , r. w. floyd , v. pratt , r. l. rivest , and r. e. tarjan . time bounds for selection . , 7(4):448461 , 1973 . d. bremner , t. m. chan , e. d. demaine , j. erickson , f. hurtado , j. iacono , s. langerman , m. ptracu , and p. taslakian . necklaces , convolutions , and _ x _ + _ y_. , 69:294314 , 2014 . r. c. buck . partition of space . , 50:541544 , 1943 . a. butman , p. clifford , r. clifford , m. jalsenius , n. lewenstein , b. porat , e. porat , and b. sach . pattern matching under polynomial transformation . , 42(2):611633 , 2013 . t. m. chan . all - pairs shortest paths with real weights in @xmath547 time . , 50(2):236243 , 2008 . s. chechik , d. larkin , l. roditty , g. schoenebeck , r. e. tarjan , and v. vassilevska williams . better approximation algorithms for the graph diameter . in _ proceedings of the 25th annual acm - siam symposium on discrete algorithms ( soda ) _ , pages 10411052 , 2014 . chen , p .- h . hsu , and k .- approximate matching for run - length encoded strings is 3sum - hard . in _ combinatorial pattern matching _ , volume 5577 of _ lecture notes in computer science _ , pages 168179 . e. w. dijkstra . some beatiful arguments using mathematical induction . , 13:18 , 1980 . d. p. dubhashi and a. panconesi . . cambridge university press , 2009 . h. edelsbrunner , j. orourke , and r. seidel . constructing arrangements of lines and hyperplanes with applications . , 15(2):341363 , 1986 . p. erds and p. turn . on a problem of sidon in additive number theory , and on some related problems . , 1(4):212215 , 1941 . j. erickson . bounds for linear satisfiability problems . , 1999 . m. l. fredman . how good is the information theory bound in sorting ? , 1(4):355361 , 1976 . m. l. fredman . new bounds on the complexity of the shortest path problem . , 5(1):8389 , 1976 . a. gajentaan and m. h. overmars . on a class of @xmath548 problems in computational geometry . , 5:165185 , 1995 . j. hartmanis and r. e. stearns . on the computational complexity of algorithms . , 117:285306 , 1965 . z. jafargholi and e. viola . , 3xor , triangles . , abs/1305.3827 , 2013 . lambert . sorting the sums @xmath549 in @xmath548 comparisons . , 103(1):137141 , 1992 . s. meiser . point location in arrangements of hyperplanes . , 106(2):286303 , 1993 . f. meyer auf der heide . a polynomial linear search algorithm for the @xmath0-dimensional knapsack problem . , 31(3):668676 , 1984 . f. p. preparata and m. i. shamos . . springer , new york , ny , 1985 . m. ptracu . towards polynomial lower bounds for dynamic problems . in _ proceedings 42nd acm symposium on theory of computing ( stoc ) _ , pages 603610 , 2010 . m. ptracu and r. williams . on the possibility of faster sat algorithms . in _ proceedings of the 21st annual acm - siam symposium on discrete algorithms ( soda ) _ , pages 10651075 , 2010 . l. roditty and v. vassilevska williams . fast approximation algorithms for the diameter and radius of sparse graphs . in _ proceedings 45th acm symposium on theory of computing ( stoc ) _ , pages 515524 , 2013 . m. a. soss , j. erickson , and m. h. overmars . preprocessing chains for fast dihedral rotations is hard or even impossible . , 26(3):235246 , 2003 . r. williams . faster all - pairs shortest paths via circuit complexity . in _ proceedings of the 46th annual acm symposium on theory of computing ( stoc ) _ , 2014 . technical report available as arxiv:1312.6680 . v. vassilevska williams and r. williams . subcubic equivalences between path , matrix and triangle problems . in _ proceedings 51th annual ieee symposium on foundations of computer science ( focs ) _ , pages 645654 , 2010 . v. vassilevska williams and r. williams . finding , minimizing , and counting weighted subgraphs . , 42(3):831854 , 2013 . for the sake of completeness we shall review a standard divide and conquer dominating pairs algorithm of preparata and shamos @xcite and give a short proof of lemma [ lem : redblue ] due to chan @xcite . we are given @xmath0 red and blue points in @xmath550 , at least one of each color , and wish to report all pairs @xmath551 where @xmath552}$ ] is red , @xmath553}$ ] is blue and @xmath554 for each @xmath555 $ ] . when @xmath556 the algorithm simply reports every pair of points , so assume @xmath557 . find the median @xmath558 on the last coordinate in @xmath480 time @xcite and partition @xmath143 into disjoint sets @xmath559 , @xmath560 of size at most @xmath561 , where @xmath562 furthermore , there can not be a red @xmath563 and blue @xmath564 such that @xmath565 . were a dominating pair only if @xmath566 for all @xmath555 $ ] , then we would break ties the other way , letting red points precede blue points . ] at this point all dominating pairs are in @xmath567 or @xmath560 , or have one point in each , in which case the blue point is necessarily in @xmath559 and the red in @xmath560 . we make three recursive calls to find dominating pairs of each variety . the first two calls are on @xmath561 points in @xmath103 . the third recursive call is on all blue points in @xmath559 and all red points in @xmath560 ; after stripping their last coordinate they lie in @xmath568 .

structural parameters ( half - light radius , mean effective surface brightness , and sersic index , parameterizing the light profile shape ) are derived for a sample of galaxies in the rich cluster ac118 at : so far the largest ( n=93 ) sample of galaxies at intermediate - redshift with structural parameters measured in the near - infrared . the parameters are obtained in two optical wavebands ( and ) and in theband , corresponding approximately to , and rest - frame . the distributions of at match those for the coma cluster ( i.e. for the local universe ) both in the optical and in the nir . theband distribution is of particular interest , since the nir light mimics the mass distribution of galaxies . the similarity of the distributions for the two clusters ( ac118 and coma ) proves that the galaxies at the bright end of the luminosity function did not significantly change their sizes since to the present epoch . the ratio of the optical to the nir half - light radius shows a marked trend with the shape of the light profile ( sersic index ) . in galaxies with ( typical bright ellipticals ) , while the average ratio is 0.8 for galaxies with lower ( typical disk systems ) . moreover , the nir sersic index is systematically larger than in the optical for . these results , translated into optical and optical - nir color gradients , imply that the optical color gradients at.3 are similar to those of nearby galaxies . the optical - nir color gradients are in the average larger , ranging from -0.73 mag / dex for to -0.35 mag / dex for . models with ` pure age ' or ` pure metallicity ' gradients are unable to reconcile our color gradients estimates with observations at , but we argue that the combined effects of age and metallicity might explain consistently the observed data : passive evolution ( plus the possible effect of dust absorption ) may account for the differences between the optical and nir structural properties . the lack of any major change in since suggests that merging involving bright galaxies did not play a significant role in the last.4 gyr (=0.3 , , h km smpc ) . the results of the present paper will be applied to the study of the scaling laws in subsequent works .

astro-ph0112040

our knowledge on the evolution of galaxies comes from the study of the properties of families of galaxies at different redshifts . observable quantities like magnitudes , colors , structural parameters , velocity dispersions and line strengths are combined into well - established relations like , for instance , the color - magnitude relation ( e.g. bower , lucey , & ellis 1992 ; kodama et al . 1998 ) , the mg-@xmath21 relation @xcite , the fundamental plane ( dressler et al . 1987 ; djorgovski & davies 1987 ; jrgensen , franx , & kjrgaard 1996 ) and its projection in the @xmath0 @xmath1 plane , that is the kormendy relation ( kormendy 1977 ; capaccioli , caon , donofrio 1992 ; ziegler et al . these relations can now be studied over a significant fraction of the age of the universe . studies of the properties of early - type galaxies in rich clusters lead to a large amount of observational evidence in favor of the passive evolution scenario , in which an early ( @xmath222 ) and short ( @xmath161 gyr ) burst of star formation is followed by a long period where the aging of stellar populations drives the observed properties of galaxies through intermediate redshifts until the present epoch ( kodama & arimoto 1997 ; stanford , eisenhardt , & dickinson 1998 ; jrgensen et al . 1999 ; kelson et al . 2000a , and references therein ) . on the other side , the alternative scenario of the hierarchical merging seems also consistent with most of the existing data ( e.g. kauffmann & charlot 1998 ) , indicating either that the present observational evidence is not sufficient to discriminate between the two pictures or that galaxy evolution is driven by both mechanisms . the half - light radius and the mean surface brightness of a galaxy measure the size and the density of the luminous matter . these quantities are related both to the gravitational potential of the galaxy and to the properties of the stellar populations . the relations existing between the structural parameters of galaxies have been recently studied at intermediate redshifts , both in the form of @xmath0 vs. @xmath1 ( * ? ? ? * and references therein ) and as luminosity - size relation @xcite . @xcite found that disks at @xmath220.5 are in the average @xmath16 1.5 mag brighter than in local galaxies . this was contrasted by the results of @xcite who found no luminosity evolution in disks up to @xmath130.9 . @xcite found that the kormendy relation of cluster galaxies at @xmath23 and @xmath24 is consistent with the passive evolution scenario . the kormendy relation has also proven to be a powerful tool to test the expansion of the universe by the measurement of the cosmological dimming ( i.e. for the tolman test ) . @xcite studied the kormendy relation up to @xmath25 , finding that the data are consistent with the cosmic expansion together with passive evolution . finally , the structural parameters enter the fundamental plane of early - type galaxies , of which the kormendy relation is a two - dimensional projection . due to its very small scatter , the fundamental plane is a powerful tool to study the evolution of the m / l ratio of galaxies belonging to distant clusters . the recent determinations of the fundamental plane at intermediate redshifts constrain the epoch of star formation in early - type galaxies at large redshifts ( @xmath222 for @xmath17=0.3 or @xmath225 for @xmath17=1 and @xmath18 ; see van dokkum & franx 1996 ; jrgensen et al . 1999 ; kelson et al . however , these studies have to face with the problem of the progenitor bias , i.e. with our ignorance of the relationship between local galaxies and their high redshift counterparts ( see van dokkum & franx 2001 for a thorough discussion ) . a clue for the interpretation and for the applications of the scaling laws is the dependence of the structural parameters on wavelength . the basic distinction between optical and nir is that the optical wavebands are very sensitive to the effects of metallicity ( through line - blanketing ) and to the content of younger populations ( dust absorption included ) , while the nir wavebands are dominated by the old , more quiescent , stellar populations and are virtually unaffected by dust absorption . for these reasons the nir is the best tracer of the mass distribution inside a galaxy . local galaxies are smaller at nir wavebands than in the visible , with a similar , though weaker , trend also within the optical wavelength range ( scodeggio et al . 1998 ; pahre , de carvalho , & djorgovski 1998 ) . this behavior is confirmed by the relatively large optical - nir color gradients measured in early - type galaxies , the optical - optical gradients being usually much smaller , except when the @xmath26band is involved ( peletier et al . 1990a ; peletier , valentijn , & jameson 1990b ; hinkley @xmath27 i m 2001 ; nelson et al . 2001 ) . small optical color gradients were recently measured in distant ( up to @xmath131 ) cluster early - type galaxies by @xcite and by @xcite . these authors concluded that metallicity is the primary factor driving the color gradients . the derivation of the structural parameters at intermediate redshifts is affected by the problem of the small angular size of the galaxies , whose half - light radii are typically less than 1 arcsec . for this reason many works were based on imaging taken with the hst . it has been demonstrated , however , that reliable structural parameters can be derived from ground - based photometry as well , on high resolution ( @xmath28/pxl ) and exceptional seeing ( @xmath29 ) conditions , provided that the properties of the psf are carefully taken into account @xcite . in this paper we obtain the first multi - band large set of structural parameters for galaxies at an intermediate redshift ( @xmath3 ) . the data are derived from ground - based ( eso - ntt ) photometry in two optical ( @xmath4 and @xmath5 ) and one nir ( @xmath30 ) wavebands . the derived values for the structural parameters will be published in a forthcoming paper addressed to the study of galaxy evolution by means of the scaling laws ( busarello et al . 2001a , in preparation ) . the paper is organized as follows . the sample and the data used in this study are presented in [ samples ] . the derivation of the structural parameters is outlined in [ strsec ] . the structural parameters from hst imaging are derived in [ hst_hst ] , where the results from sersic models and bulge - disk decomposition are compared . in [ sec_hst_r ] we use hst photometry for a subset of galaxies to verify the reliability of ground - based structural parameters . the structural parameters are presented in [ op_nir ] , where we make a comparison between optical and nir properties at @xmath9 and with the values at @xmath31 . in [ secgrad ] we analyze the optical and nir structure of galaxies in terms of internal color gradients . the results are discussed in [ conc ] . ac118 ( abell 2744 ) is a rich ( ` coma - like ' ) galaxy cluster at redshift @xmath3 that was studied by several authors in relation to its galaxy population , x - ray and radio properties and gravitational lensing @xcite . new photometric data and medium resolution spectroscopy has been collected during three observing runs ( october 1998 - september 2000 ) at the eso new technology telescope . the data of interest for the present work are the imaging in the @xmath4 and @xmath5 wavebands obtained with the emmi instrument and the imaging in the @xmath30 waveband obtained with sofi . the photometry is complemented with hst archive data consisting of a single pointing on the cluster center . the relevant information on the data is given in table [ ac118_data ] . the reduction of the @xmath6band image is described in @xcite , while the reduction of the ntt optical data will be presented elsewhere ( busarello et al . 2002b , in preparation ) . the hst data were retrieved from the stsci archive and reduced following the steps described in @xcite . the ntt photometry was used to calibrate the hst imaging into the johnson @xmath32band as outlined in appendix a. the samples used in the present study consist in : a ) 50 galaxies in the hst field ; b ) 37 galaxies in the hst field in common with the ground - based @xmath32band sample ; c ) 58 galaxies in common in the @xmath33 and @xmath32band ; d ) 93 galaxies in common in the @xmath32 and @xmath6band . samples c and d are the results of two selection steps . the first selection was based on photometric redshifts @xcite , as derived from the above photometry plus @xmath34 and @xmath26band ( new ntt data and ntt archive data respectively ) . the reliability of the photometric redshifts was verified on 75 galaxies with available spectra ( merluzzi et al . 2002 , in preparation ) . morphology is not a selection criterion in this study , since a morphological classification is not possible from our data . the photometric redshifts algorithm flags the galaxies according to whether their colors are consistent with early- or late - type template spectra . we did not use this classification as a selection criterion , and retained in the samples all the galaxies with the correct photometric redshift . we remark , however , that only 7 galaxies out of 93 in sample d are ` late - type ' according to their photometric properties ( from @xmath35 to @xmath30 ) . the second selection step is based on the reliability of the measurement of the structural parameters and is the result of the analysis presented in [ simsec ] . this lead to include in the sample the galaxies brighter than @xmath5=21 and @xmath4=20.5 , and to exclude objects with an estimated @xmath0 smaller than 0.15 arcsec . samples a and b were selected in order to obtain 1 ) reliable structural parameters in the ntt @xmath32band and 2 ) the largest number of galaxies in the hst field with the same range of sizes and luminosities of the ground - based samples . this lead to include galaxies brighter than @xmath5=21 and to exclude objects with @xmath0 smaller than 0.15 arcsec both in the ntt and in the hst image . a further ( a posteriori ) selection was done for all the samples on the basis of the results of the fitting : objects with very large uncertainties ( @xmath36100% ) were not considered in the analysis . this however lead to exclude only 2% of the objects . morphological classification from hst imaging by @xcite is available for 21 out of 93 galaxies in sample d ( 17 out of 58 in sample c ) . out of these , 6 ( 6 ) are classified as ellipticals , 10 ( 7 ) as s0 , and 5 ( 4 ) as spirals . these numbers reflect the overall distribution of morphological types in ac118 , which is dominated by early - type ( mainly s0 ) galaxies @xcite . the question of morphological types will be resumed in the conclusions . the relevant information on the samples is summarized in table 2 . since the typical galaxy sizes at @xmath130.3 are smaller than @xmath37 , it is crucial to take into account the effects of the point spread function ( psf ) when deriving structural parameters from ground - based data . this means that a proper model for the galaxy surface brightness ( sb ) has to be convolved with the psf and fitted to the observed sb . this is usually performed : 1 ) by one - dimensional fitting algorithms ( e.g. saglia et al . 1997 ) or 2 ) by two - dimensional methods @xcite . in case 1 ) the integrated light curve or the sb profile are derived by an elliptical fit to the galaxy isophotes and then fitted with the adopted model . in the two - dimensional approach , a direct fit of the galaxy sb is performed . advantages and drawbacks of both methods have been largely discussed in previous works ( see kelson et al . 2000b , and references therein ) . in the present study we adopt the two - dimensional fitting approach for two reasons : 1 ) many galaxies in ac118 belong to overlapping systems , and only the 2d approach can deal properly with these cases ; 2 ) isophotal fitting of the ground - based images should be performed on a very small number of pixels , resulting in poorly constrained isophotes . the sb of distant galaxies is well described by two - dimensional models with constant ellipticity and position angle . in this study , we use two kinds of models : the sersic model ( sersic 1968 ) and the ( @xmath38 ) bulge + disk ( @xmath39 ) model . since the @xmath39 decomposition requires high resolution and signal to noise ratios ( see ratnatunga , griffiths , & ostrander 1999 ) , this model is applied only to the hst data . ground - based structural parameters are derived by fitting the sersic models . the models used for the psf of the ntt and hst images are presented in [ psfsec ] while the fitting algorithm and the galaxy models are described in [ fitsec ] . the limits of applicability of the fitting algorithm are set in [ simsec ] . to model the psf of the hst image , we used the tiny tim 5.0 software package @xcite . since the psf depends on the position on the field , different models were created for each galaxy at the corresponding chip location . effects of psf under - sampling and broadening from image jittering are smaller than few percents of @xmath0 for the typical sizes of galaxies at this redshift and therefore these were ignored ( see kelson et al . 2000b ) . to model the psf of the ntt images we proceeded as follows . stars in the cluster field were selected by the stellar index parameter @xmath40 of s - extractor @xcite and from the aperture magnitude . to minimize star galaxy classification errors , we considered only objects with @xmath41 . we selected sources with @xmath42 mag in the sofi field and @xmath43 mag in the region not covered from the infrared imaging . such criteria allowed to obtain a sufficient @xmath44 to model the psf . the reliability of the selection is also supported by the photometric redshift algorithm : the colors of the candidate stars do not match galaxy template spectra at any redshift . the psf of the ntt images is well described by a sum of two gaussians convolved with the pixel function ( see appendix [ app_gpsf ] for details ) . the model has nine relevant parameters : the center coordinates , the central intensity , the full width at half maximum ( fwhm ) of the two gaussians in the two perpendicular directions , the ratio of the central intensities , and the local value of the background . the fits were performed by a @xmath45 minimizing algorithm based on the levenberg - marquardt method @xcite . typically , the residuals from the fitting are smaller than 5% . the uncertainties on fitting parameters were estimated by numerical simulations on models with gaussian noise . for each pixel , the standard deviation of the noise was computed by the distribution of the fit residuals in the adjacent pixels ] . figure [ psffit ] shows the values of the psf parameters in the @xmath33 , @xmath32 and @xmath6band as a function of the distance to the image centers . in general the psf tends to become broader as this distance increases . the effect is very weak in the @xmath6band image , while significant variations are found in the optical images . the trends of the parameters were modeled with empirical analytic curves , that were then used to obtain the psf model relative to each galaxy position . galaxy images were fitted by the following model : @xmath46 where @xmath7 is the galaxy brightness distribution , that is described by a set of parameters @xmath47 , @xmath48 is the psf model , @xmath49 is the value of the local background , and the symbol @xmath50 denotes the convolution . for the sersic models , the parameters @xmath51 are : the effective major semi - axis @xmath52 , the central surface brightness @xmath53 , the sersic index @xmath2 , the axial ratio @xmath54 , the position angle @xmath55 , the coordinates of the photometric center and the local value of the background . in the case of the @xmath39 decomposition , the parameters that enter the fit are those of two sersic models with @xmath56 and @xmath57 for the bulge and the disk respectively . the fit parameters are used to derive the total magnitude @xmath58 ( e.g. saglia et al . 2000 ) , @xmath0 and @xmath1 . for sersic models , @xmath0 is computed as the equivalent radius of the ellipse that encloses half of the total galaxy light , i.e. @xmath59 . for the @xmath39 decomposition , @xmath0 is calculated from the growth curve of circular aperture magnitudes . in both cases , we have @xmath60 . for the sersic models we found that both methods of computing @xmath0 agree within @xmath61 . therefore , the difference of the two definitions can be ignored in the following analysis . the parameters @xmath47 and @xmath49 were derived by a least squares fit , by minimizing the function @xmath62 \ , \label{chieq}\ ] ] where @xmath63 is the observed brightness distribution , @xmath64 is an index running on image pixels and @xmath65 is a penalty function introduced to avoid ` not allowed ' regions of the parameter space ( i.e. unrealistic values that can be found on local minima of @xmath45 ) . the @xmath45 minimization was performed by the levenberg - marquardt algorithm . the convolution with the psf was computed straightly in the real domain in order to avoid high frequencies problems when convolving galaxy models by the fast fourier transform . bad pixels , degraded regions and pixels contaminated by background objects , were excluded from summation in equation [ chieq ] . to this aim , a masking image was created interactively for each galaxy . overlapping objects were fitted simultaneously . uncertainties on structural parameters are usually estimated either from the formal errors on fitting results or by numerical simulations . the formal uncertainties are derived by the topology of the @xmath45 function near to the solution and therefore they do not take into account the various sources of errors on the estimated parameters ( e.g. smith et al . numerical simulations give more reliable uncertainties but they depend on the adopted galaxy model and are much more expensive in terms of computation time . as a compromise between accuracy and amount of computation time , we used a first order approximation : the uncertainties on the fit parameters were estimated from the derivatives of the @xmath45 function taking into account the noise in the galaxy image and the uncertainties on the psf parameters . details are given in appendix [ app_err ] . to test the performance of the fitting procedure and to set the limits to its applicability , we created two sets of simulations based on sersic @xmath66 models convolved with the ntt @xmath32band psf and with the typical noise of ground - based data . to take into account the uncertainties on the psf , all the simulations were generated with the same psf but the fits were performed with different psf models whose parameters were generated in agreement with the corresponding uncertainties ( see [ psfsec ] ) . the first set of simulations consist in @xmath67 galaxies spanning the full magnitude range in the @xmath32band ( 19 mag @xmath68 mag ) . two values of @xmath0 were considered : @xmath69 and @xmath70 , that correspond to the sizes of a small and a typical galaxy in a cluster at @xmath71 ( e.g. kelson et al . in figure [ sim_mtot ] we give the results of the simulations in terms of differences between the input model parameters and the output from the fitting procedure . the quantities shown are @xmath0 , the axis ratio @xmath54 ( both given as logarithms ) , @xmath58 , @xmath1 , @xmath2 , and the ` fundamental plane parameter ' , @xmath72 , that is the combination of the structural parameters entering the fundamental plane . it can be seen that the scatter of the fit parameters around the ` true ' values is relatively small for @xmath7322 where it amounts at most to 25% in @xmath0 , 0.5 mag in @xmath1 , 0.15 mag in @xmath58 , @xmath74 in @xmath75 and @xmath76 in @xmath54 . the scatter of @xmath2 is higher , and varies from @xmath77 at @xmath78 mag up to @xmath79 at @xmath80 mag . objects fainter than @xmath80 mag are affected by larger and systematic differences between input and computed parameters . the second set of simulations was created in order to test the fitting algorithm with respect to the galaxy size . to this aim , a wide range of values of @xmath0 was explored ( @xmath81 to @xmath82 ) for three sersic models with @xmath83 respectively . for each value of @xmath0 , a suitable value of @xmath1 was derived from the kormendy relation , whose slope and the zero point were calibrated a posteriori from the present sample . in figure [ sim_re ] , we show the comparison between true and estimated parameters . an inspection of figure [ sim_re ] shows that the parameters are recovered with a relatively small scatter ( see above ) and without systematic errors for @xmath84 ( @xmath85 ) for the whole range of @xmath2 values . the shape of light profile is subject to a small systematic error ( @xmath86 ) for @xmath87 below @xmath88 ( @xmath89 ) . we also notice that the scatter increases for large @xmath2 values , due to the effect of the uncertainties on the psf for galaxies with more concentrated profiles . the above results were used to derive the selection of the samples in this study . all the samples include the galaxies with @xmath84 in each band ; in the @xmath32band only galaxies with @xmath9021 mag were included as a conservative limit ; in the @xmath33band a more restrictive selection in magnitude ( @xmath9120.5 mag ) was necessary due to the lower s / n , while no further restriction was needed for the @xmath6band photometry . structural parameters were derived for @xmath92 galaxies in the hst image ( sample a of table [ tab_sel ] ) , 42 of these objects are cluster members according to the photometric redshifts . galaxies were fitted by both sersic and @xmath39 models , and a flag @xmath93 was assigned to each galaxy for the different cases that were encountered during the fitting . for @xmath94 galaxies , both models gave small fitting residuals , but the @xmath39 decomposition produced meaningless values of the structural parameters for the bulge or for the disk component ( e.g. @xmath95 ) . for such galaxies a possible second physical component is too weak or too small to be detected . however , except for one object that was excluded from the analysis , acceptable global parameters ( @xmath0 and @xmath1 ) were obtained from the @xmath39 fit so that such galaxies were retained in the sample and were flagged with @xmath96 . for @xmath97 objects , both sersic and @xmath39 models gave small fitting residuals and acceptable estimates of the parameters . they were flagged with @xmath98 . @xmath99 was used to indicate @xmath100 galaxies for which residuals from the sersic model shown evidence for a second component that was properly described with the @xmath39 decomposition . the flags @xmath98 and @xmath99 were assigned by visual inspection of the galaxy fits as illustrated in figure [ bd_s ] . finally , @xmath101 galaxies , of which some objects with very distorted morphology and some bcgs with very large residuals ( @xmath102 ) , were flagged with @xmath103 and excluded from the present analysis . fitting models suitable for such objects will be discussed in a forthcoming work . in figure [ bd_s_par ] , we compare the structural parameters @xmath0 , @xmath1 , @xmath58 and @xmath75 obtained by different models . table [ tab_bd ] shows the medians and the standard deviations of the differences in fitting parameters , and the relative bootstrap confidence intervals . by weighting each point in figure [ bd_s_par ] with the fitting uncertainties ( [ fitsec ] ) , we did not find significant variations of the estimates given in table [ tab_bd ] . this also holds for the other comparisons of structural parameters in the present work ( see 4 and 5 ) . it is clear from table [ tab_bd ] and from figure [ bd_s_par ] that the mean differences of the parameters have no significant biases , also when galaxies with different flag @xmath93 are considered . on the other side , the scatter in the structural parameters depends on the @xmath93 value : as expected , minimum values are obtained for @xmath98 when @xmath39 and sersic models give an equivalent description of the galaxy image . the scatter varies from @xmath104 to @xmath105 for @xmath0 , from @xmath106 mag to @xmath107 mag for @xmath1 , and from @xmath108 mag to @xmath109 mag for @xmath58 . one important fact is the robustness of the @xmath75 parameter against the adopted fitting method : due to the correlation between the errors in @xmath0 and in @xmath1 ( hamabe & kormendy 1987 ; jrgensen , franx , & kjrgaard 1995 ) , its scatter varies from @xmath110 for @xmath98 to @xmath111 for @xmath96 . the previous results are fully consistent with the findings of kelson et al . ( 2000a ) , although they find a smaller dispersion ( @xmath112 ) between the @xmath75 parameters derived by different fitting methods . the results shown in figure [ bd_s_par ] and table [ tab_bd ] give a measure of the intrinsic uncertainties on the structural parameters of the galaxies in the present sample and will be used as a reference in the following sections . the 2d fits for a representative subsample of galaxies in the hst and ntt @xmath32band images are shown in figure [ hst_r ] . although the residuals from the fit are more evident in the hst images due to the higher resolution and to the higher @xmath44 , both hst and ground - based images can be suitably fitted with the @xmath66 models , also when overlapping objects are considered . to test the reliability of the ground - based structural parameters , we compared the estimates obtained by fitting sersic models to the hst and the ntt images of galaxies in the sample b of table [ tab_sel ] . since both filters cover approximately the same spectral range ( rest - frame @xmath33band ) , we can rule out any difference between the two datasets due to a possible waveband dependence of the structural parameters . the structural parameters are compared in figure [ conf_hst_r ] , where we plot the same quantities as in figure [ bd_s_par ] . the medians and the standard deviations of the differences in @xmath113 , @xmath1 , @xmath58 and @xmath75 are given in table [ tab_hst_r ] . table [ tab_hst_r ] and figure [ conf_hst_r ] prove that the mean differences of the parameters have no significant biases . the scatter between the two sets of measurements is relatively small : it amounts to @xmath114 for @xmath0 , @xmath115 mag for @xmath1 , @xmath116 mag for @xmath58 and @xmath117 for the @xmath75 parameter . it is worth to stress again that the @xmath75 parameter has significantly smaller dispersion with respect to the other quantities . the comparison with the results of the previous sections proves that the differences in the parameters derived from hst and from ntt photometry are of the same order of those found when applying different fitting methods to the hst image . in other words , the difference between the structural parameters derived from hst and from ntt data is consistent with the precision with which these quantities can be measured ( see kelson et al . 2000a ; jrgensen et al . we thus conclude that reliable structural parameters can be still derived for typical sizes and luminosities of cluster galaxies at redshift @xmath71 by using ground - based data taken in ordinary observing conditions ( i.e. seeing @xmath118 and pixel scale @xmath119pxl ) . the @xmath33 and @xmath32band structural parameters for @xmath120 galaxies ( sample c of table [ tab_sel ] ) are compared in figure [ v_r_par ] and table [ tab_v_r ] . to allow a direct comparison , @xmath1 and @xmath58 of @xmath33band galaxies were corrected with the observed @xmath121 colors ( busarello et al . 2002b , in preparation ) . @xmath33 and @xmath32band parameters show no significant systematic differences and the scatter between the two sets of data is of the same order as those found in the preceding sections , i.e. equal to the typical uncertainty of structural parameter measurements . the mean ratio @xmath122 is consistent with that found in local early - type galaxies ( @xmath123 , scodeggio 2001 ; see also scodeggio et al . we defer further considerations to [ secgrad ] , where the results will be discussed in terms of color gradients . our @xmath32band structural parameters can be directly compared to those derived at @xmath124 by @xcite in the same wavelength range . the full consistency of the two data sets is apparent from figure [ phare ] ( upper panel ) where the distributions of half - light radii are compared ( the surface brightness is not shown for brevity reasons ) . the @xmath32 and @xmath6band structural parameters for the 93 galaxies in sample d ( table [ tab_sel ] ) are compared in figure [ r_k_par ] and table [ tab_r_k ] . nir @xmath0 and @xmath1 are significantly smaller than the relative optical parameters , while the @xmath2 values are systematically higher . on the average , such differences amount to -0.125 in @xmath113 , to -0.5 mag in @xmath1 and to + 0.5 in @xmath2 . a deeper inspection reveals the existence of two distinct subsamples which are separated by the value of the sersic parameter . for low @xmath2 values ( @xmath125 ) , the differences of @xmath6 to @xmath32band effective radii amount to -0.09 in @xmath113 , and the differences in the sersic index are higher ( @xmath86 instead of 0.5 ) . on the other side , for @xmath126 the sersic indices are nearly the same in the two wavebands ( @xmath127 ) while the @xmath6band half - light radii are much smaller than in the optical ( -0.2 dex ) . the confidence limits in table [ tab_r_k ] show that the separation into the two classes is highly significant . according to the kolmogorov - smirnov test , the probability that the two samples come from the same distribution is less than 2% for both @xmath128 and @xmath129 . the results of the previous sections make us confident that the systematic differences found between the structural parameters are real and do not originate from biases induced by the fitting method . no other measurements of @xmath6band ( rest - frame @xmath8 ) structural parameters are available at @xmath9 so that no direct comparison is possible . again we defer further considerations to [ secgrad ] . to make a comparison with the structural properties of nearby galaxies , we made use of the sample of 70 galaxies belonging of the coma cluster from the compilation by @xcite . the distribution of the @xmath33band @xmath0 from @xcite is shown in figure [ phare ] ( notice that our @xmath32band corresponds to the rest - frame @xmath33band ) . the radii of @xcite were scaled to @xmath3 by adopting @xmath130 , @xmath18 ( different cosmologies do not significantly alter the result ) . the distributions of nir @xmath0 are also compared in figure [ phare ] ( the radii of pahre 1999 were scaled as above ) . it is worth noticing that , although our rest - frame waveband is different from the @xmath6band in @xcite , the two wavebands actually probe the same stellar populations , so that the direct comparison has an immediate physical interpretation . the medians of the distributions in the nir are @xmath131 and @xmath132 for coma and ac118 respectively , while the corresponding values in the optical are @xmath133 and @xmath134 . we thus conclude that the median of the half - light radii at @xmath9 is fully consistent with the median at @xmath31 both in the optical and in the nir . this result indicates that the sizes of bright galaxies did not substantially change since @xmath71 to the present epoch . since the nir probes directly the mass distribution inside galaxies , and assuming that changes in mass produce corresponding changes in size , we argue that during the last @xmath135 gyr=0.3 , @xmath18 , h@xmath19 km s@xmath20mpc@xmath20 ] the bright cluster galaxies did not accrete a significant mass fraction and that ( dissipation - less ) merging did not play a major role ( see also nelson et al . the role played by the sersic index in the relation between nir and optical structural properties ( 6.1 ) suggests to discuss the combined behavior of half - light radii and sersic indices in terms of internal color gradients . the available measurements of optical - nir color gradients at intermediate redshifts concern 6 field early - type galaxies studied in @xmath137 by @xcite . unfortunately , no one of the galaxies is close enough in @xmath138 to our sample to allow a significant comparison . the comparison of recent measurements of optical color gradients at intermediate - redshifts with those in nearby galaxies shows little or no evolution since @xmath139 , supporting the interpretation in terms of metallicity gradients @xcite . we define the color gradient as @xmath140 , where @xmath141 is the surface brightness in the band @xmath142 , and the differences are computed between an inner ( @xmath143 ) and an outer ( @xmath144 ) radius . if the galaxy surface brightness in the waveband @xmath64 is described by the sersic law @xmath145 we obtain @xmath146^{-1 } \cdot \\ & \hspace{-2.5 cm } \cdot \left [ b_1 \cdot \left ( ( r_m / r_{e,1 } ) ^{1/n_1 } - ( r_m / r_{e,1 } ) ^{1/n_1 } \right ) + b_2 \cdot \left ( ( r_m / r_{e,2 } ) ^{1/n_2 } - ( r_m / r_{e,2 } ) ^{1/n_2 } \right ) \right ] \ , \end{array } \label{gmacumb}\end{aligned}\ ] ] where @xmath147 is a function of @xmath2 ( @xmath148 , see caon , capaccioli , & donofrio 1993 ) and the indices @xmath149 denote the two wavebands . in order to make a direct comparison with previous works @xcite we set @xmath150 and @xmath151 . equation [ gmacumb ] is more suitable for the present analysis than the approximation of @xcite , in that it allows to address the case of different values of @xmath2 and it is exact for arbitrary ratios of the half - light radii . we also verified that this formula gives a very close approximation to the linear fit to the color profiles . the @xmath33 , @xmath32 and @xmath6band structural parameters derived in the previous sections were substituted in equation [ gmacumb ] to derive the implied color gradients . the uncertainties on the gradients were derived from propagation of the errors on @xmath0 and @xmath2 , taking into account the covariance term . the @xmath121 and @xmath152 color gradients derived by means of equation [ gmacumb ] are shown in figure [ grad ] as a function of @xmath2 , and the relative median values are given in table [ tab_grad ] . the average optical ( @xmath121 ) color gradient is @xmath153 mag / dex , in agreement with the @xmath154 gradients derived by @xcite at @xmath1550.37 and with the optical ( @xmath156 ) gradients by @xcite at @xmath157 for cluster early - type galaxies ( see figure 6 in @xcite and tables 2 and 3 in tamura & ohta 2000 ) . the @xmath121 gradients present a weak trend with @xmath2 : galaxies with @xmath158 have larger negative gradients , with an average of -0.19 mag / dex , while for @xmath159 the average gradient is consistent with zero or slightly positive ( + 0.07 mag / dex ) . the wide majority of the galaxies have negative @xmath152 color gradients : the median value for the whole sample is @xmath160 mag / dex . the trend with @xmath2 is in this case stronger : the average @xmath152 color gradient is @xmath161 mag / dex for @xmath162 and @xmath163 mag / dex for @xmath164 . furthermore , the results of 6.1 indicate that the color gradients are mainly due to the difference in @xmath2 for @xmath158 and to the difference in @xmath0 for @xmath159 . the optical - nir color gradients are thus related to the relevance of the disk component , as it is apparent from figure [ grad ] : galaxies with lower bulge - to - disk ratio ( @xmath165 ) have large @xmath152 color gradients , while bulge - dominated galaxies have moderate gradients . in order to probe the origin of color gradients with the large wavelength baseline provided by the @xmath6band , we adopted two different models in which color gradients are driven by ( a ) ` pure age ' or ( m ) ` pure metallicity ' gradients . we used synthetic colors from the code of @xcite to simulate early type galaxy spectra . an exponential star formation rate with time scale of 1 gyr was adopted . in model ( a ) we assumed that the inner stellar population ( @xmath166 ) is 6.25 gyr old at @xmath3 this value corresponds to the mean luminosity - weighted age of stellar populations in ac118 galaxies as inferred from the color magnitude relation ( see merluzzi et al . 2001 ) . the stellar population age at the outer radius ( @xmath167 ) was adjusted to match the observed color gradients , and the metallicity was fixed to @xmath168 . in model ( m ) the inner stellar population is characterized by a metallicity @xmath169 @xcite and the observed color gradients were used to derive the outer @xmath170 value . age was fixed to 6.25 gyr . from the median values of the @xmath152 color gradients we derived the outer age and metallicity ( @xmath171 and @xmath172 ) for models ( a ) and ( m ) respectively . such values were then used to predict the @xmath121 color gradients at @xmath3 and the @xmath121 and @xmath152 gradients at @xmath14 . table [ tab_grad_th ] gives the predicted @xmath121 and @xmath152 gradients along with the gradients observed both at @xmath9 and at @xmath31 . optical - nir color gradients were measured for 12 nearby bright elliptical galaxies by peletier et al . ( 1990a , b ) . from their data we obtain an average @xmath152 gradient of @xmath173 , with a broad distribution of values ranging from @xmath174 to @xmath175 . @xcite estimated an average @xmath176 gradient of -0.18 by applying the formula of @xcite to the ratio of the relative half - light radii , while an average @xmath137 gradient of @xmath177 can be inferred in the same way from the coma cluster @xmath8 and @xmath4 effective radii given by @xcite . the results shown in table [ tab_grad_th ] may be summarized as follows . the pure age model is able to explain the evolution of the @xmath152 color gradient between @xmath9 and @xmath31 , but it fails to describe the @xmath121 gradient at @xmath9 . conversely , the pure metallicity model is able to reconcile the @xmath121 and @xmath152 color gradients at @xmath9 , but it fails to describe the evolution of the @xmath152 gradient as a function of look - back time . in this study we derived optical and nir structural parameters for galaxies belonging to the cluster ac118 at @xmath3 from ground - based photometry . the present data set constitutes the first large sample of structural parameters determined at intermediate redshifts in different wavebands and , in particular , it is the first such sample in the near - infrared . the samples consist in @xmath178=93 objects studied in the @xmath5 and @xmath30 wavebands and in @xmath178=58 galaxies studied in the @xmath33band . the reliability of the data was verified by comparison with the results from hst photometry for a subsample of galaxies : the scatter between measurements from hst- and from ground - based photometry is equivalent to the scatter between parameters determined with two different fitting methods from hst images , namely of @xmath1635% in @xmath0 . given that these values of the scatter are generally claimed to be a measure of the ` intrinsic uncertainty ' in structural parameter measurements , we can state that reliable structural parameters can be still derived from ground - based photometry in ordinary observing conditions at least up to @xmath9 . since morphology was not a selection criterion in this study , some remarks are needed for the comparison of our results with other works . @xcite established that @xmath179 of the galaxies in the core of ac118 are early - types while @xmath180 are spirals . they also found a significant rise in the fraction of s0 galaxies with respect to the local density - morphology relation . morphological classification from @xcite is available for 21 galaxies in sample d. the number of spirals is 5 , of which 3 have @xmath1811.5 while the other two have @xmath182 . only one of the spirals has ` late - type ' colors , while the others lie on the red sequence of ac118 . it is interesting to notice that all the remaining galaxies with ` late - type ' colors lie outside of the cluster core and have @xmath183 . they are likely disk galaxies dominated by young stellar populations . these facts are in agreement with a scenario in which spirals are accreted into the cluster from the field , cease to form stars , become gas - deficient objects and eventually undergo a morphological transformation @xcite . the remaining galaxies with morphological information consist in 10 s0 , eight of which with @xmath184 and two with @xmath185 , and 6 ellipticals , for which we find @xmath182 . we stress that the galaxies with large disks and late - type colors are only @xmath186 of galaxies in sample d. the main feature concerning the multi - band properties of galaxies is that these are much more concentrated in the nir than in the optical . the mean ratio of nir to optical half - light radii ( @xmath187 ) is not different from that found in local samples of early - type galaxies ( @xmath106 , e.g. pahre 1999 ) , but the situation is complicated by the systematic differences found in the shape of the brightness distribution : galaxies with low @xmath2 value have a light profile more peaked in the center in the @xmath30-band ( @xmath188 ) , while for @xmath182 a lower @xmath189 is observed ( @xmath190 ) . such trends can be interpreted as due to the increase of the disk fraction when @xmath2 approaches unity . to discuss the combined effects of half - light radii and sersic indices , we derived the optical and nir color gradients by means of equation 4 . most of galaxies in ac118 do not show significant optical color gradients , with a median value of @xmath153 . this result is consistent with previous estimates of optical color gradients at intermediate redshifts @xcite . conversely , strong negative gradients are found in optical nir colors : the median of the @xmath152 gradients amounts to @xmath160 . moreover , this effect depends on the shape of the light profile : the @xmath152 gradient increases from @xmath191 for galaxies with low @xmath2 value to @xmath192 for @xmath182 . again this result can be explained by an increase in the disk fraction when @xmath165 . to interpret these results in terms of the properties of the stellar populations , we introduced a pure age and a pure metallicity model . we used the @xmath152 gradients to calibrate both models and to predict @xmath121 color gradients at @xmath3 . moreover , the evolution of color gradients as inferred by the models was compared with the measurements by @xcite for a sample of nearby elliptical galaxies . the age model is not able to reconcile our optical and nir estimates at @xmath3 since a significant negative gradient is predicted in @xmath121 ( @xmath193 ) . on the other hand , the predicted color gradient agree with the measurements at @xmath194 . conversely , the pure metallicity model gives correct results at @xmath3 but predicts too strong nir gradients at @xmath194 . since the age model is able to describe the evolution of color gradients , whereas the metallicity model correctly explains the dependence of color gradients on wavelength , we argue that a model in which effects of age and metallicity are consistently combined could explain optical and nir color gradients from @xmath3 up to @xmath194 . finally , we remark that dust likely affects the optical - nir color gradients . dust is actually presents in the core of most early - type galaxies @xcite and is known to produce high central color gradients in bulges of s0 and spiral galaxies ( e.g. peletier et al . 1999 ) . the distributions of both the optical and nir half - light radii of galaxies at @xmath9 are found to be consistent with those of coma early - type galaxies . the result for the nir waveband is the most significant , since the nir probes directly the mass distribution inside galaxies . the observed distributions thus indicate that the sizes of bright galaxies did not substantially change since @xmath71 to the present epoch . this result suggests that ( dissipation - less ) merging did not play a major role in galaxy evolution during the last @xmath135 gyr ( see also nelson et al . 2001 ) . we are grateful to adam stanford who provided us with his own photometry of ac118 , that was used to check the calibration of our @xmath5-band data . we thank the referee for his / her helpful comments and suggestions and dr . m. donofrio helpful discussions . the observations at eso were collected during the guaranteed time of the osservatorio astronomico di capodimonte . this work is partially 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 . michele massarotti is partly supported by a murst - cofin grant . ac118 was observed on november 1994 with the hst wide - field planetary camera 2 in the f702w filter . two exposures of 2300 s and one exposure of 1900 s for a total integration time of 6500 s are available for a single pointing . the hst images were calibrated into the johnson @xmath32band by means of the ntt photometry . we proceeded as follows . the frames were smoothed to the same resolution of ground - based images by a convolution with the ntt @xmath32band psf . aperture photometry was then performed with the sextractor software @xcite in the same aperture of the ground data . figure [ hstcalib ] shows the difference of @xmath5 and f702w instrumental magnitudes versus @xmath121 colors . the data points were fitted with the equation : @xmath195 we found @xmath196 and @xmath197 , and that the root mean square of the residuals to the fitted relation varies from 0.05 mag for @xmath198 mag to 0.1 mag for the whole sample . the few galaxies ( 9% ) which deviate more than @xmath199 mag from the fit , turned out to be objects contaminated by bright neighbors . in the same figure we also plot the synthetic transformation to the @xmath5 waveband of holtzman et al . ( 1995 , see their equation 9 and table 10 ) : @xmath200 where @xmath201 is the exposure time , @xmath202 is the gain ratio of the wfpc2 chips , @xmath203 is an aperture correction term derived from the encircled energy measurements of @xcite , and @xmath204 is the long exposure correction term from hill et al . the same gain ratio @xmath205 was adopted for each chip , since the effects of gain differences are very small ( @xmath206 mag ) when compared to the other sources of uncertainty . we adopted @xmath207 mag and @xmath208 mag . a further shift of @xmath209 mag was needed to better match equation [ syneq ] to our data . this is in agreement with previous results of calibration of hst images with ground - based photometry ( see jrgensen et al . the psf of ntt ground - based images was described by the following model : @xmath210 \ ! r \ , ( x , y ) \ \ , \label{psf_eq}\end{aligned}\ ] ] where @xmath48 is the psf brightness distribution , @xmath211 are two - dimensional gaussian functions and @xmath5 is a box function equal to 1 on the pixel area and 0 elsewhere . the convolution in equation [ psf_eq ] was computed from the one - dimensional convolution of a gaussian with a rectangular function . for the one - dimensional case , the following expression is obtained : @xmath212 + erf \left [ \frac{a/2-x } { \sqrt{2 } \sigma } \right ] \right\ } \ \ , \label{gausrec}\end{aligned}\ ] ] where g(x ) is is a normal gaussian function of width parameter @xmath21 , @xmath213 is a rectangle of width @xmath214 , and @xmath215 is the error function . the analytical expression of the ground - based psf is obtained by substituting the formula [ gausrec ] in equation [ psf_eq ] . galaxy structural parameters were derived by minimizing the right side of equation [ chieq ] , i.e. by solving the following system of equations : @xmath216 where @xmath45 is a function of the fitting parameters @xmath217 , @xmath218 are the brightness values of image pixels , and @xmath219 are the @xmath220 parameters for the ground - based data . the changes @xmath221 , @xmath222 of @xmath218 and @xmath219 determine variations @xmath223 of the solutions @xmath224 of equations [ nlsist ] . at the first order , @xmath223 are obtained from the following formulae : @xmath225 where the symbol @xmath226 denotes matrix multiplication , @xmath227 , @xmath228 and @xmath229 are column vectors of components @xmath230 , @xmath231 and @xmath232 respectively , @xmath233 is the inverse of the @xmath45 hessian matrix , @xmath234 and @xmath7 are rectangular matrices of components @xmath235 and @xmath236 respectively . all quantities are computed at @xmath237 , @xmath238 and @xmath239 . + the noise of image pixels and the uncertainties on psf parameters ( see [ psfsec ] ) were used to estimate the quantities @xmath221 and @xmath222 . uncertainties on fitting parameters were then derived by equations [ err ] . allen , s.w . 2000 , , 315 , 269 andreon , s. 2001 , , 547 , 623 barger , a.j . , aragn - salamaca , a. , ellis , r.s . , couch , w.j . , smail , i. , & sharples , r.m . 1996 , , 279 , 1 bertin , e. , arnouts , s. 1996 , , 117 , 393 bower , r.g . , lucey , j.r . , & ellis , r.s . 1992 , , 254 , 601 bruzual , g.a . , & charlot , s. 1993 , , 405 , 538 caon , n. , capaccioli , m. , & donofrio , m. 1993 , , 265 , 1013 capaccioli , m. , caon , n. , & donofrio , m. 1992 , , 259 , 323 couch , w.j . , barger , a. j. , smail , i. , ellis , r. s. , & sharples , r. m. 1998 , , 497 , 188 djorgovski , s. , & davies , m. 1987 , , 313 , 59 dresseler , a. , lynden - bell , d. , burstein , d. , davies , r.l . , faber , s.m . , terlevich , r.j . , & wegner , g. 1987 , , 313 , 42 goudfrooij , p. , & de jong , t. 1995 , , 298 , 784 govoni , f. , enlin . t.a . , feretti , l. , & giovannini , g. 2001 , , 369 , 441 hamabe , m. , & kormendy , j. 1987 , iau symp . t. de zeeuw , dordreech : reidel , 379 hill , r. j. 1998 , , 496 , 648 hinkley s. , & i m , m. 2001 , , 560 , l41 holtzman , j. a. et al . 1995 , , 107 , 156 jrgensen , i. , franx , m. , & kjrgaard , p. 1995 , , 273 , 1097 jrgensen , i. , franx , m. , & kjrgaard , p. 1996 , , 280 , 167 jrgensen , i. , franx , m. , hjorth , j. , & van dokkum , p. g. 1999 , , 308 , 833 kauffmann , g. , & charlot , s. 1998 , , 294 , 705 kelson , d.d . , illingworth , g.d . , van dokkum , p.g . , & franx , m. 2000a , , 531 , 184 kelson , d.d . , illingworth , g.d . , van dokkum , p.g . , & franx , m. 2000b , , 531 , 137 kodama , t. , & arimoto , n. 1997 , , 320 , 41 kodama , t. , arimoto , n. , barger , a.j , & aragn - salamanca , a. 1998 , , 334 , 99 kormendy , j. 1977 , , 218 , 333 krist , j. 1999 , `` the tiny tim user s guide '' , version 5.0 , stsci , baltimore lubin , l. m. , & sandage , a. 2001 , , 121 , 2289 massarotti , m. , iovino , a. , & buzzoni , a. 2001 , , 368 , 74 merluzzi , p. , busarello , g. , massarotti , m. , & la barbera , f. 2001 , in asp conf . ser . , in press ( astro - ph/0110328 ) nelson , a.e . , simard , l. , zaritsky , d. , dalcanton , j.j . , & gonzalez , a.h . 2001 , preprint ( astro - ph/0110582 ) pahre , a.m. , de carvalho , r.r . , & djorgovski , s.g . 1998 , , 116 , 1606 pahre , m.a . 1999 , , 124 , 127 peletier , r.f . , davies , r.l . , davies , l.e . , illingworth , g.d . , & cawson , m. 1990a , , 100 , 1091 peletier , r.f . , valentijn , e.a . , & jameson , r.f . 1990b , , 233 , 62 peletier , r.f . , vazdekis , a. , arribas , s. , del burgo , c. , garca - lorenzo , b. , gutirrez , c. , mediavilla , e. , & prada , f. 1999 , , 310 , 863 press , w.h . , flannery , b.p . , teukolsky , s.a . , & vetterling , w.t . 1986 , numerical recipes ( cambridge : cambridge univ . press ) ratnatunga , k.u . , griffiths , r.e . , & ostrander , e.j . 1999 , , 118 , 86 saglia , r.p . , bertschinger , e. , baggley , g. , burstein , d. , colless , m. , roger , l.d . , mcmahan , r.k . , jr . , & wegner , g. 1997 , , 109 , 79 saglia , r.p . , maraston , c. , greggio , l. , bender , r. , & ziegler , b. 2000 , , 360 , 911 schade , d. , lilly , s.j . , le fevre , o. , hammer , f. , & crampton , d. 1996 , , 464 , 79 scodeggio , m. , gavazzi , g. , belsole , e. , pierini , d. , & boselli , a. 1998 , , 301 , 1001 scodeggio , m. 2001 , , 121 , 2413s sersic , j.l . 1968 , atlas de galaxias australes , observatorio astronomico , cordoba simard , l. , koo , d.c . , faber , s.m . , sarajedini , v.l . , vogt , n.p . , phillips , a.c . , gebhardt , k. , illingworth , g.d . , & wu , k.l . 1999 , , 519,563 smith , r.j . , lucey , j.r . , hudson , m.j . , & steel , j. 1997 , , 291 , 461 sparks , w.b . , & jrgensen , i. 1993 , , 105 , 1753 stanford , s.a . , eisenhardt , p.r.m . , & dickinson , m. 1998 , , 492 , 461 tamura , n. , kobayashi , c. , arimoto , n. , kodama , t. , & ohta , k. 2000 , , 119 , 2134 tamura , n. , & ohta , k. 2000 , , 120 , 533 van dokkum , p.g . , & franx , m. 1996 , , 281 , 985 van dokkum , p.g . , & franx , m. 2001 , , 553 , 90 williams , l.l.r . , navarro , j.f . , & bartelmann m. 1999 , , 527 , 535 wu , x - p . , xue , y - j . , & fang l - z . 1999 , , 524 , 22 ziegler , b.l . , & bender , r. 1997 , , 291 , 527 ziegler , b.l . , saglia , r.p . , bender , r. , belloni , p. , greggio , l. , & seitz , s. 1999 , , 346 , 13 cccccc v & ntt - emmi & 2700 & 9.@xmath24015@xmath2418.@xmath2406 & 0.267 & 1.2 + r & & 1800 & & & 1.0 + k & ntt - sofi & 15900 & 5@xmath2425@xmath240 & 0.292 & 0.8 + f702w & hst - wfpc2 & 6500 & @xmath243 & 0.1 & + [ ac118_data ] lcccc & & & & + wavebands & @xmath244 & @xmath5 , @xmath244 & @xmath4 , @xmath5 & @xmath5 , @xmath30 + & & & & + phot . redshift & @xmath245 & @xmath245 & @xmath246 & @xmath246 + & & & & + selection & @xmath43 , @xmath84 & @xmath43 , @xmath84 & @xmath43 , @xmath247 , @xmath84 & @xmath43 , @xmath84 + & & & & + n & 50 & 37 & 58 & 93 + & & & & + fitting models & @xmath39 , @xmath66 & @xmath66 & @xmath66 & @xmath66 + & & & & + cccccccccccc & & & & & & & & & & & + -1 & @xmath248 & @xmath249 & & @xmath250 & @xmath251 & & @xmath252 & @xmath253 & & @xmath254 & @xmath255 + & & & & & & & & & & & + 0 & @xmath256 & @xmath257 & & @xmath258 & @xmath259 & & @xmath260 & @xmath261 & & @xmath262 & @xmath263 + & & & & & & & & & & & + 1 & @xmath264 & @xmath265 & & @xmath266 & @xmath267 & & @xmath268 & @xmath269 & & @xmath270 & @xmath271 + & & & & & & & & & & & + -1,0,1 & @xmath272 & @xmath273 & & @xmath274 & @xmath275 & & @xmath276 & @xmath277 & & @xmath278 & @xmath279 + & & & & & & & & & & & + lcc & & + @xmath280 & @xmath281 & @xmath282 + & & + @xmath283 & @xmath284 & @xmath285 + & & + @xmath286 & @xmath287 & @xmath288 + & & + @xmath289 & @xmath290 & @xmath291 + & & + @xmath292 & @xmath293 & @xmath294 + & & + @xmath295 & @xmath296 & @xmath297 + & & + lcc & & + @xmath280 & @xmath298 & @xmath299 + & & + @xmath283 & @xmath300 & @xmath301 + & & + @xmath286 & @xmath302 & @xmath303 + & & + @xmath289 & @xmath304 & @xmath305 + & & + @xmath292 & @xmath306 & @xmath279 + & & + @xmath295 & @xmath307 & @xmath308 + & & + lcccccccccccc & & & & & & & & & & & & + @xmath280 & & @xmath309 & & @xmath310 & & @xmath311 & & @xmath312 & & @xmath313 & & @xmath314 + & & & & & & & & & & & & + @xmath283 & & @xmath315 & & @xmath316 & & @xmath317 & & @xmath318 & & @xmath319 & & @xmath320 + & & & & & & & & & & & & + @xmath286 & & @xmath321 & & @xmath322 & & @xmath323 & & @xmath324 & & @xmath325 & & @xmath326 + & & & & & & & & & & & & + @xmath289 & & @xmath327 & & @xmath328 & & @xmath329 & & @xmath330 & & @xmath331 & & @xmath332 + & & & & & & & & & & & & + @xmath292 & & @xmath333 & & @xmath334 & & @xmath335 & & @xmath336 & & @xmath337 & & @xmath338 + & & & & & & & & & & & & + @xmath295 & & @xmath339 & & @xmath340 & & @xmath341 & & @xmath342 & & @xmath343 & & @xmath344 + & & & & & & & & & & & & + & & & + age & -0.26 & -0.04 & -0.49 & -0.13 + @xmath349 & & & & + & & & + metallicity & -0.15 & -0.10 & -0.49 & -0.41 + @xmath350 & & & & + & & & & + observed & -0.1 & -0.02 & -0.49 & -0.14 + & & & & +

we take hydrodynamic initial conditions in relativistic heavy ion collisions from the color glass condensate ( cgc ) picture through the factorization formula . gluon distributions produced from the cgc are found to provide good initial conditions for the hydrodynamic simulations in au + au collisions at relativistic heavy ion collider ( rhic ) energies . we reproduce the centrality , rapidity , and energy dependences of multiplicity within this approach . we also investigate the energy loss of high partons in the dense thermalized medium created from colliding two cgc s . we find that our results on the centrality dependence of nuclear modification factors for pions and back - to - back correlation for charged hadrons at midrapidity are consistent with the rhic data up to semicentral events . whereas , our approach in which jets are calculated from perturbative qcd processes predicts less suppression at forward rapidity region compared to the brahms data in au+au collisions at rhic .

nucl-th0404039

experiments of high energy @xmath3 , @xmath4 , and @xmath5 collisions are currently performed at the relativistic heavy ion collider ( rhic ) in brookhaven national laboratory for comprehensive understanding of the highly hot / dense matter , the quark gluon plasma ( qgp ) . at collider energies , collisions of two relativistic nuclei involve many aspects of physics according to the relevant energy or time scale . there already exist many theoretical approaches to understand numerous rhic data . we consider in this work , particularly , the physics of the gluon saturation in a colliding nucleus , hydrodynamic evolution of produced matter , and the energy loss of hard partons in the medium . our goals are to combine them and to take a first step toward a unified understanding of the dynamical aspect of high energy heavy ion collisions . first data reported by the star collaboration at rhic @xcite revealed that the observed large magnitude of elliptic flow for hadrons is naturally explained by hydrodynamics . this suggests that large pressure possibly in the partonic phase is built at the very early stage ( @xmath6 fm/@xmath7 ) in au+au collisions at rhic . this is one of the strongest indications of early thermalization of the qgp at rhic @xcite . these hydrodynamic predictions also give good agreements with mass dependences of the second harmonic coefficient of azimuthal distribution @xmath8 in the low @xmath1 region @xcite . this is in contrast to the case in the lower energy collisions where the hydrodynamics always overpredicts the data @xcite . hydrodynamics also predicts that the scaled elliptic flow , which is defined as the second harmonics @xmath9 divided by initial spatial eccentricity @xmath10 , becomes almost constant around 0.2 @xcite . the experimental data reaches the hydrodynamic limit for the first time in central and semicentral collisions at rhic energies @xcite . moreover , @xmath9 as a function of pseudorapidity @xcite can be reproduced by hydrodynamics only in @xmath11 @xcite . all these analyses indicate that a high dense matter created in heavy ion collisions at rhic energies achieves ( local ) thermal equilibrium state in the low @xmath1 region , around midrapidity , and up to semicentral collisions . on the other hand , jet production rate becomes large enough to measure high @xmath1 power - law tails at rhic . these minijets go across the expanding matter rather than participate in the dense medium in the case of heavy ion collisions . during traveling through the expanding medium , energetic partons interact with soft matter and lose their energies dynamically ( jet quenching ) . thus , high @xmath1 hadrons coming from fragmentation of minijets can be good probes of the bulk matter @xcite . it was observed that nuclear modification factors for hadrons in the high @xmath12 gev/@xmath7 region are considerably smaller than unity in central collisions @xcite and that it increases with decreasing centrality and eventually reaches around unity in peripheral collisions . disappearance of the away - side peak in azimuthal correlation functions for high @xmath1 hadrons is also observed in central au+au collisions @xcite . this can be described by perturbative qcd ( pqcd ) based parton models with parton energy loss in the dense medium @xcite . cronin enhancement of the hadron spectra and existence of the back - to - back correlation at mid - rapidity in @xmath4 collisions at rhic @xcite support that suppression of the yield and disappearance of the away - side peak in au+au collisions at mid - rapidity are due to the final state interaction . the radiative parton energy loss induced by the medium has been studied extensively @xcite and found that the energy loss is proportional to the gluon rapidity density of the medium . theoretical analyses reveal that a large parton density is required to account for observed suppressions of single as well as dihadron spectra at rhic @xcite . from the above discussion , the current rhic data strongly suggest that the parton density created in heavy ion collisions at rhic is dense enough to cause both large elliptic flow of bulk matter and large suppression of high-@xmath1 hadrons . one of the important problems in the physics of heavy ion collisions is to reveal the origin of this dense matter . the bulk particle production in high energy hadronic collisions is dominated by the small @xmath13 modes in the nuclear wave function , where @xmath13 is a momentum fraction of the incident particles . therefore , an origin of the large density could be traced to the initial parton density at small @xmath13 inside the ultra - relativistic nuclei before collisions @xcite . it is well known that gluon density increases rapidly with decreasing @xmath13 by the bfkl cascade until gluons begin to overlap in phase space @xcite where nonlinear interaction becomes important . these gluons eventually form the color glass condensate ( cgc ) @xcite . this phenomenon is characterized by a saturation scale " @xmath14 which corresponds to the gluon rapidity density per unit transverse area . when @xmath14 is large ( @xmath15 ) , the coupling becomes weak ( @xmath16 ) . in addition , the gluon occupation number in the wave function is huge , @xmath17 . therefore this can be studied by an weak coupling classical method known as mclerran - venugopalan ( mv ) model @xcite . it has been developed to compute the classical parton distributions of nuclei based on this model . moreover , renormalization group methods which systematically incorporate quantum corrections to the classical effective theory were developed @xcite . there is an evidence for gluon saturation in deep inelastic scattering at hera . models based on the saturation picture fit the hera data well in the moderate @xmath18 and low @xmath13 @xcite . this effective theory can be applied to the collisions of nuclei and its solution can be used for initial conditions of further evolution of the system created in heavy ion collisions @xcite . perturbative solutions of the gluon production for collisions of two nuclei @xcite as well as @xmath19 collisions @xcite have been computed . non - perturbative classical solutions of the yang - mills equations were also obtained by the real - time numerical simulations on the lattice @xcite . the calculations based on the cgc picture have been compared to various rhic data . remarkably , the cgc results on the global observables , e.g. , the centrality , rapidity and energy dependences of charged hadron multiplicities agree with the rhic data under the assumption of parton - hadron duality @xcite . the cgc picture is also applied for studies of particle correlations and azimuthal anisotropies @xcite , and valence quark distributions at small-@xmath13 @xcite . it has been shown that the classical wave function in the mv model contains cronin enhancement and that quantum evolution with respect to @xmath13 makes the spectrum suppressed @xcite . suppression of nuclear modification factor at forward rapidity region in @xmath4 collisions @xcite in the cgc picture is also consistent with the recent rhic data @xcite . the above facts show that the cgc , hydrodynamics , and the energy loss of hard partons are the key ingredients to describe the rhic physics . one expects that they are indeed closely related with each other . for example , the cgc could be a good initial condition for a thermalized state because it produces a large number of gluons . thus a large number of gluons are responsible for large suppression of jet spectra . in order to understand systematically the whole of the stages of relativistic heavy ion collisions , it is indispensable to construct a unified and dynamical framework in which the above different ideas are consistently incorporated . in this work , we perform hydrodynamic simulations with the initial condition taken from the cgc and simulate parton energy loss in the fluid elements . this approach will lead us to get deeper understanding of the dynamical aspect of heavy ion collisions . in addition , some of the problems which are inherent in a particular approach can be removed . for instance , one conventionally parametrizes initial conditions in the hydrodynamic simulations shortly after a collision of two nuclei so as to reproduce final observed spectra . hydrodynamic framework itself does not provide an initial condition . hence one has to choose it from infinite possible candidates . any initial conditions which reproduce the final hadron spectra are equivalent , so it is hard to discard even one of them within the hydrodynamic approach . this is the ambiguity problem of hydrodynamic model in heavy ion collisions . a good example can be found in ref . @xcite : two completely different initial conditions end up almost the same rapidity distribution . therefore it is desired to take an initial condition which is obtained by a reliable theory . on the other hand , most of the calculations based on the cgc do not include final state interactions . according to estimation of initial parton productions from a classical lattice yang - mills simulation based on the cgc @xcite , the transverse energy per particle is roughly @xmath20 gev , while the experimental data yields @xmath21 gev @xcite . the elliptic flow from a classical yang - mills simulation is inconsistent with rhic data @xcite . those discrepancies , which are mainly due to the lack of collective effects , will be removed by introducing further time evolutions through , e.g. , hydrodynamics . for the calculation of parton energy loss , one needs time evolution of parton density . bjorken expansion @xcite is assumed for the evolution of parton density in almost all work except ref . @xcite in which hydrodynamic simulations are used to estimate the energy loss of partons for the first time . we have previously established a model ( the hydro+jet model ) in which hydrodynamic evolutions are combined with explicitly traversing non - thermalized partons with large transverse momenta @xcite . here the production of high @xmath1 partons is described by a pqcd parton model . these high @xmath1 partons come from parton distribution function in relatively large @xmath13 region . systematic studies based on the hydro+jet model have been performed for @xmath1 spectra @xcite , back - to - back correlation functions @xcite , pseudorapidity dependence of nuclear modification factors @xcite , and elliptic flow @xcite . for the initial condition of hydrodynamic simulations , we simply parametrized the initial shape of energy density in the previous work . instead , we will employ the cgc picture and use gluons produced from melting cgc for initial conditions for hydrodynamic simulations in this paper . the number and the energy density distributions for produced gluons are calculated from the @xmath0 factorization formula @xcite . although the produced gluons will reach a thermalized state through the dissipative processes in the realistic situations , the description of non - equilibrium phenomena is beyond the scope of the present paper . instead , we assume thermalization for gluons produced from the cgc . this paper is organized as follows . in sec . [ section : model ] , we discuss how the gluon distribution produced from the cgc is used for the initial condition in the hydrodynamics . in sec . [ sec : hydrojet ] , we briefly summarize the hydro+jet model . in sec . [ sec : result ] , we present hydrodynamic results on ( pseudo)rapidity distributions for charged hadrons and the transverse momentum distributions for pions , kaons , and protons . we show that the cgc provides good initial conditions in the hydrodynamic model . we also show that , with almost keeping the shape of rapidity distribution during the hydrodynamical evolution , transverse collective flow is generated by pressure gradient perpendicular to the collision axis . we study the centrality dependence of the nuclear modification factor for neutral pions and the back - to - back correlations for charged hadrons . the previous analysis on the high @xmath1 hadron spectra in the forward rapidity region is revisited . in the final section , we give a summary and a discussion about the further improvements in our approach . there exists a lot of extensive work for the description of gluon productions in nuclear collisions in the saturation regime where nonlinear effects become important . perturbative solutions for the collision of two nuclei in the mv model were obtained in refs . however , those solutions are limited for the relatively high momentum and can not be used for the calculation of total gluon multiplicities . in ref . @xcite , analytic expression for the gluon multiplicity is obtained . non - perturbative classical boost invariant solutions are currently available on the lattice @xcite and it is highly desired to use the lattice results for obtaining reliable initial conditions . however , one of our purposes in this paper is to investigate the rapidity dependence of final hadrons . therefore , we employ the @xmath0 factorization formula along the line of work done by kharzeev , levin , and nardi ( kln ) @xcite for the computation of the gluon rapidity distribution which is to be the initial condition for sequential hydrodynamic evolution . the number of produced gluons in the @xmath0-factorization formula is given by @xcite @xmath22 where @xmath23 and @xmath24 and @xmath1 are a rapidity and a transverse momentum of a produced gluon . running coupling @xmath25 is evaluated at the scale @xmath26 . the unintegrated gluon distribution @xmath27 is related to the gluon density of a nucleus by @xmath28 there are several models / parametrizations in the literature which provide a gluon distribution in the saturation region . in principle , the @xmath13 dependence of unintegrated distribution functions in eq . ( [ eq : ktfac ] ) should be solutions to the nonlinear quantum evolution equations such as balitsky - kovchegov equation @xcite or , more generally , jimwlk equation @xcite . leaving this analysis for the future work , we follow basically the kln approach which captures the essential features of the gluon saturation physics in a simple manner . in the mv model @xcite , the gluon distribution below the saturation scale @xmath14 is logarithmically suppressed @xmath29 compared to the perturbative distribution @xmath30 . instead , motivated by the kln approach , we use a simplified assumption about the unintegrated gluon distribution function : @xmath31 where @xmath32 . we introduce a small regulator @xmath33 gev/@xmath7 in order to have a smooth distribution in the forward rapidity region @xmath34 at rhic . other regions are not affected by introducing a small regulator . as we will discuss below , the above distribution depends on the transverse coordinate @xmath35 through @xmath14 . we are interested in the centrality and rapidity dependences of total yields which are dominated by the low transverse momenta . the above simple form should be fine for this purpose , although the effect of anomalous dimension resulting from the presence of the cgc modifies the @xmath1 dependence of the produced gluons as shown in ref . an overall constant factor @xmath36 is determined by fitting the multiplicity of charged hadron at midrapidity at @xmath37 gev for the most central collision . saturation momentum of a nucleus @xmath38 in @xmath39 collisions is obtained by solving the following implicit equation with respect to @xmath40 at fixed @xmath13 and @xmath41 @xmath42 where @xmath43^b \right\}.\ ] ] the number of participants is given by @xmath44 we take a simple perturbative form for the gluon distribution for a nucleon @xmath45 where @xmath46 gev . @xmath47 is used to control saturation scale in eq . ( [ eq : saturation ] ) @xcite . we choose @xmath48 for @xmath49 ( @xmath50 for @xmath51 or @xmath52 for @xmath53 ) so that average saturation momentum in the transverse plane yields @xmath54 gev@xmath55 in au+au collisions at impact parameter @xmath56 at rhic . similar to the kln approach , @xmath57 dependence of the saturation scale is motivated by the golec - biernat wsthoff model @xcite . the factor @xmath58 shows that gluon density becomes small at @xmath59 . @xmath60 usually depends on the scale @xmath18 . here we take @xmath61 as in the kln approach @xcite . we obtain the rapidity distribution for produced gluons at each transverse point @xmath35 by performing the integral of eq . ( [ eq : ktfac ] ) numerically . the transverse energy distribution @xmath62 is also obtained by weighting the transverse momentum of gluons @xmath1 in the integration with respect to @xmath1 in eq . ( [ eq : ktfac ] ) . we cut off the integral range of @xmath1 in eq . ( [ eq : ktfac ] ) since only the low @xmath1 partons are assumed to reach the local thermal equilibrium . instead , hard partons are included by assuming usual pqcd @xmath2 processes in which dglap evolution is included as it is important in the high @xmath18 reactions . it is unclear which value we have to use for the cut off momentum @xmath63 , because it contains a nonperturbative nature and also it would depend on the choice of unintegrated wave function . we set the cut off momentum at @xmath64 gev/@xmath7 which corresponds roughly to the maximum saturation scale at @xmath65 at the origin @xmath66 in central au+au collisions at rhic . gev at @xmath67 fm . ( b ) gluon number and energy densities as functions of a transverse coordinate . parameters are @xmath68 , @xmath48 , and @xmath69 . as we will discuss in the next subsection , these parameters correspond to an initialization of hydrodynamic simulations which matches the number density of gluons produced by collisions of two cgc s with the parton distribution in hydrodynamic simulations at initial time . the value of @xmath70 is chosen so that we reproduce the multiplicity of final hadrons as discussed in sec . iv . in this procedure , the transverse energy distributions represented by dashed lines are not used as hydrodynamic initial conditions . in the next subsection , we will also discuss the other matching procedure in which the transverse energy distribution is used instead of the gluon number density . in that case , the gluon distributions are factor 1.6 smaller than the results in these figures . , title="fig:",width=316 ] gev at @xmath67 fm . ( b ) gluon number and energy densities as functions of a transverse coordinate . parameters are @xmath68 , @xmath48 , and @xmath69 . as we will discuss in the next subsection , these parameters correspond to an initialization of hydrodynamic simulations which matches the number density of gluons produced by collisions of two cgc s with the parton distribution in hydrodynamic simulations at initial time . the value of @xmath70 is chosen so that we reproduce the multiplicity of final hadrons as discussed in sec . iv . in this procedure , the transverse energy distributions represented by dashed lines are not used as hydrodynamic initial conditions . in the next subsection , we will also discuss the other matching procedure in which the transverse energy distribution is used instead of the gluon number density . in that case , the gluon distributions are factor 1.6 smaller than the results in these figures . , title="fig:",width=316 ] figure [ fig : dndy_g ] ( a ) shows rapidity distributions of produced gluons obtained from eq . ( [ eq : ktfac ] ) . assuming free streaming in the longitudinal direction until @xmath71 fm/@xmath7 , we also obtain the number density and the energy density as functions of a transverse coordinate as shown in fig . [ fig : dndy_g ] ( b ) . here the direction of the @xmath13-axis is the same as that of the impact parameter vector . solid ( dashed ) line corresponds to the number ( energy ) distribution of produced gluons . regardless of the range of the parameter @xmath49 - 0.3 , we find @xmath0 factorization formula gives almost the same shape of rapidity distribution . the initial transverse energy per particle is estimated to be @xmath72 gev at @xmath73 . this is within a range estimated in a numerical simulation of the classical yang - mills equation @xcite . in order to obtain a correct initial condition for hydrodynamics , one needs non - equilibrium description for the collisions of heavy nuclei . especially , whether thermalization is achieved within a few fm/@xmath7 is a longstanding problem in the physics of heavy ion collisions . the boltzmann equation with only the elastic parton - parton scattering predicts that the system never reaches a thermal state within a reasonable time scale in heavy ion collisions @xcite . inelastic processes such as @xmath74 are quite responsible for the equilibration of the system @xcite . although the thermalization process can alter initial conditions , the centrality dependence of total hadron multiplicities is approximately preserved in the cgc initial conditions @xcite . qcd plasma instabilities may also play an important role in the thermalization of matter from anisotropic initial conditions and these may help to speed up the equilibration of the plasma @xcite . a question of chemical equilibration has been addressed in refs . @xcite . leaving these problems for the future work , we assume that the system of gluons initially produced from the cgc reaches a kinematically as well as chemically local equilibrate state at a short time scale . we further assume that during being thermalized , the shape of the rapidity distribution is not changed . thus , we take initial conditions from gluon distribution obtained in the previous subsection based on the cgc . we consider two simple ways to provide initial conditions from the cgc , i.e. , matching of number density ( ic-@xmath60 ) and matching of energy density ( ic-@xmath75 ) . we will compare charged particle distributions from these initial conditions with each other in sec . [ sec : dndy ] . assuming bjorken s ansatz @xmath76 @xcite where @xmath77 is the space - time rapidity @xmath78 , we obtain the number density and the energy density for gluons at a space - time point @xmath79 from eq . ( [ eq : ktfac ] ) @xmath80 we take two steps to specify initial conditions in hydrodynamic simulations by using eqs . ( [ eq : g_density ] ) or ( [ eq : et_density ] ) . firstly , we assume that these partons are thermally equilibrated . secondly , the gluon number density simply represents the @xmath81 parton number density , i.e. , the qgp . although the initial time @xmath82 could depend on incident energies ( or even on the space coordinate through the saturation scale ) , we assume a common @xmath82 for @xmath83 and 200 gev . we note that final particle multiplicity is not sensitive to the initial time @xmath82 within a range @xmath84 fm/@xmath7 compared to the energy dependence of the number of initial gluons . let us recall the relations among thermodynamical variables , i.e. , energy density @xmath75 , temperature @xmath85 , and number density @xmath60 for the massless free parton system : @xmath86 where @xmath87 , @xmath88 , @xmath89 , and @xmath90(= 486 mev/@xmath91 ) is a bag constant which is fixed by matching pressure between the qgp phase and the hadron phase . thus , we obtain the initial condition in the three - dimensional ( 3d ) space from the number density ( ic-@xmath60 ) : @xmath92 or , from the energy density ( ic-@xmath75 ) @xmath93}{47.5\pi^2}\right\}^{1/4}.\ ] ] when the temperature at ( @xmath82 , @xmath94 ) is below the critical temperature @xmath95 mev , the all thermodynamic variables in the site are set to zero . figure [ fig : ini_dens ] ( a ) shows the initial parton and energy distributions at the initial time @xmath96 fm/@xmath7 as a function of space - time rapidity @xmath97 . here @xmath98 and @xmath99 . it should be noted that the assumption of the thermalization for gluons produced from the cgc is to reduce the transverse energy per particle from @xmath100 gev to @xmath101 gev within our present parameters . this change should be obtained by non - equilibrium descriptions of pre - thermalization stages . we also plot an initial condition which was employed previously in ref . @xcite . here we integrate the energy density over all fluid elements with @xmath102 mev . the previous initial condition also reproduces the rapidity distribution of final charged hadrons . for longitudinal profiles , it consisted of two regions , i.e. , boost invariant region near midrapidity @xmath103 and gaussian shape in forward / backward rapidity regions . here @xmath104 is a local space - time rapidity @xcite which is shifted from the global midrapidity according to the difference of the thickness of colliding nuclei at @xmath105 . the cgc initial condition has no boost invariant region due to the @xmath13 dependence of unintegrated gluon distribution functions . it is interesting to see the transverse profile from the cgc and to compare with other parametrizations since transverse presure gradient causes radial flow in hydrodynamics . three initial conditions in the transverse direction are compared in fig . [ fig : ini_dens ] ( b ) . in ref . @xcite , transverse profile for the initial energy density scales with the number density of the binary collisions ( b. c. scaling ) . on the other hand , the flat profile with smearing near the edge of a nucleus is employed in ref . note that , in ref . @xcite , the initial condition is designed for @xmath56 fm collisions and that cylindrical symmetry is explicitly assumed . for initial energy density taken from ref . @xcite , we take account of correction factors 1.1 and @xmath106 coming from the differences of collision energy and initial time respectively . the value of maximum energy density varies from 13 to 43 gev/@xmath91 . however , these initial conditions work at least for pion @xmath1 spectrum by adjusting thermal freezeout temperature . in this section , we summarize essential features of the hydro+jet model @xcite . once an initial condition in a hydrodynamic simulation is assigned at an initial time @xmath82 , one describes the space - time evolution of thermodynamic variables by numerically solving hydrodynamic equations @xmath107 with the help of the eos , @xmath108 . we are assuming baryon free fluids here , so pressure @xmath109 is a function of energy density @xmath75 only . in the ideal hydrodynamics , the energy momentum tensor is @xmath110 , where @xmath111 is the four fluid velocity . assuming massless free parton system as discussed in the previous section , the eos for the qgp phase is @xmath112 . for the hadron phase , we employ the partial chemical equilibrium model @xcite which describes the early chemical freezeout picture by introducing chemical potential for each hadron . both the phase transition temperature and the chemical freezeout temperature are taken as @xmath113 mev . as a usual prescription , the bag constant @xmath114 mev/@xmath91 is chosen by matching condition for pressure @xmath115 . it should be emphasized that , at collider energies , relativistic coordinate in the time and longitudinal directions @xmath116 is a substantial choice to describe the evolution in the whole space - time as discussed in ref . @xcite . within hydrodynamic approaches , the studies of centrality and rapidity dependences of observables simultaneously at the collider energies are only accomplished by fully 3d hydrodynamic simulations with the relativistic coordinate . the statistical model and blast wave model analyses tell us that the kinetic ( thermal ) freezeout temperature @xmath117 is smaller than the chemical freezeout temperature @xmath118 @xcite . whereas , both thermal and chemical equilibrium is assumed in the conventional eos . this means that both freezeouts happen simultaneously and that , consequently , one can not reproduce the @xmath1 spectra and the particle ratio at the same time . by taking account of the chemical potential accompanied with the number conservation of each stable hadron between chemical freezeout and kinetic freezeout , we obtain the eos denoting the picture of early chemical freezeout . note that this eos results in the almost @xmath117 independent @xmath1 slope for pions @xcite . for the hard part of the model , we generate momentum spectra of hard partons by pythia 6.2 @xcite in which @xmath2 pqcd hard processes are included . initial and final state radiations are taken into account for the enhancement of higher - order contributions associated with multiple small - angle parton emission . eks98 nuclear shadowing @xcite is used assuming the impact parameter dependence as in ref . we employ the model in ref . @xcite to take into account the multiple initial state scatterings , in which initial @xmath119 is broadened proportional to the number of scatterings . initial transverse positions of jets at an impact parameter @xmath120 are determined randomly according to the probability specified by the number of binary collision distribution . jets are freely propagated up to the initial time @xmath82 of hydrodynamic simulations neglecting the possible interactions in the pre - thermalization stages . jets are assumed to travel with a straight line trajectory in a time step . jets lose their energies through gluon emission induced by the dense medium . a lot of work has already done by many authors @xcite . here we employ the gyulassy - levai - vitev ( glv ) formula @xcite based on an opacity expansion approach which is a relevant formalism for heavy ion collisions . in the original formula , the amount of energy loss is expanded into opacity . here we take the first order term and neglect the kinematics of emitted gluons @xmath121 where @xmath122 includes the strong running coupling and color casimir factors and @xmath123 is a parton density which is to be taken from full 3d hydrodynamic simulations in our approach . @xmath124(=0.5gev ) is a screening mass and @xmath125(=3fm ) is a typical length of a medium which can be identified with a lifetime of partonic phase . @xmath126 and @xmath127 are a position and an initial four momentum of a jet . the existence of @xmath128 comes from the so - called lpm effect @xcite . in the static medium case , @xmath129 is replaced by an initial energy of a jet @xmath130 . in the actual simulations , the coefficient @xmath122 in eq . ( [ eq : glv ] ) is regarded as an adjustable parameter roughly corresponding to an averaged value over quarks and gluons up to some factor . our strategy here is that a constant @xmath122 is set by fitting the phenix data on @xmath131 for neutral pions in the most central events at midrapidity @xcite and that we next look at the centrality and rapidity dependences . the main reasons are the following . first , we neglect fluctuations of the radiated gluon number @xcite in our calculations . an absolute value of the energy loss is about factor two smaller when one takes into account the fluctuations of the radiated gluon number at rhic @xcite . however , the @xmath1 dependence of @xmath131 is found to be the same at rhic energies @xcite . second , in our present hydrodynamic calculations , it is assumed that massless free gas for quarks ( @xmath132 ) and gluons . however , it is commonly believed that gluons are dominant components in the early stages of collisions at collider energies because of the gluon dominance of the parton structure function at small-@xmath13 . thus inclusion of specie dependent factors in @xmath122 may lead to discrepancy between our hydro+jet results and data . in this section , we first study the centrality dependence of longitudinal and transverse distribution . next , the results for the centrality dependence of the nuclear modification factors and back - to - back correlations are presented . finally , we study the nuclear modification factor at forward rapidity . let us summarize our model parameters before showing the results . in the following , we use the parameters @xmath48 and @xmath49 in the gluon distribution of a nucleon . @xmath36 depends on how to match initial distributions . for ic-@xmath75 ( ic-@xmath60 ) , @xmath133 . in the hydrodynamic calculation , the kinetic freezeout temperature of @xmath134 mev and the initial time of @xmath135 fm/@xmath7 are taken . we will not analyze the detailed @xmath117 dependence , but will make a short comment on that . a parameter @xmath136 in the parton energy loss formula is fixed by fitting the nuclear modification factor for neutral pions from phenix data at 10% centrality @xcite . gev at @xmath67 fm are compared to the corresponding transverse energy distributions obtained from hydrodynamic simulations with cgc initial condition . , width=316 ] ( pseudo)rapidity distributions of all hadrons and transverse energy in central au+au collisions from the hydrodynamic simulation are shown in fig . [ fig : dndy_final ] . here ic-@xmath60 is used . as one can compare fig . [ fig : dndy_final ] with fig . [ fig : ini_dens ] ( a ) , the effect of the hydrodynamic afterburner is to reduce the transverse energy per particle due to @xmath137 work . we find that @xmath138 gev and @xmath139 gev . recall that our cgc initial condition from eq . ( [ eq : ktfac ] ) yields @xmath100 gev and that it becomes @xmath140 gev after assuming a thermal state from fig . [ fig : ini_dens ] ( a ) . the consequence of the hydrodynamic evolution on the shape of rapidity distribution is to make it wider very slightly . final shape of the rapidity distribution looks like gaussian which is consistent with recent data from brahms @xcite . our result supports that the kln calculation @xcite which is based on the assumption of parton - hadron duality is a good approximation for the discussion of the shape of pseudorapidity distribution . the number of initial gluons at midrapidity is @xmath141 ( see fig . [ fig : ini_dens ] ( b ) ) , which can be comparable with the number of final hadrons at midrapidity ( @xmath142 ) . one see in fig . [ fig : dndy_final ] that pseudorapidity distributions for both number and transverse energy are almost flat at mid - rapidity in the range of @xmath143 , although rapidity distribution has no flat region . the main difference between the previous @xcite and the present initial conditions is whether boost invariant region exists or not ( see fig . [ fig : ini_dens ] ( a ) ) . one may ask why the previous initial conditions with boost invariant region near midrapidity can also result in the gaussian shape in rapidity distributions . it would be instructive here to discuss about the boost invariance for thermodynamic quantities and particle distributions within a hydrodynamic framework . non - boost invariant @xmath144 recently observed by brahms @xcite reminds us the so - called landau picture @xcite . boost invariance for @xmath144 is equivalent to the boost invariance of thermodynamic quantities in the original bjorken assumption @xcite in which the length in the @xmath97 direction is assumed to be infinite . however , this is not always true if boost - invariant region for thermodynamic variables is finite . in such a case , rapidity distribution can have _ no _ boost - invariant region . let us consider a simple example . rapidity distribution can be related to the phase - space distribution function via cooper - frye formula @xcite : @xmath145 consider , e.g. , a finite bjorken cylinder @xmath146 with a radius @xmath147 for massless pions . assuming no transverse flow , we have @xmath148 in the case of flat temperature profile . thus the rapidity distribution becomes @xmath149 where @xmath150 -1)$ ] . this shows rapidity distribution for massless pions in a fluid element at @xmath77 becomes @xmath151 . after the above distribution is folded in the finite boost - invariant region @xmath152 , one obtain @xmath153.\ ] ] the resultant rapidity distribution still looks like a gaussian shape when @xmath154 . the flat region starts to appear when @xmath155 . for more detailed discussions including finite mass effects , see ref . we actually reproduced the phobos ( @xmath156130 gev ) and the brahms ( @xmath156200 gev ) data by using such a bjorken like initial condition in ref @xcite ( for an initial condition , see also fig . [ fig : ini_dens ] ( a ) ) . . here contribution only from hydrodynamic components is taken into account . @xmath134 mev is used for all centralities . data are taken from ref . @xcite , width=336 ] in fig . [ fig : dndeta200 ] , pseudorapidity distributions of charged hadrons in au + au collisions at both @xmath83 and 200 gev are compared with the phobos data @xcite . figures 4 ( a ) and ( b ) ( figures 4 ( c ) and ( d ) ) show the results of the initial condition which matches the number ( energy ) density distributions for partons between the cgc and hydrodynamics . impact parameters in each panel are , from top to bottom , 2.4 , 4.5 , 6.3 , 7.9 , and 9.1 fm ( 2.5 , 4.4 , 6.4 , 7.9 , and 9.1 fm ) for @xmath157 200 ( 130 ) gev . these impact parameters are evaluated from the average number of participants at each centrality estimated by phobos @xcite . we are not interested in fit for all rapidity region , because we do not include baryon chemical potentials and both the cgc and hydrodynamics break down when density becomes small . the results within the range @xmath158 - 4 are satisfactory for reproduction of data . therefore , kln @xmath0 factorization approach in the cgc provides very good initial conditions for the hydrodynamic simulations which reproduce rapidity , centrality and energy dependences . it should be also emphasized that it is not easy to parametrize such initial conditions which reproduce the data with the same quality as the cgc initial conditions presented here . in the calculations , we use @xmath49 to get the best fit . however , we find that the results with the range of @xmath49 - 0.3 are still within experimental error bars . note that a good description of hera data has been obtained by parametrization of the saturation scale @xmath159 with @xmath160 - 0.3 @xcite . kinetic freeze - out temperature @xmath117 which is conventionally fixed by the slope of the transverse momentum spectra is a free parameter in hydrodynamics . however , @xmath1 slope for pions becomes insensitive to @xmath117 when one considers early chemical freezeout @xcite . but , we need to check the @xmath117 dependence on the multiplicity . when we change the kinetic freeze - out temperature , we can also reproduce the data by @xmath161 = 3.5 , 3.6 , 3.7 , and 3.88 for @xmath117 = 80 , 100 , 120 , and 140 mev , respectively , for ic-@xmath60 . we also investigate the @xmath82 dependence on the particle multiplicity . we find that the particle yield with @xmath162 fm/@xmath7 is only 4% larger than the case @xmath96 fm/@xmath7 regardless of the centralities as well as energies . this difference can be understood by the @xmath137 work effect . mean transverse momenta @xmath163 for pions , kaons and protons as a function of @xmath164 are compared with the phenix data @xcite in fig . [ fig : meanpt ] . since the mean transverse momenta reflects the low @xmath1 physics especially radial flow , contribution from hydrodynamic components at midrapidity is considered in this calculations . although our results are slightly smaller than the data in central and semicentral regions , the overall trend is consistent with the data . for pions , the semihard spectrum starts to be comparable with the soft spectrum around @xmath165 - 2 gev/@xmath7 in this approach . as we will see in the next subsection , we reproduce the @xmath1 spectra for pions by including contribution from jets . so the deviation for pions can be filled by the semihard spectrum . while semihard components for kaons and protons are very small in low and intermediate @xmath1 regions . a little more radial flow is needed to gain the mean transverse momentum in central collisions . more detail discussion will be given in the next subsection . in this subsection , we compute the centrality and rapidity dependences of high @xmath1 spectra and nuclear modification factors in the forward rapidity region . all results are obtained by using the initial condition ic-@xmath60 . in our model , high @xmath1 jets suffer interaction with the local parton density whose evolution is governed by hydrodynamics with the cgc initial conditions . we only take into account parton energy loss in the deconfined matter @xmath166 . figure [ fig : dndpt ] shows that the @xmath1 spectra for identified hadrons from the cgc+hydrojet model are compared with the phenix data @xcite for several centrality bins . the final @xmath1 spectrum in our approach is the sum of the contribution from fluid elements and from fragmentation of jets . first of all , we see that spectra obtained with the cgc initial condition yield the similar results with the initial condition employed previously @xcite . agreement for the pion spectra with the phenix data is perfect . as mentioned before , hydrodynamic component is insensitive to the choice of @xmath117 for the pion spectrum in the chemically frozen hydrodynamics . slopes of kaons and protons in the model prediction are steeper than that of the data . this indicates that more radial flow is needed for kaons and protons to reduce yields in the low @xmath1 region and to enhance in the intermediate @xmath1 region simultaneously , even after inclusion of the jet components into hydrodynamics . we have checked that freeze - out temperature of @xmath167 mev still underestimates the slope . rhic data suggest that initial transverse flow at the time of thermalization @xmath82 seems to be important as studied by kolb and rapp @xcite . this observation is natural because one expects that many rescatterings lead to thermalization . initial flow profile can be calculated from the classical yang - mills fields . therefore , implementation of yang - mills results into the initial condition in the hydrodynamic simulations could considerably improve ( modify ) the hydro results on the transverse momentum distributions . discrepancy in the high @xmath1 for the proton spectrum might be improved by considering the recombination among hard partons and soft partons @xcite . in fig . [ fig : raa45 ] , we investigate the centrality dependence of the nuclear modification factor for @xmath168 @xmath169 together with the phenix data @xcite . our results only account for the data up to mid - central events and fail to reproduce the data in peripheral collisions . although this is reasonable because neither cgc nor hydrodynamic description can be reliable in small particle densities , let us study more detailed mechanism . in au + au collisions at @xmath37 is compared with the phenix data @xcite . , width=336 ] in the recent studies @xcite , it is claimed that the centrality dependence of single and dihadron suppression spectra even in peripheral collisions is described by jet quenching in the partonic phase . they assume in their calculation that the parton density is scaled with the number of participants and that the system expands according to ( 1 + 1)-dimensional bjorken expansion @xcite . line integral for the loss of parton energy is evaluated at all densities . this means that there exists only the partonic phase even in very small partonic density and that the effects of phase transition or the hadronic phase are neglected in those calculations . in order to study discrepancy between their results and our results in peripheral collisions , we also estimate the centrality dependence of @xmath131 assuming ( 1 + 1)-dimensional bjorken expansion with the same initial conditions as the 3d ones . in this calculation , the transverse shape of the initial condition is unchanged and the magnitude of the parton density decreases as @xmath170 . note that our transverse profile scales as @xmath171 at mid - rapidity . pqcd calculations with ( 1 + 1)-d bjorken expansion are performed until the maximum partonic temperature reaches @xmath172 mev . when jet quenching happens in @xmath166 , the result of this ( 1 + 1)-d bjorken calculation yields almost the same results as our full 3-d calculations as plotted in the dotted line in fig . [ fig : raa45 ] . however , when we do not impose a restriction on temperature in the parton energy loss as @xmath166 , the result becomes closer to the data ( see solid line in fig . [ fig : raa45 ] ) . the difference between these two calculations suggests that there exists a contribution to energy loss besides the purely partonic phase . this can be understood as follows . the density of the system essentially decreases like @xmath173 even in our 3d hydrodynamics . in the case of most central collisions , initial density is so high and most energy loss is likely to occur in the early stages of the collisions . on the other hand , initial energy loss alone is not enough to lose energy of each parton due to small life time of the partonic phase in peripheral collisions . hence energy loss should last longer until the density of the system becomes small . from above considerations , hadronic interactions @xcite seem to be required as well in order to understand the suppression in peripheral collisions , when the bulk matter scales as the @xmath174 which is consistent with the centrality dependence of the experimental data contrast to the number of participants scaling . in au + au collisions at @xmath37 is compared with the star data @xcite . , width=336 ] in fig . [ fig : iaa ] , we plot the ratio of the strength of correlation in au+au and in @xmath3 collisions @xmath175 as a function of the number of participants @xmath176 together with the experimental data from star @xcite . @xmath177 is the azimuthal pair distribution per trigger particle , @xmath178 here @xmath179 and @xmath180 are , respectively , the relative azimuthal angle and pseudorapidity between a trigger particle and an associated particle . charged hadrons in @xmath181 gev/@xmath7 and in @xmath182 gev/@xmath7 are defined to be trigger particles and associated particles respectively . in the calculation of @xmath177 , back ground is subtracted as in ref . near side correlation is defined as @xmath183 radian and away side as @xmath184 radian . our results for the near side correlation show @xmath185 for all centralities , because interactions of jets after their hadronization are not included . however the data show @xmath186 for the near side in peripheral collisions . the results on the away side correlation for @xmath187 deviate from the data , which is consistent with the behavior of @xmath131 as discussed in fig . [ fig : raa45 ] . recently , the pseudorapidity dependence of nuclear modification factors in both au+au and @xmath188+au collisions observed by brahms @xcite attracts many theoretical interests in the context of the saturation physics , since one can go deeply into the small @xmath13 region at forward / backward rapidity . previously , we investigated the nuclear modification factor in au+au collisions @xmath189 within the hydro+jet model @xcite . in ref . @xcite , rapidity dependence of initial parton density was assumed to be flat in @xmath190 and @xmath191 was less than unity just because of the difference of @xmath1 slope of semihard spectra between midrapidity and forward / backward rapidity . however , the initial parton density from the cgc has no flat region . therefore , we should revisit the rapidity dependence of the nuclear modification factor , because partons travel almost straight line and parton energy loss may directly probe the medium at each rapidity bin . is compared with the brahms data @xcite on @xmath192 . note that minijets are produced via pqcd hard @xmath193 processes , not @xmath194 . , width=336 ] the cgc+hydrojet result for charged hadrons is compared with the brahms data @xcite in fig . [ fig : reta ] . we obtain @xmath195 in the soft region @xmath196 gev/@xmath7 which is dominated by the hydrodynamic component . this is the same result as previous work even though initial rapidity distribution is different . the reason is the following . both initial conditions lead to reproduce the phobos / brahms data on @xmath197 in which multiplicity at @xmath198 is almost the same as at @xmath199 . @xmath1 distributions at @xmath200 and 2.2 are similar to each other below @xmath201 gev/@xmath7 in @xmath3 collisions @xcite . we also find that the effect of radial flow at @xmath202 is also the same as at midrapidity . hence @xmath131 in the low @xmath1 regions are insensitive to pseudorapidity within this range . one may worry about the effect of pseudorapidity dependent jacobian @xmath203 on @xmath1 distributions . however , this is already canceled in the calculation of @xmath131 before obtaining its ratio @xmath191 in eq . ( [ eq : reta ] ) . situation is different in the high @xmath1 part . our result starts to deviate from unity at @xmath204 gev/@xmath7 since @xmath1 slope from pqcd hard collisions becomes much steeper for larger rapidity . on the other hand , experimental data deviates prior to our result . note that the discrepancy between the present result and our previous result @xcite comes from the initial longitudinal shape of the energy density shown in fig . [ fig : ini_dens ] ( a ) . in the previous parametrization of the initial energy density , dynamical evolution at midrapidity was almost the same as at the _ space - time rapidity _ @xmath205 . thus @xmath206 came from purely the difference of @xmath1 slope between midrapidity and forward rapidity . while , in the initial condition from the cgc , initial energy density at midrapidity is slightly larger than at @xmath207 . therefore , the slope " effect is compensated by the dynamical effect . this result indicates that @xmath1 spectrum in forward rapidity region is already suppressed by the initial state effect . if this is the case , instead of using conventional pqcd approach , we may need to include @xmath13-evolution for the calculation of @xmath1 spectrum up to @xmath208 gev/@xmath7 that corresponds to @xmath209 at @xmath210 @xcite . as demonstrated in ref . @xcite , study of the back - to - back correlations at forward rapidity will clarify the influence of parton saturation in the cgc . we developed a dynamical model for the description of the time evolution of high dense matter created in heavy ion collisions at high energies on the basis of the cgc in colliding nuclei , hydrodynamic evolution for produced bulk matter , and the parton energy loss in the medium . in this approach , collisions of two cgc s are calculated for the initial condition of hydrodynamic simulations . thermalized dense partons produced from colliding two cgc s are evolved hydrodynamically and also used to evaluate energy loss of high-@xmath1 partons . we have shown that hydrodynamic simulations with the cgc initial conditions describe centrality , rapidity , and energy dependences of charged hadron multiplicity very well . the best agreement for the transverse momentum distribution was obtained for pions with the hydrodynamic component and the quenched hard jet component . however , our results on kaons and protons indicated that we need other mechanisms to understand kaon and proton transverse momentum spectra from low to high-@xmath1 . the centrality dependence of nuclear modification factors and the back - to - back correlation were found to be consistent with the data up to the centrality of @xmath211 - 150 . finally we compared the nuclear modification factor at @xmath212 to brahms data by using this new initial conditions . we found that , when the qgp parton density at the forward rapidity is smaller than at mid - rapidity , suppression of hard jet from pqcd calculation which is supplemented by dglap evolution is not enough to account for the data . it suggests that the small-@xmath13 evolution will be necessary for the calculation of particle spectra at forward rapidities at rhic . our main objective has been developing a consistent dynamical framework for the time evolution of both soft matter and hard jets . in order to meet this goal , many remains are to be done in the near future . each component should be improved to obtain a better description of the whole of the stages of heavy ion collisions . for initial conditions , more sophisticated unintegrated gluon distribution should be used as mentioned in sec . [ section : model ] . numerical solutions of the classical yang - mills equations in ( 2 + 1 ) dimension @xcite can be used for transverse profile in the boost invariant hydrodynamics . quantum evolution is important for the lhc energy and forward / backward rapidity regions at rhic . ( 3 + 1)-d yang - mills solutions in which renormalization group evolution of the color charges is implemented are strongly demanded . the ideal massless qgp eos is employed in the deconfined phase in our present hydrodynamic model . the eos in hydrodynamics should be consistent with lattice qcd results . for example , one can parametrize recent lattice results by a resonance gas model for hadronic phase @xcite . in connection with jet quenching phenomena , it is argued that the small non - perturbative screening mass which is observed in lattice simulations @xcite leads to reduction of the amount of jet quenching in the vicinity of phase transition region @xcite . therefore , one expect that jet quenching would strongly depend on the eos @xcite . one of the important goals in the jet tomography is to constrain the evolving gluon density . therefore , removing the free parameter @xmath122 in eq . ( [ eq : glv ] ) will be one of the most important future work within our approach in order to see that gluon density provided by the hydrodynamic simulation is truly consistent with the energy loss of partons . in order to achieve this goal , we need to extend our hydrodynamics and use more complicated formula for the parton energy loss at the same time . for the more realistic description of chemical compositions of quarks and gluons in the deconfinement phase , we can solve hydrodynamic equations together with rate equations for number densities of quarks and gluons @xcite . this also affects jet quenching through the different cross section among @xmath213 , @xmath214 , and @xmath215 processes . the kinematic effect of radiated gluons , fluctuations , or higher order terms in the opacity expansion in glv formula should be taken into account in the energy loss of hard partons . the authors are grateful to a. dumitru , u. heinz , j. jalilian - marian , d. kharzeev , l. mclerran , d. teaney , and r. venugopalan for useful comments . the work of t.h . was supported by riken . research of y.n . was supported by the doe under contract no . de - fg03 - 93er40792 . k. h. ackermann _ et al . _ , star collaboration , phys . lett . * 86 * , 402 ( 2001 ) . p. huovinen , in _ quark gluon plasma 3 _ , edited by r.c . hwa and x.n . wang ( world scientific , singapore , 2004 ) ; p. f. kolb and u. heinz , in _ quark gluon plasma 3 _ , edited by r.c . hwa and x.n . wang ( world scientific , singapore , 2004 ) . b. b. back _ et al . _ , phobos collaboration , phys . rev . lett . * 89 * , 222301 ( 2002 ) . t. hirano , phys . c * 65 * , 011901 ( 2002 ) . m. gyulassy , i. vitev , x. n. wang , and b. w. zhang , in _ quark gluon plasma 3 _ , edited by r.c . hwa and x.n . wang ( world scientific , singapore , 2004 ) ; a. kovner and u. a. wiedemann , in _ quark gluon plasma 3 _ , edited by r.c . hwa and x.n . wang ( world scientific , singapore , 2004 ) . k. adcox _ et al . _ , phenix collaboration , phys . rev . lett . * 88 * , 022301 ( 2002 ) ; s. s. adler _ et al . _ , phys . lett . * 91 * , 072301 ( 2003 ) . i. arsene _ et al . _ , brahms collaboration , phys . lett . * 91 * , 072305 ( 2003 ) . c. adler _ et al . _ , star collaboration , phys . * 90 * , 082302 ( 2003 ) . t. hirano and y. nara , phys . lett . * 91 * , 082301 ( 2003 ) . x. n. wang , phys . b * 579 * , 299 ( 2004 ) . b. b. back _ et al . _ , phobos collaboration , phys . lett . * 91 * , 072302 ( 2003 ) . s. s. adler _ et al . _ , phenix collaboration , phys . * 91 * , 072303 ( 2003 ) . j. adams _ et al . _ , star collaboration , phys . lett . * 91 * , 072304 ( 2003 ) . r. baier , y.l . dokshitzer , s. peign , and d. schiff , phys . b * 345 * , 277 ( 1995 ) ; r. baier , y. l. dokshitzer , a. h. mueller , s. peigne , and d. schiff , nucl . phys . * b483 * , 291 ( 1997 ) ; r. baier , y. l. dokshitzer , a. h. mueller , s. peign , and d. schiff , nucl . * b484 * , 265 ( 1997 ) ; r. baier , y. l. dokshitzer , a. h. mueller , and d. schiff , nucl . b * 531 * , 403 ( 1998 ) ; r. baier , d. schiff , and b.g . zakharov , ann . nucl . part . sci . * 50 * , 37 ( 2000 ) . x. f. guo and x. n. wang , phys . lett . * 85 * , 3591 ( 2000 ) ; x. n. wang and x. f. guo , nucl . phys . a * 696 * , 788 ( 2001 ) ; e. wang and x. n. wang , phys . lett . * 87 * , 142301 ( 2001 ) . * 89 * , 162301 ( 2002 ) . p. arnold , g. d. moore and l. g. yaffe , jhep * 0111 * , 057 ( 2001 ) ; jhep * 0206 * , 030 ( 2002 ) ; jhep * 0301 * , 030 ( 2003 ) . e. iancu and r. venugopalan , in _ quark gluon plasma 3 _ , edited by r.c . hwa and x.n . wang ( world scientific , singapore , 2004 ) . l. d. mclerran and r. venugopalan , phys . d * 49 * , 2233 ( 1994 ) ; * 49 * , 3352 ( 1994 ) ; * 50 * , 2225 ( 1994 ) . j. jalilian - marian , a. kovner , l. mclerran , and h. weigert , phys . d * 55 * , 5414 ( 1997 ) ; j. jalilian - marian , a. kovner , a. leonidov , and h. weigert , nucl . b * 504 * , 415 ( 1997 ) , phys . d * 59 * , 014014 ( 1999 ) , phys . d * 59 * , 034007 ( 1999 ) , [ erratum - ibid . d * 59 * , 099903 ( 1999 ) ] ; j. jalilian - marian , a. kovner , and h. weigert , phys . rev . d * 59 * , 014015 ( 1999 ) . a. kovner , j. g. milhano , and h. weigert , phys . d * 62 * , 114005 ( 2000 ) ; a. kovner and j. g. milhano , phys . d * 61 * , 014012 ( 2000 ) ; h. weigert , nucl . a * 703 * , 823 ( 2002 ) . a. h. mueller , phys . lett . * b475 * , 220 ( 2000 ) ; a. dumitru and m. gyulassy , phys . lett . * b494 * , 215 ( 2000 ) ; j. bjoraker and r. venugopalan , phys . c * 63 * , 024609 ( 2001 ) ; j. serreau and d. schiff , jhep * 0111 * , 039 ( 2001 ) . y. v. kovchegov and d. h. rischke , phys . c * 56 * , 1084 ( 1997 ) ; s. g. matinyan , b. mller and d. h. rischke , _ ibid . _ c * 56 * , 2191 ( 1997 ) ; c*57 * , 1927 ( 1998 ) ; xiao - feng guo , _ ibid . _ d * 59 * , 094017 ( 1999 ) . d. kharzeev and m. nardi , phys . b * 507 * , 121 ( 2001 ) ; d. kharzeev and e. levin , _ ibid . _ b * 523 * , 79 ( 2001 ) ; d. kharzeev , e. levin , and m. nardi , arxiv : hep - ph/0111315 ; d. kharzeev , e. levin and m. nardi , nucl . a * 730 * , 448 ( 2004 ) . j. l. albacete , n. armesto , a. kovner , c. a. salgado and u. a. wiedemann , phys . * 92 * , 082001 ( 2004 ) ; d. kharzeev , y. v. kovchegov , and k. tuchin , phys . d * 68 * , 094013 ( 2003 ) ; j. jalilian - marian , y. nara , and r. venugopalan , phys . lett . b * 577 * , 54 ( 2003 ) . j. p. blaizot , f. gelis , and r. venugopalan , hep - ph/0402256 ; hep - ph/0402257 ; e. iancu , k. itakura , and d. n. triantafyllopoulos , hep - ph/0403103 ; r. baier , a. h. mueller , and d. schiff , hep - ph/0403201 . a. szczurek , acta phys . b * 34 * , 3191 ( 2003 ) . i. balitsky , nucl . b * 463 * , 99 ( 1996 ) . y. v. kovchegov , phys . d * 60 * , 034008 ( 1999 ) . e. iancu , k. itakura , and l. mclerran , nucl . a * 708 * , 327 ( 2002 ) . j. randrup and s. mrowczynski , phys . c * 68 * , 034909 ( 2003 ) ; p. arnold , j. lenaghan , and g. d. moore , jhep * 0308 * , 002 ( 2003 ) ; p. romatschke and m. strickland , phys . d * 68 * , 036004 ( 2003 ) ; phys . d * 69 * , 065005 ( 2004 ) ; s. mrowczynski , a. rebhan and m. strickland , arxiv : hep - ph/0403256 . t. s. biro , e. van doorn , b. muller , m. h. thoma , and x. n. wang , phys . c * 48 * , 1275 ( 1993 ) : p. levai , b. muller , and x. n. wang , phys . c * 51 * , 3326 ( 1995 ) : d. k. srivastava , m. g. mustafa , and b. muller , phys . c * 56 * , 1064 ( 1997 ) : d. m. elliott and d. h. rischke , nucl . a * 671 * , 583 ( 2000 ) . t. hirano and k. tsuda , phys . rev . c * 66 * , 054905 ( 2002 ) . k. morita , s. muroya , c. nonaka and t. hirano , phys . c * 66 * , 054904 ( 2002 ) . see , for example , e. v. shuryak , nucl . a661 * , 119 ( 1999 ) ; u. heinz , _ ibid . _ * a661 * , 140 ( 1999 ) . t. sjostrand , p. eden , c. friberg , l. lonnblad , g. miu , s. mrenna , and e. norrbin , comput . commun . * 135 * , 238 ( 2001 ) . k. j. eskola , v. j. kolhinen , and p. v. ruuskanen , nucl . * b535 * , 351 ( 1998 ) ; k. j. eskola , v. j. kolhinen , and c. a. salgado , eur . phys . j. c. * 9 * , 61 ( 1999 ) . f. karsch , k. redlich , and a. tawfik , phys . b * 571 * , 67 ( 2003 ) . o. kaczmarek , f. karsch , e. laermann , and m. lutgemeier , phys . d * 62 * , 034021 ( 2000 ) . a. dumitru and r. d. pisarski , phys . b * 525 * , 95 ( 2002 ) .

we describe a seemingly unnoticed feature of the text - book maxwell - lorentz system of classical electrodynamics which challenges its formulation in terms of an initial value problem . for point - charges , even after appropriate renormalization , we demonstrate that most of the generic initial data evolves to develop singularities in the electromagnetic fields along the light cones of the initial charge positions . we provide explicit formulas for the corresponding fields , demonstrate how this phenomenon renders the initial value problem ill - posed , and show how such bad initial data can be ruled out by extra conditions in addition to the maxwell constraints . these extra conditions , however , require knowledge of the history of the solution and , as we discuss , effectively turn the maxwell - lorentz system into a system of delay equations much like the fokker - schwarzschild - tetrode equations . for extended charges such singular light fronts persist in a smoothened form and , as we argue , yield physically doubtful solutions . our results also apply to some extent to expectation values of field operators in quantum field theory .

1602.04685

in classical electrodynamics , the dynamics of @xmath0 charges and their corresponding electromagnetic fields is governed by the lorentz equations @xmath1,\end{aligned}\ ] ] the maxwell equations @xmath2 and the maxwell constraints @xmath3 for @xmath4 . in our notation , @xmath5 denote the position and momentum of the @xmath6th charge at time @xmath7 . for simplicity we give all charges the same mass @xmath8 and rigid electric charge density @xmath9 and use units such that the speed of light equals one and the vacuum permittivity equals @xmath10 . note that by virtue of , the constraints at @xmath11 imply that they hold for all times @xmath12 . contrary to the text - book presentation , see , e.g. , @xcite , in which one employs only one total electric and magnetic field , it will be convenient for our discussion to associate with each charge @xmath6 an individual electric and magnetic field @xmath13 . thanks to the linearity of the maxwell equations in the field degrees of freedom , the equations of motion - coincide with the one given in text - books when setting @xmath14 . other choices of @xmath15 allow to switch on or off the interaction of the @xmath16-th field on the @xmath6-th charge . for arbitrary @xmath15 and smooth and compactly supported @xmath17 it has been proven that the coupled system of equations - has a well - posed initial problem for any initial data @xmath18 with reasonably regular fields @xmath19 fulfilling the constraints ; see @xcite . spinning charges were discussed in @xcite and the semi - relativistic system was considered in @xcite . very early , however , it was observed , e.g. , in @xcite , that replacing the charge density @xmath20 by a dirac delta distribution @xmath21 ( for simplicity , setting the total electric charge equal one ) renders the self - interaction summand @xmath22 on the right - hand side of the lorentz equation , and thereby , also the coupled system of equations - , ill - defined . the reason for this is that , in the point - charge case @xmath23 , the maxwell fields @xmath24 are not entirely smooth anymore but have a second order pole at @xmath25 which is exactly where they would have to be evaluated in @xmath22 . in order to distinguish the case of general @xmath20 from the point - charge case of @xmath23 , we use the convention that lower - case fields @xmath26 solve the equations - for @xmath23 , which then implies the relation @xmath27 , where @xmath28 denotes the convolution and @xmath29 is a solution to the free maxwell equations , i.e. - for @xmath30 . to see the divergent behavior of @xmath26 , thanks to the linearity , it suffices to regard a special solution to - for a fixed charge trajectory @xmath31 . in the following we drop the index @xmath6 to keep the notation slim . two well - known solutions of - are the advanced and retarded linard - wiechert fields @xmath32=({\boldsymbol{e}}^\pm_t,{\boldsymbol{b}}^\pm_t)$ ] , where the square bracket notation emphasizes the functional dependence on the charge trajectory @xmath33 . they are given by @xmath34}{|{\boldsymbol{x } } -{\boldsymbol{q}}|(1\pm { { \boldsymbol{n}}}\cdot{\boldsymbol{v}})^3}\bigg|^\pm , \nonumber \\ { \boldsymbol{b}}^{\pm}_t({\boldsymbol{x } } ) & : = \mp { \boldsymbol{n}}^{\pm}\wedge{\boldsymbol{e}}^{\pm}_t({\boldsymbol{x } } ) , \label{eq : lw}\end{aligned}\ ] ] where we have used the abbreviations @xmath35 cf . all other solutions @xmath36 to - for the same trajectory @xmath33 can then be represented as @xmath37+(1-\lambda){\boldsymbol{f}}^+_t[{\boldsymbol{q}},{\boldsymbol{p}}]+{\boldsymbol{f}}^0_t\end{aligned}\ ] ] for @xmath38 $ ] , where @xmath39 is a solution to the corresponding homogeneous equations , i.e. , - for @xmath40 . for smooth @xmath39 , the explicit expressions in imply that all corresponding fields @xmath36 are smooth on @xmath41 where they admit the discussed singular behavior that renders the term @xmath22 in ill - defined for @xmath23 . to still make sense out of this ill - defined self - interaction , an informal mass renormalization argument is usually employed , see @xcite , which effectively replaces the problematic term @xmath22 with the finite abraham - lorentz - dirac back reaction @xmath42 . in the non - relativistic regime , the latter may be approximated by @xmath43 , with @xmath44 denoting the electric charge . this procedure cures the original problem , however , introduces a dynamical instability as for almost all but very special initial accelerations , which now must be provided along with initial positions and momenta , the corresponding charge trajectories approach the speed of light exponentially fast . nevertheless , it was shown that the subset of physically sensible solutions can be well approximated in certain regimes by a dynamically stable version that was suggested by landau and lifschitz ; see @xcite . after replacing the ill - defined term @xmath22 appropriately or simply omitting it by setting @xmath45 , which often can be justified as its renormalized version is usually small ( e.g. , for small acceleration , jerk , and electric charge ) , one might hope that there are no further obstacles in arriving at a solution theory for the maxwell - lorentz system - in the point - charge limit @xmath46 . a general proof of the well - posedness of the corresponding initial value problem , however , is difficult and remains open . the first two difficulties are obvious : 1 ) the charges must not collide , otherwise @xmath47 in blows up ; and 2 ) the charges must not approach the speed of light too fast , otherwise the factors @xmath48 in may blow up . mathematically , difficulty 1 ) poses a similarly delicate problem as in the @xmath0-particle problem of gravitation , only now with the additional complication that the coulomb potentials in are lorentz - boosted and to be evaluated at _ delayed _ or _ advanced _ times @xmath49 as given in . difficulty 2 ) is due to the accumulation of the escaping fields along the light cone and must be excluded with an a priori bound on the charge velocities . when handled with care , it is reasonable to expect that at most only very few initial values @xmath50 lead to catastrophic events due to these two difficulties . however , there is a third difficulty which is more subtle and , to our knowledge , has not received attention yet . given a charge trajectory @xmath51 , only rather special initial fields @xmath52 give rise to solutions @xmath26 to - that are sufficiently regular outside a neighborhood of @xmath25 in order to be evaluated in the terms @xmath53 in for all times . generic initial fields will generate singular fronts in the fields traveling at the speed of light , and another charge @xmath16 having velocities below the speed of light is bound to traverse such fronts in finite time . + in section [ sec : steps ] we explain the mathematical origin of this questionable artifact and discuss how solutions with singular light fronts can be ruled out by appropriate restrictions on the initial values . in section [ sec : ml - system ] , for the point - charge case @xmath23 , we give necessary conditions for global existence of piecewise as well as globally smooth solutions to the maxwell - lorentz system - . we discuss that this point - charge phenomenon has a straight - forward analogue in quantum field theory , and furthermore , implications on the case of extended charges @xmath20 . in the latter , the singular light fronts qualitatively persist in @xmath24 , however , in a smoothened version as can be seen from the convolution relation @xmath27 . due to this additional smoothness , the singular light fronts cause no trouble concerning the solution theory anymore . nevertheless , as illustrated in a quantitative example in the end of section [ sec : admissible ] , they can cause sharp , though smooth , steps on the length scale of the diameter of @xmath20 , and therefore , in principal observable radiation . moreover , in section [ sec : admissible ] , we demonstrate that the initial value problem is ill - posed when demanding smooth global solutions , as the necessary restriction on the initial fields @xmath52 requires information about the , at @xmath11 , unknown charge trajectories @xmath54 . we introduce a mathematical procedure for finding admissible initial fields despite this fact . the latter , however , introduces an unwanted arbitrariness which , as we suggest in section [ sec : conclusion ] , can be eliminated by physical reasoning . the resulting restrictions on the initial values naturally turn the equations of motion of the maxwell - lorentz system - into a class of delay differential equations that include the fokker - schwarzschild - tetrode equations of motion of wheeler - feynman electrodynamics @xcite and the synge equations @xcite as prime examples . let us assume for a moment that , at least for a neighborhood around @xmath11 , the maxwell - lorentz system - has as a solution with actual charge trajectories @xmath54 and fields @xmath55 for @xmath4 . the goal in this section is to introduce explicit formulas for those fields @xmath56 , depending on their corresponding trajectory @xmath54 and initial field @xmath19 , in order to infer the properties of general solutions to - . since we can always retrieve from the point - particle fields @xmath26 the ones of the extended charges by convolution , @xmath57 , we will consider the point - particle case @xmath58 only . furthermore , we drop the index @xmath6 in this entire section because all the computations hold for any given charge @xmath6 . an explicit expression for @xmath36 solving - for trajectory @xmath33 and initial field @xmath59 can be found by recasting the maxwell equations in an integral form that reads @xmath60 again , the convolution is denoted by @xmath28 , and furthermore , @xmath61 is the propagator of the wave equation given by @xmath62 where @xmath63 are the advanced and retarded green s functions of the dalembert operator . see appendix [ sec : appendix ] for details on the derivation of from kirchhoff s formulas . note that by virtue of the maxwell equations , the maxwell constraint is preserved over time , which then also holds in this integral form . before we begin with an investigation of the properties of the general maxwell field , it is illustrative to look at a simple example that shows how singular light fronts arise . considering the maxwell constraint , one might think that an obvious candidate for a good initial field @xmath59 is given by the coulomb field @xmath64 plugging the explicit form of the initial field into and the actual trajectory @xmath33 into allows to compute . the corresponding solution @xmath36 to - reads @xmath65 \label{eq : coulomb_new } \\ & + { \mathds{1}}_{b^c_{{\left| t\right|}}({\boldsymbol{q}}_0)}{\boldsymbol{f}}_0 \label{eq : coulomb_old } \\ & + { \boldsymbol{r}}_t^{-\sigma(t)}[{\boldsymbol{q}}_0 , { \boldsymbol{p}}_0 ] , -{\boldsymbol{r}}_t^{-\sigma(t)}[{{\boldsymbol{q}}}_0,0 ] , \label{eq : coulomb_boundary}\end{aligned}\ ] ] using @xmath66({\boldsymbol{x}}):= \frac{\delta({\left| t\right|}- |{\boldsymbol{x } } - { \boldsymbol{q}}_0| ) } { ( 1\pm { \boldsymbol{n}}_0 \cdot { \boldsymbol{v}}_0 ) |{\boldsymbol{x } } - { \boldsymbol{q}}_0| } \begin{pmatrix } { \boldsymbol{n}}_0\pm { \boldsymbol{v}}_0 \\ - { \boldsymbol{n}}_0 \wedge { \boldsymbol{v}}_0 \end{pmatrix } , \label{eq : boundary}\end{aligned}\ ] ] together with @xmath67 where , with slight abuse of the introduced square bracket notation , this time the arguments in the square brackets in are not functions but just position and momentum @xmath68 , respectively . furthermore , @xmath69 denotes the sign of @xmath12 , i.e. , @xmath70 stands for @xmath71 if @xmath72 and for @xmath73 if @xmath74 , and @xmath75 denotes the characteristic function being one for @xmath76 in the closed ball @xmath77 of radius @xmath78 around @xmath79 and zero for @xmath76 in the open set @xmath80 . for the details regarding the computation we refer the reader to the appendix [ sec : appendix ] . the result shows that , according to , inside the light cone of space - time point @xmath81 the new advanced / retarded linard - wiechert field generated by the charge trajectory @xmath33 builds up as expected while , according to , in this region the initial coulomb field @xmath59 given in is displaced . the field @xmath59 then persists only outside of that light cone . in addition , one finds two distributions in that depend on the newtonian initial data @xmath82 only and have support exclusively on the light cone . see figure [ fig : coulomb ] for an illustration . this figure illustrates the supports of the terms in - making up the solution @xmath36 to the maxwell equations - for an initial coulomb field @xmath59 given in and some fixed charge trajectory @xmath33 . inside the light cone of @xmath83 ( blue area ) the initial field @xmath59 is displaced by the retarded / advanced linard - wiechert field @xmath84 $ ] in . outside that light cone ( white area ) the initial coulomb field persists . the distribution valued terms are located only on the light cone . ] by inspecting - more closely one thus finds that in general the maxwell field @xmath36 is not smooth on @xmath85 although the initial field @xmath59 in is . on the contrary , for most charge trajectories @xmath33 the field will express singular fronts on the light cone of @xmath86 because : 1 ) the distributions @xmath87 $ ] and @xmath88 $ ] in cancel only if @xmath89 ; and 2 ) the remaining terms and only connect continuously on this light cone if at least the acceleration @xmath90 vanishes . otherwise , the field @xmath36 will have a discontinuity there . at first sight this phenomenon may seem surprising . however , it has a rather simple explanation . morally , the initial coulomb field @xmath59 in corresponds to the field generated by a charge at rest at position @xmath79 . more precisely , @xmath59 is the retarded linard - wiechert field generated by an auxiliary charge trajectory @xmath91 fulfilling @xmath92 and @xmath93 for @xmath94 . if the actual charge trajectory @xmath33 does not connect smoothly to @xmath91 at @xmath11 but admits a kink , this sudden change of acceleration will result in a radiation field traveling along the light cone of @xmath86 . should @xmath95 be non - zero , an infinite acceleration is necessary to change the momentum from @xmath96 to @xmath95 , and the corresponding radiation gives rise to the distributions , whereas a step in the acceleration merely causes a discontinuity on the light cone . this simple example demonstrates that in order to prevent singular light fronts , a compatibility condition between the initial field @xmath59 and the actual trajectory @xmath33 has to be met . in the rest of this section , the objective is to identify a necessary compatibility condition for the general case . according to the general splitting in , also any relevant initial field @xmath59 , obeying the maxwell constraint , can be written in the form @xmath97 + ( 1-\lambda ) { \boldsymbol{f}}_{0}^+[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] + { \boldsymbol{f}}_{0}^{0},\ ] ] for some @xmath38 $ ] , @xmath98 $ ] being the linard - wiechert fields generated by a smooth auxiliary charge trajectory @xmath91 fulfilling @xmath92 . note that , given any general initial field @xmath59 , equation is merely a definition of @xmath99 which must then be a homogeneous field , i.e. , one fulfilling the maxwell constraint for @xmath40 . as this free field @xmath39 propagates independently of the charges , nevertheless , influences them , it is reasonable to assume its initial value @xmath99 to be smooth ( which implies @xmath39 to be smooth ) to avoid additional difficulties less regularity of @xmath99 and @xmath91 suffices , but this is not our focus here . plugging the actual trajectory @xmath33 and the initial field @xmath59 in the form of into the explicit expressions - above , one finds @xmath100 \label{eq : f - new } \\ & + { \mathds{1}}_{b_{{\left| t\right|}}({\boldsymbol{q}}_0 ) } \lambda \left ( { \boldsymbol{f}}_t^{-}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] - { \boldsymbol{f}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] \right ) \label{eq : f - ret } \\ & + { \mathds{1}}_{b_{{\left| t\right|}}({\boldsymbol{q}}_0 ) } ( 1- \lambda ) \big ( { \boldsymbol{f}}_t^{+}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] - { \boldsymbol{f}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] \big ) \label{eq : f - adv } \\ & + { \mathds{1}}_{b_{{\left| t\right|}}^c({\boldsymbol{q}}_0 ) } \big ( \lambda { \boldsymbol{f}}_t^{-}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] + ( 1-\lambda ) { \boldsymbol{f}}_t^{+}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] \big ) \label{eq : f - old } \\ & + { \boldsymbol{r}}_t^{-\sigma(t)}[{\boldsymbol{q}}_0 , { \boldsymbol{p}}_0]-{\boldsymbol{r}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q}}}_0 , \tilde{{\boldsymbol{p}}}_0 ] \label{eq : f - boundary } \\ & + { \boldsymbol{f}}_t^{0}. \label{eq : f - free}\end{aligned}\ ] ] the details are again given in the appendix [ sec : appendix ] . the first three terms have support inside and on the light cone of @xmath86 . the term describes the field that is generated by the actual charge trajectory @xmath33 between time 0 and @xmath12 , and terms - describe how the initial advanced and retarded linard - wiechert fields encoded in are propagated inside the light cone . depending on the sign of @xmath12 , one of the terms - will vanish and the respective other will be proportional to the difference @xmath101-{\boldsymbol{f}}^-_t[\tilde{{\boldsymbol{q}}},\tilde{{\boldsymbol{p}}}])$ ] , which according to dirac @xcite can be interpreted as the radiation emitted or absorbed by the auxiliary charge trajectory @xmath91 between time 0 and @xmath12 . moreover , the term is the propagated remainder of the initial retarded and advanced linard - wiechert fields , and therefore , only has support outside the light cone . the terms in are again the distributions given in having support on the light cone , and @xmath102 in is simply the field @xmath99 propagated from 0 to @xmath12 by the free maxwell equations , i.e. , - for @xmath40 . note that @xmath102 is as regular as @xmath103 . see figure [ fig : retarded ] for an illustration of the trajectories and supports of the terms - . an illustration of the charge trajectories @xmath33 and @xmath91 as well as supports of the corresponding terms in - for the case @xmath104 and @xmath105 . ] the solution @xmath36 in - can be recast in a more compact form @xmath106 -{\boldsymbol{f}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] \right ) \label{eq : short - new } \\ & + \lambda { \boldsymbol{f}}_t^{-}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] + ( 1- \lambda ) { \boldsymbol{f}}_t^{+}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] \label{eq : short - lw } \\ & + { \boldsymbol{r}}_t^{-\sigma(t)}[{\boldsymbol{q}}_0 , { \boldsymbol{p}}_0]-{\boldsymbol{r}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q}}}_0 , \tilde{{\boldsymbol{p}}}_0 ] \label{eq : short - dist } \\ & + { \boldsymbol{f}}_t^{0 } , \label{eq : short - free}\end{aligned}\ ] ] from which one can read off necessary compatibility conditions between the initial field @xmath59 and the charge trajectory @xmath33 that prevent the development of singular light fronts : * the distributions must cancel each other because neither , , nor contain dirac delta distributions . this is the case if and only if @xmath107 , where @xmath92 was already assumed in order to fulfill the maxwell constraint . * provided ( c1 ) is fulfilled , the field @xmath36 is continuous on @xmath108 if and only if term vanishes on the light cone of @xmath86 . this can be seen as follows : by virtue of , for all times @xmath12 the terms - are smooth everywhere except maybe on the light cone of @xmath86 as well as the points @xmath109 and @xmath110 . however , since these terms coincide with - , they must be smooth in @xmath110 as and are free fields and has only support outside of the light cone of @xmath111 . as the free field @xmath39 is smooth , and by ( c1 ) terms , vanish , the field @xmath36 is continuous on @xmath108 if and only if vanishes on the light cone of @xmath86 . this is the case if and only if the accelerations @xmath112 and @xmath113 coincide at @xmath11 . furthermore , if and only if all @xmath114-th derivatives of @xmath110 and @xmath109 for @xmath115 coincide at @xmath11 , the field @xmath36 has @xmath116 spatial derivatives on @xmath117 . finally , if and only if the trajectories @xmath110 and @xmath109 connect smoothly at time @xmath11 , the field @xmath36 is smooth on @xmath117 . it was called to our attention that also in @xcite , where a rigorous electrodynamic point - charge limit was studied in the dipole approximation , a condition relating the initial fields and initial momenta similar to ( c1 ) was needed to ensure convergence . in this chapter we discuss the implications of the observations made in section [ sec : steps ] on the fully coupled system of maxwell s and lorentz s equations - . our main interest , which will be discussed first , lies in the case of @xmath118 point - like charges , i.e. , @xmath23 , either with a properly renormalized self - interaction term @xmath22 or without it , i.e. , @xmath45 . the implications on the maxwell - lorentz system for smooth extended charges @xmath20 are considered in the end . first and foremost , we observe that in a system of at least two charges , one charge , say number 2 , will inevitably cross the light cone of the initial space - time point of another charge , say number 1 , at a time @xmath119 , which is bounded from below by the minimal distance divided by speed of light ; see figure [ fig:3 ] . charge 2 on trajectory @xmath120 is bound to cross the light cone of the initial space - time point @xmath121 of charge 1 on trajectory @xmath122 . ] thus , at @xmath123 the lorentz force felt by charge 2 must evaluate the field @xmath124 at some point on the light cone of @xmath121 . recall that for an initial field @xmath52 of the form with auxiliary charge trajectory @xmath125 , the propagated field @xmath26 is given by - . should condition ( c1 ) of section [ sec : steps ] not be satisfied , this evaluation is ill - defined because of the presence of the distributions . in this case , the dynamics will cease to exist beyond the time instant @xmath119 . hence , ( c1 ) is a necessary condition for global existence of solutions to the maxwell - lorentz system - . should condition ( c1 ) hold but not ( c2 ) , then the force on charge 2 will undergo a discontinuous jump when traversing the light cone at time @xmath119 . therefore , ( c2 ) is a necessary condition for having continuous or smooth solutions to the maxwell - lorentz system - . the following two arguments illustrate that ( c1 ) and ( c2 ) are violated for generic initial data @xmath50 obeying the maxwell constraints only . precisely , they show that global existence is not stable under arbitrarily small perturbations of the initial data . for this purpose , let us assume that @xmath126 is a global solution to the maxwell - lorentz system - for some initial value @xmath50 such that the initial fields @xmath52 are of the form for some smooth auxiliary trajectory @xmath125 and some smooth initial free field @xmath127 . recall from our discussion in section [ sec : steps ] that any relevant initial field can be written in this form , and then , it automatically fulfills the maxwell constraint . + * no - go argument ( a1 ) : * by maxwell constraints and necessary condition ( c1 ) we have @xmath128 for @xmath4 . then , perturbing the initial momentum of charge 1 by @xmath129 for any vector @xmath130 of arbitrarily small norm @xmath131 leads to a corresponding local solution @xmath132 with @xmath133 taking the form of - , whereas the contribution equals the distribution @xmath134-{\boldsymbol{r}}_t^{-\sigma(t)}[{{\boldsymbol{q}}}_0 , { { \boldsymbol{p}}}_0]$ ] , which does not vanish . in other words ( c1 ) is violated , and @xmath133 manifests a singular light front with support on the light cone of space - time point @xmath121 , as discussed in section . by virtue of - , this perturbation in the initial momentum propagates not faster than the speed of light . in particular , the perturbed field @xmath133 of charge 1 and the perturbed trajectory @xmath135 of charge 2 remain identical on @xmath136 for @xmath7 . in consequence , charge 2 is bound to touch the light cone of @xmath121 at the very same time @xmath119 as in the unperturbed solution , only now the perturbed field @xmath133 contains a singular light front consisting of distributions . in conclusion , the dynamics will cease to exist beyond time @xmath119 , as discussed above . the argument is depicted in figure [ fig : a1 ] . perturbing the initial momentum of charge 1 by @xmath137 leads to a singular front supported on the light cone of @xmath121 and , thus , to a sudden stop of the dynamics at time @xmath138 when charge 2 touches the light cone . ] * no - go argument ( a2 ) : * this time , let us assume the global solution is also smooth and @xmath139 ( for @xmath140 a similar argument can be found ) . due to condition ( c2 ) , @xmath141 and @xmath142 coincide at time @xmath11 . now , we perturb a little bit the trajectory @xmath143 that defined the initial field @xmath144 given in in an arbitrarily small neighborhood of the retarded time @xmath145 belonging to space - time point @xmath121 . due to - this causes a small perturbation @xmath146 , and we tune this perturbation such that the lorentz force on charge 1 at @xmath11 changes its value . in consequence , the potential local solution @xmath147 corresponding to this perturbed initial data violates ( c2 ) as the accelerations @xmath148 and @xmath149 do not match anymore at @xmath11 . as discussed in section , this leads to a discontinuity on the light cone of @xmath121 . however , by virtue of - the perturbed field @xmath133 of charge 1 and the perturbed trajectory @xmath135 of charge 2 remain identical on @xmath136 for @xmath7 . therefore , charge 2 is bound to hit the light cone of @xmath121 at the very same time @xmath119 as in the unperturbed solution . at this instant , due to the discontinuity of @xmath133 , the acceleration of charge 2 will undergo a likewise discontinuous jump . hence , should the perturbed solution exist globally it can only be piecewise smooth . furthermore , the discontinuity in the acceleration of charge 2 will give rise to a corresponding discontinuity in the field @xmath150 on the light cone of @xmath151 , which charge 1 is bound to cross eventually . by this mechanism , a whole network of singular light fronts is developed . the argument is depicted in figure [ fig : a2 ] . + perturbing the initial field of charge 1 , @xmath152 , by a small bump in @xmath143 at the corresponding retarded time , leads to a discontinuity of @xmath124 supported on the light cone of @xmath121 , and thus , charge 2 experiences a sudden jump in acceleration at time @xmath138 , which causes a discontinuity in @xmath150 . ] these two arguments indicate that the initial value problem of the maxwell - lorentz system - with renormalized ( or without ) self - interaction term is ill - posed for general initial values @xmath50 only fulfilling the maxwell constraints : even if a global solution is found , only a small perturbation in the initial values suffices to prevent either global existence , by ( a1 ) , or global smoothness , by ( a2 ) , of the potential solution corresponding to the perturbed initial values . one might tend to think that these are all problems connected to the point - like nature of the charges , a concept that could even be considered questionable in the classical regime . indeed , as discussed in the introduction , it is true that for the maxwell - lorentz system of smoothly extended charges those mathematical problems do not show up . nevertheless , the qualitative behavior of generation of singular light fronts for initial conditions that violate ( c1 ) or ( c2 ) remains the same . as the fields of the extended charges are of the form @xmath27 , the discussed singular fronts are now only smeared out by the charge density @xmath20 . for @xmath20 supported on the scale of the classical electron radius , i.e. , @xmath153 , the singular fronts will still result in sharp though smooth steps in the fields on the respective light cones . other charges are bound to eventually traverse such steps and will suddenly on time scales of @xmath154 divided by their respective speed start or stop to radiate , thus , leading to potentially observable though physically questionable phenomena . furthermore , it is interesting to note that the singular light fronts persist also in quantum field theory . this can readily be observed in the following toy model in which a fixed source at @xmath155 interacts with a second - quantized and massless scalar field @xmath156 by means of the interaction hamiltonian @xmath157 , where @xmath158 and @xmath159 are the bosonic creation and annihilation operators fulfilling the ccr @xmath160=\delta^3({\boldsymbol{k}}-{\boldsymbol{k}}')$ ] . formally , @xmath161 is not well - defined without an ultraviolet cut - off such as @xmath20 but for the sake of the argument it is sufficient to continue informally . let @xmath162 be the time evolution generated by @xmath161 , then for any initial unit fock state @xmath163 we get @xmath164 the expectation value of this scalar field can therefore be represented by means of kirchhoff s formulas as it was done for the maxwell field in appendix [ sec : appendix ] . in the simplest case of an initial vacuum @xmath165 one finds @xmath166 where the discontinuity on the light front shows up again as the field is build up over time starting from an initial vacuum . this behavior only disappears for special initial @xmath163 , precisely , the ground state of this toy model , which can be computed explicitly , plus smooth additional free fields . if we further allow the charge @xmath167 to move , very similar scenarios as discussed in section [ sec : steps ] can be constructed ; but this shall not be our focus here . in conclusion , for any choice of @xmath20 , and be it for mathematical or physical reasons , it seems desirable to restrict the space of initial values of the maxwell - lorentz system - beyond the maxwell constraints . if for a moment we also admit piecewise smooth solutions to the maxwell - lorentz system - , a sensible restriction on the space of initial values can be taken from condition ( c1 ) . if we require the initial value @xmath50 to comprise fields @xmath52 of the form for piecewise smooth auxiliary trajectories @xmath125 fulfilling @xmath168 , condition ( c1 ) as well as the maxwell constraints are fulfilled by definition and there seems to be no further obstacle concerning mathematical well - posedness of the respective initial value problem . if , however , we demand smooth global solutions , we would also need to comply with condition ( c2 ) . in order to do so we would have to know the derivatives of the charge trajectories @xmath54 at initial time @xmath11 . but those are unknown as they already require knowledge of a local solution in a neighborhood of @xmath11 . hence , there is no possibility to restrict the space of initial fields a priori in order to ensure well - posedness . as a workaround one may consider the following approach : given initial data @xmath50 fulfilling and ( c1 ) , it is possible to compute the solution of the maxwell - lorentz equations in a sufficiently small time interval @xmath169 . this can be done as the singular fronts live only on the light cones of the initial space - time points @xmath170 so that @xmath171 only has to be chosen smaller than the smallest time @xmath119 when some charge hits a singular front . this preliminary local solution allows to compute all derivatives of the charge trajectories @xmath54 at @xmath11 , and hence , it would allow to adapt the auxiliary trajectories @xmath125 in a neighborhood of @xmath11 to connect smoothly to @xmath54 such that ( c2 ) is fulfilled . this procedure changes the initial fields @xmath172 in a spatial neighborhood around the initial positions @xmath173 . if self - interaction is excluded , the adapted initial values @xmath174 , however , fulfill the maxwell constraints , ( c1 ) , and ( c2 ) , and therefore , should not bare any further obstacles concerning smooth global solutions . if self - interaction is included , the above procedure would have to be iterated until a fixed - point is found as the change in @xmath175 implies again a change in the initial acceleration of charge @xmath6 . + though mathematically sound , physically , this is a rather opaque procedure . it is not anymore a formulation of classical electrodynamics in terms of an initial value problem for - but in terms of an initial guess , that , first , has to be adapted in a quite arbitrary way before a global solution can be inferred at all . + so what is overlooked when naively regarding the maxwell - lorentz system - as an initial value problem ? any inhomogeneous solution @xmath26 to the maxwell equations - is of the form , which implies that the entire history of the charge trajectory @xmath54 is already encoded in the spatial dependence of the field @xmath26 ; recall the @xmath49 dependence in . now , if we set some initial field @xmath52 by hand , for which the maxwell constraint only requires that we choose it of the form with some auxiliary trajectory @xmath125 fulfilling @xmath176 , the maxwell time evolution is fooled to believe that the history of the charge trajectory is given by @xmath125 . but except for @xmath176 , the history of the auxiliary trajectory @xmath125 may have nothing in common with the actual one @xmath177 , which is to be computed . as a matter of fact , the maxwell equations propagate such an initial field @xmath52 as if it was generated by the auxiliary charge trajectory @xmath125 outside the light cone of @xmath170 while , inside , a new field is generated according to the actual trajectory @xmath54 . it is therefore not surprising that the incompatibilities between the actual charge trajectories @xmath54 and the initial fields @xmath52 of the solution - discussed in section [ sec : steps ] occur during the dynamics and that any mismatch between the actual and auxiliary charge trajectories in the sense of ( c1 ) and ( c2 ) expresses itself as a singular light front . in view of this , it would be desirable to find a formulation of classical electrodynamics that automatically avoids any such incompatibilities . this is possible and in section [ sec : conclusion ] we discuss a whole class of such formulations having two representatives that are well - known since the beginning of classical electrodynamics . but first , we end this section with a short example that illustrates quantitatively 1 ) that the phenomenon of singular light fronts can lead to significant radiation effects and 2 ) how the initial fields encode the histories of their respective charge trajectories : + _ quantitative example : _ we reconsider the introductory example from section [ sec : steps ] of a charge , referred to as charge 1 , having an initial position and momentum @xmath178 and an initial coulomb field @xmath179 , i.e. , lorentz - boosted w.r.t . whatever its future trajectory @xmath122 may be , its field @xmath124 will be of the form - as depicted in figure [ fig : coulomb ] . in the following we will use si units and charges smeared out by @xmath20 instead of point - charges , hence we regard @xmath181 . let us suppose that the other initial fields @xmath182 of all @xmath183 other charges do not comprise free fields , i.e. , @xmath184 , and are such that charge 1 experiences a large initial acceleration , say , @xmath185 . charge 2 is assumed to have initial position and momentum @xmath186 at a sufficiently large distance , say , @xmath187 , so that , initially , it moves almost freely with velocity , say , @xmath188 . eventually , it will reach the vicinity of the light cone of @xmath189 smeared out by @xmath20 . now , if charge 1 and 2 are made out of clouds of , say , @xmath190 electrons produced by electron guns and collimated to balls of @xmath191 , the acceleration of charge 2 is @xmath192 where @xmath119 is the time of arrival inside the smeared out light cone and @xmath193 is the @xmath194-component of the electric field @xmath195 computed with the help of and : @xmath196 note that , since for simplicity we assumed that the initial coulomb field @xmath179 was already properly lorentz - boosted , the distributions cancel in contrast to the introductory example in section [ sec : steps ] . in other words , ( c1 ) is fulfilled which makes the only contribution to the field . as a result of this analysis , we find that within the time of traversal of the smeared out light cone , @xmath197 , there must be a sudden rise in emission of radiation of charge 2 . according to lamor s formula , the resulting increase in power goes from almost zero to @xmath198 one may now wonder , why the flank in radiation power increase is so steep . as we discussed , the initial lorentz - boosted coulomb field of charge 1 encodes the history of a charge with constant momentum @xmath199 . this tells the maxwell dynamics that there must be a sudden change in acceleration at time @xmath11 from zero to @xmath200 in order to fit the initial data , and therefore , that a step in increase of radiation power ( smeared out by @xmath20 ) must be produced . in a more realistic scenario , however , charge 1 would first have to acquire the initial acceleration @xmath200 in the past @xmath74 ; to match the above numbers , for instance , by entering a capacitor and falling through a voltage of @xmath201 over a distance of @xmath202 . depending on the duration of the acceleration process which may take considerably longer than @xmath203 , a quite different initial field @xmath204 is produced . this time , it consists of the former lorentz - boosted coulomb field plus a radiation part that was emitted during the process of acceleration . in contrast to the above scenario , this additional radiation will hit charge 2 much earlier before entering the light cone region , and instead of generating a steep flank in radiation power of charge 2 from zero to @xmath205 there will be a respectively smoother increase . in other words , and independently of the fictitious numbers we have used above , this example shows that if the choice of initial values to the maxwell - lorentz system - imply a large initial acceleration for charge 1 , which may later on enforce charge 2 to generate significant radiation , the cause that led to this large initial acceleration lies in the history of charge 1 which is encoded in the initial field @xmath204 . hence , an unnatural choice for @xmath204 such as the lorentz - boosted coulomb field above , which would imply that charge 1 did not accelerate for @xmath74 but then suddenly does at @xmath11 , leads to the peculiar effect of the steep flank in radiation power increase of charge 2 . modelling more accurately how charge 1 acquired the initial acceleration in the past will eliminate this effect . this demonstrates the intimate connection between the history of a charge trajectory and its generated field , which is the starting point of our discussion of the formulation of classical electrodynamics in section [ sec : conclusion ] . as demonstrated , in the case of point - charges , the restriction of the solution space of the maxwell - lorentz system - to smooth solutions does not allow a formulation in terms of an initial value problem . though a potential global solution is uniquely identified by its initial data @xmath50 , only very special initial fields fulfilling the necessary condition ( c2 ) lead to smooth global solutions . furthermore , the information needed to restrict the initial data according to ( c2 ) would already require knowledge of the unknown solution . even for smooth charge distributions @xmath20 , neglect in matching the initial fields @xmath52 to the history of the charge trajectories yields rather arbitrary differences in the predictions as the example in the preceding section illustrates . these circumstances suggest that we might need to change the way we look at the solution theory for the maxwell - lorentz system . the starting point for such a consideration is the fact that the maxwell field at one time instant and the entire trajectory of the charge that generated it are intimately intertwined beyond the maxwell constraint . this can be observed best when imagining a single charge @xmath6 incoming from the remote past @xmath206 . considering , e.g. , the case @xmath104 , any auxiliary trajectory @xmath125 in the expression of the field @xmath26 in - is forgotten during a time evolution from @xmath206 to any finite time @xmath12 and so are any potential singular fronts as they escape to spatial infinity with the speed of light . concerning point - wise evaluation in any finite region of space - time , the maxwell field in - reduces to the expression @xmath207 + { \boldsymbol{f}}^0_{i , t}.\end{aligned}\ ] ] nothing changes in this argument and in the form of when the charge trajectory @xmath54 is not prescribed but also develops simultaneously to the evolution of the maxwell fields , i.e. , according to the fully coupled system - . hence , stopping the dynamics at time @xmath11 and starting it again in an initial value problem fashion dictates the natural choice for the initial field at @xmath11 . this means that the initial field @xmath52 should be of the form for a auxiliary trajectory @xmath125 that coincides with the actual one @xmath54 and that the free field @xmath127 , as it evolves independently of the charges , equals the incoming free field evolved from @xmath206 to @xmath11 . hence , in the general case for any @xmath38 $ ] , where also advanced linard - wiechert fields may occur , one would expect the maxwell field to take the form @xmath208+(1-\lambda){\boldsymbol{f}}_{i , t}^+[{\boldsymbol{q}}_i,{\boldsymbol{p}}_i ] + { \boldsymbol{f}}^0_{i , t}.\end{aligned}\ ] ] any compatibility condition , such as the maxwell constraint , ( c1 ) , and ( c2 ) , is now naturally fulfilled for all times @xmath12 . but this comes at a high price . by , the fields @xmath52 at time @xmath11 depend on the entire history of the charge trajectories which consequently means letting go of the initial value formulation of classical electrodynamics . in view of the above , however , such a step seems well grounded . in section [ sec : ml - system ] , it was already indicated when insisting on the merely mathematical property of smoothness of solutions . but there , one might even have been tempted to accept potential kinks in the charge trajectories , say , as long as they decay fast enough . however , the discussion above and in section [ sec : admissible ] shows that there is also a physical reason why the initial value formulation is questionable , namely the fact that at each time instant the entire history of a charge trajectory is already encoded in the spatial dependence of its field . therefore , when entertaining the thought that charges are incoming from the remote past , the form of the maxwell fields is already presupposed by and the space of potential solutions @xmath126 of the maxwell - lorentz system - should consequently be restricted to solutions having maxwell fields @xmath26 that fulfill . such a restriction is easily implemented in the fundamental equations of motion - . it simply means replacing the maxwell fields on the right - hand side of with the explicit form given in . this makes the maxwell equations and constraints - redundant and turns the coupled system of the odes and pdes - , only consisting of terms that are all evaluated at the same time instant @xmath12 , into the following system of odes that involve terms depending on advanced or delayed times @xmath49 as given in : @xmath209 , \nonumber \\ { \boldsymbol{f}}_{i , t}&=({\boldsymbol{e}}_{i , t},{\boldsymbol{b}}_{i , t } ) \nonumber \\ & = \rho*\big(\lambda{\boldsymbol{f}}_{i , t}^-[{\boldsymbol{q}}_i,{\boldsymbol{p}}_i]+(1-\lambda){\boldsymbol{f}}_{i , t}^+[{\boldsymbol{q}}_i,{\boldsymbol{p}}_i]\big ) + { \boldsymbol{f}}^0_{i , t}. \nonumber\end{aligned}\ ] ] here , @xmath210 denotes any given solution of the free maxwell equations . it is interesting to note that by virtue of - the free fields @xmath29 , when prescribed in the remote past , are forgotten should they have some spatial decay at spatial infinity @xcite . in this case , for @xmath211 , no self - interaction @xmath45 or @xmath212 , and for point charges @xmath23 , the system of equations is equivalent to the fokker - schwarzschild - tetrode equations @xcite as used in wheeler s and feynman s investigation of classical radiation reaction @xcite . they can be derived from a simple action principle @xcite , and furthermore , allow a derivation of dirac s radiation damping term @xmath213 without the need of a mass renormalization procedure @xcite . moreover , for @xmath104 , point charges @xmath23 , and no initial free fields , the resulting equations are equivalent to the synge equations @xcite . the nature of these equations , involving a priori unbounded state - dependent delays @xmath49 , cf . , in the definition of the linard - wiechert fields , renders a general classification of solutions very difficult . in mathematics , this problem is known as the _ electrodynamic n - body problem_. to this day , global existence has only been established when considering two repelling charges and restricting the motion of the charges to a straight line @xcite . when constraining the charge trajectories at times @xmath214 , for arbitrary large but finite @xmath215 , existence of solutions on @xmath216 $ ] was shown for @xmath0 smoothly extended charges in three dimensions @xcite . however , except for very special situations @xcite , almost nothing is known about uniqueness of solutions ; see @xcite . it may turn out that solutions can only be identified uniquely when whole stripes of trajectories are specified . nevertheless , such types of state - dependent delay differential equations are currently heavily under investigation in the contemporary mathematics literature ( see , e.g. , @xcite and the references therein ) and there is good reason to expect that their solution theory will soon be better understood . in this appendix we explain how the solution formula - for maxwell s equations - can be derived from kirchhoff s formulas . afterwards we demonstrate the main steps in the computation of the explicit expression - from which - and also - can be inferred . again we omit the charge index and restrict ourselves to the point charge case @xmath217 from which the corresponding result for general @xmath20 can be inferred ; all equalities are meant in distribution sense . the maxwell equations , taking into account the constraint , imply the following inhomogeneous wave equation @xmath218 thanks to kirchhoff s formulas @xcite , the unique solution @xmath219 to the initial value problem @xmath220 can be expressed in the form of @xmath221 with @xmath61 as given in . applying this formula to system with initial values @xmath59 and @xmath222 at initial time @xmath223 we find an expression for the unique solution given by @xmath224 these formulas can also be found in @xcite . note , that in general @xmath225 and @xmath226 do not solve maxwell s equations individually . in the limit @xmath227 and for @xmath59 having some spatial decay @xmath225 vanishes and @xmath226 coincides with the advanced / retarded linard - wiechert field @xmath228=({\boldsymbol{e}}^\pm , { \boldsymbol{b}}^\pm)$ ] as given in , which are solutions of the maxwell equations - ; see @xcite . next , we discuss the computation of - . without loss of generality we set @xmath229 , and exemplary compute the term in coming from the upper left entry in the matrix of differential operators evaluated at spacial point @xmath230 . the other two terms can be inferred analogously . we compute @xmath231 in the next step , we employ the identity @xmath232 for the matrix @xmath233 with entries @xmath234 where the indices @xmath235 denote the components of the respective vectors and we have used the abbreviations @xmath236 not to be confused with the notations @xmath237 in and @xmath238 in . for the @xmath6th component of the vector we then obtain @xmath239 \delta^3({\boldsymbol{x } } -{\boldsymbol{y}}- { \boldsymbol{q}}_{t \pm { | { \boldsymbol{y}}| } } ) \label{eq : ainside}\\ & - \int_{\partial b_{|t|}(0)}d\sigma(y ) \ , n_j \frac{1}{{| { \boldsymbol{y}}|}}l({\boldsymbol{y}})^{\pm}_{ij } \ , \delta^3({\boldsymbol{x } } -{\boldsymbol{y}}- { \boldsymbol{q}}_{t \pm { | { \boldsymbol{y}}| } } ) \label{eq : aboundary}\end{aligned}\ ] ] by partial integration and stokes theorem , where @xmath240 denotes the surfaces of the ball of radius @xmath241 around the origin and @xmath242 the respective surface measure . furthermore , we adopt the einstein summation convention regarding the latin indices . next , it is convenient to carry out a change of variables according to transformation @xmath243 having jacobi determinant @xmath244 . note that , as the trajectory @xmath33 is time - like , @xmath215 has an inverse , and therefore , @xmath245 holds . after this transformation , turns into @xmath246 \bigg|_{{\boldsymbol{y}}= { \boldsymbol{z } } - { \boldsymbol{q}}^{\pm } } \delta^3({\boldsymbol{x } } -{\boldsymbol{z } } ) \\ & = { \mathds{1}}_{b_{|t|}({\boldsymbol{q}}_0 ) } \frac{1}{1\pm { \boldsymbol{n}}^{\pm } \cdot { \boldsymbol{v}}^{\pm } } \partial_{y_j}\left [ \frac{1}{{| { \boldsymbol{y}}|}}l({\boldsymbol{y}})^{\pm } _ { ij}\right ] \bigg|_{{\boldsymbol{y}}= { \boldsymbol{x } } - { \boldsymbol{q}}^{\pm } } , \end{aligned}\ ] ] where we used the abbreviations introduced in . the derivatives can now be carried out in a straight - forward manner . the boundary term of the partial integration gives @xmath247 carrying out the analogous computations as for and for the remaining two terms in gives the following structure : @xmath248 + { \boldsymbol{r}}_t^{-\sigma(t)}[{\boldsymbol{q}}_0,{\boldsymbol{p}}_0 ] , \ ] ] where we have used the notation and . the first summand in comprises all @xmath249 integrals while the @xmath250 integrals make up the second summand . finally , in order to compute @xmath251 , given in , we assume an initial field @xmath59 as in given in terms of a smooth auxiliary trajectory @xmath252 , a homogeneous field @xmath103 , and a parameter @xmath253 $ ] . plugging this choice for @xmath59 into gives @xmath254 \label{eq : af1_1 } \\ & + ( 1- \lambda ) \begin{pmatrix}\partial_t & { \nabla \wedge}\\ -{\nabla \wedge } & \partial_t \end{pmatrix}k_{t } * { \boldsymbol{f}}_0^+[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p}}}]\label{eq : af1_2 } \\ & + \begin{pmatrix}\partial_t & { \nabla \wedge}\\ -{\nabla \wedge } & \partial_t \end{pmatrix}k_{t } * { \boldsymbol{f}}_0 ^ 0 . \label{eq : af1_3}\end{aligned}\ ] ] making use of the fact that the linard - wiechert fields @xmath255 $ ] solve maxwell s equations and that @xmath103 is a homogeneous field , the three summands - can be simplified according to @xmath256 = { \boldsymbol{f}}_t^{\pm}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] - { \boldsymbol{f}}_t^{(2)}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p}}}]\\ & \begin{pmatrix}\partial_t & { \nabla \wedge}\\ -{\nabla \wedge } & \partial_t \end{pmatrix}k_t * { \boldsymbol{f}}_{0}^{0 } = { \boldsymbol{f}}_t^{0},\end{aligned}\ ] ] where @xmath257 denotes the unique solution for the free maxwell equations to the initial value @xmath103 . thanks to we get @xmath258 - { \boldsymbol{r}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q}}}_0 , \tilde{{\boldsymbol{p}}}_0 ] - { \mathds{1}}_{b_{{\left| t\right|}}({\boldsymbol{q}}_0)}{\boldsymbol{f}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] \right ) \\ & + ( 1- \lambda ) \left ( { \boldsymbol{f}}_t^{+}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] - { \boldsymbol{r}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q}}}_0 , \tilde{{\boldsymbol{p}}}_0 ] - { \mathds{1}}_{b_{{\left| t\right|}}({\boldsymbol{q}}_0)}{\boldsymbol{f}}_t^{-\sigma(t)}[\tilde{{\boldsymbol{q } } } , \tilde{{\boldsymbol{p } } } ] \right)\\ & + { \boldsymbol{f}}_t^{0}.\end{aligned}\ ] ] plugging this result together with into , we arrive at the representation - for the field @xmath36 . * acknowledgments . * the authors express their gratitude for the helpful discussions with detlef drr , michael kiessling , diego noja , herbert spohn , and roderich tumulka . this work was funded by the elite network of bavaria through the junior research group ` interaction between light and matter ' . 29ifxundefined [ 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 @noop _ _ , ed . (: , ) @noop _ _ (: , ) @noop * * , ( ) in @noop _ _ * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ (: , ) @noop _ _ , ed . (: , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ , vol . (: , ) @noop * * , ( ) @noop _ _ (: , ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . ( , ) @noop ( )

it is shown that the inflationary era in early universe is realized due to the effect of backreaction of quantized matter fields . in fact we start by quantizing a free scalar field in the friedmann - robertson - walker space - time , and the field is fluctuating quantum mechanically around the bottom of the mass potential . we then obtain the vacuum expectation value of the energy density of the scalar field as a functional of the scale factor of the universe . by plugging it into the einstein equation , a self - consistent equation is established in which the matter fields determine the time evolution of the universe . we solve this equation by setting few conditions and find the following solution : the universe expands la de sitter with e - folding number and then it turns to shrink with a decreasing hubble parameter which rapidly goes to zero . * a new mechanism of realizing inflationary universe + * with recourse to backreaction of quantized free fields + inflation without inflaton + * * yoshinobu habara , hikaru kawai and masao ninomiya okayama institute for quantum physics , 1 - 9 - 1 kyoyama , okayama 700 - 0015 , japan department of physics , kyoto university , kyoto 606 - 8502 , japan [email protected] , + [email protected] , + [email protected] pacs numbers : 04.62.+v , 98.80.cq , 11.25.-w , 11.25.mj , 98.80.qc keywords : inflation , string theory , cosmic microwave background

1410.0644

it is believed by most astro - cosmologists as well as astro - particle physicists that the only solution to resolve the flatness and horizon problems in the universe is to think of exponential expansion called inflation @xcite right after birth of our universe . in recent years several observational results to confirm the inflation were reported @xcite . in particular , the most important evidence in the present days to provide data is the temperature fluctuation of the cosmic microwave background radiation ( cmb ) which shows us quantitatively that there was the era of the inflation in the early universe . using this data we can investigate the mechanism of the inflationary expansion in detail . the cmb is the photon emitted in the recombination era of which the wavelength is very long so that cmb reaches us without much scattering afterwards . therefore cmb carries to us the information of the early universe . the most peculiar properties of cmb is that it is homogeneous and uniform with very high accuracy and furthermore the spectrum agrees very well with the black body radiation with the temperature 2.725 [ k ] while the temperature fluctuation @xmath7 is of the order of @xmath8 . this temperature fluctuation is believed to come from the gravitational potential @xmath9 through sachs - wolfe effect @xmath10 @xcite . in the usual inflation models the gravitational potential @xmath9 is believed to be created when the hypothetical scalar particle(s ) called inflaton(s ) fluctuates quantum mechanically during rolling down classically along with the potential . there are many models proposed for creation mechanism , such as slow - roll inflation @xcite , chaotic inflation @xcite etc . in our previous papers @xcite , we show that the temperature fluctuation of cmb can be explained as the vacuum fluctuation of a large number of matter fields . here the matter fields are originated from the string excited states and/or the kaluza - klein modes on the compactified space . there appear very many kinds of fields in the exponentially expanding universe , where the hubble parameter plays the role of temperature . these fields fluctuate around the bottom of the mass potential , and generate the energy density of fluctuation . finally the fluctuation creates the gravitational potential @xmath9 through the einstein equation . although each fluctuation due to single field is very small , but enormous number of species of fields contribute and it turns out to be summed up to a huge fluctuation compatible to the present observational value . furthermore it is possible to determine the radii of the compactified spaces , as well as the string coupling constant by comparison with the observational values of cmb . the very merit of our theory is in the fact that we do not need to put any assumption or fine - tuning on the form of the matter fields potential , but we simply consider free fields . however in our above described theory there still are several problems to be resolved . in the first place the spectral index @xmath11 which indicates the distance dependence of the fluctuation or , in other words , the 2-point correlation function of the fields exceeds one , i.e. @xmath12 in our model , although the observed value of the planck is @xmath13 @xcite . as was pointed out in our foregoing papers @xcite the reason why we obtained @xmath12 is in the fact that we assumed the background metric is exactly the de sitter space - time . in order to explain the experimental value of the spectral index , we may have to introduce a time dependent hubble parameter @xmath2 , something like @xmath14 . so far we have not discussed the origin of the inflation , and simply assumed the de sitter space - time . in this paper we point out the possibility that the inflation is caused by the vacuum energy of the many fields in the expanding universe . it is the purpose of the present article that we present a self - consistent theory of the time evolution of the universe by incorporating the backreaction of the quantized matter fields . the main idea of the present paper follows our pervious articles @xcite , except that we do not assume de sitter background and incorporate the backreaction of the matter fields . in the early universe we consider @xmath15 species of the free scalar fields . the time evolution of the early universe is assumed to follow the classical einstein equation , @xmath16 on the right hand side the energy - momentum tensor of the species @xmath17 of the matter fields is regarded as the vacuum expectation value @xmath18 , in the metric @xmath19 . when one expresses @xmath18 by a functional of arbitrary metric @xmath19 , @xmath20 $ ] , the einstein equation ( [ 1-einstein.eq ] ) describes the time evolution of @xmath19 caused by the backreaction from the matter fields . here we take the vacuum of the bunch - davies type in the calculation of @xmath20 $ ] . as we will see @xmath20$]is proportional to @xmath21 . therefore in the early universe , where @xmath2 is large , we can ignore the initial state dependence of the energy - momentum tensor and simply take @xmath20 $ ] . we mainly use the eikonal approximation to make solutions of the equation of motion of the matter fields . then we perform the 1-loop momentum integration to calculate the correlation functions . by so doing the eq . ( [ 1-einstein.eq ] ) becomes a second order differential equation for @xmath2 . the present paper is organized as follows . in section [ 2 ] , we review briefly the eikonal approximation . then we investigate the solution of matter field of mass @xmath22 in two cases , that is , for the cases of @xmath23 and @xmath24 , and evaluate @xmath25 for each cases . in section [ 3 ] we assume that there exist both light and heavy particles in the universe and search for de sitter - like expanding solutions of eq . ( [ 1-einstein.eq ] ) , where @xmath2 is almost constant . then we examine the solutions when the hubble parameter is very small compared to that of the de sitter - like expanding era . from this analysis we will see that the hubble parameter becomes zero and the de sitter era ends for a certain region of the initial conditions . section [ 4 ] is devoted to conclusion and outlook . in appendix [ app - airy ] we describe in detail the calculation by using the airy function around boundary between uv and ir regimes in the eikonal approximation . in appendix [ app - damping ] we show the calculation of the correlation functions using the solution in the ir regime in the eikonal approximation . it is legitimate to assume that the background metric is homogeneous and uniform as are indicated from cmb data . we thus start with investigating the one species of the scalar field described by the action @xmath26 in the friedmann - robertson - walker space - time @xmath27 for the later convenience to apply the dimensional regularization , the space - time dimension in eq . ( [ 2-action ] ) is taken to be @xmath28 with @xmath29 . in terms of the co - moving time @xmath30 we write the differentiation as @xmath31 and then the fourier mode @xmath32 of the rescaled scalar field @xmath33 obeys the following equation of motion : @xmath34\right)\chi_k=0 . \label{2-eom}\end{aligned}\ ] ] here the potential @xmath35 $ ] is defined by @xmath36 & \equiv \left(1-\frac{\epsilon}{2}\right)\frac{a^{\prime \prime}}{a } -\epsilon \left(\frac{1}{2}-\frac{\epsilon}{4}\right ) \left(\frac{a^{\prime}}{a}\right)^2-m^2a^2 \nonumber \\ & = \left(1-\frac{\epsilon}{2}\right)a^2\left(2h^2+\dot{h}\right ) -\epsilon \left(\frac{1}{2}-\frac{\epsilon}{4}\right)a^2h^2-m^2a^2 , \label{2-potential}\end{aligned}\ ] ] where @xmath2 is the hubble parameter @xmath37 the quantity we are interested in is the @xmath38-component of the energy - momentum tensor as a functional of @xmath0 ( or @xmath2 ) , @xmath39 & = \frac{\mu^{4-n}}{2}\left\langle\dot{\phi}^2 + \frac{1}{a^2}\left(\nabla \phi \right)^2+m^2\phi^2\right\rangle \nonumber \\ & = \frac{\mu^{\epsilon}}{2a^{4-\epsilon}}\bigg\ { \left\langle \left(\chi^{\prime}\right)^2\right\rangle -\left(1-\frac{\epsilon}{2}\right)ah \left\langle \chi^{\prime}\chi + \chi \chi^{\prime}\right\rangle + \left(1-\frac{\epsilon}{2}\right)^2a^2h^2 \left\langle \chi^2\right\rangle \nonumber \\ & \qquad \qquad \quad + \left\langle \left(\nabla \chi \right)^2\right\rangle + m^2a^2\left\langle \chi^2\right\rangle \bigg\}. \label{2-energy - momentum.tensor}\end{aligned}\ ] ] here a mass scale @xmath40 is introduced to keep the mass dimension of the energy - momentum tensor . we solve the equation of motion ( [ 2-eom ] ) with recourse to the eikonal approximation by putting @xmath41 then by using the boundary condition @xmath42 and the assumption of the eikonal approximation @xmath43 ( [ 2-eom ] ) has the solution @xmath44}. \label{2-eikonal.solution}\end{aligned}\ ] ] here the wronskian is used to normalize the solution : @xmath45 the general solution for @xmath46 is given by @xmath47 in the oscillatory region , @xmath48 we then quantize the field @xmath49 canonically by using the commutation relation @xmath50 = i\delta^{3-\epsilon}(\vec{x}-\vec{y } ) . \label{2-commutation.relation}\end{aligned}\ ] ] the vacuum state is defined as usual @xmath51 so that the vacuum expectation value of an operator @xmath52 is given by @xmath53 . in the following we consider the situation where @xmath54 and @xmath55 are much smaller than @xmath2 itself , @xmath56 which means that @xmath2 and @xmath54 vary much slower than the time evolution of the universe . in this subsection we evaluate the energy density @xmath58 of a scalar field when hubble parameter is much larger than mass ; @xmath57 . it is well known that if @xmath2 is constant @xmath58 is given by @xmath59 note that this result is obtained without using the eikonal approximation . when @xmath60 and @xmath61 , the potential ( [ 2-potential ] ) is always positive , i.e. @xmath36>0 . \label{2-potential.is.positive}\end{aligned}\ ] ] then the eq . ( [ 2-oscillatory.region ] ) is not always satisfied , and we should analyze the following three momentum regions to continue the solution @xmath62 in the ultra - violet limit @xmath63 ( equivalently @xmath64 ) into the infrared region : 1 . @xmath65 + in this region the solution is oscillatory as @xmath66{k^2-v(\tau ) } } e^{-i\int_{\tau_k}^{\tau } d\tau^{\prime } \sqrt{k^2-v(\tau^{\prime})}}. \label{2-oscillation.solution}\end{aligned}\ ] ] here @xmath67 is determined from @xmath68 . 2 . @xmath69 + the solution is expressed in terms of the airy function @xmath70 3 . @xmath71 + the damping and enhancing solutions given by @xmath72{v(\tau ) -k^2}}\left\{\frac{i}{2 } e^{-\int_{\tau_k}^{\tau}d\tau^{\prime}\sqrt{v(\tau^{\prime})-k^2 } } + e^{+\int_{\tau_k}^{\tau}d\tau^{\prime}\sqrt{v(\tau^{\prime})-k^2 } } \right\}. \label{2-damping.solution}\end{aligned}\ ] ] in the above ( ii ) and ( iii ) , the boundary @xmath73 is chosen to satisfy the validity condition of the eikonal approximation ( [ 2-eikonal.condition ] ) . in fact , @xmath74 here @xmath75 is obtained from ( [ 2-comoving.time-diff ] ) and ( [ 2-potential ] ) , @xmath76 we next evaluate the vacuum expectation value of the energy density in each region . in the region ( i ) it is given by @xmath77 \nonumber \\ & \quad -\frac{1}{64\pi^2}\left(-m^4 + 2m^2h^2 - 6h^2\dot{h}-2h\ddot{h}+\dot{h}^2 \right)\log \frac{2h^2+\dot{h}-m^2}{\mu^2}+o(\epsilon ) . \label{2-em.tensor - oscillation}\end{aligned}\ ] ] only in this region there appears an ultra - violet divergence which should be renormalized . indeed in ( [ 2-em.tensor-oscillation ] ) , the @xmath78 term gives the divergence in dimensional regularization . there are five terms in the coefficient of @xmath78 . the @xmath79 and @xmath80 terms are renormalized into the cosmological constant term @xmath81 and the einstein - hilbert action @xmath82 , respectively . the rest three terms are @xmath83 , @xmath84 , and @xmath85 and from the dimensional analysis their counter terms are given by @xmath86 and @xmath87 . in the case of frw metric ( [ 2-frw.metric ] ) , these two actions give the following additional terms to the euler - lagrange equation : @xmath88 therefore the three coefficients of the @xmath78 term can be renormalized by these two counter terms . in principle we may introduce finite renormalization for the three terms @xmath79 , @xmath80 and @xmath89 . however , since we have to make the same renormalization to the two cases @xmath57 and @xmath90 of the single theory , we relegate to argue this problem after the following subsection [ 2.2 ] for the case of a small hubble parameter @xmath90 . then we make use of the minimal subtraction , and expand ( [ 2-em.tensor-oscillation ] ) for large @xmath2 assuming @xmath91 and @xmath61 : @xmath92 for the regions ( ii ) and ( iii ) the contributions to @xmath93 are described in the appendices [ app - airy ] and [ app - damping ] respectively . we here show the combined result of ( ii ) and ( iii ) calculated numerically under the conditions @xmath57 , @xmath60 and @xmath61 : @xmath94 note that the coefficient of @xmath84 does not remain stationary when we vary the artificial parameter @xmath73 , which was introduced to separate two regions ( ii ) and ( iii ) . however this coefficient is very small compared to the other terms and does not affect the physical result . in this subsection we derive the energy density in the case of a small hubble parameter , @xmath90 . in the same manner as the previous subsection [ 2.1 ] we calculate the vacuum expectation value of the energy density . again we assume that the change of the hubble parameter is fairly slow , @xmath95 in this case the potential term ( [ 2-potential ] ) of the equation of motion ( [ 2-eom ] ) is always negative , @xmath36<0 . \label{2-potential.is.negative}\end{aligned}\ ] ] then the solution of the eikonal approximation ( [ 2-eikonal.solution ] ) is expressed as @xmath96\right)}.\end{aligned}\ ] ] apparently any turning point does not exist so that the solution ( [ 2-eikonal.solution ] ) is valid in the whole region of @xmath97 . thus we obtain the energy density as @xmath98 \nonumber \\ & \quad -\frac{1}{64\pi^2}\left(-m^4 + 2m^2h^2 - 6h^2\dot{h}-2h\ddot{h}+\dot{h}^2 \right)\log \frac{m^2 - 2h^2-\dot{h}}{\mu^2}+o(\epsilon ) . \label{2-em.tensor - damping}\end{aligned}\ ] ] we then make the renormalization by using the minimal subtraction , and the expansion assuming @xmath99 and @xmath61 gives the result , @xmath100 hereby we next make use of the finite renormalization for the cosmological constant term @xmath81 and the einstein - hilbert action @xmath82 in such a way that @xmath79 and @xmath101 terms and @xmath80 and @xmath102 terms are set to zero . the reason for so doing is because the energy density for small values of @xmath2 should be zero . although we should add the same counter terms to ( [ 2-em.tensor-oscillation-approx ] ) , such terms are negligible when @xmath2 is much larger than @xmath22 . furthermore we may make a finite renormalization for @xmath86 and @xmath87 which generates an additional term @xmath103 to ( [ 2-em.tensor-damping-approx ] ) . so far we have obtained the energy density of a single scalar field for two cases . the explicit form for the case of a large hubble parameter is given by @xmath104 which is the sum of ( [ 2-em.tensor-oscillation-approx ] ) and ( [ 2-em.tensor-airy+damping ] ) . for the case of a small hubble parameter we have @xmath105 we next consider the einstein equation assuming that there are many fields with various masses . then the total energy density is written as @xmath106 where @xmath107 and @xmath108 stand for the numbers of the fields whose masses are lighter and heavier than the hubble parameter , respectively . the ratio of @xmath107 to @xmath108 may vary as the hubble parameter evolves in time . however , the coefficients of each terms in ( [ 3-energy.density-right ] ) and ( [ 3-energy.density-heavy ] ) are roughly the same , so that we may replace @xmath93 and @xmath109 with their average . then ( [ 3-total.energy0 ] ) becomes @xmath110 where @xmath15 is the total number of the species . then the @xmath38-component of the einstein equation reads @xmath111 rewriting this equation after plugging ( [ 3-energy.density-right ] ) and ( [ 3-energy.density-heavy ] ) , we obtain the equation of describing the time evolution of the hubble parameter @xmath2 : @xmath112 where @xmath22 is the typical mass of the fields heavier than the hubble parameter . a comment is in order : the mass scale @xmath40 is originally arbitrary , but as a cut - off scale in the general relativity , field theories and/or string theory , it may be natural to choose planck scale @xmath113 or string scale @xmath114 . in this subsection we analytically solve the above eq . ( [ 3-einstein.eq ] ) by using linear approximation around the de sitter solution . it is clear that the eq . ( [ 3-einstein.eq ] ) has the solution , where @xmath2 is a constant , @xmath115 which is nothing but the de sitter solution . we next consider a fluctuation @xmath116 around this solution @xmath117 , @xmath118 in the linear approximation the eq . ( [ 3-einstein.eq ] ) becomes @xmath119 where @xmath120 and @xmath121 are constants given by @xmath122 the general solution reads @xmath123 let us turn to investigate the behavior of the solution ( [ 3-einstein.eq-begin-solution ] ) by using observational values . if we assume that the matter field potential is about the forth power of the gut scale , @xmath124}$ ] @xcite , we obtain @xmath125}. \label{3-hubble}\end{aligned}\ ] ] consequently , from ( [ 3-pure.de.sitter ] ) , the number of species @xmath15 is evaluated as @xmath126 let us , for example , choose @xmath40 and @xmath22 in ( [ 3-einstein.eq ] ) as @xmath127 } , \nonumber \\ & m=10^{16}\text { [ gev ] } , \label{3-scales}\end{aligned}\ ] ] where @xmath114 is the string scale . in the case of @xmath128 , in which we do not make an additional finite renormalization to the minimal subtraction , the solution ( [ 3-einstein.eq-begin-solution ] ) becomes @xmath129 if @xmath130 satisfies @xmath131 the duration of the inflation becomes @xmath132 i.e. the e - folding number exceeds @xmath133 . in fig . [ fig : energy - backreaction_inflation1](a ) we depict the time evolution of the hubble parameter with the choices @xmath134 and @xmath135 . the present approximation @xmath60 and @xmath61 is valid in the region @xmath136 in view of ( [ 3-pure.de.sitter ] ) this is nothing but the region @xmath137 however , if we introduce an additional finite renormalization in the form of @xmath138 we can relax the condition of @xmath130 by choosing appropriate @xmath139 . for instance if we choose @xmath140 , the solution ( [ 3-einstein.eq-begin-solution ] ) reads @xmath141 @xmath2 is depicted in fig . [ fig : energy - backreaction_inflation1](b ) for @xmath142 and @xmath143 , which indicates that @xmath2 is close to @xmath117 in the region ( [ 3-inflation.era0 ] ) . : the inflation.,title="fig:",width=302 ] : the inflation.,title="fig:",width=302 ] ( -350,0)@xmath144 ( -120,0)@xmath144 ( -480,120)@xmath145 ( -380,100)(a ) ( -360,100)@xmath128 ( -360,80)@xmath134 ( -360,60)@xmath135 ( -150,100)(b ) ( -130,100)@xmath140 ( -130,80)@xmath142 ( -130,60)@xmath143 [ fig : energy - backreaction_inflation1 ] in the previous subsection [ 3.1 ] , we have found the inflationary solution has a sufficiently large e - folding number , and after the exponential expansion , the hubble parameter starts to decrease . in this subsection we show that in the above solution the hubble parameter becomes zero for large @xmath146 , which indicates that the inflation will end automatically . as we have seen in the previous subsection the eq . ( [ 3-einstein.eq ] ) becomes invalid sometime after @xmath2 starts to decrease . however the eq . ( [ 3-einstein.eq ] ) becomes valid again if @xmath2 is sufficiently small . therefore if we assume that @xmath2 continues to decrease , we can use eq . ( [ 3-einstein.eq ] ) to describe how the inflation stops . in this region @xmath147 term on the right hand side of ( [ 3-einstein.eq ] ) is much smaller than @xmath148 term on the left hand side because they balanced before @xmath2 becomes small : @xmath149 thus the eq . ( [ 3-einstein.eq ] ) reads @xmath150 where the constants @xmath151 and @xmath152 are given by @xmath153 here @xmath15 and @xmath40 are given by ( [ 3-number.of.species ] ) and ( [ 3-scales ] ) . the solution @xmath2 of ( [ 3-einstein.eq-end2 ] ) rapidly decreases as @xmath146 increases and then , with small oscillation , @xmath154 as @xmath155 for various choices of @xmath22 with @xmath156 . for example when we impose an appropriate initial condition so as to satisfy @xmath157 and @xmath158 and perform the numerical calculation for two cases @xmath159 and @xmath160 [ gev ] . the results are depicted in fig . [ fig : energy - backreaction_inflation2 ] . here fig . [ fig : energy - backreaction_inflation2](a ) and fig . [ fig : energy - backreaction_inflation2](b ) shows the results for @xmath128 and @xmath140 respectively . with the initial condition @xmath161 and @xmath162.,title="fig:",width=302 ] with the initial condition @xmath161 and @xmath162.,title="fig:",width=302 ] ( -340,0)@xmath144 ( -110,0)@xmath144 ( -490,120)@xmath163 ( -370,140)(a ) ( -350,140)@xmath128 ( -320,120)@xmath159 [ gev ] ( -350,123 ) ( -320,100)@xmath164 [ gev ] ( -350,103 ) ( -140,140)(b ) ( -120,140)@xmath140 ( -90,120)@xmath159 [ gev ] ( -120,123 ) ( -90,100)@xmath164 [ gev ] ( -120,103 ) in the present paper we investigate the effect of the backreaction to the expansion of the space - time due to the matter fields existing in the universe . on arbitrary friedmann - robertson - walker space - time , we quantize , as an example of the matter field , the scalar field and then we calculate the vacuum expectation value of the energy density . since we are not able to obtain the exact solution of the scalar field , we make use of the eikonal approximation to construct the solution . we then plug it in the energy density of the matter field . combining it with the friedmann equation we obtain a self - consistent equation that describes the time evolution of the space - time . thus we can determine the behavior of the early universe including the backreaction of the matter fields . as a result we find the following two eras in the early universe : 1 . we obtain the inflationary solution with the e - folding number @xmath1 if we make a fine tuning of few orders of magnitude for the initial condition . this initial condition may be determined by a scenario of the birth of the universe in string theory and/or quantum gravity . 2 . after the era ( 1 ) the hubble parameter will continue to decrease and then with small oscillation @xmath2 goes to zero asymptotically . then finally the inflation ends . in this manner by imposing some reasonable conditions we can make a inflation scenario without inflaton . the problems next to be discussed are why the temperature fluctuation in the observed cmb is of the order @xmath165 and very small non - gaussianity is realized ( cf . @xcite , @xmath166 ( 95`% ` cl ) ) . in our previous papers @xcite , by using the de sitter background without incorporating the backreaction , we estimated the cmb temperature fluctuation and non - gaussianity @xmath167 using our theory of many scalar fields , and obtained that if @xmath165 then @xmath168 . the difference between the present and previous papers ia in the fact that in the present paper we incorporated the effect of backreaction , which may not affect largely to the fluctuation , because in both cases the fluctuation is produced during the exponentially expanding era . furthermore in the previous papers @xcite we had the problem that the spectral index @xmath11 exceeded 1 and contradicted the observed value . however as is mentioned in section [ 1 ] in the present paper the condition for @xmath169 translated into @xmath170 and @xmath171 is satisfied in the whole range of the inflation era ( see fig [ fig : energy - backreaction_inflation1 - 2 ] ) . for the case of fig . [ fig : energy - backreaction_inflation1](a).,title="fig:",width=302 ] ( 7,37)@xmath144 ( -270,120)@xmath172 ( -80,180)@xmath128 ( -80,160)@xmath134 ( -80,140)@xmath135 furthermore we consider a new mechanism to produce the temperature fluctuation in cmb that is different from our previous papers @xcite . in @xcite , we pointed out that the fluctuation of the energy density @xmath173 in the universe creates the gravitational potential @xmath9 through the einstein equation and turns to be the temperature fluctuation of cmb . a new mechanism we would like to discuss here is that the fluctuation @xmath174 of the time @xmath175 at which the inflation ends generates the density perturbation , as in the ordinary scenario with the inflaton , @xmath176 . @xmath174 can be roughly estimated as follows . first we express @xmath175 in terms of the initial values @xmath177 and @xmath178 . we define @xmath175 as the time @xmath146 when the third term of the eq . ( [ 3-einstein.eq-begin-solution1 ] ) becomes of order @xmath179 : @xmath180 on the other hand from ( [ 3-einstein.eq-begin-solution1 ] ) we obtain @xmath181 by combining ( [ 4-inflation.ends ] ) and ( [ 4-initial.condition ] ) , we obtain @xmath182 then @xmath174 is given by @xmath183 on the other hand by taking the variation of the friedmann equation @xmath184 , we have @xmath185 from ( [ 4-fluctuation.of.time ] ) and ( [ 4-variation.of.friedmann ] ) we obtain the density perturbation as @xmath186 using the eq . ( [ 3-number.of.species ] ) , @xmath187 is evaluated as @xmath188 which is consistent with the observational value . it would be interesting to investigate this new mechanism in detail . the authors acknowledge y. sekino for discussion in the early stage of the present work . one of the authors h.k is supported by the jsps grant in aid for scientific research no . m.n is also supported by the jsps grant in aid for scientific research no . in the region @xmath69 , there is no ultra - violet divergence , so that we can take @xmath190 . the solution of the scalar field in this region is given by the eq . ( [ 2-airy.solution ] ) , @xmath191 where the lower end of the integration of @xmath192 is in fact the boundary @xmath73 obtained in eq . ( [ 2-airy.boundary ] ) , @xmath193 in order to carry out the @xmath192-integration of the correlation functions such as @xmath194 , let us evaluate @xmath67 and @xmath195 approximately under the conditions @xmath60 and @xmath61 . firstly , @xmath67 is determined from the equation @xmath196 imposing the condition @xmath60 , we can approximate the scale factor @xmath0 to be the pure de sitter one , @xmath197 , and obtain @xmath198 thus @xmath67 is evaluated as @xmath199 in the same manner @xmath195 becomes @xmath200 then we would like to calculate the four kinds of correlation functions in eq . ( [ 2-energy-momentum.tensor ] ) in order . firstly , @xmath194 is calculated as @xmath201 here ( [ app - tau_k.approx ] ) and ( [ app-b.approx ] ) are used to obtain the last line . after the change of variable @xmath202 , this integration turns to be @xmath203 here we define the upper end of the integration region as @xmath204 by assuming that @xmath205 and @xmath206 are small , this upper end can be approximated as @xmath207 then we perform the integrations including the airy functions numerically and expand them in terms of @xmath205 and @xmath206 . in the same way , we calculate the correlations @xmath208 , @xmath209 and @xmath210 as follows : @xmath211 @xmath212 @xmath213 similarly to the region ( ii ) , we do not have the ultra - violet divergence in the region @xmath71 , so that we take @xmath190 . as we stated in the subsection [ 2.1 ] , from ( [ 2-airy.boundary ] ) , we have the upper end of this region @xmath214 and the solution of the scalar field is given by eq . ( [ 2-damping.solution ] ) : @xmath215{v(\tau ) -k^2}}\left\ { \frac{i}{2}e^{-\int_{\tau_k}^{\tau}d\tau^{\prime } \sqrt{v(\tau^{\prime})-k^2 } } + e^{+\int_{\tau_k}^{\tau}d\tau^{\prime } \sqrt{v(\tau^{\prime})-k^2}}\right\}. \label{app - damping.solution}\end{aligned}\ ] ] here let us calculate @xmath216 integration of the exponent of the fourier mode @xmath217 @xmath218 under the condition @xmath60 . in this situation we can approximate the scale factor as the pure de sitter solution @xmath219 , and we obtain @xmath220 now we evaluate the four kinds of correlation functions included in the energy density ( [ 2-energy-momentum.tensor ] ) . firstly , as for the correlation @xmath194 , applying the result ( [ app-damping.exponent ] ) , we find @xmath221 by changing the integration variable as @xmath222 , and using the lower end of @xmath223 integration , @xmath224 we rewrite @xmath194 as @xmath225 furthermore we expand @xmath226 with respect to @xmath227 and @xmath206 : @xmath228 . \label{app - correlation - mass}\end{aligned}\ ] ] here , in each terms with @xmath229 , we approximate the lower end @xmath230 of the @xmath223 integration as @xmath231 here we have used ( [ 2-potential ] ) and ( [ 2-potential - prime ] ) to obtain the second equality . in practice we calculate the @xmath223 integrations numerically . in the same manner we evaluate the rest correlations @xmath208 , @xmath209 and @xmath210 . @xmath232 . \label{app - correlation - nabla}\end{aligned}\ ] ] @xmath233 \nonumber \\ & = \frac{1}{4\pi^2}\int_0^{k_0}dk\ > k^2 \bigg[\frac{1}{4}\left\{\frac{v^{\prime}}{4}(v - k^2)^{-\frac{5}{4 } } + ( v - k^2)^{\frac{1}{4}}\right\}^2 \left(\textstyle \frac{\sqrt{v}+\sqrt{v - k^2}}{\sqrt{v}-\sqrt{v - k^2 } } \right)^{-\frac{\sqrt{v}}{ah}}e^{2\frac{\sqrt{v - k^2}}{ah } } \nonumber \\ & \qquad \qquad \qquad \qquad + \left\{\frac{v^{\prime}}{4}(v - k^2)^{-\frac{5}{4 } } -(v - k^2)^{\frac{1}{4}}\right\}^2 \left(\textstyle \frac{\sqrt{v}+\sqrt{v - k^2}}{\sqrt{v}-\sqrt{v - k^2 } } \right)^{\frac{\sqrt{v}}{ah}}e^{-2\frac{\sqrt{v - k^2}}{ah}}\bigg ] \nonumber \\ & = \frac{v^2}{4\pi^2 } \int_{z_0}^1dz\ > z^2\sqrt{1-z^2}\left[\frac{1}{4 } \left\{\frac{1}{4}\frac{v^{\prime}}{v^{\frac{3}{2}}}\frac{1}{z^3}+1 \right\}^2\left(\frac{1+z}{1-z}\right)^{-\frac{\sqrt{v}}{ah } } e^{2\frac{\sqrt{v}}{ah}z } + \left\{\frac{1}{4}\frac{v^{\prime}}{v^{\frac{3}{2}}}\frac{1}{z^3}-1 \right\}^2\left(\frac{1+z}{1-z}\right)^{\frac{\sqrt{v}}{ah } } e^{-2\frac{\sqrt{v}}{ah}z}\right ] \nonumber \\ & = \frac{1}{4\pi^2}\frac{(v^{\prime})^2}{v}\times \frac{1}{16 } \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^4}\left\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\frac{\sqrt{v}}{ah } } e^{2\frac{\sqrt{v}}{ah}z } + \left(\frac{1+z}{1-z}\right)^{\frac{\sqrt{v}}{ah } } e^{-2\frac{\sqrt{v}}{ah}z}\right\ } \nonumber \\ & \quad + \frac{1}{4\pi^2}(v^{\prime})^2\sqrt{v}\times \frac{1}{2 } \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z}\left\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\frac{\sqrt{v}}{ah } } e^{2\frac{\sqrt{v}}{ah}z } -\left(\frac{1+z}{1-z}\right)^{\frac{\sqrt{v}}{ah } } e^{-2\frac{\sqrt{v}}{ah}z}\right\ } \nonumber \\ & \quad + \frac{1}{4\pi^2}v^2 \int_{z_0}^1dz\ > z^2\sqrt{1-z^2}\left\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\frac{\sqrt{v}}{ah } } e^{2\frac{\sqrt{v}}{ah}z } + \left(\frac{1+z}{1-z}\right)^{\frac{\sqrt{v}}{ah } } e^{-2\frac{\sqrt{v}}{ah}z}\right\ } \nonumber \\ & = \frac{a^4h^4}{4\pi^2 } \left(4 + 6\frac{\dot{h}}{h^2}+\frac{\ddot{h}}{h^3}-2\frac{m^2}{h^2 } \right)^2 \left(2+\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^{-1 } \nonumber \\ & \quad \times \bigg [ \frac{1}{16}\int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^4 } \bigg\{\textstyle \frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } + \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \quad -\frac{\sqrt{2}}{64}\left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right ) \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^4}\textstyle \left(\log \frac{1+z}{1-z}-2z\right)\bigg\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } -\left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \quad + \frac{1}{16\times 32 } \left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^2 \nonumber \\ & \qquad \qquad \times \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^4}\textstyle \bigg\{\frac{1}{4}\left((\sqrt{2}-8z)\log \frac{1+z}{1-z } + 2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z+8z^2\right ) \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } \nonumber \\ & \qquad \qquad \qquad \qquad \qquad \qquad \textstyle -\left((\sqrt{2}+8z)\log \frac{1+z}{1-z } -2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z-8z^2\right ) \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \quad + \cdots \bigg ] \nonumber \\ & \quad + \frac{a^4h^4}{4\pi^2 } \left(4 + 6\frac{\dot{h}}{h^2}+\frac{\ddot{h}}{h^3}-2\frac{m^2}{h^2 } \right ) \left(2+\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^{\frac{1}{2 } } \nonumber \\ & \qquad \times \bigg [ \frac{1}{2}\int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z } \bigg\{\textstyle \frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } -\left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad -\frac{\sqrt{2}}{8}\left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right ) \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z}\textstyle \left(\log \frac{1+z}{1-z}-2z\right)\bigg\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } + \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad + \frac{1}{64 } \left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^2 \nonumber \\ & \qquad \qquad \quad \times \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z}\textstyle \bigg\{\frac{1}{4}\left((\sqrt{2}-8z)\log \frac{1+z}{1-z } + 2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z+8z^2\right ) \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } \nonumber \\ & \qquad \qquad \qquad \qquad \qquad \qquad \quad \textstyle + \left((\sqrt{2}+8z)\log \frac{1+z}{1-z } -2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z-8z^2\right ) \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad + \cdots \bigg ] \nonumber \\ & \quad + \frac{a^4h^4}{4\pi^2 } \left(2+\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^2 \nonumber \\ & \qquad \times \bigg [ \int_{z_0}^1dz\ > z^2\sqrt{1-z^2 } \bigg\{\textstyle \frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } + \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad -\frac{\sqrt{2}}{4}\left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right ) \int_{z_0}^1dz\ > z^2\sqrt{1-z^2}\textstyle \left(\log \frac{1+z}{1-z}-2z\right)\bigg\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } -\left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad + \frac{1}{32 } \left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^2 \nonumber \\ & \qquad \qquad \quad \times \int_{z_0}^1dz\ > z^2\sqrt{1-z^2}\textstyle \bigg\{\frac{1}{4}\left((\sqrt{2}-8z)\log \frac{1+z}{1-z } + 2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z+8z^2\right ) \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } \nonumber \\ & \qquad \qquad \qquad \qquad \qquad \qquad \quad \textstyle -\left((\sqrt{2}+8z)\log \frac{1+z}{1-z } -2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z-8z^2\right ) \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad + \cdots \bigg ] . \nonumber \\ \label{app - correlation - kinetic}\end{aligned}\ ] ] @xmath234 \nonumber \\ & = -\frac{1}{2\pi^2}\int_0^{k_0}dk\ > k^2 \bigg[\frac{1}{4}\left\{\frac{v^{\prime}}{4}(v - k^2)^{-\frac{3}{2}}+1 \right\ } \left(\textstyle \frac{\sqrt{v}+\sqrt{v - k^2}}{\sqrt{v}-\sqrt{v - k^2 } } \right)^{-\frac{\sqrt{v}}{ah}}e^{2\frac{\sqrt{v - k^2}}{ah } } \nonumber \\ & \qquad \qquad \qquad \qquad + \left\{\frac{v^{\prime}}{4}(v - k^2)^{-\frac{3}{2}}-1\right\ } \left(\textstyle \frac{\sqrt{v}+\sqrt{v - k^2}}{\sqrt{v}-\sqrt{v - k^2 } } \right)^{\frac{\sqrt{v}}{ah}}e^{-2\frac{\sqrt{v - k^2}}{ah}}\bigg ] \nonumber \\ & = -\frac{v^{\frac{3}{2}}}{2\pi^2 } \int_{z_0}^1dz\ > z\sqrt{1-z^2}\left[\frac{1}{4 } \left(\frac{1}{4}\frac{v^{\prime}}{v^{\frac{3}{2}}}\frac{1}{z^3}+1 \right)\left(\frac{1+z}{1-z}\right)^{-\frac{\sqrt{v}}{ah } } e^{2\frac{\sqrt{v}}{ah}z } + \left(\frac{1}{4}\frac{v^{\prime}}{v^{\frac{3}{2}}}\frac{1}{z^3}-1 \right)\left(\frac{1+z}{1-z}\right)^{\frac{\sqrt{v}}{ah } } e^{-2\frac{\sqrt{v}}{ah}z}\right ] \nonumber \\ & = -\frac{1}{4\pi^2}v^{\prime}\times \frac{1}{2 } \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^2}\left\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\frac{\sqrt{v}}{ah } } e^{2\frac{\sqrt{v}}{ah}z } + \left(\frac{1+z}{1-z}\right)^{\frac{\sqrt{v}}{ah } } e^{-2\frac{\sqrt{v}}{ah}z}\right\ } \nonumber \\ & \quad -\frac{1}{4\pi^2}v^{\frac{3}{2}}\times 2 \int_{z_0}^1dz\ > z\sqrt{1-z^2}\left\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\frac{\sqrt{v}}{ah } } e^{2\frac{\sqrt{v}}{ah}z } -\left(\frac{1+z}{1-z}\right)^{\frac{\sqrt{v}}{ah } } e^{-2\frac{\sqrt{v}}{ah}z}\right\ } \nonumber \\ & = -\frac{a^3h^3}{4\pi^2 } \left(4 + 6\frac{\dot{h}}{h^2}+\frac{\ddot{h}}{h^3}-2\frac{m^2}{h^2 } \right ) \nonumber \\ & \quad \times \bigg [ \frac{1}{2}\int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^2 } \bigg\{\textstyle \frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } + \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \quad -\frac{\sqrt{2}}{8}\left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right ) \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^2}\textstyle \left(\log \frac{1+z}{1-z}-2z\right)\bigg\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } -\left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \quad + \frac{1}{64 } \left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^2 \nonumber \\ & \qquad \qquad \times \int_{z_0}^1dz\ > \frac{\sqrt{1-z^2}}{z^2}\textstyle \bigg\{\frac{1}{4}\left((\sqrt{2}-8z)\log \frac{1+z}{1-z } + 2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z+8z^2\right ) \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } \nonumber \\ & \qquad \qquad \qquad \qquad \qquad \qquad \textstyle -\left((\sqrt{2}+8z)\log \frac{1+z}{1-z } -2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z-8z^2\right ) \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \quad + \cdots \bigg ] \nonumber \\ & \quad -\frac{a^3h^3}{4\pi^2 } \left(2+\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^{\frac{3}{2 } } \nonumber \\ & \qquad \times \bigg [ 2\int_{z_0}^1dz\ > z\sqrt{1-z^2 } \bigg\{\textstyle \frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } -\left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad -\frac{\sqrt{2}}{2}\left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right ) \int_{z_0}^1dz\ > z\sqrt{1-z^2}\textstyle \left(\log \frac{1+z}{1-z}-2z\right)\bigg\{\frac{1}{4 } \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } + \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad + \frac{1}{16 } \left(\frac{\dot{h}}{h^2}-\frac{m^2}{h^2}\right)^2 \nonumber \\ & \qquad \qquad \quad \times \int_{z_0}^1dz\ > z\sqrt{1-z^2}\textstyle \bigg\{\frac{1}{4}\left((\sqrt{2}-8z)\log \frac{1+z}{1-z } + 2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z+8z^2\right ) \left(\frac{1+z}{1-z}\right)^{-\sqrt{2}}e^{2\sqrt{2}z } \nonumber \\ & \qquad \qquad \qquad \qquad \qquad \qquad \quad \textstyle + \left((\sqrt{2}+8z)\log \frac{1+z}{1-z } -2\left(\log \frac{1+z}{1-z}\right)^2 - 2\sqrt{2}z-8z^2\right ) \left(\frac{1+z}{1-z}\right)^{\sqrt{2}}e^{-2\sqrt{2}z}\bigg\ } \nonumber \\ & \qquad \qquad + \cdots \bigg ] . \label{app - correlation - cross}\end{aligned}\ ] ]

for let denote the supremum of the numbers such that every-hlder continuous function is of bounded variation on a set of hausdorff dimension . kahane and katznelson ( 2009 ) proved the estimate and asked whether the upper bound is sharp . we show that in fact . let and denote the hausdorff and upper minkowski dimension , respectively . the upper bound on is a consequence of the following theorem . let\}$ ] be a fractional brownian motion of hurst index . then , almost surely , there exists no set $ ] such that and is of bounded variation . furthermore , almost surely , there exists no set $ ] such that and is-hlder continuous for some . the zero set and the set of record times of witness that the above theorems give the optimal dimensions . we also prove similar restriction theorems for deterministic self - affine functions and generic-hlder continuous functions . finally , let\}$ ] be a two - dimensional brownian motion . we prove that , almost surely , there is a compact set $ ] such that and is non - decreasing in each coordinate . it remains open whether is best possible .

1504.04789

let @xmath22 $ ] denote the set of @xmath3-hlder continuous functions @xmath23\to { \mathbb{r}}$ ] . in 2009 , kahane and katznelson @xcite proved the following result and asked whether it is sharp . for every @xmath0 there exists a function @xmath24 such that if @xmath9 $ ] and @xmath25 is of bounded variation , then the hausdorff dimension satisfies @xmath26 . is the above result the best possible ? we answer this question negatively and determine the optimal bound . let @xmath27}\{\dim_{\mathcal{h}}a : f|_{a } \textrm { is of bounded variation}\},\ ] ] so that the above theorem states @xmath28 , see figure [ f : v ] . [ t : va ] for all @xmath0 we have @xmath29 ; ; ; ; ( 6.7,0 ) ; \to [ 0,1]$ ] with hlder exponent . its graph consists of 6 affine copies of itself.,scaledwidth=90.0% ] kahane and katznelson also asked about dimensions of sets @xmath30 such that the restriction to @xmath30 is hlder continuous . ( see the next section for related results . ) we present two constructions , one deterministic and one stochastic , of functions that are not hlder on any set of high enough dimension . first we consider self - affine functions . these are constructed in definition [ d : sa ] below , see figure [ f:23 ] for illustration . [ t : sa0 ] there is a dense set @xmath31 with the following property . for each @xmath32 there is a self - affine function @xmath33 such that for all @xmath9 $ ] 1 . if @xmath34 is @xmath14-hlder continuous for some @xmath15 , then @xmath35 ; 2 . if @xmath34 is of bounded variation , then @xmath36 . for a stochastically self - affine process , fractional brownian motion ( see definition [ d : fbm ] ) , we prove the following . [ t : fbm0 ] let @xmath0 and let @xmath8\}$ ] be a fractional brownian motion of hurst index @xmath3 . then , almost surely , for all @xmath9 $ ] 1 . if @xmath37 is @xmath14-hlder continuous for some @xmath15 , then @xmath35 ; 2 . if @xmath37 is of bounded variation , then @xmath38 . [ c : fbm ] let @xmath0 and let @xmath8\}$ ] be a fractional brownian motion of hurst index @xmath3 . then @xmath39 let @xmath40 be the zero set of @xmath16 and let @xmath41 : b(t)=\max_{s\in [ 0,t ] } b(s)\}$ ] be the set of record times of @xmath16 . it is classical that , almost surely , @xmath42 , see ( * ? ? ? * chapter 18 ) . for the record , let us state the following , more subtle fact . [ p : r ] almost surely , @xmath43 . we could not find a reference for this in the literature , and include a proof in section [ s : fbm ] . clearly @xmath40 and @xmath44 witness that theorem [ t : fbm0 ] and corollary [ c : fbm ] are best possible . simon @xcite proved that a standard linear brownian motion is not monotone on any set of positive lebesgue measure . theorem [ t : fbm0 ] for @xmath45 with hausdorff dimension in place of upper minkowski dimension is due to balka and peres @xcite . the methods used there do not extend to minkowski dimension or to general exponents @xmath3 . related results in the discrete setting , concerning non - decreasing subsequences of random walks , can be found in @xcite . now we consider higher dimensional brownian motion . let @xmath46 and @xmath23\to { \mathbb{r}}^d$ ] . we say that @xmath47 is _ non - decreasing _ on a set @xmath9 $ ] if all the coordinate functions of @xmath48 are non - decreasing . [ t:2dbm ] let @xmath8\}$ ] be a standard two - dimensional brownian motion . then , almost surely , there exists a compact set @xmath18 $ ] such that @xmath16 is non - decreasing on @xmath49 and @xmath19 . corollary [ c : fbm ] ( or ( * ? ? ? * theorem 1.2 ) ) implies that , almost surely , the @xmath50-dimensional brownian motion @xmath16 can not be non - decreasing on any set of hausdorff dimension larger than @xmath51 . the following problem remains open in all dimensions @xmath46 . let @xmath46 and let @xmath52 be a standard @xmath50-dimensional brownian motion . what is the supremum of the numbers @xmath53 such that , almost surely , @xmath16 is non - decreasing on some set of hausdorff dimension @xmath53 ? finally , we prove restriction theorems for a generic @xmath3-hlder continuous function ( in the sense of baire category ) , see the following section for the details . let @xmath54 $ ] denote the set of continuous functions @xmath23\to { \mathbb{r}}$ ] endowed with the maximum norm . elekes ( * ? ? ? * theorems 1.4 , 1.5 ) proved the following restriction theorem . assume that @xmath55 . for a generic continuous function @xmath56 $ ] ( in the sense of baire category ) for all @xmath9 $ ] 1 . if @xmath48 is @xmath14-hlder continuous , then @xmath57 ; 2 . if @xmath48 is of bounded variation , then @xmath58 . kahane and katznelson ( * ? ? ? * theorems 2.1 , 3.1 ) , and independently mth ( * ? ? ? * theorems 1.4 , 1.5 ) proved that the above result is sharp . [ t : kkm ] let @xmath55 . for every @xmath56 $ ] there are compact sets @xmath59 $ ] such that 1 . @xmath60 and @xmath48 is @xmath14-hlder continuous ; 2 . @xmath61 and @xmath62 is of bounded variation . in other words there is always a set @xmath30 with the given properties and dimension , and for generic functions there is no @xmath30 of larger dimension . let us recall that the _ @xmath14-variation _ of a function @xmath63 is defined as @xmath64 in the theorems above , bounded variation can be generalized to finite @xmath14-variation for all @xmath65 by similar methods . for the following result see mth ( * ? ? ? * theorem 5.2 ) . [ t : m ] let @xmath65 and @xmath56 $ ] . then there is a compact set @xmath9 $ ] such that @xmath66 and @xmath48 has finite @xmath14-variation . our initial interest came from questions of kahane and katznelson @xcite on restrictions of hlder continuous functions . first we need the following definition . let @xmath67 be the set of @xmath3-hlder continuous functions @xmath63 . for all @xmath0 and @xmath65 define @xmath68 } \sup_{a\subset [ 0,1 ] } \left\{\dim_{\mathcal{h}}a : f|_{a}\in c^{\beta}(a)\right\ } , \\ v(\alpha,\beta)&=\inf_{f\in c^{\alpha}[0,1 ] } \sup_{a\subset [ 0,1 ] } \left\{\dim_{\mathcal{h}}a : v^{\beta}(f|_{a})<\infty\right\}. \end{aligned}\ ] ] replacing hausdorff dimension by upper minkowski dimension yields @xmath69 } \sup_{a\subset [ 0,1 ] } \left\{\overline{\dim}_{\mathcal{m}}a : f|_{a}\in c^{\beta}(a)\right\ } , \\ \overline{v}(\alpha,\beta)&=\inf_{f\in c^{\alpha}[0,1 ] } \sup_{a\subset [ 0,1 ] } \left\{\overline{\dim}_{\mathcal{m}}a : v^{\beta}(f|_{a})<\infty\right\}.\end{aligned}\ ] ] as the hausdorff dimension is smaller than or equal to the upper minkowski dimension , @xmath70 and @xmath71 . if @xmath72 then @xmath73 for all @xmath74 $ ] , so @xmath75 . for the following theorem see ( * ? ? * theorems 5.1 , 5.2 ) . for all @xmath76 we have @xmath77 [ q : kk ] are the above bounds optimal ? we answer this question negatively and find the sharp bounds , which generalizes theorem [ t : va ] . [ t : answer ] for all @xmath0 we have @xmath78 in [ s : slte ] we prove restriction theorems for functions which satisfy certain scaled local time estimates . this allows us to prove the following more general version of theorems [ t : sa0 ] and [ t : fbm0 ] , see [ s : kk , s : fbm ] , respectively . [ t : sa ] there is a dense set @xmath31 with the following property . for each @xmath79 there is a self - affine function @xmath80 $ ] such that for all @xmath9 $ ] 1 . [ eq : sa1 ] if @xmath81 for some @xmath15 , then @xmath35 ; 2 . [ eq : sa2 ] if @xmath82 for some @xmath65 , then @xmath83 . [ t : fbm ] let @xmath0 and let @xmath8\}$ ] be a fractional brownian motion of hurst index @xmath3 . then , almost surely , for all @xmath9 $ ] 1 . if @xmath84 for some @xmath15 , then @xmath35 ; 2 . [ eq : fbm2 ] if @xmath85 for some @xmath65 , then @xmath86 . the zero set of @xmath16 and the following result ( see ( * ? ? ? * theorem 4.3 ) ) with lemma [ l : ub ] witness that theorem [ t : fbm ] is sharp for all @xmath87 . [ t : a ] let @xmath0 and @xmath88 be fixed . then there is a compact set @xmath9 $ ] ( which depends only on @xmath3 and @xmath14 ) such that @xmath89 and if @xmath23\to { \mathbb{r}}$ ] is a function and @xmath90 such that for all @xmath91 $ ] we have latexmath:[\[\label{eq : mod } @xmath48 has finite @xmath14-variation . in section [ s : kk ] we prove theorem [ t : answer ] by using [ t : sa ] to obtain the sharp upper bounds for @xmath93 and @xmath94 . theorem [ t : fbm ] may be used there instead of theorem [ t : sa ] . finally , theorems [ t : kkm ] , [ t : m ] and [ t : a ] provide the optimal lower bounds for @xmath95 and @xmath96 . in section [ s:2d ] we consider higher dimensional brownian motion and prove theorem [ t:2dbm ] . in order to do so , we establish a general limit theorem for random sequences with i.i.d . increments , which is of independent interest . finally , in section [ s : generic ] we study generic @xmath3-hlder continuous functions in the sense of baire category . for @xmath0 let @xmath97 $ ] be the set of functions @xmath23\to { \mathbb{r}}$ ] such that for all @xmath91 $ ] we have @xmath98 let us endow @xmath97 $ ] with the maximum metric , then it is a complete metric space and hence we can use baire category arguments . we show that a generic @xmath99 $ ] witnesses @xmath100 and @xmath101 for all @xmath14 simultaneously . [ t : gen ] let @xmath0 . for a generic @xmath99 $ ] for all @xmath9 $ ] 1 . [ eq : gen1 ] if @xmath102 for some @xmath103 , then @xmath57 ; 2 . [ eq : gen2 ] if @xmath104 for some @xmath65 , then @xmath105 . let @xmath9 $ ] be non - empty and @xmath106 . a function @xmath107 is called ( uniformly ) _ @xmath3-hlder continuous _ if there exists a constant @xmath108 such that @xmath109 for all @xmath110 . for the definitions of @xmath54 $ ] , @xmath111 $ ] , and @xmath112 $ ] see section [ s : rel ] . the diameter of @xmath30 is denoted by @xmath113 . for all @xmath114 the _ @xmath115-hausdorff content _ of @xmath30 is @xmath116 the _ hausdorff dimension _ of @xmath30 is defined as @xmath117 let @xmath118 denote the cardinality of the set @xmath119 . let @xmath120 be an integer . for all @xmath121 and @xmath9 $ ] define @xmath122 the _ upper minkowski dimension _ of @xmath30 is defined as @xmath123 it is easy to show that this definition is independent of the choice of @xmath124 and we have @xmath125 for all @xmath9 $ ] . for more on these concepts see @xcite . [ d : sa ] a compact set @xmath126 is called _ self - affine _ if for some @xmath120 there are injective and contractive affine maps @xmath127 such that @xmath128 a continuous function @xmath56 $ ] is _ self - affine _ if @xmath129 is a self - affine set . [ d : fbm ] let @xmath0 . the process @xmath130 is called a _ fractional brownian motion of hurst index @xmath3 _ if * @xmath16 is a gaussian process with stationary increments ; * @xmath131 and @xmath132 has standard normal distribution for every @xmath133 ; * almost surely , the function @xmath134 is continuous . the covariance function of @xmath16 is @xmath135 . it is well known that almost surely @xmath16 is @xmath53-hlder continuous for all @xmath136 , see [ l : ub ] below . for more information see ( * ? ? ? * chapter 8) and ( * ? ? ? * chapter 18 ) . let @xmath137 be a _ complete _ metric space . a set is _ somewhere dense _ if it is dense in a non - empty open set , otherwise it is _ nowhere dense_. we say that @xmath138 is _ meager _ if it is a countable union of nowhere dense sets , and a set is called _ co - meager _ if its complement is meager . by baire s category theorem a set is co - meager iff it contains a dense @xmath139 set . we say that the _ generic _ element @xmath140 has property @xmath141 if @xmath142 is co - meager . let @xmath143,d_{{\mathcal{h}}})$ ] be the set of non - empty compact subsets of @xmath144 $ ] endowed with the _ hausdorff metric _ , that is , for each @xmath145 $ ] we have @xmath146 where @xmath147 . then @xmath143,d_{h})$ ] is a compact metric space , see @xcite for more on this concept . let @xmath148 stand for the support of the measure @xmath149 . for @xmath150 let @xmath151 and @xmath152 denote the lower and upper integer part of @xmath153 , respectively . in this section we prove restriction theorems for functions satisfying a scaled local time estimate . first we need some notation . let @xmath154 $ ] and an integer @xmath120 be fixed . let @xmath121 and @xmath155 . a _ time interval of order @xmath156 _ is of the form @xmath157 let @xmath158 . a _ value interval of order @xmath156 _ is of the form @xmath159 for all @xmath160 define @xmath161 where @xmath162 is the set of time intervals of order @xmath156 . clearly , @xmath163 . [ d : an ] for a function @xmath23\to { \mathbb{r}}$ ] the _ scaled local time _ @xmath164 is the number of order @xmath156 intervals in @xmath165 in which @xmath47 takes at least one value in @xmath166 : @xmath167 it is easy to see that if @xmath47 is @xmath3-hlder continuous then for every @xmath168 , for some @xmath169 we have @xmath170 , since the function can not visit too many different value intervals in any given time interval . finally , for each @xmath171 define @xmath172 thus the set @xmath173 includes @xmath3-hlder functions with scaled local times which are not much larger than the minimal values possible given their continuity . we shall see below that the self - affine functions we define , as well as fractional brownian motion belong ( almost surely ) to this class . the main goal of this section is to prove theorems [ t : holder ] and [ t : variation ] . for the notation @xmath173 and @xmath174 see definition [ d : an ] and , respectively . [ t : holder ] let @xmath120 be an integer and let @xmath154 $ ] . let @xmath175\to { \mathbb{r}}$ ] be a function such that @xmath176 for all large enough @xmath156 . assume that @xmath15 and @xmath9 $ ] such that @xmath48 is @xmath14-hlder continuous . then @xmath35 . moreover , @xmath177 assume that @xmath9 $ ] and @xmath178 are fixed such that @xmath179 is @xmath14-hlder continuous . choose @xmath180 such that @xmath176 for all @xmath181 . clearly it is enough to prove . by decomposing @xmath30 into finitely many pieces of small enough diameters , we may assume that @xmath48 is @xmath53-hlder continuous with hlder constant @xmath182 , that is , for all @xmath110 we have @xmath183 for all @xmath121 let @xmath184 let @xmath185 . assume that @xmath186 such that @xmath187 and @xmath188 . now we will prove that @xmath189 let us fix an arbitrary @xmath190 for some @xmath191 . as @xmath192 , in order to show it is enough to prove that @xmath193 , therefore @xmath194 for some @xmath158 . as @xmath195 , we have @xmath196 for @xmath197 , which yields . hence follows . fix an integer @xmath198 and let @xmath156 be an arbitrary integer such that @xmath199 . for all @xmath200 let @xmath201 . let @xmath202 be the minimal number such that @xmath203 . note that @xmath204 for every @xmath205 , thus such a @xmath202 exists . then the recursion and @xmath206 yield that @xmath207 therefore @xmath208 . applying repeatedly and using that @xmath209 and @xmath210 we obtain that @xmath211 hence follows , and the proof is complete . the notation @xmath173 and @xmath174 are given in definition [ d : an ] and , respectively . [ t : variation ] let @xmath120 be an integer , @xmath154 $ ] and @xmath65 . let @xmath23\to { \mathbb{r}}$ ] be a function such that @xmath176 for all large enough @xmath156 . assume that @xmath212 $ ] is such that @xmath48 has finite @xmath14-variation . then @xmath213 . moreover , @xmath214 if the theorem holds for @xmath215 , then it holds for every @xmath216 . thus we may assume that @xmath217 , so @xmath218 . suppose that @xmath219 $ ] such that @xmath179 has finite @xmath14-variation . choose @xmath180 such that @xmath176 for all @xmath181 . clearly it is enough to prove . by decomposing @xmath30 into finitely many pieces of small enough diameters , we may assume that the @xmath14-variation of @xmath179 is at most @xmath182 , that is , @xmath220 let @xmath186 such that @xmath221 and @xmath188 . assume that @xmath222 and @xmath223 contains @xmath224 sub - intervals in @xmath225 . first we prove that @xmath226 let @xmath227 be the sub - intervals of @xmath223 in @xmath225 such that for every @xmath228 there is an _ even _ @xmath229 such that @xmath230 . we may assume that @xmath231 , otherwise we switch to odd integers @xmath232 and repeat the same proof . for all @xmath233 choose an @xmath234 such that @xmath235 . we may assume that @xmath236 whenever @xmath237 . let @xmath238 , and if @xmath239 is defined then let @xmath240 if the minimum exists . as @xmath195 , for all @xmath241 we have @xmath242 so if @xmath243 then @xmath244 is defined and @xmath245 . thus the length of the defined sequence @xmath246 satisfies @xmath247 . by construction @xmath248 for all @xmath249 , which implies . index the elements @xmath250 , and assume that each @xmath251 contains @xmath252 intervals of @xmath225 , so @xmath253 . inequality yields that @xmath254 therefore @xmath255 now assume that @xmath256 are fixed such that and @xmath257 , we prove that @xmath258 for all @xmath259 let @xmath260 applying inequality for @xmath261 and @xmath262 , and using that @xmath263 and @xmath264 imply that for every @xmath265 we have @xmath266 as @xmath267 , the above inequality implies that @xmath268 then @xmath269 , @xmath270 and imply . finally , let @xmath156 be an arbitrary integer with @xmath271 . let @xmath272 and @xmath273 , then @xmath274 , so @xmath275 . applying for @xmath276 easily yields that @xmath277 as @xmath278 , inequality follows . the proof is complete . the main goal of this section is to prove [ t : sa , t : answer ] . first we define a family of self - affine functions @xmath279 , which will be used in section [ s : generic ] as well . let @xmath280 be fixed integers such that @xmath281 is odd . for every @xmath282 and @xmath283 define the one - to - one affine map @xmath284 as @xmath285 see figure [ f : maps ] . as the @xmath286 are contractions , hutchinson s contraction mapping theorem ( * ? ? ? * page 713 ( 1 ) ) implies that there is a unique , non - empty compact set @xmath126 such that @xmath287 ( 0,0 ) rectangle ( 6,6 ) ; ( 0,0 ) rectangle ( 0.4,2 ) ; ( 0.4,2 ) rectangle ( 0.8,4 ) ; ( 0.8,4 ) rectangle ( 1.2,6 ) ; ( 0,0 ) ( 0.35,1.75 ) ; ( 0.4,2 ) ( 0.75,3.75 ) ; ( 0.8,4 ) ( 1.15,5.75 ) ; ( 1.2,4 ) rectangle ( 1.6,6 ) ; ( 1.6,2 ) rectangle ( 2,4 ) ; ( 2,0 ) rectangle ( 2.4,2 ) ; ( 1.2,6 ) ( 1.55,4.25 ) ; ( 1.6,4 ) ( 1.95,2.25 ) ; ( 2,2 ) ( 2.35,0.25 ) ; ( 2.4,0 ) rectangle ( 2.8,2 ) ; ( 2.8,2 ) rectangle ( 3.2,4 ) ; ( 3.2,4 ) rectangle ( 3.6,6 ) ; ( 2.4,0 ) ( 2.75,1.75 ) ; ( 2.8,2 ) ( 3.15,3.75 ) ; ( 3.2,4 ) ( 3.55,5.75 ) ; ( 3.6,4 ) rectangle ( 4,6 ) ; ( 4,2 ) rectangle ( 4.4,4 ) ; ( 4.4,0 ) rectangle ( 4.8,2 ) ; ( 3.6,6 ) ( 3.95,4.25 ) ; ( 4,4 ) ( 4.35,2.25 ) ; ( 4.4,2 ) ( 4.75,0.25 ) ; ( 4.8,0 ) rectangle ( 5.2,2 ) ; ( 5.2,2 ) rectangle ( 5.6,4 ) ; ( 5.6,4 ) rectangle ( 6,6 ) ; ( 4.8,0 ) ( 5.15,1.75 ) ; ( 5.2,2 ) ( 5.55,3.75 ) ; ( 5.6,4 ) ( 5.95,5.75 ) ; ( 0,0 ) rectangle ( 6,6 ) ; ( 0,0 ) ( 0.08,2 ) ; ( 0.08,2 ) ( 0.16,0 ) ; ( 0.16,0 ) ( 0.24,2 ) ; ( 0.24,2 ) ( 0.32,0 ) ; ( 0.32,0 ) ( 0.4,2 ) ; ( 0.4,2 ) ( 0.48,4 ) ; ( 0.48,4 ) ( 0.56,2 ) ; ( 0.56,2 ) ( 0.64,4 ) ; ( 0.64,4 ) ( 0.72,2 ) ; ( 0.72,2 ) ( 0.8,4 ) ; ( 0.8,4 ) ( 0.88,6 ) ; ( 0.88,6 ) ( 0.96,4 ) ; ( 0.96,4 ) ( 1.04,6 ) ; ( 1.04,6 ) ( 1.12,4 ) ; ( 1.12,4 ) ( 1.2,6 ) ; ( 2.4,0 ) ( 2.48,2 ) ; ( 2.48,2 ) ( 2.56,0 ) ; ( 2.56,0 ) ( 2.64,2 ) ; ( 2.64,2 ) ( 2.72,0 ) ; ( 2.72,0 ) ( 2.8,2 ) ; ( 2.8,2 ) ( 2.88,4 ) ; ( 2.88,4 ) ( 2.96,2 ) ; ( 2.96,2 ) ( 3.04,4 ) ; ( 3.04,4 ) ( 3.12,2 ) ; ( 3.12,2 ) ( 3.2,4 ) ; ( 3.2,4 ) ( 3.28,6 ) ; ( 3.28,6 ) ( 3.36,4 ) ; ( 3.36,4 ) ( 3.44,6 ) ; ( 3.44,6 ) ( 3.52,4 ) ; ( 3.52,4 ) ( 3.6,6 ) ; ( 4.8,0 ) ( 4.88,2 ) ; ( 4.88,2 ) ( 4.96,0 ) ; ( 4.96,0 ) ( 5.04,2 ) ; ( 5.04,2 ) ( 5.12,0 ) ; ( 5.12,0 ) ( 5.2,2 ) ; ( 5.2,2 ) ( 5.28,4 ) ; ( 5.28,4 ) ( 5.36,2 ) ; ( 5.36,2 ) ( 5.44,4 ) ; ( 5.44,4 ) ( 5.52,2 ) ; ( 5.52,2 ) ( 5.6,4 ) ; ( 5.6,4 ) ( 5.68,6 ) ; ( 5.68,6 ) ( 5.76,4 ) ; ( 5.76,4 ) ( 5.84,6 ) ; ( 5.84,6 ) ( 5.92,4 ) ; ( 5.92,4 ) ( 6,6 ) ; ( 1.2,6 ) ( 1.28,4 ) ; ( 1.28,4 ) ( 1.36,6 ) ; ( 1.36,6 ) ( 1.44,4 ) ; ( 1.44,4 ) ( 1.52,6 ) ; ( 1.52,6 ) ( 1.6,4 ) ; ( 1.6,4 ) ( 1.68,2 ) ; ( 1.68,2 ) ( 1.76,4 ) ; ( 1.76,4 ) ( 1.84,2 ) ; ( 1.84,2 ) ( 1.92,4 ) ; ( 1.92,4 ) ( 2,2 ) ; ( 2,2 ) ( 2.08,0 ) ; ( 2.08,0 ) ( 2.16,2 ) ; ( 2.16,2 ) ( 2.24,0 ) ; ( 2.24,0 ) ( 2.32,2 ) ; ( 2.32,2 ) ( 2.4,0 ) ; ( 3.6,6 ) ( 3.68,4 ) ; ( 3.68,4 ) ( 3.76,6 ) ; ( 3.76,6 ) ( 3.84,4 ) ; ( 3.84,4 ) ( 3.92,6 ) ; ( 3.92,6 ) ( 4,4 ) ; ( 4,4 ) ( 4.08,2 ) ; ( 4.08,2 ) ( 4.16,4 ) ; ( 4.16,4 ) ( 4.24,2 ) ; ( 4.24,2 ) ( 4.32,4 ) ; ( 4.32,4 ) ( 4.4,2 ) ; ( 4.4,2 ) ( 4.48,0 ) ; ( 4.48,0 ) ( 4.56,2 ) ; ( 4.56,2 ) ( 4.64,0 ) ; ( 4.64,0 ) ( 4.72,2 ) ; ( 4.72,2 ) ( 4.8,0 ) ; it is easy to see that @xmath288 is a graph of a function @xmath289\to [ 0,1]$ ] , and @xmath279 can be approximated as follows . let @xmath290\}$ ] be the diagonal of @xmath144 ^ 2 $ ] and define @xmath291 as @xmath292 if @xmath293 denotes the @xmath294 iterate @xmath295 , then @xmath296 is the graph of a piecewise linear function @xmath297 which converges uniformly to @xmath279 as @xmath298 . clearly @xmath279 is a self - affine function with @xmath299 and @xmath300 , and the definition yields that @xmath279 is hlder continuous with exponent @xmath301 . figure [ f : new ] shows the piecewise linear function @xmath302 , which approximates @xmath303 . define the set @xmath304 then @xmath305 is a countable dense subset of @xmath306 , since every rational @xmath307 is in @xmath305 by @xmath308 and @xmath309 . for all @xmath32 fix @xmath276 such that @xmath310 , and define @xmath311 $ ] . fix @xmath312 such that @xmath313 . we use the scaled local times with @xmath314 . by [ t : holder , t : variation ] it is enough to show that @xmath315 for all @xmath316 . let us fix @xmath316 . clearly @xmath317 , and the construction of @xmath318 yields that for every @xmath319 , @xmath320 , and @xmath321 we have @xmath322 similarly , @xmath323 the above and @xmath324 yield that for all @xmath325 we have @xmath326 thus @xmath315 , and the proof is complete . note that the bound we get for @xmath327 does not use the @xmath328 factor . it is possible to go through the proof [ t : holder ] with such a stronger assumption , which would slightly improve the bounds on @xmath174 with @xmath329 in place of @xmath330 error term . let @xmath0 . by [ t : kkm , t : m , t : a ] it is enough to prove that @xmath331 for the first inequality let @xmath332 be arbitrary , then @xmath333 $ ] . suppose that @xmath334 is @xmath14-hlder continuous on some set @xmath9 $ ] . theorem [ t : sa ] yields that @xmath335 , thus @xmath336 . since @xmath305 is dense in @xmath337 , we obtain that @xmath338 . for the second inequality define @xmath339 and let @xmath340 be arbitrary . then @xmath333 $ ] . assume that @xmath334 has finite @xmath14-variation on some @xmath30 . theorem [ t : sa ] and @xmath341 imply that @xmath342 , so @xmath343 . as @xmath305 is dense in @xmath344 , we have @xmath345 . let @xmath0 be fixed and let @xmath346\to { \mathbb{r}}$ ] be a fractional brownian motion of hurst index @xmath3 . we think of @xmath16 as a random function from @xmath144 $ ] to @xmath347 . the main goal of this section is to prove theorem [ t : fbm ] . by theorems [ t : holder ] and [ t : variation ] it is enough to prove the following proposition . for the notation @xmath348 see definition [ d : an ] . [ p : key ] let @xmath0 and let @xmath346\to { \mathbb{r}}$ ] be a fractional brownian motion of hurst index @xmath3 . then , almost surely , @xmath349 for all @xmath156 large enough . first we define a discrete ( truncated ) scaled local time . for all @xmath121 let @xmath350 for all @xmath160 let @xmath351 clearly , @xmath352 . for a function @xmath23\to { \mathbb{r}}$ ] the _ discrete scaled local time _ @xmath353 is the number of points in @xmath354 which are mapped to @xmath166 by @xmath47 : @xmath355 for every @xmath171 define @xmath356 first we need to prove some lemmas . to avoid technical difficulties we assume that the domain of @xmath16 is extended to @xmath357 when necessary . let @xmath358 be the probability space on which our fractional brownian motion is defined , and let @xmath359 be the natural filtration . if @xmath360 $ ] is a stopping time then define the @xmath361-algebra @xmath362 for all stopping times @xmath363 and integers @xmath160 and @xmath169 let @xmath364 [ l : sln ] there is a finite constant @xmath365 depending only on @xmath3 such that for every bounded stopping time @xmath363 and integers @xmath160 and @xmath169 we have , almost surely , @xmath366 pitt ( * ? ? ? * lemma 7.1 ) showed that the property of _ strong local nondeterminism _ holds for fractional brownian motion , that is , there is a constant @xmath367 such that for all @xmath368 , almost surely , @xmath369 let us fix @xmath133 . as @xmath16 is gaussian , almost surely the conditional distribution @xmath370 is normal , and implies that its density function is bounded by @xmath371 . therefore , almost surely , @xmath372 where @xmath373 . applying for finitely many @xmath374 implies that @xmath375 where @xmath376 . the proof is complete . [ l : snmpq ] there is a finite constant @xmath377 depending only on @xmath3 such that for all @xmath378 and @xmath379 we have @xmath380 let @xmath381 be the constant in lemma [ l : sln ] , clearly we may assume that @xmath382 . we will show that @xmath383 satisfies the lemma . we define stopping times @xmath384 . let @xmath385 . if @xmath386 is defined for some @xmath387 then let @xmath388 be the first time such that the contribution of @xmath16 to @xmath389 on @xmath390\cap { \mathcal{l}}_{n , m , p}$ ] is at least @xmath391 if such a time exists , otherwise let @xmath392 . then @xmath382 and the definition of stopping times yield that @xmath393 note that we may assume that @xmath394 and hence the above conditional probabilities are defined , otherwise we are done immediately . therefore it is enough to prove that @xmath395 for all @xmath396 . the definition of @xmath397 , lemma [ l : sln ] , and the conditional markov s inequality imply that , almost surely , @xmath398 therefore the tower property of conditional expectation yields that @xmath399 which completes the proof . the following lemma is a discrete version of proposition [ p : key ] . [ l : dkey ] almost surely , @xmath400 for all large enough @xmath156 . we give an upper bound for @xmath401 by applying lemma [ l : snmpq ] with @xmath402 to each of the relevant @xmath403 . since @xmath160 , @xmath404 and @xmath405 , there are at most @xmath406 possibilities to choose @xmath403 . therefore lemma [ l : snmpq ] implies that @xmath407 therefore @xmath408 , so the borel cantelli lemma implies that @xmath409 the following lemma is well known , see the more general ( * ? ? ? * corollary 7.2.3 ) . [ l : ub ] almost surely , we have @xmath410 now we are ready to prove proposition [ p : key ] . by lemmas [ l : dkey ] and [ l : ub ] we can choose a random @xmath180 such that , almost surely , for every @xmath181 we have the following properties : 1 . [ i1 ] @xmath411 } |b(t)|<n-1 $ ] ; 2 . [ i2 ] @xmath400 ; 3 . [ i3 ] @xmath412 for all @xmath413 ; 4 . [ i4 ] @xmath414 . fix a path of @xmath16 for which the above properties hold . let us fix an arbitrary @xmath181 , it is enough to prove that @xmath415 . let @xmath160 , @xmath404 and @xmath158 be given , we need to show that @xmath416 property [ i1 ] yields that if @xmath417 with @xmath418 then @xmath419 . therefore [ i2 ] implies that for all @xmath417 we have @xmath420 let @xmath421 be a time interval of order @xmath156 such that @xmath422 and @xmath423 , then [ i3 ] yields that @xmath424 finally , , and [ i4 ] imply that @xmath425 hence holds , and the proof of proposition [ p : key ] is complete . we include here a proof of proposition [ p : r ] , which we could not find in the literature . recall that @xmath8\}$ ] is a fractional brownian motion of hurst index @xmath3 . the lower bound is quite elementary , while the upper bound relies on a first moment computation and on a result of molchan . he studied the distribution of the maximal value of fractional brownian motion on @xmath426 $ ] and of the time @xmath427 when it is achieved . specifically , molchan ( * ? ? ? * theorem 2 ) proved the following . @xmath428 as @xmath429 . with probability one @xmath16 is @xmath53-hlder continuous for every @xmath136 and maps @xmath44 to the non - degenerate interval @xmath430\}]$ ] , thus @xmath431 . therefore it is enough to prove that @xmath432 almost surely . using that @xmath427 and @xmath433 have the same distribution and @xmath434\}$ ] is also a fractional brownian motion , for all @xmath435 we obtain @xmath436 \neq\emptyset)&={\mathbb{p}}(\tau_{\max}(t ) \geq t-{\varepsilon})={\mathbb{p}}(\tau_{\max}(1)\geq 1-{\varepsilon}/t ) \\ & = { \mathbb{p}}(\tau_{\max}(1)\leq { \varepsilon}/t)=({\varepsilon}/t)^{1-\alpha+o(1)},\end{aligned}\ ] ] with the @xmath437 term tending to @xmath438 as @xmath439 . let @xmath440 be the number of intervals @xmath441 $ ] which intersect @xmath44 . let @xmath442 be arbitrary and fix @xmath443 such that @xmath444 \neq\emptyset ) \leq ( { \varepsilon}/t)^{1-\alpha-\delta}$ ] whenever @xmath445 . using this above for @xmath446 and the trivial bound @xmath182 for @xmath447 , for every large enough @xmath281 we obtain @xmath448 where @xmath449 is a finite constant depending only on @xmath450 . by markov s inequality @xmath451 for any @xmath281 large enough . applying this for @xmath452 yields that @xmath453 thus the borel cantelli lemma implies that , almost surely , @xmath454 for all large enough @xmath156 . therefore @xmath455 . as @xmath442 was arbitrary , @xmath456 almost surely . the aim of this section is to prove theorem [ t:2dbm ] . the idea is to find ( in a greedy manner ) large sets along which a simple random walk in @xmath457 is monotone . since the scaled simple random walk converges to brownian motion , this gives sets along which @xmath16 is monotone . to control the dimension of the limit sets we estimate the energy of the discrete sets and apply a version of frostman s lemma to bound the dimension . given a simple random walk @xmath458 , define the _ greedy increasing subset _ by @xmath459 our first task is to prove tightness for the number and structure of record times in @xmath460 . since our argument may apply in similar situations , we state some of our arguments in the more general context of sums of i.i.d.variables with power law tails . before focusing on the case of random walks , we prove theorem [ t : acc ] , a limit theorem , which will allow us to transfer estimates from the random walk setting to brownian motion . as theorem [ t : acc ] below is a quite general result about random sequences with i.i.d . increments , we hope that it will find further applications . fix @xmath0 . let @xmath461 be some random variable , and assume that there are @xmath462 such that for all @xmath171 we have @xmath463 let @xmath464 be an i.i.d . sequence with the law of @xmath363 . define @xmath465 the number of steps before reaching @xmath156 is denoted by @xmath466 . the following lemma is fairly standard . [ l : m_n_tight ] there are constants @xmath467 such that for all @xmath468 we have 1 . [ cl1 ] @xmath469 . [ cl2 ] @xmath470 . [ cl3 ] @xmath471 , and more generally , for all integers @xmath472 we have @xmath473 claim [ cl1 ] is given by ( * ? ? ? * lemma 4.2 ) . inequality and @xmath474 imply that @xmath475 so [ cl2 ] holds . the first bound of [ cl3 ] follow easily from [ cl2 ] . the general bound holds since the first @xmath476 with @xmath477 ( if there is such @xmath476 ) is a stopping time . applying [ cl2 ] to the sequence starting at time @xmath476 completes the proof . let @xmath149 be a non - atomic _ mass distribution _ on a metric space @xmath478 , ( that is , a borel measure on @xmath479 with @xmath480 ) . for @xmath481 , define the _ @xmath53-energy of @xmath149 _ by @xmath482 for the following theorem see ( * ? ? ? * theorem 4.27 ) . [ t : energy ] let @xmath149 be a non - atomic mass distribution on a metric space @xmath479 with @xmath483 . then @xmath484 . consider the set @xmath485 , endowed with lebesgue measure @xmath486 . we next estimate the @xmath53-energy of @xmath486 . [ l : energy_small ] let @xmath487 . there is a finite constant @xmath488 , such that for all @xmath156 we have @xmath489 the argument is to consider the contribution to the energy from pairs @xmath490 with distance at various scales , and the largest scale will dominate the rest . up to a factor of @xmath491 we may restrict the integral to @xmath492 . we split the integral on @xmath493 into several parts . note that @xmath494 is a disjoint union of unit intervals . let @xmath495 be the contribution to @xmath496 from pairs @xmath497 and @xmath498 where @xmath499 and @xmath500 . the number of such pairs @xmath501 is at most @xmath502 , so lemma [ l : m_n_tight ] [ cl3 ] yields that @xmath503 for @xmath504 let @xmath505 be the contribution to @xmath496 from pairs @xmath497 and @xmath498 where @xmath499 and @xmath506 . let @xmath507 denote the number of such pairs @xmath501 . for such @xmath501 the contribution from @xmath497 and @xmath498 to @xmath505 is at most @xmath508 where we used @xmath509 . thus @xmath510 . lemma [ l : m_n_tight ] [ cl3 ] yields that for every @xmath511 , conditioned on @xmath512 , the expected number of @xmath513 in @xmath514 $ ] is at most @xmath515 . lemma [ l : m_n_tight ] [ cl3 ] also implies that the expected number of @xmath516 in @xmath517 equals @xmath518 . therefore @xmath519 and we obtain that @xmath520 this partition gives the identity @xmath521 . as @xmath522 whenever @xmath523 , we have @xmath524 . with the bounds above , the largest @xmath202 dominates the sum and we arrive at the inequality @xmath525 . we will wish to work with rescaled sets . for all @xmath171 let @xmath526 define the measure @xmath527 , that is , @xmath528 times the lebesgue measure restricted to @xmath529 . [ l : mu_tight ] let @xmath487 and @xmath530 . then there exist @xmath531 depending on @xmath532 so that with probability at least @xmath533 we have @xmath534 ) \leq n \quad \text { and } \quad { \mathcal{e}}_\gamma(c_n,\mu_n ) \leq n_{\gamma}.\ ] ] let @xmath535 $ ] and let @xmath180 be arbitrary . for all @xmath171 the following three inequalities hold . markov s inequality and @xmath536 yield that @xmath537 lemma [ l : m_n_tight ] [ cl3 ] and [ cl1 ] yield that @xmath538 lemma [ l : energy_small ] yields that @xmath539 . indeed , @xmath540 and since @xmath541 on @xmath529 , changing to @xmath542 gives a further factor of @xmath543 . now markov s inequality implies that @xmath544 the above three inequalities with large enough @xmath545 complete the proof . theorem [ t : acc ] is concerned with a sequence of sequences of i.i.d . variables @xmath546 satisfying . that is , for each fixed @xmath156 the variables @xmath546 are i.i.d . , but there could be arbitrary dependencies between variables with different numbers @xmath156 . the superscript @xmath156 is also the parameter used for scaling sums of the @xmath156th sequence . thus we denote @xmath547 for all @xmath171 define @xmath548 [ t : acc ] with the notations above , almost surely , @xmath549 has an accumulation point @xmath49 in the hausdorff metric such that @xmath550 . let @xmath542 be @xmath528 times the lebesgue measure on @xmath529 . fix @xmath530 . for some finite constants @xmath551 for all @xmath156 define the event @xmath552 ) \leq n_0 \text { and } { \mathcal{e}}_{\gamma}(c_n,\mu_n ) \leq n_{\gamma } \textrm { for all } 0<\gamma<\alpha\}.\ ] ] since the map @xmath553 is non - decreasing , applying lemma [ l : mu_tight ] for a sequence of parameters @xmath554 with @xmath555 implies that there are constants @xmath556 such that @xmath557 for all @xmath156 . let @xmath558 , then @xmath559 . since @xmath530 was arbitrary , it is enough to prove that the theorem is satisfied whenever @xmath560 holds . assume that @xmath560 holds , then there is a random subsequence @xmath561 such that the events @xmath562 hold for all @xmath200 . since @xmath563)\leq n_0 $ ] for all @xmath511 , by passing to a subsequence we may assume that @xmath564 weakly , where @xmath149 is a measure on @xmath144 $ ] satisfying @xmath565)\leq n_0 $ ] . similarly , we may assume that @xmath566 in the hausdorff metric for some compact set @xmath18 $ ] . as @xmath567 $ ] is close in the hausdorff metric to @xmath568 , we get @xmath569 . for all @xmath487 we obtain @xmath570 for the first inequality see e.g. ( * ? ? ? * lemma 2.2 ) . theorem [ t : energy ] now implies that @xmath571 , and the proof is complete . we now apply theorem [ t : acc ] to random walks on @xmath457 and thus prove theorem [ t:2dbm ] . for each @xmath156 , let @xmath572 be a simple random walk on @xmath457 , and define the rescaled random walks by @xmath573 . it is well known that it is possible to construct the walks @xmath572 and two - dimensional brownian motion @xmath8\}$ ] on the same probability space so that @xmath574 converges uniformly to @xmath16 on the interval @xmath144 $ ] , see e.g. ( * ? ? ? * theorem 3.5.1 ) or @xcite . we henceforth assume such a coupling . recall that for each walk we construct the _ greedy increasing subset _ by @xmath575 for every @xmath156 , this sequence has i.i.d . increments with the law of @xmath576 we use the notation @xmath577 if @xmath578 as @xmath579 . we need the following known estimate . [ t : greedy ] let @xmath458 be a two - dimensional simple random walk . let @xmath363 be the hitting time of the positive quadrant : @xmath580 . then there is a @xmath90 so that @xmath581 for the above theorem see the general result of denisov and wachtel ( * ? ? ? * theorem 1 ) , or a bit weaker one due to varopoulos ( * ? ? ? * ( 0.3.3 ) and ( 0.4.1 ) ) . on exit times of planar brownian motion from cones see evans ( * ? ? ? * corollary 5(i ) ) or the somewhat weaker ( * ? ? ? * lemma 10.40 ) . in the brownian case the exponent @xmath21 can be calculated by mapping the complement of @xmath582 onto a half plane by the conformal map @xmath583 and using the conformal invariance of planar brownian motion . the continuous case can be transformed to the discrete one by coupling . for the history of similar estimates and for further references see denisov and wachtel @xcite . recall that @xmath572 are two - dimensional simple random walks so that the rescaled walks @xmath574 converge uniformly to a brownian motion @xmath16 . let @xmath363 be the hitting time of the positive quadrant by @xmath584 , that is , @xmath585 . for every @xmath586 , the greedy increasing subsequence of @xmath572 has i.i.d . increments , distributed as @xmath363 . by theorem [ t : greedy ] we have @xmath587 with some @xmath90 . thus we can apply theorem [ t : acc ] with @xmath588 . this yields that , almost surely , there is an accumulation point @xmath49 of @xmath549 in the hausdorff metric such that @xmath19 . as @xmath589 uniformly , @xmath16 is non - decreasing on @xmath49 . this completes the proof . for a higher dimensional simple random walk @xmath590 define the hitting time @xmath591 then @xmath592 for some @xmath593 , see ( * ? ? ? * theorem 1 ) . our argument proves an analogue of theorem [ t:2dbm ] in higher dimensions with this @xmath3 instead of @xmath21 . the goal of this section is to prove theorem [ t : gen ] . first we need some preparation . the following lemma is probably well known . however , we could not find an explicit reference for its second claim , so we outline the proof . [ l : ext ] let @xmath594 and @xmath595 . assume that @xmath596 and @xmath63 is a function such that for all @xmath110 we have @xmath597 then @xmath47 extends to @xmath598 satisfying the above inequality for all @xmath599 . if @xmath30 is closed then @xmath119 can be chosen to be linear on the components of @xmath600 . as @xmath47 admits a unique continuous extension to the closure of @xmath30 which clearly satisfies , we may assume that @xmath30 is closed . let @xmath223 be any component of @xmath600 , it is enough to prove that @xmath47 extends to @xmath601 such that holds . if @xmath602 or @xmath603 for some @xmath604 then @xmath605 works . now let @xmath606 for some @xmath607 and let @xmath119 be the linear extension of @xmath47 to @xmath223 . the concavity of the function @xmath608 implies that @xmath609 for all @xmath610 , the straightforward calculation is left to the reader . let @xmath617 such that the oscillation of @xmath47 on @xmath618 $ ] is at most @xmath619 for all @xmath620 . let @xmath621 be the piecewise linear function passing through the points @xmath622 , then clearly @xmath623 . applying lemma [ l : ext ] @xmath476-times we obtain that @xmath624 $ ] . we can choose @xmath625 such that @xmath626 satisfies @xmath627 . hence for all @xmath91 $ ] we have @xmath628 let @xmath629 , then it is easy to see that every horizontal line segment of the graph of @xmath630 ( if there are any ) can be replaced by two line segments of nonzero slopes such that the resulting function @xmath613 satisfies @xmath631 and for all @xmath91 $ ] we have @xmath616 clearly @xmath615 , and the proof is complete . now we are ready to prove the first part of theorem [ t : gen ] . the concept of the proof is similar to that of ( * ? ? ? * theorem 1.4 ) , but the technical details are much more difficult and in order to create appropriate hlder continuous functions some new ideas are needed as well . let @xmath632 be arbitrarily fixed , and define @xmath633 : \dim_{\mathcal{h}}\{f = g\}\leq 1-\beta \textrm { for all } g\in c_1^{\beta}[0,1]\},\ ] ] where we use the notation @xmath634 : f(x)=g(x)\}$ ] . first we show that it is enough to prove that @xmath635 is co - meager in @xmath97 $ ] . indeed , since co - meager sets are closed under countable intersection , this implies that for a countable dense set @xmath636 the set @xmath637 is co - meager in @xmath97 $ ] . now assume that @xmath638 and @xmath9 $ ] such that @xmath48 is @xmath14-hlder for some @xmath15 , we need to prove that @xmath57 . choose a sequence @xmath639 such that @xmath640 and fix an @xmath171 . as @xmath48 is @xmath14-hlder , there is an @xmath612 such that for all @xmath641 with @xmath642 the function @xmath643 is @xmath644-hlder with hlder constant @xmath182 . let @xmath645 such that @xmath646 for all @xmath647 . then @xmath648 are @xmath644-hlder with hlder constant @xmath182 , so by lemma [ l : ext ] there are functions @xmath649 $ ] such that @xmath650 for all @xmath511 . therefore @xmath651 implies that @xmath652 for all @xmath511 , thus the countable stability of hausdorff dimension yields @xmath653 . this holds for all @xmath171 , so @xmath57 . now let @xmath632 be fixed , and for all @xmath654 define @xmath655 : { \mathcal{h}}_{\infty}^{1-\beta+1/n}(\{f = g\})\leq 1/m \textrm { for all } g\in c_1^{\beta}[0,1 ] \right\}.\ ] ] clearly @xmath656 , thus it is enough to show that each @xmath657 contains a dense open subset of @xmath97 $ ] . assume that @xmath658 , @xmath659 , @xmath625 , and a piecewise linear function @xmath660 $ ] with nonzero slopes are given such that for all @xmath91 $ ] we have @xmath661 by lemma [ l : dense ] it is enough to find a function @xmath99 $ ] and @xmath662 such that @xmath663 where @xmath664 denotes the closed ball in @xmath97 $ ] centered at @xmath47 with radius @xmath224 . now we define @xmath47 . we can fix integers @xmath665 such that @xmath666 is odd and @xmath667 let @xmath668 and let @xmath669 $ ] be the self - affine function defined in section [ s : kk ] . we will approximate @xmath670 by re - scaled copies of @xmath671 . as @xmath671 is @xmath53-hlder continuous , there exists @xmath672 such that for all @xmath91 $ ] we have @xmath673 assume that @xmath674 such that @xmath670 is linear on each interval @xmath675 $ ] with nonzero slopes . let @xmath676 for all @xmath677 , then @xmath678 for all @xmath679 . let us define @xmath680 and @xmath681 such that @xmath682 for all @xmath620 and @xmath683 let @xmath684 now we are ready to define @xmath47 . if for some @xmath620 , @xmath685 , and @xmath686 we have @xmath687 note that the linearity of @xmath670 implies that if @xmath153 satisfies , then @xmath688 now we prove that @xmath689 \quad \textrm{and } \quad f\in b(f_0,r_0/2).\ ] ] as the range of @xmath670 is @xmath144 $ ] , the definition of @xmath47 and @xmath690 imply that for all @xmath691 $ ] we have @xmath692 thus it is enough to prove for that @xmath99 $ ] . assume that @xmath91 $ ] and @xmath492 , we need to prove that @xmath693 . we consider three cases . _ first case _ : suppose that @xmath694 . then , @xmath660 $ ] and the definition of @xmath690 imply that @xmath695 _ second case _ : assume that @xmath490 are _ adjacent points _ , that is , @xmath696 $ ] for some @xmath697 and @xmath698 . then the definitions of @xmath47 and @xmath699 , inequality , the definition of @xmath690 , and @xmath700 yield that @xmath701 _ third case _ : suppose that there exists @xmath702 such that @xmath703 and @xmath704 are adjacent points . the triangle inequality , the above inequality , the definition of @xmath690 , and @xmath700 imply that @xmath705 by the definition of @xmath706 for all @xmath490 at least one of the above three cases holds , which concludes the proof of . finally , we prove that holds for some @xmath662 . let us define @xmath442 as @xmath707 since @xmath708 , we have @xmath709 . thus by @xmath710 we can fix an @xmath711 such that for all @xmath200 and @xmath712 we have @xmath713 since @xmath714 , we can define @xmath715 let us fix an integer @xmath716 such that @xmath717 let us define @xmath662 as @xmath718 then clearly @xmath719 . let us fix @xmath720 , it is enough to show that @xmath721 . fix @xmath620 and @xmath685 arbitrarily , and let @xmath722 $ ] . let @xmath723 $ ] , by the subadditivity of @xmath724 it is enough to show that @xmath725 assume that @xmath121 and @xmath726 is a closed interval . we divide @xmath223 into @xmath727 non - overlapping closed intervals of equal length , the resulting intervals are called the _ elementary intervals of @xmath223 of level @xmath156_. assume that @xmath728 and let @xmath729 be an elementary interval of @xmath730 of level @xmath731 . now we show that @xmath732 intersects at most @xmath733 many first level elementary intervals of @xmath729 . let us decompose @xmath729 into @xmath666 non - overlapping intervals of equal length , let @xmath734 be one of them . let @xmath735 be two nonconsecutive first level elementary intervals of @xmath729 , it is enough to show that @xmath732 can not intersect both @xmath736 and @xmath737 . assume to the contrary that there are @xmath738 and @xmath739 such that @xmath740 and @xmath741 . if there are @xmath742 first level elementary intervals of @xmath729 between @xmath736 and @xmath737 , then @xmath743 therefore @xmath723 $ ] yields that @xmath744 on the other hand side , the definition of @xmath613 , @xmath47 and @xmath224 imply that @xmath745 the above inequalities and yield that @xmath746 , which is a contradiction . therefore @xmath747 intersects at most @xmath748 many elementary intervals of @xmath730 of level @xmath749 . since the length of these intervals is less than @xmath750 , the definition of @xmath361 and inequality yield that @xmath751 hence holds , and the proof is complete . in order to prove the second part of theorem [ t : gen ] we need a bridge between the notions of @xmath3-hlder continuity and @xmath14-variation . the following lemma is ( * ? ? ? * lemma 4.1 ) , see also ( * ? ? ? * lemma 4.1 ) . clearly it is enough to prove the theorem for a countable dense set of parameters @xmath14 . since co - meager sets are closed under countable intersection , it is enough to show the statement for an arbitrary fixed @xmath65 . for all @xmath99 $ ] let @xmath757 $ ] be given such that @xmath758 . fix an arbitrary @xmath759 and let @xmath760 . it is enough to prove that @xmath761 for a generic @xmath99 $ ] . applying lemma [ l : bv ] we obtain that for all @xmath99 $ ] there are sets @xmath762 such that as @xmath765 and @xmath763 are @xmath53-hlder continuous , theorem [ t : gen ] and the definition of @xmath53 imply that for a generic @xmath99 $ ] for all @xmath171 we have @xmath766 inequalities , , and the countable stability of hausdorff dimension yield that @xmath761 for a generic @xmath99 $ ] . the proof is complete .

using data from the nasa spacecraft _ kepler _ , we study solar - like oscillations in red - giant stars in the open cluster ngc6811 . we determine oscillation frequencies , frequency separations , period spacings of mixed modes and mode visibilities for eight cluster giants . the oscillation parameters show that these stars are helium - core - burning red giants . the eight stars form two groups with very different oscillation power spectra ; the four stars with lowest-values display rich sets of mixed modes , while this is not the case for the four stars with higher . for the four stars with lowest , we determine the asymptotic period spacing of the mixed modes ,p , which together with the masses we derive for all eight stars suggest that they belong to the so - called secondary clump . based on the global oscillation parameters , we present initial theoretical stellar modeling which indicate that we can constrain convective - core overshoot on the main sequence and in the helium - burning phase for thesem stars . finally , our results indicate less mode suppression than predicted by recent theories for magnetic suppression of certain oscillation modes in red giants .

1703.01097

the nasa spacecraft _ kepler _ @xcite has provided photometric time - series of unprecedented quality for solar - type and red - giant stars , and asteroseismic analysis of these data have led to remarkable results as reviewed by . the asteroseismic analysis allows for a determination of the evolutionary stage of evolved stars , as period spacings of mixed oscillation modes ( especially @xmath1 ) separate hydrogen - shell - burning and helium - core - burning red giants @xcite . mixed modes are non - radial modes which behave both like @xmath5-modes , probing the outer layers of the star , and like @xmath6-modes , probing the deep interior and hence the evolutionary stage of the star . @xcite showed that @xmath2p-@xmath7 ( period spacing , large frequency separation ) diagrams using both the median period spacing , derived from the observed oscillation spectra , and the asymptotic period spacing from theoretical models , separate red giants of different mass and evolutionary stage . the evolution of the asymptotic period spacing as a function of mass and age depends on the amount of convective - core overshoot , which means that period spacings of evolved stars offer the opportunity of constraining theoretical stellar models , including the amount of convective - core overshoot on the main sequence for 1.24.0m@xmath4 stars @xcite . it has also been found that non - radial modes are suppressed in a subset of the red giants observed with _ kepler _ . this has recently been suggested to be associated with strong internal magnetic fields @xcite , which means that asteroseismology of red giants might also offer the opportunity of investigating magnetic fields within giant stars . evolved stars in open clusters observed with _ kepler _ have been studied using asteroseismology by several groups , e.g. , @xcite . ngc6811 is a cluster of intermediate age ( @xmath31gyr ) which contains pulsating stars , binaries and red giants , as discussed in detail by @xcite . the presence of both oscillating red giants and detached eclipsing binaries in the same cluster , sharing common parameters such as age , distance and metallicity , offers unique possibilities for testing theoretical stellar models . as described in @xcite , our long - term aim is to combine asteroseismology of giant stars with information derived from detached eclipsing binaries , in order to test asteroseismic scaling relations and challenge stellar models . the asteroseismic scaling relations relate the global properties of solar - like oscillations ( frequency of maximum power @xmath8 and large frequency separation @xmath7 ) to stellar mass , radius and effective temperature ( see sect . they are used for deriving stellar properties of ( faint ) stars for which only this global asteroseismic information is available and are therefore important to test . in this paper we present an analysis of _ kepler _ light - curves of eight oscillating red giants in ngc6811 to derive the global oscillation parameters to be used for testing the asteroseismic scaling relations , as presented in sections25 . from the asteroseismic parameters we derive masses and evolutionary stages for these stars , we present comparisons of asymptotic period spacings @xmath2p with initial stellar modeling , and we use those period spacings to show that convective - core overshoot both on the main sequence and in the helium - core - burning phase can be constrained for these stars using the asteroseismic parameters ( sections 6 and 7 ) . another main result is that we find that the power spectra of the eight stars show very different oscillation properties despite having very similar stellar properties . we discuss in sect.8 if this could be explained by the magnetic effects mentioned above . we used _ kepler _ long - cadence data ( @xmath330 min . sampling , * ? ? ? * ) to study eight red giants in the open cluster ngc6811 . five stars ( kic9716522 , 9655101 , 9655167 , 9534041 , 9716090 ) are listed as red giants in @xcite , kic9532903 , 9776739 had their first asteroseismic study by and were identified as cluster members by @xcite . kic9409513 was found in our own investigations of the cluster , and is also included in @xcite who furthermore show that it is unlikely to find more cluster - member giants . the giants have @xmath9 allowing spectroscopic studies @xcite . each time series includes more than 60,000 data points and spans @xmath31400 days ( bjd 2454953 - 2456391 , _ kepler _ quarters q1 q17 ) from data - release 22 . the data were extracted from _ kepler _ pixel - data using reduction software kindly provided to us by s. bloemen . as the standard _ kepler _ light - curves are not optimized for photometry in crowded fields , as in an open cluster , we manually defined large pixel apertures for each star in order to optimize the light curves . the extracted light - curves were filtered with the kasoc filter @xcite to prepare the data for asteroseismic analysis . the filter operates with two timescales , one long and one short , to remove trends and signals that do not originate from the stellar oscillations . to test the robustness of the filtering , we have worked with two versions of the filtered data , one with long and short timescales of 30d and 3d , and one with 10d and 1d , see @xcite for details . quantitative comparisons show that the power spectra of the two filtered versions differ noticeably only at very low frequencies ( below @xmath10hz ) . we do not use this very low - frequency part of the power spectra in our analysis and we find consistent results using the two versions of the time series , as the results of our asteroseismic analysis differ only within the 1@xmath11 uncertainties . we adopt the mean values from the two versions as our final values . the list of analyzed stars is found in table [ tab.1 ] and the power spectra of their light curves are presented in fig . [ fig.power ] . the stars are arranged in order of increasing large frequency spacing @xmath0 and frequency of maximum power , @xmath8 , in both table [ tab.1 ] and fig . [ fig.power ] . we will describe how we determine these asteroseismic parameters below , however already at this point we notice the remarkable feature in the power spectra of the eight stars shown in fig . [ fig.power ] ; the top four stars all have higher amplitudes and richer oscillation spectra as compared to the four bottom stars . we return to this below . the stellar backgrounds originating from convection ( granulation , activity ) were fitted in the power spectra using a two - component model @xcite , including the attenuation factor @xmath12 which is needed with _ kepler _ long - cadence data , of the following form : @xmath13 here , @xmath14 and @xmath15s for the _ kepler _ long - cadence data . @xmath16 is the white - noise level , @xmath17 is the frequency , and @xmath18 and @xmath19 are the timescales and amplitudes of each component . we include a gaussian envelope to account for the oscillation signal and to determine the frequency of maximum power ( @xmath8 ) , such that the power spectrum is modelled using this expression @xcite : @xmath20 here , @xmath21 and @xmath22 are the height and width of the envelope . three examples of power spectra with their background fits and gaussian envelopes superimposed are shown in fig . [ fig.loglog ] . it is known that fitting the stellar background is a difficult task , see e.g. @xcite for a discussion . we have tried different versions of the background fit , but found the model above to give the best results . this is in agreement with @xcite who empirically , on the basis of extensive statistical tests on a number of red giants in the open cluster ngc6819 observed with _ kepler _ , found this model to be optimal for stars with @xmath8 in the range of our stars . we found that the value of @xmath8 is affected by the choice of background model and may differ by several @xmath23hz from one model to the other . the difference is within the uncertainties for most stars , but for the two stars in table [ tab.1 ] with @xmath8 above 100@xmath23hz , differences in @xmath8 between background models were up to 5@xmath23hz . we trust the model above ( eq . 1 ) for three reasons ; 1 ) it is found by @xcite to be optimal for this type of stars , 2 ) the fit converges better ( faster ) than for the other models and 3 ) the results show the expected correlation between @xmath0 and @xmath8 , see for example @xcite , which is not the case when we use other models . the frequency of maximum power @xmath8 was determined for the eight stars using the fitting procedure described in the previous section . to estimate the uncertainty on @xmath8 we split each time series ( two time series for each star , filtered using two different timescales as described above ) in two halves , fitted each half separately using eq . 2 , and used the largest differences between the values of @xmath8 derived from the fit to the full series and from the fits to the half series for our uncertainty estimates . the estimates were taken as the largest difference divided by @xmath24 , however in some cases the derived uncertainty is below 1.0@xmath23hz which seems optimistic when comparing to @xcite who also determined @xmath0 and @xmath8 for these stars , using an automated pipeline . we therefore applied a lower limit on the uncertainty on @xmath8 of 1.0@xmath23hz . for a couple of the stars we did try to split the time series in four instead of two and found consistent results whether we used the same method as above or the standard deviation of the mean value . the results are shown in table [ tab.1 ] . we note that our @xmath8 values generally compare well with those of @xcite except for kic9534041 , where the errorbars overlap at the 2@xmath11 level only . for kic9655167 @xcite quotes a higher @xmath8 value than ours , but also a larger error of nearly 6.0@xmath23hz . we finally note that our uncertainty values are comparable to those of @xcite for similar stars in ngc6819 . solar - like oscillations in red giants follow approximately the asymptotic relation @xcite @xmath25 where @xmath17 is the frequency , @xmath26 is the radial order of the oscillations and @xmath27 is the angular degree , @xmath0 is the large frequency separation between modes with the same @xmath27-value , @xmath28 is a dimensionless parameter which is sensitive to the surface layers , and @xmath29 , where @xmath30 is the small frequency separation between modes of @xmath31 and @xmath32 . to determine the large frequency separation @xmath7 , we first applied a method described in @xcite , which was developed with the aim of analyzing solar - like oscillations in _ kepler _ data . for a range of trial @xmath7-values , the region of the power spectrum containing the oscillations is for each @xmath7-value cut in sections of @xmath7/2 . these sections are then stacked and the highest peak in the summed spectrum is found . this peak will be at its maximum height when the correct @xmath7 value is used , as the @xmath33 modes in this case will add up and create a single , very strong peak in the summed spectrum . in principle , the mixed @xmath1 modes could bias the results . however , our later refinement using only @xmath31 frequencies showed that this was not the case . we then found the frequencies for the individual oscillation modes following a method largely based on @xcite , and detected in this way between 28 and 51 frequencies in the eight analyzed stars . solar - like oscillations are stochastic which means that each individual oscillation mode is represented by a number of peaks in the power spectrum , together forming a lorentzian shape . following @xcite , the power spectra were smoothed using @xmath34 s smooth function , taking care that the smoothing was as mild as possible to ensure that the mean positions of the oscillation modes remained unaffected . in the smoothed power spectra , we detected the oscillation frequencies using an automated procedure where each frequency value in the end , as an extension of the method presented in @xcite , was determined from a gaussian fit to the top part of the corresponding peak in the power spectrum . each oscillation frequency was then manually inspected , using for each star the power spectra from both filtered versions of the time series . the signal - to - noise ( @xmath35 ) values of the individual modes were estimated from the amplitude spectra as illustrated in fig . [ fig.sn ] . for each star , we estimated the background in the amplitude spectrum using median boxcar smoothing on the low- and high - frequency side of the oscillation signal , and used linear interpolation between the low- and high - frequency parts to estimate the background in the region of the spectrum where the oscillations are found , see fig . [ fig.sn ] . we subtracted the background signal and determined the amplitudes of all peaks in the residual spectrum , and used the median peak amplitude at the low- and high - frequency side to estimate the typical amplitude of a noise peak in these two frequency regions . as the noise is higher at low frequencies than at high frequencies , we again used linear interpolation to estimate the noise at the position of a given peak in the region of the oscillations . the @xmath35-value was then calculated as the ratio between the amplitude of the suspected oscillation peak and the noise estimate at the given frequency . we estimated the uncertainties on the frequencies by first dividing each time series in two halves and determined frequency values from these different spectra . as we have two versions of the time series for each star , we obtain four estimates for each frequency , and took the _ rms_-scatter of these four values as our first guess of the uncertainty on the given frequency . we then used the fact that we have a number of detected frequencies for each star ; we expect that the frequency uncertainty correlates with the reciprocal value of the @xmath35 . we therefore plotted the frequency uncertainties as a function of @xmath36 , fitted a line , and used that line to estimate frequency uncertainty at a given @xmath35 . we have compared the uncertainties derived in this way to those of @xcite , who use mcmc - techniques on similar data for similar stars in the open cluster ngc6819 , and found the results to be quite comparable . we note that the search for frequencies in these stars is not a blind search , as the frequency peaks follow eq.3 , at least approximately . in the process of searching for the frequencies , we took advantage of eq.3 , in the sense that the @xmath37 modes form a regular pattern , providing additional support for modes with low @xmath35 . a small number of the retained modes are therefore not significant when seen as isolated peaks , having @xmath35-ratios in amplitude of @xmath38 or even a bit lower , however they fit in the expected mode structure and are therefore included , although a few of them may be noise peaks . in table [ tab.2 ] we list the frequencies ( with uncertainties ) , @xmath35-values and mode - identification for one of the stars , kic9532903 . the frequencies for the remaining seven stars can be found in the online appendices and in the results section of the kasoc data base . for each star , we then constructed chelle diagrams using the first value of the large frequency separation determined above . an example of such an chelle diagramme is shown in the left panel of fig . [ fig.echelle ] . in the chelle diagramme , the oscillation modes are separated along the x - axis into ridges according to the degree ( @xmath27-value ) of the modes , and along the y - axis according to the radial order ( @xmath26-value ) of the modes . in this diagramme , modes of @xmath39 can therefore be identified . we are particularly interested in the mixed - mode structure at @xmath1 , which we will return to in the following section . here , we mention that the final value of the large frequency separation is determined from a weighted linear fit to the four central @xmath31 modes , using the uncertainties on the individual frequencies as weights and for the final error estimate , see table [ tab.1 ] . although we only have measurements for eight stars in ngc6811 , we can still investigate the relation between @xmath0 and @xmath8 . fitting a power law to the data , we find the following relation : @xmath40 the coefficients found here are slightly different from those of other studies , e.g. @xcite , or @xcite who found a constant of @xmath41 and a power of @xmath42 for the giants in ngc6819 . however , the difference is in line with expectations since the relationship between @xmath0 and @xmath8 is mass - dependent . the small frequency separation ( @xmath43 ) is found from the central @xmath37 modes , see the left panel of fig . [ fig.echelle ] . finally the value of @xmath28 is determined from the @xmath31 modes using eq . 3 . these values are listed for completeness in table [ tab.1 ] , along with surface gravities determined from the asteroseismic parameters , but are not used for the analysis in the present paper . the chelle diagrams are shown for all eight stars in fig . [ fig.allechelle ] . the four leftmost panels plot the chelle diagrams of the stars with rich oscillation spectra , while the rightmost panels show the four stars with less pronounced @xmath1 modes . the differences already noted in the power spectra are seen from the chelle diagrams as well . there is a wealth of information present in the oscillation spectra and chelle diagrams for these stars , however the detailed asteroseismic analysis ( including amplitudes , line widths , mode lifetimes , comparison of individual frequencies with model frequencies etc . ) will be presented in a forthcoming paper , where the results for the eclipsing binary stars in the cluster will aid the analysis by pinpointing the cluster isochrone and place additional constraints on the theoretical stellar models . the mixed oscillation modes in evolved stars are characterized by a near equidistance in period @xcite . the so - called observed period spacing @xmath2p@xmath44 , which is some average of the period differences between consecutive pairs of modes , differ from the period spacing of the pure @xmath6-modes , @xmath2p . however , in some cases the latter can be inferred from the observed value @xcite . @xcite showed that the observed period spacing can be used for distinguishing between hydrogen - shell - burning giants with @xmath2p@xmath44 values of @xmath350 seconds and helium - core - burning giants with @xmath2p@xmath44 values of @xmath3100 300 seconds and that the period spacings depend on both mass and evolution . we have determined @xmath2p@xmath44 for our eight giants , again using an analysis largely based on @xcite . in the left panel of fig . [ fig.echelle ] we show , as already discussed , the echelle diagramme for one of the stars ( kic9532903 ) , separating the @xmath37 modes from the mixed @xmath1 modes . in the rightmost panel of fig . [ fig.echelle ] , we show a histogram of the period differences between neighboring @xmath1 modes . the x - axis plots the period spacing of neighboring modes @xmath45 while the y - axis plots a measure of the power ( amplitude ) of the modes represented in each bin . if a pair of modes ( @xmath46,@xmath47 ) have mode heights in the smoothed power spectrum of ( @xmath48,@xmath49 ) , then that pair contributes a value of @xmath50 to the sum of the power in the corresponding bin . in this way , the analysis will be dominated by pairs where both modes have high signal - to - noise , and the region in the power spectrum where the oscillation power is highest is also weighted highest . the histogram is then smoothed , and a gaussian fit is used for determining the value of the average period difference , @xmath2p@xmath44 . this is a different approach than the one used by @xcite , however we have checked that the differences in the obtained values are only marginal . the @xmath2p@xmath44-values are listed in table [ tab.1 ] . they are in the range of @xmath3100 150 seconds , which implies that these red giants are burning helium in their cores @xcite . four of the stars have a sufficiently high number of detectable , mixed @xmath1 modes to allow a determination of the asymptotic period spacing @xmath2p @xcite . we used an empirical method described by @xcite to determine @xmath2p ; fig . [ fig.truedp ] plots the individual period spacings versus the mean value of the two frequencies which gives the period spacing , for the four stars for which @xmath2p is determined . following @xcite , the data are fitted using the asymptotic period spacing of the @xmath6-modes , @xmath2p , and a series of lorentzian profiles separated in frequency by @xmath0 , and for which the width and depth , and the changes with frequency in the width and depth , are fitted as free parameters . the fits are overplotted in fig . [ fig.truedp ] and the asymptotic period spacings listed in table [ tab.1 ] . we tested the robustness of these fits , for example by varying initial guesses and omitting data points , and found the fits to be very stable . we compare the derived period spacings to theoretical models below . for the remaining four stars , we were not able to obtain stable fits to the individual period spacings versus frequency , due fewer frequencies and much lower oscillation amplitudes of especially the @xmath1 modes . for these stars , we set lower limits on the asymptotic period spacings based on the few individual period spacings we deemed trustworthy ( i.e. , where the individual frequency peaks in the power spectra have relatively high amplitude ) . these limits are listed in table [ tab.1 ] as well . we note that compared the method of @xcite for deriving @xmath2p to the method of which relies on the asymptotic relation . the conclusion of is that the two methods provide very similar results . our measurements are furthermore in good agreement with the predictions for @xmath2p@xmath44 and @xmath2p in @xcite , their fig . 1 ( lower panel with overshoot , values in the range @xmath51s for @xmath2p@xmath44 and @xmath52s for @xmath2p ) , except for kic9409513 , where our measured @xmath2p@xmath44 seems higher than expected . finally , we remark that the fitted profiles for the two most evolved stars are different from the profiles of the two least evolved stars , as the more evolved stars have more narrow minima . this is related to the properties of the mixed modes as the stars evolve , however detailed studies of this effect requires individual modeling of the stars in question and will therefore be done at a later stage , when we can include constraints from the binary stars in the cluster . from the asteroseismic parameters , we can estimate the masses and radii of the eight red giants by using the relations between the asteroseismic parameters and stellar mass and radius , see e.g. @xcite : @xmath53 @xmath54 for the solar values , we used @xmath55hz , @xmath56hz @xcite and @xmath57k . the effective temperatures listed in table [ tab.1 ] are , for five of the stars , derived using the so - called sme - method in @xcite . since all stars have not been subject to a common spectroscopic study , we made use of other literature results for subsamples of the stars to put all stars on a common temperature scale . specifically , kic9534041 and 9716522 were analysed in @xcite together with three stars from @xcite . the mean @xmath58 difference of the three latter stars between the two studies ( 34k ) was used to adjust the @xmath58 of kic9534041 and 9716522 to our adopted temperature scale , resulting in effective temperatures of 5027 and 4826k , respectively . for kic9409513 , we found its effective temperature along with that of kic9695101 in the apokasc catalog @xcite , and used the temperature difference for kic9695101 to place kic9409513 on our scale . although it is desirable to measure the effective temperatures of all the stars in a homogeneous way for later detailed studies , the temperatures obtained using the corrections described here are sufficient for our current analysis . the results for the masses and radii are listed in table [ tab.1 ] , indicating masses for the red giants of just above 2m@xmath4 . the uncertainties on mass and radius are obtained using propagation of errors for @xmath8 , @xmath7 , and @xmath58 where for the latter we adopted an uncertainty of 100k in agreement with @xcite . the scaling relations provide a model - indepen- + dent estimate of stellar properties that can be refined from our knowledge of stellar evolution . with this in mind , we fit the observed asteroseismic and atmospheric quantities to a grid of basti @xcite isochrones using the bayesian stellar algorithm ( basta , * ? ? ? * ; * ? ? ? we take into account deviations in the scaling relations as a function of metallicity , effective temperature , and evolutionary phase using the corrections from serenelli et . al . 2017 ( in preparation ) , and determine the stellar properties given in table [ tab.3 ] . these results indicate a mass - sequence among the eight stars , where lower @xmath59 means higher mass , and that the masses are slightly higher than 2m@xmath4 , in agreement with those derived from the scaling relations above . in this we have assumed a common age for all eight stars of @xmath60gyr and solar metallicity following the results of @xcite . effective temperature offsets could influence the absolute values of the masses obtained by asteroseismology , but not the basic results of a mass sequence in the cluster . we have done tests of our obtained masses by increasing / decreasing the @xmath58 values by 50k , and found variations no larger than 0.03m@xmath4 in the individual stars and the persistence of a mass sequence as a function of evolution . we note that the formal model uncertainty estimates in table [ tab.3 ] become unrealistically low for the bottom couple of stars , however as we only use the models for corroborating the estimated masses , this is not important for the present paper . the stellar properties determined in the previous section are based on the acoustic global seismic parameters @xmath8 and @xmath7 , which are sensitive to the ratio of the stellar mass and radius . more detailed information about the stellar core can be obtained from the analysis of gravity modes that penetrate deep into the stellar interior . @xcite studied the correlations between period spacings in red giants and the characteristics of the helium core , based on theoretical models . the stellar masses derived above imply that the red giants in ngc6811 have masses close to the so - called transition mass , separating stars into those that go through a helium flash and those that ignite helium under non - degenerate conditions . this means that these stars have relatively low helium - core masses and belong to the secondary clump @xcite . this conclusion is corroborated by our four derived asymptotic period spacings in the range 180 220s , which is in agreement with the values expected for masses close to the transition mass , see fig . 1 in @xcite . period spacings of red giants have the potential of constraining convective - core overshoot in theoretical stellar models @xcite . to have a first look at this for ngc6811 , we calculated sequences of models of evolved stars in order to compare our derived values of @xmath8 , @xmath7 and @xmath2p with those of the models , in a similar procedure as used by @xcite . the models were computed with the garching stellar evolution code ( garstec , weiss & schlattl 2008 ) at solar metallicity using the @xcite mixture , the opal equation of state @xcite , opal opacities for high temperatures @xcite and those of @xcite for low temperatures , the nacre compilation of nuclear reaction rates @xcite including the updated @xmath61 reaction from @xcite , and the mixing - length theory of convection as described in @xcite . overshoot is implemented as a diffusive process using an exponential decay of the convective velocities based on 2d simulations of stellar atmospheres @xcite , with the diffusion constant given by : @xmath62 here @xmath63 is the distance to the convective border , @xmath64 is a constant derived from the mixing - length theory convective velocities , @xmath65 is the pressure scale height and @xmath66 is an efficiency parameter . the code includes a geometrical cut - off factor to limit the amount of overshoot if convective cores are small and is calibrated to prevent the survival of the pre - main - sequence convective core appearing from non - equilibrium @xmath67he burning in a solar model . under this prescription the calibrated main - sequence overshoot efficiency from isochrone fits to open cluster turn - offs is @xmath68 ( see , e.g. , * ? ? ? * ; * ? ? ? * for details ) , and it is the value we used for main - sequence overshoot in all the models where this effect is taken into account . however , in the present analysis we removed the geometrical cut - off during the helium - burning phase to allow a higher degree of control in the amount of mixing beyond the formal convective boundary to reproduce the measured period spacing values ( see below ) . for this reason the efficiency values quoted for the core helium - burning phase are not directly comparable to the @xmath68 efficiency applied in the main sequence . the asymptotic period spacings for selected models in the helium - core - burning phase were determined from the integral of the brunt - visl frequency , while estimates of the large frequency separation and frequency of maximum power were obtained from the scaling relations . we have calculated models of 2.2m@xmath4 stars , based on the masses derived from the scaling relations above , and 2.4m@xmath4 models for comparison . the models have been computed with 1 ) overshoot both on the main sequence ( @xmath68 ) and in the core - helium - burning phase ( see above for details ) , 2 ) with overshoot only in either the main - sequence ( @xmath68 ) or in the helium - burning phase ( @xmath66 variable ) , and 3 ) without any overshoot at all . in fig . [ fig.models1 ] and [ fig.models2 ] we compare three of the models to the values of @xmath0 and @xmath8 for our eight stars . two of the models ( green and blue ) include convective overshoot on both the main sequence and in the core - helium - burning phase but differ in mass , while the last model ( red ) has the same mass as the green , but has no overshoot . a model with overshoot only in the core - helium - burning phase is very similar to the model without any overshoot , just as a model with overshoot in the main sequence only is very similar to the model with overshoot in both phases . these models are therefore not shown in fig . [ fig.models1 ] and [ fig.models2 ] . these plots indicate that the 2.2m@xmath4 model with overshoot matches the data better than the corresponding 2.4m@xmath4 model for the least evolved stars , and that , assuming that a mass of 2.2m@xmath4 is indeed representative for these stars , a model without overshoot , or with overshoot in the core - helium - burning phase only , can not reproduce the measured values of @xmath8 and @xmath0 . convective - core overshoot on the main sequence ( with an efficiency of the order of the calibrated @xmath68 value ) therefore seems necessary for matching the observed asteroseismic parameters of @xmath8 and @xmath0 for the secondary - clump stars in ngc6811 . however , we can not constrain overshoot in the core - helium - burning phase because @xmath8 and @xmath0 are not sufficient in this regard . we note that it is possible that a sequence of models with slightly different masses and overshoot may also match our measurements . it is , however , beyond the scope of the present paper to explore this any further . in fig . [ fig.sequence ] we compare the measured values of ( @xmath0,@xmath2p ) with theoretical models with different amounts of convective - core overshoot in the helium - core - burning phase . the black model has overshoot on the main sequence only , using the calibrated @xmath68 value , while the other models ( labeled _ all os1 , os2 _ and _ os3 _ ) also have increasing amounts of overshoot in the helium - burning phase , resulting in different sizes of the convective cores . the figure shows that models with overshoot on the main sequence only does not reproduce the measured combinations of @xmath0 and @xmath2p , at least not in all parts of the diagram . the models can be tweaked by adding overshoot in the core - helium - burning phases , which have been done for three additional models also included in the figure ; the green model ( labeled _ all os 2 _ ) fits two of the data points reasonably well , whereas the black model seems to be the better fit for the two more evolved stars with lowest @xmath0 . we would not expect , however , to find a single evolutionary track which matches all the data points because we are in fact observing a sequence of masses , as mentioned above . despite this , we havent yet found a scenario which explains all the measurements . further refinement is beyond the scope of the the present paper , as the eclipsing binaries will provide further constraints for the stellar models , however fig . [ fig.sequence ] demonstrates the potential of red - giant period - spacings for constraining the amount of mixing beyond the core in the helium - burning phase as well as in the main sequence . the models presented here display so - called _ breathing pulses _ which are sudden changes in evolution due to instabilities at the core , resulting in mixing of additional helium into the core and hence increasing the time spent in the helium - burning phase slightly . it is still a matter of debate whether these instabilities actually occur in stars , see e.g. @xcite ; fig . [ fig.sequence ] suggests that further investigation , using measured ( @xmath0,@xmath2p)-values and dedicated modeling , will provide input to this discussion . the breathing pulses seen in our fig . [ fig.sequence ] seems very large and have a significant effect on the evolutionary tracks at low @xmath0 ; @xcite show a diagram similar to ours ( their fig . 19 ) , however their models are not shown down to sufficiently low @xmath0 , and we therefore can not judge if their models display breathing pulses or not and how their models compare with our measurements at low @xmath0 . for their model with maximal overshoot , it seems unlikely that they would be able match our data points as the predicted period spacings are too high . we finally note that according to our measured period spacings , the two most evolved stars ( with lowest @xmath8 and @xmath0 ) are still burning helium in their cores , as agb stars will have much lower period spacings , similar to those of the rgb stars . @xcite describes how the @xmath2p - values will very quickly ( within 1 myr ) drop at the end of core - helium burning , which is also found by @xcite , who discuss how stars leaving the secondary clump to become agb stars are in fact still burning helium in their cores ; this is supported by our 2.2m@xmath4 model which indicate that the most evolved of these two stars still has about 10 % helium left in its core . as is seen in fig . [ fig.power ] , the power spectra of our eight stars split them in two groups with the four most evolved stars ( lowest @xmath8 ) showing rich spectra while the other four show much fewer @xmath1 modes , which is the reason we were not able to derive the asymptotic period spacings for the latter . the stellar parameters for these stars are very similar , especially for the six stars with @xmath0 values above 7@xmath23hz , rendering an explanation of the observed differences based only on structural differences unlikely . it is therefore reasonable to look for other explanations and , as mentioned in the introduction , mode suppression of @xmath1 and higher order modes have been suggested to arise due to strong internal magnetic fields . before we discuss this in further detail , we investigate whether our four stars with highest @xmath0 actually do show suppressed modes . this is done by measuring the squared mode visibilities , following the methods described in @xcite . for this , we used versions of the power spectra where the background signal described in sect . 3 had been subtracted . these power spectra showed minor , positive offsets from zero in the regions of the oscillations ; in order to correct for this , we identified the regions of the four central orders of the @xmath69 modes , which have very small oscillation amplitudes , and used for each star the median value of these regions to correct the zeropoints . for each of the eight stars , we extracted a region of the power spectrum containing the four central @xmath31 peaks , see fig . [ fig.central ] . these regions were then folded using the large frequency separations , resulting in the diagrams shown in fig . [ fig.collapse ] . again following @xcite , the folded diagrams were divided in regions associated with @xmath39 , and the summed power in each region were calculated . the mode visibilities of the non - radial , @xmath70 modes were calculated as the summed power in the region corresponding to the given @xmath27-value , divided by the summed power in the region of @xmath31 . the results are given in table [ tab.4 ] for @xmath1 and @xmath32 ( we have not done this for the few suspected @xmath69 modes ) . comparing the visibilities of @xmath1 and @xmath32 we see differences between the stars for @xmath1 ( first data column in table [ tab.4 ] ) , while this is not the case for @xmath32 . the values of the mode visibilities for @xmath1 and @xmath32 ( first and third data column in table [ tab.4 ] ) can be directly compared with the theoretical predictions in . these authors predict @xmath1 visibilities for non - suppressed stars of 1.54 , and @xmath32 visibilities of 0.58 . for two of our stars , kic9532903 and kic9776739 , the visibilities for @xmath1 mathces the theoretical predictions , which is not the case for the remaining six stars . for @xmath32 all but one of our stars matches reasonably well with the theoretical predictions . for one star , kic9716909 , the squared visibility for @xmath32 is higher than for the remaining stars , however we assume that this is due to a lack of power in @xmath31 , see fig . [ fig.collapse ] . for comparison , @xcite published predictions for magnetic suppression of @xmath1 and @xmath32 modes for core - helium burning giants . according to their fig . 1 , the squared visibilities relative to those of stars with no mode suppression should be in the range @xmath71 for @xmath1 and @xmath72 for @xmath32 for core - helium burning stars with masses between 2.0 and 2.5@xmath73 and radial mode lifetimes of 10 - 30 days . in fig . [ fig.central ] we show the smoothed and un - smoothed power spectra for each of the eight stars . we have fitted lorentzian profiles to a number of isolated modes in each of the eight stars ( not shown ) and find that the mode lifetimes are in the range of 10 - 30 days ( see table [ tab.1 ] ) , and our results are therefore directly comparable to the predictions . to compare with the result of @xcite , we have in data column two in table [ tab.4 ] divided the measured visibilities with the mean value for @xmath1 for kic9532903 and kic9776739 . although we may see mode suppression of @xmath1 for some of the stars , it is much less than predicted by the magnetic suppression theory , and for @xmath32 we do not see any suppression at all . this allows for two different interpretations of our measurements . previous investigations of @xmath1 visibilities for red - giant stars observed with _ kepler _ show that the stars separate into two branches @xcite stars on the branch with low visibilities are referred to as having suppressed modes . however , the scatter among stars on the branch corresponding to unsuppressed modes is much larger than predicted by theory due to differences in stellar parameters . therefore , one interpretation of our measurements is that none of the stars show suppressed modes by the definition in the literature , but there are small but real star - to - star differences of unknown physical origin . alternatively , the four least evolved stars might be showing suppressed modes of @xmath1 , but much less than predicted by the current theory of magnetic suppression . this may be related to the issue mentioned by @xcite that the theory assumes close to full suppression @xcite ; however we do observe mixed @xmath1 modes in all our stars ( see fig . [ fig.allechelle ] and fig . [ fig.central ] ) , which can only be the case if there is power in the @xmath6-modes . in the following , we consider what conclusions might be reached by accepting the current theory of magnetic suppression for either of these two scenarios , starting with the interpretation of the four least evolved stars showing suppressed modes while the others do not . we return to the other interpretation later . @xcite make theoretical predictions for what the observations of suppressed modes ( or not ) in helium - core - burning giants can tell us about the destruction of internal magnetic fields during the evolution of giant stars . the basic idea is that the magnetic field will likely be consumed in convective regions in the star after the main sequence . therefore , a magnetic field is only able to survive in regions of the star that does not become convective at any time after the main sequence . according to the model calculations by @xcite , their fig . 6 , there are no such regions once the star reaches the helium - burning phase if it is below the transition mass and goes through the helium flash . but if the star is above the transition mass there will be a radiative region above the helium - burning convective core which is left untouched by convection where a magnetic field could survive . in fig . [ fig.basti ] we show the size of the helium core for the basti - models introduced above . two model sequences are shown , one with overshoot on the main sequence , and one without . as we have shown , overshoot on the main sequence is necessary to match our observations , indicating that the stars discussed in this paper have masses close to the transition mass in the models , which is about 2.1m@xmath4 for the models with overshoot . according to fig . [ fig.power ] and table [ tab.4 ] , we see suppressed modes for the lowest - mass stars and no mode suppression for the higher - mass stars . therefore , within the theory of magnetic suppression , the difference between the two groups is not related to the transition mass , since that would create the opposite scenario . instead , our observations of four stars with suppressed modes in their early stages of helium core - burning provides evidence that they are indeed above the transition mass . by extension , the four more evolved stars are also above the transition mass as their later evolutionary stages dictate that they reached helium core - burning earlier , which happened because they were more massive . to explain why the four most evolved stars do not show mode suppression we need to adopt a more speculative part of the analysis of @xcite . the suggestion is that the consumption by the convective core of the part of the magnetic field inside the core , will also destabilize and thereby remove the magnetic field from the radiative regions above the convective core . if this process takes some time , then it can explain why the four most evolved stars in our sample do not show suppressed modes while the four least evolved ones do ; the convective helium - burning core destroys the magnetic field , but not immediately . alternatively , since our model comparisons in fig . [ fig.sequence ] suggest that the transition between suppressed and non - suppressed modes happens at a mass close to where the models show the maximum extension of the core , it could be that at this point the convective core has grown big enough to eliminate the last regions that has remained radiative since the main sequence , see fig . 6 of @xcite . the fact that the stars fall in two groups with very different oscillation characteristics may be by chance , as mode suppression is only seen in a subset of the red giants . the masses of our targets are similar to the ones showing close to 50% occurrence rates for suppressed modes among rgb stars @xcite . assuming there is no evolution effect ( no change since the rgb ) and a 50% occurrence rate for suppressed modes , there would be a @xmath74 probability of reaching our observed grouping by chance . returning to the alternative interpretation of no mode suppression , it seems from @xcite that the intrinsic scatter among the stars whith no suppression is comparable to the differences we see between our stars . therefore , we are perhaps just seeing stars at different sides of a population showing a range of visibilities . our sample is much more homogeneous than the sample in @xcite as our cluster stars share common properties such as chemical composition and age , but as already noted by @xcite the observed variation of visibilities is much larger than predicted due to differences in stellar parameters . thus , the physical cause for the different visibilities is currently unknown . according to stello et al . ( 2016 ) , around 50% of giant stars in our mass range show suppressed modes on the rgb . if none of our eight stars , which are all in the later secondary clump phase , show magnetically suppressed modes , this suggests that the magnetic fields in red - giant stars disappears between the rgb and secondary clump phases . if not , there is only a @xmath74 probability of finding no star with mode suppression in our sample . regardless of which scenario turns out to be the correct , the eight stars presented in this paper constitutes an interesting sample for investigating mode suppression . we have analyzed data from _ kepler _ for eight red - giant stars in the open cluster ngc6811 . from the asteroseismic analysis , we find these eight stars to be helium - core - burning , secondary - clump stars . two stars are on the verge of evolving into the agb - phase . comparison with stellar models indicate that these stars can be used for constraining the stellar models , including the amount of convective - core overshoot in the main - sequence and helium - core - burning phases . the possible division of the eight stars in two groups with different mode visibilities points to some evolutionary effect being at play . this may be mode suppression by internal magnetic fields , as discussed in recent literature , however we observe a lower amount of suppression than currently predicted by theory . as a next step , we will extend the theoretical analysis to include comparisons with the observed period spacings as well as with the individual oscillation frequencies ( the chelle diagrammes ) , and we will combine the asteroseismic constraints with those emerging from an ongoing analysis of eclipsing binary stars in ngc6811 . funding for the stellar astrophysics centre is provided by the danish national research foundation ( grant agreement no . : dnrf106 ) . the research was supported by the asterisk project ( asteroseismic investigations with _ song _ and _ kepler _ ) funded by the european research council ( grant agreement no . : 267864 ) . in loving memory of nikolaj holmbo arentoft . these stars now sing for you . angulo , c. , arnould , m. , rayet , m. et al . 1999 , nuclear physics a , 656 , 3 ballot , j. , barban , c. , and vant veer - menneret , c. 2011 , a&a , 531 , a124 bedding , t. r. , mosser , b. , huber , d. et al . 2011 , , 471 , 608 borucki , w. j. , koch , d. , basri , g. et al . 2010 , science , 327 , 977 brogaard , k. , jessen - hansen , j. , handberg , r. et al . 2016 , an special issue _ reconstruction of the milky way s history : spectroscopic surveys , asteroseismology and chemo - dynamical models _ , guest editors c. chiappini , j. montalbn , and m. steffen , in press buysschaert , b. , beck , p. g. , corsaro , e. et al . 2016 , a&a , 588 , a82 cantiello , m. , fuller , j. and bildsten , l. 2016 , , 824 , 14 cassisi , s. 2005 , eprint arxiv : astro - ph/0506161 chaplin , w. j. , miglio , a. 2013 , , 51 , 353 christensen - dalsgaard , j. , arentoft , t. , brown , t. m. et al . 2008 , j. phys . 118 , 012039 christensen - dalsgaard , j. 2011 , arxiv:1106.5946v1 constantino , t. , campbell , t. w. , christensen - dalsgaard , j. et al . 2015 , , 452 , 123 corsaro , e. , stello , d. , huber , d. et al . 2012 , , 757 , 190 freitag , b. , ludwig , h. g. , & steffen , m. 1996 , a&a , 313 , 497 ferguson , j. w. , alexander , d. r. , allard , f. et al . 2005 , , 623 , 585 formicola , a. , imbriani , g. , costantini , h. et al . 2004 , physics letters b , 591 , 61 fuller , j. , cantiello , m. , stello , d. et al . 2015 , science , 350 , 423 girardi , l. 1999 , , 308 , 818 gough d. o. 1986 , in osaki , y. , ed . , hydrodynamic and magnetodynamic problems in the sun and stars . university tokyo press , tokyo , p. 117 grevesse , n. , sauval , a. j. 1998 , space science reviews , 85 , 161 handberg , r. , brogaard , k. , miglio , a. , et al . 2017 , , submitted handberg , r. , lund , m. n. 2014 , , 445 , 2698 hekker , s. , basu , s. , stello , d. et al . 2011 , a&a , 530 , a100 hjrringgaard , j. g. , silva aguirre , v. , white , t. r. et al . 2017 , , 464 , 3713 huber , d. , beddring , t. r. , stello , d. et al . 2010 , , 723 , 1607 huber , d. , bedding , t.r . , stello , d. et al . 2011 , , 743 , 143 iglesias , c. a. , rogers , f. j. 1996 , , 464 , 943 kallinger , t. , de ridder , j. , hekker , s. , et al . 2014 , a&a , 570 , a41 kippenhahn , r. , weigert , a. , weiss , a. 2012 , stellar structure and evolution ( springer - verlag berlin heidelberg , 2012 ) koch , d. g. , borucki , w. j. , basri , g. , et al . 2010 , , 713 , l79 magic , z. , serenelli , a. , weiss , a. and chaboyer , b. 2010 , , 718 , 1378 miglio , a. , brogaard , k. , stello , d. et al . 2012 , , 419 , 2077 molenda - akowicz , j. , sousa , s. g. , frasca , a. et al . 2013 , , 434 , 1422 molenda - akowicz , j. , brogaard , k. , niemczura , e. et al . 2014 , , 445 , 2446 montalbn , j. , miglio , a. , noels , a. et al . 2013 , , 766 , 118 mosumgaard , j. r. 2014 , bachelor project , university of aarhus mosser , b. , belkacem , k. , goupil , m .- j . 2010 , a&a , 517 , a22 mosser , b. , barban , c. , montalbn , j. et al . 2011 , a&a , 532 , a86 mosser , b. , elsworth , y. , hekker , s. et al . 2012a , a&a , 537 , a30 mosser , b. , goupil , m. j. , belkacem , k. et al . 2012b , a&a , 540 , a143 mosser , b. , belkacem , k. , pinon , c. et al . 2016 , in prep . pietrinferni , a. , cassisi , s. , salaris , m. et al . 2004 , , 612 , 168 pinsonneault , m.h . , elsworth , y. , epstein , c. et al . 2014 , , 215 , 23 rogers , f. j. , nayfonov , a. 2002 , , 576 , 1064 rogers , f. j. , swenson , f. j. , iglesias , c. a. 1996 , , 456 , 902 sandquist , e. , jessen - hansen , j. , shetrone , m. d. et al . 2016 , , 831 , 36 silva aguirre , v. , ballot , j. , serenelli , a. m , & weiss , a. 2011 , a&a , 529 , 63 silva aguirre , v. , davies , g. r. , basu , s. et al . 2015 , , 452 , 2127 silva aguirre , v. , lund , m. n. , anita , h. m. et al . 2017 , , accepted stello , d. , chaplin , w. j. , basu , s. et al . 2009 , , 400 , l80 stello , d. , meibom , s. , gilliland , r. l. 2011 , , 739 , 13 stello , d. 2012 , in astronmical society of the pacific conference series , vol . 462 , progress in solar / stellar physics with helio- and asteroseismology , ed . h. shibahashi , m. takata & a. e. lynas - gray , 200 stello , d. , huber , d. , bedding , t. r. et al . 2013 , , 765 , l41 stello , d. , cantiello , m. , fuller , j. et al . 2016 , nature , 529 , 364 stello , d. , cantiello , m. , fuller , j. et al . 2016 , , 33 , 6 tassoul m. 1980 , apjs 43 , 469 vandakurov y. v. 1967 , azh . 44 , 786 weiss , a. , schlattl , h. 2008 , , 316 , 99 ccccccccccccc kic & @xmath0 ( @xmath23hz ) & @xmath43 ( @xmath23hz ) & @xmath28 & @xmath8 ( @xmath23hz ) & @xmath2p@xmath44 ( s ) & @xmath2p ( s ) & t@xmath75 ( k ) & m@xmath76(m@xmath4 ) & r@xmath76(r@xmath4 ) & log@xmath77 & @xmath78 ( d ) + 9716522 & [email protected] & [email protected] & [email protected] & [email protected] & 147@xmath798 & 179@xmath794 & 4826@xmath79100 & [email protected] & [email protected] & 2.64 & 29@xmath797 + 9409513 & [email protected] & [email protected] & [email protected] & [email protected] & 190@xmath793 & 197@xmath795 & 4985@xmath79100 & [email protected] & [email protected] & 2.76 & 29@xmath7913 + 9532903 & [email protected] & [email protected] & [email protected] & [email protected] & 135@xmath793 & 227@xmath796 & 5008@xmath79100 & [email protected] & [email protected] & 2.88 & 20@xmath793 + 9776739 & [email protected] & [email protected] & [email protected] & [email protected] & 136@xmath793 & 215@xmath793 & 4952@xmath79100 & [email protected] & [email protected] & 2.88 & 20@xmath794 + 9655101 & [email protected] & [email protected] & [email protected] & [email protected] & 134@xmath793 & @xmath80170 & 5005@xmath79100 & [email protected] & [email protected] & 2.91 & 18@xmath795 + 9655167 & [email protected] & [email protected] & [email protected] & [email protected] & 104@xmath798 & @xmath80140 & 4924@xmath79100 & [email protected] & [email protected] & 2.91 & 23@xmath798 + 9534041 & [email protected] & [email protected] & [email protected] & [email protected] & 120@xmath792 & @xmath80155 & 5027@xmath79100 & [email protected] & [email protected] & 2.93 & 19@xmath792 + 9716090 & [email protected] & [email protected] & [email protected] & [email protected] & 100@xmath793 & @xmath80125 & 4980@xmath79100 & [email protected] & [email protected] & 2.94 & 14@xmath795 + ccc @xmath17 ( @xmath23hz ) & sn & mode i d + [email protected] & 1.7 & @xmath1 + [email protected] & 2.2 & @xmath32 , @xmath81 + [email protected] & 3.1 & @xmath31 , @xmath82 + [email protected] & 2.9 & @xmath1 + [email protected] & 2.2 & @xmath1 + [email protected] & 3.2 & @xmath1 + [email protected] & 2.1 & @xmath32 , @xmath82 + [email protected] & 4.2 & @xmath31 , @xmath83 + [email protected] & 2.0 & @xmath1 + [email protected] & 6.1 & @xmath1 + [email protected] & 4.6 & @xmath1 + [email protected] & 3.1 & @xmath1 + [email protected] & 4.4 & @xmath32 , @xmath83 + [email protected] & 10.4 & @xmath31 , @xmath84 + [email protected] & 3.4 & @xmath69 ? + [email protected] & 7.1 & @xmath1 + [email protected] & 10.3 & @xmath1 + [email protected] & 7.2 & @xmath1 + [email protected] & 8.6 & @xmath32 , @xmath84 + [email protected] & 13.4 & @xmath31 , @xmath85 + [email protected] & 2.9 & @xmath69 ? + [email protected] & 6.3 & @xmath1 + [email protected] & 18.2 & @xmath1 + [email protected] & 8.6 & @xmath1 + [email protected] & 13.7 & @xmath32 , @xmath85 + [email protected] & 17.9 & @xmath31 , @xmath86 + [email protected] & 3.9 & @xmath69 ? + [email protected] & 1.4 & @xmath69 ? + [email protected] & 5.6 & @xmath1 + [email protected] & 17.4 & @xmath1 + [email protected] & 15.1 & @xmath1 + [email protected] & 12.5 & @xmath32 , @xmath86 + [email protected] & 19.7 & @xmath31 , @xmath87 + [email protected] & 4.7 & @xmath69 ? + [email protected] & 16.2 & @xmath1 + [email protected] & 17.4 & @xmath1 + [email protected] & 10.4 & @xmath1 + [email protected] & 10.1 & @xmath32 , @xmath87 + [email protected] & 12.6 & @xmath31 , @xmath88 + [email protected] & 7.2 & @xmath1 + [email protected] & 14.0 & @xmath1 + [email protected] & 7.4 & @xmath1 + [email protected] & 10.2 & @xmath32 , @xmath88 + [email protected] & 8.3 & @xmath1 + [email protected] & 9.5 & @xmath1 + [email protected] & 7.0 & @xmath32 , @xmath89 + [email protected] & 6.8 & @xmath31 , @xmath90 + [email protected] & 2.0 & @xmath1 + [email protected] & 5.8 & @xmath1 + [email protected] & 6.4 & @xmath1 + [email protected] & 4.3 & @xmath1 + cccc kic & m@xmath91(m@xmath4 ) & r@xmath91(r@xmath4 ) & t@xmath91 ( k ) + & & & + 9716522 & 2.33@xmath92 & 12.21@xmath93 & 4886@xmath94 + 9409513 & 2.35@xmath95 & 10.61@xmath96 & 4990@xmath94 + 9532903 & 2.21@xmath95 & 9.00@xmath97 & 5003@xmath98 + 9776739 & 2.21@xmath99 & 8.91@xmath100 & 4990@xmath101 + 9655101 & 2.23@xmath102 & 8.75@xmath103 & 4990@xmath104 + 9655167 & 2.15@xmath105 & 8.49@xmath106 & 5003@xmath107 + 9534041 & 2.11@xmath108 & 8.24@xmath109 & 4977@xmath110 + 9716090 & 2.10@xmath111 & 8.07@xmath111 & 4990@xmath112 + & & & + cccc kic & @xmath113 & @xmath114 & @xmath115 + & & & + 9716522 & [email protected] & [email protected] & [email protected] + 9409513 & [email protected] & [email protected] & [email protected] + & & & + 9532903 & [email protected] & [email protected] & [email protected] + 9776739 & & [email protected] & [email protected] + & * 1.58@xmath116 * & & + & & & + 9655101 & [email protected] & [email protected] & [email protected] + 9655167 & [email protected] & [email protected] & [email protected] + & & & + 9534041 & [email protected] & [email protected] & [email protected] + 9716090 & [email protected] & [email protected] & [email protected] +

we are able to extend the investigation of the color - morphology - density - radius relations , for bright and faint galaxies , to and to very low density regions , probing the transition region between cluster and field galaxies , and finding a smooth variation between these two populations . we investigate the environmental variation of galaxy properties ( and their relations ) , such as color , spectral type and concentration . our sample comprises 6,415 galaxies that were previously selected as cluster members from 152 systems with . this sample is further divided in complete subsamples of 5,106 galaxies with ( from clusters at ) and 1,309 galaxies with ( from objects at ) . we characterize the environment as a function of the local galaxy density and global cluster related parameters , such as radial distance , substructure , x - ray luminosity and velocity dispersion . for a sample of field galaxies we also trace their environmental dependence using a local galaxy density estimate . our main findings are : ( i ) the fraction of discs is generally higher than the ones for blue and star - forming galaxies , indicating a faster transformation of color and star - formation compared to structural parameters . ( ii ) regarding the distance to the cluster center we find a small variation in the galaxy populations outside the virial radius . once within that radius the fractions of each population change fast , decreasing even faster within . ( iii ) we also find a small increase in the fraction of blue faint galaxies within , before decreasing again to the most central bin . ( iv ) our results do not indicate a significant dependence on cluster mass , except for the disc fraction in the core of clusters . ( v ) the relations between galaxy properties also point to no dependence on cluster mass , except for the scatter of the color stellar mass relation . our results corroborate a scenario on which pre - processing in groups leads to a strong evolution in galaxy properties , before they are accreted by large clusters . [ firstpage ] surveys galaxies : clusters : environment galaxies : evolution .

1310.6309

galaxies in the local universe can be classified in two broad types , according to their structure or properties related to their star formation history . elliptical and s0 galaxies tend to be redder , to be bulge - dominated and show little star formation , being called early - type galaxies . spirals are bluer , show high star formation rate ( sfr ) and are termed late - type . the presence of these two main types have a strong correlation with environment , with the former being predominant in the most dense regions of the universe ( central parts of groups and clusters of galaxies ) , while the latter are more common in low - density environments , called field @xcite . those results can arise from differences in the intrinsic galaxy formation process ( galaxy properties also show strong dependence on stellar mass ) . another natural interpretation of this morphology - density relation is that galaxies are transformed as they move from the field to the central parts of clusters . several different processes acting within groups and clusters could be important for the transformation of late - type into early - type galaxies as they move from low to high density regions . these include ram - pressure stripping , strangulation , mergers and tidal stripping from the cluster potential @xcite . the open question regarding all these mechanisms is to know where they become relevant , as we consider infalling galaxies into groups or clusters . as these systems boundaries are hard to place it is nearly impossible to mark the transition between the field and groups / clusters . recent work has shown that the variation of galaxy types extend to very low density regions where the cluster influence should me minimal @xcite . these results became possible after the very large projects such as the 2df galaxy redshift survey ( 2dfgrs ) and sloan digital sky survey ( sdss ) became available . according to these works there is a critical density for which the connection between sfr and environment first takes place . as this critical density is compatible with regions outside the cluster virial radius , it is suggested that galaxies are first processed in groups @xcite . besides this pre - processing @xcite also suggests that _ overshooting _ ( @xcite ; another indirect mechanism ) is responsible for transforming galaxies , as well as ram pressure due to low density gas in the outskirts of clusters . the morphology - density ( and color - density ) relation is known to have a strong dependence on mass , with red early - type objects being the most massive . that agrees with the scenario of hierarchical growth of structure , from which we expect the most massive halos to be in the most dense parts of the universe . as a consequence we expect the morphology - density relation to be a morphology - halo mass relation however , results in the literature are contradictory . for instance , @xcite and @xcite find no dependence on the fraction of early - type galaxies ( bulge - dominated and red galaxies ) , and properties of the cluster red - sequence ( rs ) , with parent halo mass . on the other hand , @xcite find a small dependence between morphology and halo mass . these differences may be associated to different halo mass ranges . the work of @xcite is based on a narrow cluster mass range ( @xmath7 ) , using simulated data , while the results of @xcite are restricted to clusters with @xmath8 km s@xmath9 . @xcite consider systems spanning a much wider mass range ( @xmath10 ) . another recent work @xcite based on even more massive systems than @xcite ( @xmath11 ; but a different lower mass cut ) find no significant variation of galaxy type fraction with halo mass . this work is the fourth of a series aiming to investigate cluster and galaxies properties at low redshifts ( @xmath12 ) . our main goal is to study galaxy properties in a variety of environments , from the central parts of groups / clusters to the field . for that we obtain the variation of the fractions of galaxy populations ( traced by different properties ) with local and global environment , characterized by the local galaxy density , distance to the center of the parent cluster , cluster mass ( traced by the velocity dispersion and x - ray luminosity ) and substructure . besides investigating the environmental variation of the galaxy populations we check its dependence on galaxy cluster mass . we also investigate the differences in the field and group / cluster galaxies and the transition between such populations . all the analysis is performed on two different luminosity ranges ( @xmath2 and @xmath3 ) , so that we can assess the impact of environment for giant and dwarf galaxies separately . the environmental influence is also investigated in different ranges of galaxy stellar mass . this paper is organized as follows : @xmath132 has the data description , where we also discuss the local galaxy density estimates , the field sample and the separation of the galaxy populations . in @xmath133 we present the environmental variation of different galaxy populations according to different environment tracers . in @xmath134 we show how the relations between different galaxy properties depend on environment . in @xmath135 we summarize and discuss our main results . the cosmology assumed in this work considers @xmath140.3 , @xmath150.7 , and h@xmath16 @xmath17 @xmath18 @xmath19 mpc@xmath9 , with @xmath17 set to 0.7 . for simplicity , in the following we are going to use the term `` cluster '' to refer loosely to groups and clusters of galaxies , except where explicitly mentioned . in the first paper of this series ( hereafter paper i , @xcite ) we defined a cluster sample from the supplemental version of the northern sky optical cluster survey ( nosocs , @xcite ) . for that we used data from the 5th sloan digital sky survey ( sdss ) release . this sample comprises 7,414 systems well sampled in sdss dr5 ( details in paper i ) . nosocs has its origin on the digitized version of the second palomar observatory sky survey ( poss - ii ; dposs , @xcite ) . in @xcite and @xcite the photometric calibration and object classification for dposs , respectively , are described . the supplemental version of nosocs @xcite goes deeper ( @xmath20 ) , but covers a smaller region than the main nosocs catalog @xcite . for a subset of the 7,414 nosocs systems with sdss data we extracted a subsample of low redshift galaxy clusters ( @xmath1 ) this subsample comprises 127 clusters , for which we had enough spectra in sdss for spectroscopic redshift determination , as well as to select cluster members and perform a virial analysis , obtaining estimates of velocity dispersion , physical radius and mass ( @xmath21 , @xmath22 , @xmath23 , @xmath24 and @xmath25 ; details in paper i ) . this low - redshift sample was complemented with more massive systems from the cluster infall regions in sdss ( cirs ) sample ( @xcite , hereafter rd06 ) . cirs is a collection of @xmath1 x - ray selected clusters overlapping the sdss dr4 footprint . the same cluster parameters listed above were determined for these 56 cirs clusters . in the second paper of this series ( hereafter paper ii , @xcite ) we investigated the scaling relations of clusters using this combined sample of 183 clusters at @xmath1 , except for three systems that are not used for having biased values of @xmath21 and mass due to projection effects . these 3 clusters are abell 1035b , abell 1291a , and abell 1291b . we plan to investigate the properties of these interacting clusters in detail in a future work . details about this low - redshift sample and the estimates obtained for the clusters can be found in papers i and ii . for the clusters with at least five galaxy members within r@xmath26 we also have a substructure estimate , based on the ds , or @xmath27 test @xcite . details about this test can be found in paper i. the redshift limit of the sample ( @xmath28 ) is due to incompleteness in the sdss spectroscopic survey for higher redshifts , where galaxies fainter than @xmath29 are missed , biasing the dynamical analysis ( see discussion in section 4.3 of @xcite ) . we eliminated interlopers using the `` shifting gapper '' technique @xcite , applied to all galaxies with spectra available within a maximum aperture of 2.50 h@xmath9 mpc . we also estimated x - ray luminosity ( @xmath30 , using rosat all sky survey data ) , optical luminosity ( @xmath31 ) and richness ( n@xmath32 , @xcite ) . the centroid of each nosocs cluster is a luminosity weighted estimate , which correlates well with the x - ray peak ( see @xcite ) . in @xcite ( hereafter paper iii ) we investigated the connection between galaxy evolution and the dynamical state of galaxy clusters , indicated by their velocity distributions . here we focus on the investigation of the variation of galaxy properties ( and their relations ) , such as color , spectral class , stellar mass , and concentration , with the environment . note that we only use galaxies that are selected as cluster members by the `` shifting gapper '' technique . for the current work , we have also implemented one modification to the `` shifting gapper '' technique , regarding the radial offset between two consecutive galaxies . in previous papers of this series , if this quantity was greater than 1.0 mpc the `` shifting gapper '' procedure was stopped . then , all galaxies with a radial distance from the cluster centre greater or equal to this galaxy s radial distance were eliminated as interlopers ( see @xcite ) . this maximum offset between consecutive galaxies is now set to be 0.75 @xmath33 , where @xmath34 is the biweight estimate of the typical separation of the inner galaxies ( within 0.50 h@xmath9 mpc ) . this modification was motivated by the fact that for a few systems the total number of members ( n@xmath35 ) was much larger than the number of members within r@xmath26 ( n@xmath36 ) . these were small mass groups that have a large cluster nearby ( after @xmath37 2.0 mpc ) . with the new version of this code some lower mass systems end up not having a minimum of ten group members selected to run the virial analysis . due to that our sample is reduced from 180 groups and clusters to 152 , for which we have 6,415 galaxies , being 5,106 with @xmath2 ( from clusters at @xmath1 ) and 1,309 galaxies with @xmath3 ( from objects at @xmath4 ) . our clusters span the range @xmath38 km s@xmath9 , or the equivalently in terms of mass , @xmath39 . when comparing literature results regarding the environmental dependence of galaxy properties one central issue is the definition of environment . that can be local ( associated to the galaxy neighborhood ) or global ( related to the large scale structure ) . the galaxy environment measure is further complicated by different selection criteria , issues related to the survey geometry , as well as luminosity limits . @xcite investigate the impact of group environment on different scales , concluding that not only the surrounding density matters . @xcite present a multiscale model - independent approach to measure the galaxy density . their method is used to investigate the variation of the galaxy ( @xmath40 ) color distribution ( with -21.5 @xmath41 -20 ) on multiscale density . @xcite compare the results from twenty different published environment definitions . they classify the methods to measure environment in two groups , those based on nearest neighbours and those using fixed apertures . they conclude that the local environment is best probed by nearest neighbours methods , while the large - scale environment is best measured by apertures . considering the results from @xcite we decide to adopt a nearest neighbour method to estimate the local environment . the authors also mention that the choice of @xmath42 - the rank of the density - defining neighbour - is very important , as the environment measure may loose power in the case @xmath42 is larger than the number of galaxies residing in the halo . hence , we chose to work with the @xmath43 local galaxy density estimator , as @xmath44 is typically smaller than the number of galaxies we have per cluster and is a common estimate in the literature . to estimate local galaxy densities we proceed as follows . for every galaxy member we compute its projected distance , d@xmath45 , to the 5th nearest galaxy found around it , within a maximum velocity offset of 1000 @xmath46 @xmath19 ( relative to the velocity of the galaxy in question ) . the local density @xmath43 is simply given by 5/@xmath47d@xmath48 , and is measured in units of galaxies / mpc@xmath49 . density estimates are also obtained relatively only to galaxies brighter than a fixed luminosity range , which we adopt as @xmath50 . on what regards the global environment we consider the distance to the center of the parent cluster and its mass ( traced by the x - ray luminosity and velocity dispersion ) . we also inspect possible differences relative to the dynamical state of clusters , indicated by the substructure estimate . the fiber collision issue affects the sdss spectroscopic sample and the derived density estimates . due to a mechanical restriction spectroscopic fibers can not be placed closer than 55 arcsecs on the sky . an algorithm used for target selection randomly chooses which galaxy gets a fiber , in case of a conflict @xcite . this problem is reduced by spectroscopic plate overlaps , but fiber collisions still lead to a @xmath37 6% incompleteness in the main galaxy sample . our approach to fix this problem is similar to the one adopted by @xcite and @xcite . for galaxies brighter than @xmath51 18 with no redshifts we assume the redshift of the nearest neighbour on the sky ( generally the galaxy it collided with ) . this may result in some nearby galaxies to be placed at high redshift , artificially increasing their estimated luminosities . due to that the collided galaxies also assume the magnitudes of their nearest neighbours , resulting in an unbiased luminosity distribution . notice that the fraction of fixed galaxies is quite small ( at most @xmath52 , at highest densities ) ; the above correction procedure has been shown to accurately match the multiplicity function of groups in mock catalogues @xcite ; and the velocity distribution ( relative to group centre ) of the original and fixed samples are consistent @xcite . hence , the local galaxy density estimates take in account the fiber collision issue . for every galaxy we want to estimate density , either group member or belonging to the field ( see next section ) in the original spectroscopic sample , we use the fixed sample to find d@xmath45 and estimate @xmath43 . besides computing local galaxy densities to group / cluster members we also do that for a galaxy field sample . this sample is constructed as follows . from the whole dr7 data set we select galaxies that would not be associated to a group or cluster , considering the cluster catalog from @xcite . to be conservative , we consider a galaxy to belong to the _ field _ if it is not found within 4.0 mpc and not having a redshift offset smaller than 0.06 of any cluster from @xcite . we select more than 60,000 field galaxies , but work with a smaller subset ( randomly chosen ) for which we derived local density estimates . in the end we use 2,986 field galaxies at @xmath53 0.100 with @xmath2 , and 1,829 at @xmath53 0.045 with @xmath3 ( the same order of cluster galaxies ) . for these objects we compute density estimates in the same way as done for the cluster members . for the current work we consider the stacked properties of cluster and field galaxies ( in this case , regarding local density only ) . we do that considering the radial offset ( in units of @xmath23 ) , absolute magnitudes , colors and local densities of all member galaxies coming from the 152 clusters ( or the field ) . our sample consists of 5,106 bright member galaxies with @xmath2 ( at @xmath1 ) , 1,309 faint members with @xmath3 ( at @xmath4 ) , 2,986 bright field galaxies ( @xmath2 , @xmath1 ) and 1,829 faint field galaxies ( @xmath3 , @xmath4 ) . to compute the absolute magnitudes of each galaxy ( in @xmath54 bands ) we consider the following formula : @xmath55 ( @xmath56 is one of the four sdss bands we considered ) , where dm is the distance modulus ( considering the redshift of each galaxy ) , @xmath57 is the k@xmath58correction and @xmath59 ( @xmath60 , @xcite ) is a mild evolutionary correction applied to the magnitudes ( for each galaxy redshift ) . the k@xmath58corrections are obtained directly from the sdss database , for every object in each band . rest - frame colors are also obtained for all objects . galaxies can generally be separated in two groups according to their structure and star formation activity . early - type ( e and s0 ) galaxies are redder , have little star formation and high concentration , while late - type galaxies are bluer , show active star formation and are less concentrated . here we use three galaxy parameters to separate galaxies in two populations and investigate the variation of the fraction of each population ( specified by each of the three parameters ) with environment . we use the ( @xmath40 ) color , the concentration index @xmath61 ( defined as the ratio of the radii enclosing 90 per cent and 50 per cent of the galaxy light in the r - band , @xmath62 ) and a spectral classification called @xmath63 ( which is based on a pca analysis of the sdss spectrum database ) . we call blue / red galaxies those with ( @xmath40 ) values smaller / greater than 2.2 @xcite . galaxies are considered to have low - concentration ( and are called disc - dominated objects ) if they have @xmath64 ( bulge - dominated galaxies have values greater than 2.6 ) . the parameter @xmath63 ranges from about @xmath580.35 to 0.5 for early- to late - type galaxies ( or passive to star - forming ) , with the former having values typically below @xmath580.05 ( spectra with lower - temperature absorption features ) . late - type objects have spectra with higher temperature features or emission lines . note that the separation value for @xmath61 is in agreement to what was adopted by @xcite and @xcite . in fig . [ fig : gals_sep1 ] we show the correlation between ( @xmath40 ) color with @xmath63 and @xmath61 for bright galaxies ( @xmath2 ) in the groups / clusters regions ( _ i.e. _ , all galaxies , members and interlopers ) . the horizontal lines indicate the color separation , while the solid vertical lines indicate the median values ( for @xmath63 and @xmath61 ) of galaxies with ( @xmath40 ) @xmath65 . in the bottom panel the dashed vertical line shows the actual value adopted to split galaxies regarding concentration ( @xmath66 ) . [ fig : gals_sep2 ] is analogous , but for faint galaxies ( @xmath3 ) . as we can see from these two figures , for the data used here , @xmath67 gives the median value of galaxies with ( @xmath40 ) @xmath65 . we consider @xmath66 for concordance with the literature ( actually some authors even consider higher values of @xmath61 ) . note also that we show only galaxies in cluster regions and not from the whole sky ; and concentration does not provide a sharp cut as color or @xmath63 . our choice @xmath66 provides a good compromise between completeness and concentration of the two broad populations ( early and late type galaxies ) . a smaller value of @xmath61 would increase the completeness of early - type objects at the cost of a larger contamination by late - types and vice - versa ( see discussion in @xcite ) . note that this choice of @xmath66 is good enough to split galaxies in two broad populations ( early and late type galaxies ) , as is the goal of next section ( @xmath133 ) . in @xmath134 we investigate the environmental dependence of the relations between different galaxy properties . in that case , we analyze the relations for different subsamples according to concentration . there we decided to consider three ranges of concentration . we call low concentration galaxies or _ classical discs _ objects with @xmath68 , while we consider _ classical bulges _ galaxies with @xmath69 . objects with @xmath70 are called intermediate types . in that case we have purer samples of bulges and discs ( avoiding contamination from the other type ) and consider an intermediate population that can give clues to galaxy transformation regarding environment and stellar mass . in the next section we investigate the environmental variation of the fractions of blue , high @xmath63 ( we call those star - forming ) and low concentration galaxies ( we call those disc - dominated ) , which we refer as @xmath71 and @xmath72 , respectively . instead of separating galaxy populations using only one parameter ( _ e.g. _ color ) we work with these three parameters as `` red '' does not necessarily imply `` bulge '' ( or high - concentration ) , due to differences in the relations between color and morphology with density . ) color with the parameters @xmath63 and @xmath73 for bright galaxies ( @xmath2 ) in group / clusters regions . the horizontal lines indicate the color separation , while the solid vertical lines indicate the median value of galaxies with ( @xmath40 ) @xmath65 . in the bottom panel the dashed vertical line shows the actual value adopted to split galaxies according to concentration ( @xmath66).,width=336 ] ) . the horizontal and vertical lines are the same.,width=336 ] in this section we investigate the variation of galaxy properties ( given by fractions of populations ) with the environment , characterized by the local galaxy density , normalized radial distance to the parent cluster , presence of substructure , as well as its x - ray luminosity and velocity dispersion . in @xmath133.5 the environmental influence is also investigated in different ranges of galaxy stellar mass . when considering the local galaxy density we use cluster members and field galaxies . for the remaining tracers of environment we use only cluster members . in fig . [ fig : denbins ] we show how the fractions of blue ( @xmath74 ) , high @xmath63 ( or star - forming , @xmath75 ) , and low concentration ( or disc - dominated , @xmath72 ) galaxies depend on local galaxy density ( @xmath43 ) , for the bright regime ( @xmath2 ) . @xmath74 is shown in the top panel , while @xmath75 is in the middle panel and @xmath72 in the bottom . red open circles show cluster members , while blue filled circles indicate field galaxies . fig . [ fig : denbins2 ] is analogous , but showing faint galaxies ( @xmath3 ) . from both figures we can see the results of the color- and morphology - density relations , with a small fraction of blue and late - type galaxies ( characterized by @xmath63 and @xmath61 ) in the most dense regions of the universe . these fractions grow continuously to the lower density regime . however , the pace is reduced when we leave clusters and reach the field for bright galaxies , while for faint objects the fractions actually become approximately constant . for instance , for bright objects we have @xmath76 and @xmath77 at the highest and lowest density bins , for cluster members , respectively . for field galaxies we go from @xmath78 to @xmath79 from the highest to lowest density bins . for bright cluster members we still have @xmath80 and @xmath81 , and @xmath82 and @xmath83 , in the highest and lowest density bins , respectively . the equivalent fractions to field galaxies are @xmath84 and @xmath85 , and @xmath86 and @xmath87 . the fact that the variation with density still holds to the lowest density bins in the field , for bright galaxies , indicates that other factors not directly related to clusters may act to affect these galaxies ( such as tidal effects from neighbour galaxies ) . note that in the transition between clusters and the field we might have some contamination on both sides ; _ i.e. _ , field galaxies that are erroneously classified as cluster members and possibly cluster galaxies that are wrongly called field objects ( the former would belong to systems not listed in @xcite ) . however , we do not expect these contaminations to be high as the `` shifting gapper '' technique @xcite was extensively tested elsewhere ( with its member list being well compared to other methods ; see @xcite ) ; and the cluster catalog of @xcite is complete for rich systems in the local universe . so , the expected smooth transition from clusters to the field is not affected in our results . for bright galaxies ( fig . [ fig : denbins ] ) we can see that the highest density field galaxies show larger fractions ( @xmath71 and @xmath72 ) than cluster galaxies at similar density , indicating that even the cluster low density environment plays a role on further diminishing the fraction of blue and star - forming bright galaxies @xcite . for faint objects ( fig . [ fig : denbins2 ] ) that is not true , as at the same density cluster and field galaxies show the same proportion of blue and star - forming objects . we also notice that for the highest density bins the fractions ( @xmath74 and @xmath75 ) do not become smaller than @xmath88 for faint galaxies , while it is @xmath89 for the bright regime ( for @xmath72 the corresponding values are @xmath83 and @xmath88 ) . that could be due to the formation of small ( low luminosity ) galaxies in the central parts of clusters , from tidal debris of galaxy mergers or as consequence of ram pressure stripping of larger galaxies @xcite . three other interesting features stand out in fig . [ fig : denbins2 ] . first , there is nearly no environmental variation for faint field objects , except for the most dense bin . that indicates the factors acting on bright field galaxies are not equally effective for the faint objects . second , the environmental variation is present in all bins for cluster members ( red open circles ) ; indicating the effectiveness of the cluster environment even in the outskirts . the third point is also seen in fig . [ fig : denbins ] . @xmath74 and @xmath75 show very similar behavior to each other , while @xmath72 is generally higher , especially for cluster objects . that goes in line with other results in the literature suggesting that color and star - formation are affected much faster than structural parameters @xcite . and @xmath72 with local galaxy density ( @xmath43 ) . red open circles are for cluster members and blue filled circles for field galaxies . these results consider bright galaxies ( @xmath2).,width=336 ] ) .,width=336 ] figures [ fig : radbins ] and [ fig : radbins2 ] are analogous to the ones shown in the section above , but now exhibiting the variations of @xmath71 and @xmath72 with the normalized distance to the parent cluster ( @xmath90 ) . the results are only shown to cluster members . [ fig : radbins ] is for bright and fig . [ fig : radbins2 ] is for faint galaxies . the color - morphology - radius imprint is clearly seen in both figures . differently than what is seen for cluster galaxies as a function of @xmath43 ( red open circles ) the values of @xmath71 and @xmath72 stop growing ( or grow much slower ) after a given radius ( @xmath91 ) . it is also possible to devise another abrupt change at another radius ( @xmath92 ) . from @xmath91 to the central part of clusters the decline in @xmath71 and @xmath72 becomes steeper within @xmath92 ( in agreement with @xcite ) . for bright galaxies , in the most central bin we have almost no blue and high @xmath63 ( @xmath93 ) galaxies , but the fraction of low @xmath61 objects is @xmath94 . @xmath74 and @xmath75 grow to @xmath95 to roughly the virial radius ( approximated by @xmath23 ) . outside the virial radius they grow to @xmath88 , remaining nearly constant after that . @xmath72 grows to @xmath96 at @xmath97 , being @xmath98 for larger radii . in other words , the population of bright galaxies do not change significantly for objects falling into a cluster until they are within its virial radius . within the cluster the population changes faster when in the core ( @xmath92 ) . that indicates the tidal effects ( due to the cluster potential ) and the interaction to the icm ( denser within @xmath23 ) are very effective on transforming the galaxy populations , which would also lead to the so called anemic spirals @xcite . a similar trend is seen for faint galaxies , but different regimes may take place . from large to small radius it seems that @xmath71 and @xmath72 are roughly constant until @xmath99 ( @xmath100 for @xmath74 ) , falling to a lower plateau between @xmath99 and @xmath91 ( @xmath101 for @xmath74 ) and showing a steep decline until @xmath102 ( @xmath74 decreases to @xmath103 ) . in the very central region ( within @xmath102 ) there is a slight rise in the values of @xmath71 and @xmath72 before reaching the minimum value in the most central bin . if that rise in the central part is real it is probably associated to the formation of small galaxies in the core of clusters , from tidal debris or due to ram pressure stripping of larger galaxies @xcite . the variation in the four regimes mentioned above for @xmath75 and @xmath72 is similar to @xmath74 , being slightly noisier for @xmath72 . against local density @xmath71 have a minimum at @xmath89 and @xmath88 for bright and faint objects , respectively ( the corresponding values are @xmath88 and @xmath83 for @xmath72 ) . regarding the physical radius the minimum values of @xmath71 are @xmath104 and @xmath105 for bright and faint objects , respectively ( for @xmath72 those values are @xmath94 and @xmath106 ) . within @xmath107 , for faint objects , and at all radius for bright galaxies , it can also be seen that @xmath72 is generally higher than @xmath74 and @xmath75 , indicating that infalling galaxies have their color and spectral properties transformed quicker than their structure , corroborating the results from previous section . one final result regarding radius deserves our attention . as seen in @xcite the variation of galaxy types reach far distances from the cluster centers . here we show this variation to be real , for bright and faint galaxies , up to @xmath108 , suggesting that pre - processing by groups , _ overshooting _ and possibly ram pressure in the outskirts of those systems act to transform the galaxy population . and @xmath72 with normalized distance to the cluster center ( @xmath90 ) . only cluster members are shown . these results consider bright galaxies ( @xmath2).,width=336 ] ) .,width=336 ] in figures [ fig : subbins ] and [ fig : subbins2 ] we show the variations of @xmath71 and @xmath72 as a function of @xmath43 and @xmath90 , as in the figures above . however , here we show the results for bright member galaxies coming from clusters with the presence or not of substructure ( blue filled and red open circles , respectively ) . substructure is estimated using the @xmath27 ( or dressler and schectman ) test ( details in paper i ) . it seems that in the most dense regions and within the virial radius there is no significant difference in the values of @xmath74 and @xmath75 , but there might be more low concentration ( @xmath72 ) galaxies in systems with substructure . for instance , at @xmath109 @xmath110 and @xmath88 for systems with and without substructure , respectively . for lower density regions and outside @xmath23 we have slightly higher fractions ( @xmath71 and @xmath72 ) for clusters with substructure . these results agree to the findings of paper iii , for which the systems are classified as gaussian and non - gaussian . , but splitting the sample in galaxies from clusters with or without substructure.,width=336 ] ) .,width=336 ] in this section we investigate the possible dependence of the galaxy population on the parent cluster mass , which we indicate by x - ray luminosity and velocity dispersion . on what follows we show the results obtained for systems at low and high mass ranges . in figures [ fig : lxbins ] and [ fig : lxbins2 ] we show the variations of @xmath71 and @xmath72 as a function of @xmath43 and @xmath90 . however , here we show the results for bright galaxies coming from clusters with low ( @xmath111 ) or high ( @xmath112 ) x - ray luminosity ( used as a mass indicator ) , blue filled and red open circles , respectively . from these plots we do not find any significant difference between the results for low or high @xmath30 systems . the exception is @xmath72 showing different results ( although the results are noisier than for @xmath74 and @xmath75 ) , but only in the central ( within @xmath23 ) and most dense bins ( especially the most central and dense ) . for instance , at @xmath113 @xmath114 and @xmath115 for systems with low and high x - ray luminosity , respectively . at @xmath116 we find @xmath117 and @xmath118 , for low and high @xmath30 , respectively . note that @xmath75 is also slightly higher for low @xmath30 systems in the most central bin . , but splitting the sample in galaxies from clusters with low or high gas content ( @xmath111 or @xmath112).,width=336 ] ) .,width=336 ] in figure [ fig : vdispbins ] we show again the variations of @xmath71 and @xmath72 as a function of @xmath90 , but now we split the galaxies in two mass ranges traced by velocity dispersion ( having approximately the same number of galaxies on each set ) . namely groups with velocity dispersion smaller than 400 km s@xmath9 ( blue filled circles ) and clusters with @xmath119 650 km s@xmath9 ( red open circles ) . according to this separation there are 84 low mass and 16 high mass systems from the original 152 in our sample . there is roughly the same number of galaxies in the two samples , namely 1,685 galaxies in the groups and 1,497 in the high mass clusters . the median value for @xmath23 of the groups is 0.99 mpc , with a minimum of 0.51 mpc and maximum value of 1.46 mpc . the median mass is 1.19 @xmath120 , with minimum and maximum values of 0.16 and 3.84 @xmath120 , respectively . for the high mass sample the median , minimum and maximum values of @xmath23 are 2.00 , 1.76 and 2.39 mpc , respectively . in terms of mass the corresponding values are 9.82 , 6.68 and 16.93 @xmath120 . there is no large difference in the results for low and high mass systems , except for @xmath72 in the two most central bins and @xmath75 in the central bin . at @xmath121 @xmath122 for low mass systems and @xmath123 for high mass objects hence , we find no large differences regarding mass ( using @xmath30 and @xmath124 ) , except for @xmath72 and @xmath75 in the very central parts . , but splitting the sample in galaxies from clusters with small or high mass , traced by the velocity dispersion . blue filled circles are the results for groups ( @xmath125 400 km s@xmath9 ) and red empty circles for clusters ( @xmath126 650 km s@xmath9).,width=336 ] it is well known that galaxy stellar mass and environment are related , with massive galaxies preferentially residing in the densest regions of the universe . many galaxy properties also shown a correlation with stellar mass @xcite , so that it is important to try disentangling the connection between galaxy properties to environment and stellar mass . in order to do that we investigate in this section the environmental dependence of the galaxy population at fixed stellar mass . to obtain galaxy stellar mass estimates we used the stellar population synthesis code starlight @xcite . this code fits an observed spectrum with a linear combination of a number of template spectra with known properties ( see paper iii for details ) . as shown in @xmath134 most bright galaxies are massive , while most faint objects have small masses ( the overlap for our sample occurs around @xmath127 ) . hence , the results shown next consider both bright and faint galaxies , with a combined sample having @xmath128 . in figures [ fig : den_stmassbins ] and [ fig : rad_stmassbins ] we show the variations of @xmath71 and @xmath72 as a function of @xmath43 and @xmath90 . however , the results are now displayed for four different stellar mass ranges : @xmath129 ( circles ) , @xmath130 ( squares ) , @xmath131 ( triangles ) and @xmath132 ( hexagons ) . cluster members are show in red ( filled symbols ) , while field galaxies are in blue ( open symbols ) . as expected , the largest fractions of blue , star - forming and discs are seen for the lower mass galaxies ( the first two mass bins , @xmath133 ) . the environmental variation for these objects is negligible for lower density regions , becoming relevant once galaxies inhabit regions typical of cluster outskirts . towards larger densities and cluster centers the variation in galaxy populations ( for these low mass objects ) becomes stronger ( in agreement with @xcite ) . however , the environmental dependence is steeper in terms of @xmath74 and @xmath75 , being weaker ( flatter ) for @xmath72 . on the contrary , the most massive galaxies have small values of @xmath71 and @xmath72 and also smaller variation with density and cluster radius . these results corroborate the idea that star formation is halted first in higher mass galaxies and that their environmental variation is small when compared to low mass galaxies . that is probably due to a combination of intrinsic galaxy evolution and the longer time these massive systems spend inside clusters . , but splitting the sample according to the galaxy stellar mass . we consider four intervals : @xmath129 ( circles ) , @xmath130 ( squares ) , @xmath131 ( triangles ) and @xmath132 ( hexagons ) . as before , cluster members results are in red ( filled symbols ) and field results in blue ( open symbols ) . all galaxies with @xmath128 ( bright and faint ) are considered.,width=336 ] , but splitting the sample according to the galaxy stellar mass . we consider four intervals : @xmath129 ( circles ) , @xmath130 ( squares ) , @xmath131 ( triangles ) and @xmath132 ( hexagons ) . all galaxies with @xmath128 ( bright and faint ) are considered.,width=336 ] there are well know relations between different photometric and spectral parameters of galaxies , which can be used to understand their formation and evolution . so , as important as to see how those different parameters change with environment it is to investigate the environmental dependence of these relations . that is the goal of the current section , where we investigate relations involving spectral type , color , stellar mass and size , for subsamples obtained relative to concentration , density , cluster radial distance and mass . in fig . [ fig : ec_mstel ] we show the connection between the spectral type ( @xmath63 ) and stellar mass in five local galaxy density bins and for different ranges of concentration . density increases left - right and concentration from bottom - top . the @xmath43 bi - weight estimate of galaxies in each column is indicated in the top panels , ranging from @xmath134 to @xmath135 galaxies mpc@xmath136 . in the lower panels we show low concentration galaxies ( @xmath68 ) , while intermediate types ( @xmath70 ) are displayed in the middle panels and bulges ( @xmath69 ) are shown in the top panels . red small points indicate bright galaxies ( @xmath2 ) and blue large crosses are for faint objects ( @xmath3 ) . the results consider only group members . in @xmath132.3 we call low concentration galaxies ( @xmath68 ) as _ classical discs _ and the high concentration objects as _ classical bulges_. from now on , for simplicity , we are calling these two populations just as discs and bulges , respectively . from fig . [ fig : ec_mstel ] we can see that bulge - dominated objects ( @xmath69 ) are mostly passive objects ( @xmath137 ) . the exception is for a few star - forming galaxies found mainly in the lowest density bin . those objects are mainly faint and with a small stellar mass content . however , a few bright star - forming objects ( red points ) are also found at higher local densities . for disc - dominated objects ( @xmath68 ) most points are consistent with star - forming low mass galaxies ( @xmath138 ) . we can also see that as density increases the median stellar mass too . the most dramatic environmental variation is seen for intermediate type galaxies , which become more massive and , most important , passive as we go from the left to right panels . investigating field galaxies ( figure [ fig : ec_mstel2 ] ) we first notice the five density ranges are much different , ranging from @xmath139 to @xmath140 galaxies mpc@xmath136 . hence , the highest density bin is comparable to the second poorest for clusters ( typical of their outskirts ) . nonetheless , we can see the behavior for galaxies at similar densities ( in groups or field , in figures [ fig : ec_mstel ] and [ fig : ec_mstel2 ] ) show similar relations . for field galaxies we notice that star - forming objects are dominant even for intermediate types ( @xmath70 ) and there is a steeper variation of @xmath63 as a function of @xmath141 for intermediate types in the three lowest density bins . figure [ fig : gr_mstel ] is similar to the previous two , but we show the relation between the rest - frame color @xmath142 ( a rough star formation indicator ) and stellar mass in five local galaxy density bins and for different ranges of concentration . as before , density increases from left to right . differently from previous two figures we only display the binned results ( considering stellar mass ) on each panel . we use different symbols for the three ranges of concentration . blue triangles are for low concentration galaxies ( @xmath68 ) , red squares represent intermediate types ( @xmath70 ) and dark circles display bulges ( @xmath69 ) . in the top panels we have the results for group members and in the bottom for field galaxies . the most striking variations are seen for intermediate concentration objects ( @xmath70 ) , both for field and group galaxies . as density increases ( left to right panels ) these objects ( shown by red squares ) become progressively more massive and most important red . we can see the low mass intermediate type galaxies assume redder colors as density increases . that is true even for the field galaxies , which are restricted to lower density environments . we can see a similar behavior for low mass discs ( blue triangles , @xmath68 ) , but the effect is not as strong as for the red squares , especially for field galaxies . bulges ( dark circles , @xmath69 ) are mostly massive and red objects . for groups that is true for all environments , even in the outskirts ( low densities ) , while we see the lower mass field bulges are bluer in the low density environments ( first three density bins ) . we also investigate how the rest - frame color @xmath142 correlates with galaxy size ( indicated by the petrosian radius in the @xmath143band ) . that is shown for group members in fig . [ fig : gr_rpet ] in the same five density bins and for the three concentration ranges as above . as in fig . [ fig : ec_mstel ] red small points are for bright galaxies and blue large crosses for faint . [ fig : gr_rpet2 ] is analogous to fig . [ fig : gr_rpet ] , but showing the results for field galaxies . from the group members , as we go bottom - up and left - right we can see galaxies forming a tight relation in the color - size space , with large and more concentrated galaxies becoming redder ( see @xcite ) . in other words , as galaxies get to denser environments and grow in size , their star formation decreases . besides that we see that the fraction of blue galaxies is also larger for the middle and bottom panels . if we consider @xmath144 as a typical color for red galaxies @xcite and call blue galaxies with @xmath145 ( typical color minus 0.10 ) we see there is only a small number of blue objects that are also bulges in the three most dense bins ( upper right panels ) . in all the other panels the number of blue objects increases , even for bulges in the two lowest density bins ( upper left panels ) . with a few exceptions , objects with @xmath146 kpc are only seen in the two most dense bins , for intermediate types and bulges . an interest exception is a large disc ( @xmath147 kpc ) seen in the lower right panel . considering the field galaxies ( fig . [ fig : gr_rpet2 ] ) we see the color - radius sequence is starting to form for bulges in high densities ( upper right ) , but still with many blue galaxies and with a small proportion of large objects ( @xmath146 kpc ) . in all the remaining panels we see a large fraction of blue galaxies and nearly no galaxies with @xmath146 kpc . fig . [ fig : gr_mstel2 ] is similar to the ones right above , but now we separate galaxies according to the distance to the center of their parent group . we show the relation between @xmath142 and @xmath141 , also split according to @xmath61 and for bright ( red ) and faint ( blue ) objects . the red sequence is easily seen for bulges ( upper panels ) and it may be seen for intermediate types in the central parts of groups ( middle right panels ) . it is interesting to see that for bulges in the outskirts ( upper left ) the red sequence has a much larger scatter and there is a non - negligible fraction of blue bulges . in the outskirts the discs ( lower left ) are mainly blue . in general , the results of this figure corroborate what was seen in fig . [ fig : gr_mstel ] , splitting galaxies according to local density . besides the dependence on environment traced by local galaxy density or radial distance ( to the parent cluster ) we have also investigated if the relations between physical properties of galaxies depend on the parent cluster mass , traced by velocity dispersion ( @xmath124 ) . [ fig : ec_magrvd ] is similar to fig . [ fig : gr_mstel ] , but now galaxies are separated accordingly to the velocity dispersion of the parent group / cluster . we consider five bins of ( @xmath124 ) . the @xmath124 bi - weight estimate in each column is indicated in the top panels , ranging from @xmath148 to @xmath149 km s@xmath9 . the lower panels have low concentration galaxies ( @xmath68 ) , the middle panels have intermediate types ( @xmath70 ) , while the top panels display bulges , with @xmath69 . as before , red small points indicate bright galaxies ( @xmath2 ) and blue large crosses are for faint objects ( @xmath3 ) . the results consider only group members . from this figure , the main difference regarding @xmath124 may only be the scatter of the red sequence , which might be smaller for higher mass clusters . more massive clusters may also have more faint low mass bulges ( blue crosses ) , as seen in the upper panels . considering galaxies to first be pre - processed in groups , their star formation and color transformation may be further reduced when their parent groups are accreted into larger clusters . those results would contradict the ones from @xcite . they find no dependence of several different properties of the cluster rs ( including the scatter ) with parent halo mass . however , the comparison is not straightforward as their sample is restricted to clusters with @xmath150 km s@xmath9 and for galaxies within @xmath151 . from this plot we conclude that there is no large difference ( expect for the scatter in the red sequence and large number of low mass bulges ) in the relations of galaxy properties for groups and massive clusters , corroborating the idea that galaxies are pre - processed in groups before those are assembled in larger systems . but no large modification occurs after that , except for the tightening of the relations involving luminosity ( color - magnitude , for instance ) . that may be due to a gradual decrease in the star formation rate of galaxies as they reside in larger structures in this work we investigate the impact of environment on the abundance of galaxy populations and on the relations of their physical properties ( such as color , stellar mass , size and spectral classification ) . our sample consists of 6,415 cluster members split in two luminosity and redshift ranges , being 5,106 bright galaxies ( @xmath2 ; at @xmath1 ) and 1,309 faint galaxies ( @xmath3 ; at @xmath4 ) . these galaxies are selected from a total of 152 clusters in the local universe with velocity dispersion in the range @xmath38 km s@xmath9 ( in terms of mass being @xmath39 ) . the local environment is traced by @xmath43 ( a nearest neighbour method ) , while the global environment is related to cluster parameters , such as distance to the cluster center , substructure , x - ray luminosity and velocity dispersion . the analysis regarding the local environment also shows density estimates ( @xmath43 ) of field galaxies . besides the influence of the environment we investigated the dependence of galaxy populations relative to galaxy stellar mass . our main results are : 1 . the fractions of blue ( @xmath74 ) , high @xmath63 ( or star - forming , @xmath75 ) , and low concentration ( or disc - dominated , @xmath72 ) galaxies show a strong variation with local galaxy density , from cluster cores to the field . for bright cluster members @xmath74 ranges from @xmath89 to @xmath77 , while for field galaxies the variation is from @xmath88 to @xmath79 . the results are similar for @xmath75 , but the typical values are higher for @xmath72 ( ranges from @xmath88 to @xmath83 for bright member galaxies and from @xmath152 to @xmath87 for field galaxies ) . there is a smooth transition between cluster and field galaxies , and at similar densities the fractions are higher for field galaxies . 2 . for faint objects there is a strong variation with density in @xmath74 , @xmath75 and @xmath72 for cluster members . however , for field galaxies these fractions are nearly constant . @xmath74 and @xmath75 range from @xmath153 to @xmath154 for cluster members , increasing at most to @xmath155 for field objects . @xmath72 changes from @xmath83 to @xmath154 for cluster members , being that the field value . note that for all the parameters there is a minimum fraction ( @xmath88 for @xmath74 and @xmath75 and @xmath83 for @xmath72 ) in the most dense regions , indicating that the formation of dwarfs in the central parts of clusters might be effective . 3 . for bright and faint galaxies @xmath74 and @xmath75 are similar and generally lower than @xmath72 , especially for cluster objects . that corroborate other results in the literature indicating that color and star - formation are affected faster than structural properties @xcite . 4 . considering the distance to the cluster center , for bright galaxies , the populations do not change significantly until they fall within the virial radius . another abrupt variation seems to happen at @xmath156 ( in agreement with @xcite ) . within that radius the values of @xmath74 , @xmath75 and @xmath72 decrease faster . we also find the clusters cores to be nearly devoided of bright , blue and star - forming objects ( @xmath157 ) . 5 . for faint objects a similar behavior is seen , but with different regimes . from large to small radius @xmath71 and @xmath72 are approximately constant until @xmath99 , then decreasing , but remaining roughly constant between @xmath99 and @xmath91 . then there is a steep decline until @xmath102 . within that radius there is a slight rise in the values of @xmath71 and @xmath72 before reaching the minimum value in the most central bin . that rise might be associated to the formation of small galaxies in the core of clusters , from tidal debris or due to ram pressure stripping of larger galaxies @xcite . 6 . as a function of radius @xmath72 is also generally higher than @xmath74 and @xmath75 , corroborating the slower change in structural parameters . we also show that the dependence with radius goes to large radius ( @xmath108 ) , although the variation is much stronger within the virial radius . that agrees with @xcite , suggesting that some processes like _ overshooting _ and ram pressure may be relevant in the outskirts . 7 . separating the clusters according to the presence of substructure we find that within the virial radius @xmath72 seems to be higher for systems with substructure . outside @xmath23 and for lower density regions there are slightly higher fractions for objects with substructure . 8 . investigating the dependence to cluster mass , traced by @xmath30 or velocity dispersion we find no significant difference between the results for groups and massive clusters . the exceptions are @xmath72 and @xmath75 in the most central bins , which agrees with other results in the literature ( _ e.g. _ @xcite ) . regarding galaxy stellar mass we find the largest fractions of blue , star - forming and discs are seen for the lower mass galaxies . the environmental variation for these objects is negligible for lower density regions , becoming relevant once galaxies reach regions typical of cluster outskirts , increasing further to higher densities . the variation of @xmath72 is not as steep as for @xmath74 and @xmath75 . on the other hand , the most massive galaxies have small values of @xmath71 and @xmath72 and also smaller variation with density and cluster radius . these results corroborate the idea that star formation is halted first in higher mass galaxies and that their environmental variation is small when compared to low mass galaxies . that is probably due to a combination of intrinsic galaxy evolution and the longer time these massive systems spend inside clusters . the relation between @xmath63 and stellar mass shows a tight connection for bulges , both at low and high densities . for discs the relation may be forming at high densities for cluster members . however , the most dramatic environmental variation is seen for intermediate type galaxies . they become more massive and , most important , passive as density increases . we also find that for the most dense regions ( either in clusters or field ) there is a non - negligible population of bulges that are classified as star - forming ( those are mainly faint and low - mass galaxies ) . for field objects the fractions of star - forming are higher and the connection between @xmath63 and @xmath141 is steeper than for cluster galaxies . the connection between rest - frame color @xmath142 and stellar mass also indicates the largest environmental variation is seen for intermediate type objects ( @xmath70 ) when compared to classical discs ( @xmath68 ) and bulges ( @xmath69 ) . galaxies called intermediate type progressively become more massive and especially redder as density increases . that is true even for the field galaxies ( restricted to lower density environments ) . 13 . the color - size relation also shows a tight connection as concentration and density increases . however , large galaxies ( @xmath146 kpc ) are generally seen for intermediate types and cluster bulges in the two highest density bins . the fraction of blue galaxies is also larger for systems in low - density regions or discs in high densities ( both for field and cluster galaxies ) . 14 . separating the color - mass ( galaxy stellar mass ) relation according to the mass of the parent system we find no significant difference between the results for group or high mass cluster galaxies . the only exception being the scatter of the color - mass relation and an excess of low mass bulges in massive clusters . those results also indicate that pre - processing in groups should be very effective . a controversial issue in the literature regards the variation of galaxy populations with parent halo mass ( see @xcite ) . although some works find no variation with halo mass , others indicate a significant dependence . part of this discrepancy may be due to different mass and radial ranges sampled by each study . in the current work we have objects from small groups to large clusters ( @xmath39 ) , and spanning radial distances up to @xmath108 . we find that only in the very central parts there is a small difference in @xmath72 and @xmath75 for low and high mass clusters . that agrees with @xcite , but slightly disagrees with @xcite and @xcite ( although they find a small dependence of the spiral fraction on halo mass ) . for the first case the mass range sampled is very different ( @xcite only have large clusters ) , while in the second the morphological indicator ( as well as the member selection ) used is different than ours . however , all these results reinforce the idea that galaxy transformation happens predominantly on group scales , with large clusters being the result of an hierarchical aggregation of smaller systems ( with no significant further morphological transformation , @xcite ) . the results from @xcite are complementary to this idea , indicating that star formation is quenched in the group scale , but morphological transformation is a separate process , occurring in clusters . in other words , environmental galaxy transformations can be divided in two steps . first , star formation is halted in discs residing in relatively low density environments . second , a morphological transformation from disc to bulge dominated systems occur at higher densities . the authors consider groups as low density and high mass clusters as high density environments . but groups can be as dense as clusters ( especially compact groups , cg ) . for instance , @xcite consider that pre - processing is even more effective in compact groups compared to loose groups ( lg ) . that could be due to the high densities and low velocity dispersion typical of cgs . note that @xcite are interested on classical bulge plus disc galaxies ( sample with 12500 objects ) . our results add to the above conclusions , as we find no significant differences between the values of @xmath74 , @xmath75 and @xmath72 for groups and clusters ( except for @xmath72 in the most central regions ) . we also find that @xmath72 is larger than @xmath74 and @xmath75 , for bright and faint galaxies . hence , in agreement to @xcite , our results point to a scenario on which local density is the main driver for galaxy evolution and not the parent halo mass . this evolution happens as pre - processing in the group scale with star formation quenching and , to a lesser extent , morphological transformation . however , the second effect takes longer for being effective ( @xmath158 ) and shows a small segregation between groups and clusters in their central regions ( most dense in the universe ) ; as @xmath72 is slightly higher for groups than for high mass clusters . that can be seen from figures [ fig : lxbins ] , [ fig : lxbins2 ] and [ fig : vdispbins ] ( so , both as a function of density and radius ) . what we see in the current work does not contradict @xcite or @xcite . the former consider a sample of classical bulge plus disc galaxies only , while we have galaxies of all morphologies . @xcite work with high mass clusters , consider the most central region ( @xmath151 ) only and are restricted to red sequence galaxies , while we consider a broader mass range and a larger region around clusters . the pre - processing in groups can be explained by different mechanisms , such as mergers , strangulation and ram pressure , that can accelerate star formation , remove the gas reservoir and destroy galactic discs @xcite . internal processes associated to the stellar mass are also relevant to halt star formation . within larger systems galaxy mergers are less probable , but galaxies may still be transformed due to the cumulative effects of successive weak encounters , tidal stripping by the cluster gravitational potential and interaction with the denser icm . the smooth transition between cluster and field , seen in figures [ fig : denbins ] and [ fig : denbins2 ] , indicate that although cluster edges are hard to define , density is indeed a key parameter for galaxy transformation . for bright galaxies ( figure [ fig : denbins ] ) the larger fractions for the highest density bin of field galaxies ( compared to cluster galaxies at similar densities ) indicate the significance of cluster environment related effects even in their outskirts @xcite . for bright field galaxies density is a key parameter for their evolution ( although we can not discard or distinguish internal feedback process ) . that is not true for faint field objects ( figure [ fig : denbins2 ] ) , as low luminosity field galaxies are almost independent of environment . for cluster members the dependence on local density is very strong , both for bright and faint galaxies . considering galaxy stellar mass ( figures [ fig : den_stmassbins ] and [ fig : rad_stmassbins ] ) we find the largest fractions of blue , star - forming and discs for lower mass galaxies . environment is not relevant for these objects until they are in regions with local density typical of cluster outskirts . the environmental influence becomes stronger for higher densities for these low mass galaxies . on the contrary , the most massive galaxies show small values of @xmath71 and @xmath72 , as well as a smaller variation with density and cluster radius . comparing figures [ fig : denbins ] and [ fig : radbins ] we see that bright cluster galaxies outside @xmath23 have local densities of about 2 galaxies mpc@xmath136 . in other words , outside the virial radius the local density drops very fast , which explains in part the strong transition we see once galaxies are within this radius . within the virial radius another transition is seen in the inner parts , indicating that cluster related processes ( tidal effects and ram pressure ) further influence the galaxy populations . for faint galaxies , another characteristic region is still seen at @xmath159 . using detailed simulations @xcite could investigate which processes are most common at these regions . they found that at @xmath160 _ overshooting _ and pre - processing by groups explain the cold gas and star formation radial trends . but to larger radius ( reaching @xmath161 ) those two process can not fully explain the hot gas radial trends , so that ram pressure is also relevant . their results indicate that at large radius the ram pressure due to the icm can strip the hot gas haloes both for low and high mass galaxies . the cold gas can be stripped only within @xmath91 . @xcite show that even in low density regions the division of galaxy populations in two broad families is valid . we corroborate these results here , inspecting galaxies at different densities , but we also split them in cluster members and field galaxies . even for the latter this division is still valid . when moving to higher densities we can see the red , early - type sequence is well established with a few transition galaxies still being noticed , such as bulges classified as star - forming ( or blue ) . that is also seen as function of cluster radius , with the red sequence becoming narrower from the outer to inner regions of clusters . in agreement to @xcite we find that cluster mass is not a key parameter for establishing the red sequence . but we find evidence that for high mass systems the red sequence show a reduced scatter compared to small groups . that result indicates that star formation is shut down and galaxies reach the red sequence already in the group scale , but the job is finished in higher mass systems , when the red sequence becomes tighter . note again that @xcite is restricted to the central region and , most important , to higher mass clusters . hence , our main results point to the variation of galaxy populations from very low to the most dense regions of the universe , with a stronger variation after galaxies become part of groups / clusters ( being found within @xmath23 ) . the picture that comes out favors a pre - processing in the group scale , affecting first the star - formation and in a second stage galaxy structure . our results indicate that local density is the main driver for galaxy evolution . the parent halo mass may be relevant only to the most central regions . however , that is a controversial issue in the literature , possibly due to selection effects of the samples considered . in agreement with @xcite and @xcite , we also find an environmental dependence out to very large radius , corroborating the idea that pre - processing by groups and _ overshooting _ act to transform galaxy properties . the possible effect of ram pressure at large radius can only be tested using high resolution numerical simulations , as seen in the work of @xcite . nonetheless , our main findings are in good agreement to what is found in semi - analytic models . albr thanks for the support of cnpq , grants 306870/2010- 0 and 478753/2010 - 1 . paal thanks the support of cnpq , grant 304692/2011 - 5 . we are also thankful to r. de carvalho and f. la barbera for important suggestions regarding this manuscript . this research has made use of the sao / nasa astrophysics data system , and the nasa / ipac extragalactic database ( ned ) . funding for the sdss and sdss - ii was 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 . a list of participating institutions can be obtained from the sdss web site http://www.sdss.org/.

complex enriques surfaces with a finite group of automorphisms are classified into seven types . in this paper , we determine which types of such enriques surfaces exist in characteristic 2 . in particular we give a one dimensional family of classical and supersingular enriques surfaces with the automorphism group isomorphic to the symmetric group of degree five .

1512.06923

we work over an algebraically closed field @xmath2 of characteristic 2 . complex enriques surfaces with a finite group of automorphisms are completely classified into seven types . the main purpose of this paper is to determine which types of such enriques surfaces exist in characteristic 2 . recall that , over the complex numbers , a generic enriques surface has an infinite group of automorphisms ( barth and peters @xcite ) . on the other hand , fano @xcite gave an enriques surface with a finite group of automorphisms . later dolgachev @xcite gave another example of such enriques surfaces . then nikulin @xcite proposed a classification of such enriques surfaces in terms of the periods . finally the second author @xcite classified all complex enriques surfaces with a finite group of automorphisms , geometrically . there are seven types @xmath3 of such enriques surfaces . the enriques surfaces of type @xmath4 or @xmath5 form an irreducible one dimensional family , and each of the remaining types consists of a unique enriques surface . the first two types contain exactly twelve nonsingular rational curves , on the other hand , the remaining five types contain exactly twenty nonsingular rational curves . the enriques surface of type @xmath4 ( resp . of type @xmath6 ) is the example given by dolgachev ( resp . by fano ) . we call the dual graphs of all nonsingular rational curves on the enriques surface of type @xmath7 the dual graph of type @xmath7 ( @xmath8 ) . in positive characteristics , the classification problem of enriques surfaces with a finite group of automorphisms is still open . especially the case of characteristic 2 is most interesting . in the paper @xcite , bombieri and mumford classified enriques surfaces in characteristic 2 into three classes , namely , singular , classical and supersingular enriques surfaces . as in the case of characteristic @xmath9 , an enriques surface @xmath10 in characteristic 2 has a canonical double cover @xmath11 , which is a separable @xmath12-cover , a purely inseparable @xmath13- or @xmath14-cover according to @xmath10 being singular , classical or supersingular . the surface @xmath15 might have singularities , but it is @xmath16-like in the sense that its dualizing sheaf is trivial . in this paper we consider the following problem : _ does there exist an enriques surface in characteristic @xmath17 with a finite group of automorphisms whose dual graph of all nonsingular rational curves is of type @xmath18 or @xmath19 @xmath20 _ note that if enriques surface @xmath21 in any characteristic has the dual graph of type @xmath7 ( @xmath22 ) , then the automorphism group @xmath23 is finite by vinberg s criterion ( see proposition [ vinberg ] ) . we will prove the following table [ table1 ] : [ ] in table [ table1 ] , @xmath24 means the existence and @xmath25 means the non - existence of an enriques surface with the dual graph of type @xmath26 . in case of type @xmath27 , the construction of such enriques surfaces over the complex numbers works well in characteristic 2 ( theorems [ ithm ] , [ iithm ] , [ vithm ] ) . the most difficult and interesting case is of type @xmath6 . we give a 1-dimensional family of classical and supersingular enriques surfaces with a finite group of automorphisms whose dual graph is of type @xmath6 ( theorems [ main ] , [ main2 ] ) . we remark that this family is non - isotrivial ( theorem [ non - isotrivial ] ) . recently the authors @xcite gave a one dimensional family of classical and supersingular enriques surfaces which contain a remarkable forty divisors , by using a theory of rudakov and shafarevich @xcite on purely inseparable covers of surfaces . we employ here the same method to construct the above classical and supersingular enriques surfaces with the dual graph of type @xmath6 . it is known that there exist enriques surfaces in characteristic 2 with a finite group of automorphisms whose dual graphs of all nonsingular rational curves do not appear in the case of complex surfaces ( ekedahl and shepherd - barron@xcite , salomonsson@xcite ) . see remark [ extra ] . the remaining problem of the classification of enriques surfaces in characteristic 2 with a finite group of automorphisms is to determine such enriques surfaces appeared only in characteristic 2 . the plan of this paper is as follows . in section [ sec2 ] , we recall the known results on rudakov - shafarevich s theory on derivations , lattices and enriques surfaces . in section [ sec3 ] , we give a construction of a one dimensional family of classical and supersingular enriques surfaces with the dual graph of type @xmath28 . moreover we show the non - existence of singular enriques surfaces with the dual graph of type @xmath19 ( theorem [ non - existvii ] ) . in section [ sec4 ] , we discuss other cases , that is , the existence of singular enriques surfaces of type @xmath29 and the non - existence of other cases ( theorems [ ithm ] , [ non - existi ] , [ iithm ] , [ non - existii ] , [ vithm ] , [ non - existvi ] , [ non - existiii ] ) . in appendices a and b , we give two remarks . as appendix a , we show that the covering @xmath16 surface of any singular enriques surface has height @xmath30 . as appendix b , we show that for each singular enriques surface with the dual graph of type @xmath31 its canonical cover is isomorphic to the kummer surface of the product of two ordinary elliptic curves . * acknowledgement . * the authors thank igor dolgachev for valuable conversations . in particular all results in section [ sec4 ] were obtained by discussion with him in soeul and kyoto , 2014 . they thank him that he permits us to give these results in this paper . the authors also thank matthias schtt and hiroyuki ito for pointing out the non - existence of singular enriques surfaces with the dual graph of nonsingular rational curves of type @xmath28 . let @xmath2 be an algebraically closed field of characteristic @xmath32 , and let @xmath21 be a nonsingular complete algebraic surface defined over @xmath2 . we denote by @xmath33 a canonical divisor of @xmath21 . a rational vector field @xmath34 on @xmath21 is said to be @xmath35-closed if there exists a rational function @xmath36 on @xmath21 such that @xmath37 . a vector field @xmath34 for which @xmath38 is called of additive type , while that for which @xmath39 is called of multiplicative type . let @xmath40 be an affine open covering of @xmath21 . we set @xmath41 . affine varieties @xmath42 glue together to define a normal quotient surface @xmath43 . now , we assume that @xmath34 is @xmath35-closed . then , the natural morphism @xmath44 is a purely inseparable morphism of degree @xmath35 . if the affine open covering @xmath45 of @xmath21 is fine enough , then taking local coordinates @xmath46 on @xmath47 , we see that there exsit @xmath48 and a rational function @xmath49 such that the divisors defined by @xmath50 and by @xmath51 have no common divisor , and such that @xmath52 by rudakov and shafarevich @xcite ( section 1 ) , divisors @xmath53 on @xmath47 give a global divisor @xmath54 on @xmath21 , and zero - cycles defined by the ideal @xmath55 on @xmath47 give a global zero cycle @xmath56 on @xmath21 . a point contained in the support of @xmath56 is called an isolated singular point of @xmath34 . if @xmath34 has no isolated singular point , @xmath34 is said to be divisorial . rudakov and shafarevich ( @xcite , theorem 1 , corollary ) showed that @xmath57 is nonsingular if @xmath58 , i.e. , @xmath34 is divisorial . when @xmath57 is nonsingular , they also showed a canonical divisor formula @xmath59 where @xmath60 means linear equivalence . as for the euler number @xmath61 of @xmath21 , we have a formula @xmath62 ( cf . katsura and takeda @xcite , proposition 2.1 ) . now we consider an irreducible curve @xmath63 on @xmath21 and we set @xmath64 . take an affine open set @xmath47 above such that @xmath65 is non - empty . the curve @xmath63 is said to be integral with respect to the vector field @xmath34 if @xmath66 is tangent to @xmath63 at a general point of @xmath65 . then , rudakov - shafarevich @xcite ( proposition 1 ) showed the following proposition : [ insep ] @xmath67 if @xmath63 is integral , then @xmath68 and @xmath69 . @xmath70 if @xmath63 is not integral , then @xmath71 and @xmath72 . a lattice is a free abelian group @xmath73 of finite rank equipped with a non - degenerate symmetric integral bilinear form @xmath74 . the signature of a lattice is the signature of the real vector space @xmath75 equipped with the symmetric bilinear form extended from the one on @xmath73 by linearity . a lattice is called even if @xmath76 for all @xmath77 . we denote by @xmath78 the even unimodular lattice of signature @xmath79 , and by @xmath80 or @xmath81 the even _ negative _ definite lattice defined by the cartan matrix of type @xmath80 or @xmath81 respectively . we denote by @xmath82 the orthogonal direct sum of lattices @xmath73 and @xmath83 . let @xmath84 be the orthogonal group of @xmath73 , that is , the group of isomorphisms of @xmath73 preserving the bilinear form . in characteristic 2 , a minimal algebaic surface with numerically trivial canonical divisor is called an enriques surface if the second betti number is equal to 10 . such surfaces @xmath21 are divided into three classes ( for details , see bombieri and mumford @xcite , section 3 ) : * @xmath33 is not linearly equivalent to zero and @xmath85 . such an enriques surface is called a classical enriques surface . * @xmath86 , @xmath87 and the frobenius map acts on @xmath88 bijectively . such an enriques surface is called a singular enriques surface . * @xmath86 , @xmath87 and the frobenius map is the zero map on @xmath88 . such an enriques surface is called a supersingular enriques surface . let @xmath21 be an enriques surface and let @xmath89 be the quotient of the nron - severi group of @xmath21 by torsion . then @xmath89 together with the intersection product is an even unimodular lattice of signature @xmath90 ( cossec and dolgachev @xcite , chap . ii , theorem 2.5.1 ) , and hence is isomorphic to @xmath91 . we denote by @xmath92 the orthogonal group of @xmath89 . the set @xmath93 has two connected components . denote by @xmath94 the connected component containing an ample class of @xmath21 . for @xmath95 with @xmath96 , we define an isometry @xmath97 of @xmath89 by @xmath98 the isometry @xmath97 is called the reflection associated with @xmath99 . let @xmath100 be the subgroup of @xmath92 generated by reflections associated with all nonsingular rational curves on @xmath21 . then @xmath94 is divided into chambers each of which is a fundamental domain with respect to the action of @xmath100 on @xmath94 . there exists a unique chamber containing an ample class which is nothing but the closure of the ample cone @xmath101 of @xmath21 . it is known that the natural map @xmath102 has a finite kernel ( dolgachev @xcite , theorems 4 , 6 ) . since the image @xmath103 preserves the ample cone , we see @xmath104 . therefore @xmath105 is finite if the index @xmath106 $ ] is finite . thus we have the following proposition ( see dolgachev @xcite , proposition 3.2 ) . [ finiteness ] if @xmath100 is of finite index in @xmath107 , then @xmath23 is finite . over the field of complex numbers , the converse of proposition [ finiteness ] holds by using the torelli type theorem for enriques surfaces ( dolgachev @xcite , theorem 3.3 ) . now , we recall vinberg s criterion which guarantees that a group generated by finite number of reflections is of finite index in @xmath92 . let @xmath108 be a finite set of @xmath109-vectors in @xmath89 . let @xmath110 be the graph of @xmath108 , that is , @xmath108 is the set of vertices of @xmath110 and two vertices @xmath99 and @xmath111 are joined by @xmath112-tuple lines if @xmath113 . we assume that the cone @xmath114 is a strictly convex cone . such @xmath110 is called non - degenerate . a connected parabolic subdiagram @xmath115 in @xmath110 is a dynkin diagram of type @xmath116 , @xmath117 or @xmath118 ( see @xcite , p. 345 , table 2 ) . if the number of vertices of @xmath115 is @xmath119 , then @xmath120 is called the rank of @xmath115 . a disjoint union of connected parabolic subdiagrams is called a parabolic subdiagram of @xmath110 . we denote by @xmath121 a parabolic subdiagram which is a disjoint union of two connected parabolic subdiagrams of type @xmath122 and @xmath123 , where @xmath124 is @xmath125 , @xmath126 or @xmath127 . the rank of a parabolic subdiagram is the sum of the rank of its connected components . note that the dual graph of singular fibers of an elliptic fibration on @xmath21 gives a parabolic subdiagram . for example , a singular fiber of type @xmath128 , @xmath129 or @xmath130 defines a parabolic subdiagram of type @xmath131 , @xmath132 or @xmath133 respectively . we denote by @xmath134 the subgroup of @xmath92 generated by reflections associated with @xmath135 . [ vinberg](vinberg @xcite , theorem 2.3 ) let @xmath108 be a set of @xmath109-vectors in @xmath89 and let @xmath110 be the graph of @xmath108 . assume that @xmath108 is a finite set , @xmath110 is non - degenerate and @xmath110 contains no @xmath112-tuple lines with @xmath136 . then @xmath134 is of finite index in @xmath92 if and only if every connected parabolic subdiagram of @xmath110 is a connected component of some parabolic subdiagram in @xmath110 of rank @xmath137 (= the maximal one ) . finally we recall some facts on elliptic fibrations on enriques surfaces . [ multi - fiber](dolgachev and liedtke @xcite , theorem 4.8.3 ) let @xmath138 be an elliptic fibration on an enriques surface @xmath21 in characteristic @xmath17 . then the following hold . @xmath67 if @xmath21 is classical , then @xmath36 has two tame multiple fibers , each is either an ordinary elliptic curve or a singular fiber of additive type . @xmath70 if @xmath21 is singular , then @xmath36 has one wild multiple fiber which is a smooth ordinary elliptic curve or a singular fiber of multiplicative type . @xmath139 if @xmath21 is supersingular , then @xmath36 has one wild multiple fiber which is a supersingular elliptic curve or a singular fiber of additive type . as for the number of multiple fibers in each case , it is given in bombieri and mumford @xcite , proposition 11 . let @xmath140 be a multiple fiber of @xmath141 . if @xmath21 is classical , the multiple fiber @xmath140 is tame . therefore , the normal bundle @xmath142 of @xmath143 is of order 2 ( cf . katsura and ueno @xcite , p. 295 , ( 1.7 ) ) . on the other hand , neither the picard variety @xmath144 of the multiplicative group @xmath145 nor @xmath146 of the supersingular elliptic curve @xmath147 has any 2-torsion point . therefore , @xmath143 is either an ordinary elliptic curve or a singular fiber of additive type . now , we consider an exact sequence : @xmath148 then , we have the long exact sequence @xmath149 if @xmath21 is either singular or supersingular , we have @xmath150 . note that in our case the canonical divisor @xmath151 is linearly equivalent to 0 . since @xmath140 is a multiple fiber , by the serre duality theorem , we have @xmath152 therefore , we see that the natural homomorphism @xmath153 is an isomorphism . if @xmath21 is singular , then the frobenius map @xmath154 acts bijectively on @xmath155 . hence , @xmath154 acts on @xmath156 bijectively . therefore , @xmath143 is either an ordinary elliptic curve or a singular fiber of multiplicative type . if @xmath21 is supersingular , then the frobenius map @xmath154 is the zero map on @xmath155 . hence , @xmath154 is also a zero map on @xmath156 . therefore , @xmath143 is either a supersingular elliptic curve or a singular fiber of additive type . let @xmath138 be an elliptic fibration on an enriques surface @xmath21 . we use kodaira s notation for singular fibers of @xmath36 : @xmath157 [ singular - fiber ] let @xmath138 be an elliptic fibration on an enriques surface @xmath21 in characteristic @xmath17 . then the type of reducible singular fibers is one of the following : @xmath158 @xmath159 consider the jacobian fibration @xmath160 of @xmath36 which is a rational elliptic surface . it is known that the type of singular fibers of @xmath36 coincides with that of @xmath161 ( cf . liu - lorenzini - raynaud @xcite , theorem 6.6 ) . now the assertion follows from the classification of singular fibers of rational elliptic surfaces in characteristic 2 due to lang @xcite , @xcite ( also see ito @xcite ) . in this section , we construct enriques surfaces in characteristic 2 whose dual graph of all nonsingular rational curves is of type vii . the method to construct them is similar to the one in katsura and kondo @xcite , 4 . we consider the nonsingular complete model of the supersingular elliptic curve @xmath147 defined by @xmath162 for @xmath163 , the addition of this elliptic curve is given by , @xmath164 the @xmath165-rational points of @xmath147 are given by @xmath166 the point @xmath167 is the zero point of @xmath147 , and these points make the cyclic group of order five : @xmath168 now we consider the relatively minimal nonsingular complete elliptic surface @xmath169 defined by @xmath170 with a parameter @xmath171 . this surface is a rational elliptic surface with two singular fibers of type @xmath172 over the points given by @xmath173 , and two singular fibers of type @xmath174 over the points given by @xmath175 . here , @xmath176 is a primitive cube root of unity . we consider the base change of @xmath169 by @xmath177 . then , we have the elliptic surface defined by @xmath178 we consider the relatively minimal nonsingular complete model of this elliptic surface : @xmath179 the surface @xmath15 is an elliptic @xmath16 surface . from @xmath15 to @xmath180 , there exists a generically surjective purely inseparable rational map . we denote by @xmath181 the algebraic surface whose coefficients of the defining equations are the square roots of those of @xmath180 . then , @xmath181 is also a rational surface , and we have the frobenius morphism @xmath182 . @xmath154 factors through a generically surjective purely inseparable rational map from @xmath181 to @xmath15 . by the fact that @xmath181 is rational we see that @xmath15 is unirational . hence , @xmath15 is a supersingular @xmath16 surface , i.e. the picard number @xmath183 is equal to the second betti number @xmath184 ( cf . shioda @xcite , p.235 , corollary 1 ) . the discriminant of the elliptic surface @xmath185 is given by @xmath186 and the @xmath187-invariant is given by @xmath188 therefore , on the elliptic surface @xmath185 , there exist two singular fibers of type @xmath189 over the points given by @xmath190 , and two singular fibers of type @xmath191 over the points given by @xmath175 . the regular fiber over the point defined by @xmath192 is the supersingular elliptic curve @xmath147 . the elliptic @xmath16 surface @xmath193 has ten sections @xmath194 @xmath195 given as follows : @xmath196 these ten sections make the cyclic group of order 10 , and the group structure is given by @xmath197 with @xmath198 , the zero section . the images of @xmath199 ( resp . @xmath200 ) ( @xmath201 ) on @xmath180 give sections ( resp . multi - sections ) of @xmath169 . the intersection numbers of the sections @xmath202 @xmath195 are given by @xmath203 where @xmath204 is kronecker s delta . on the singular elliptic surface @xmath205 , we denote by @xmath206 the fiber over the point defined by @xmath207 . @xmath206 is an irreducible curve and on @xmath206 the surface @xmath205 has only one singular point @xmath208 . the surface @xmath15 is a surface obtained by the minimal resolution of singularities of @xmath205 . we denote the proper transform of @xmath206 on y again by @xmath206 , if confusion does nt occur . we have nine exceptional curves @xmath209 @xmath210 over the point @xmath208 , and as a singular fiber of type @xmath211 of the elliptic surface @xmath185 , @xmath206 and these nine exceptional curves make a decagon @xmath212 clockwisely . the blowing - up at the singular point @xmath208 gives two exceptional curves @xmath213 and @xmath214 , and they intersect each other at a singular point . the blowing - up at the singular point again gives two exceptional curves @xmath215 and @xmath216 . the exceptional curve @xmath215 ( resp . @xmath216 ) intersects @xmath213 ( resp . @xmath214 ) transeversely . exceptional curves @xmath215 and @xmath216 intersect each other at a singular point , and so on . by successive blowing - ups , the exceptional curve @xmath217 finally appears to complete the resolution of singularity at the point @xmath208 , and it intersects @xmath218 and @xmath219 transeversely . summerizing these results , we see that @xmath206 intersects @xmath213 and @xmath214 transversely , and that @xmath209 intersects @xmath220 @xmath221 transversely . we choose @xmath213 as the component which intersects the section @xmath222 . then , 10 sections above intersect these 10 curves transversely as follows : [ cols="^,^,^,^,^,^,^,^,^,^,^",options="header " , ] [ ] the surface @xmath15 is a supersingular @xmath16 surface with artin invariant @xmath30 . the elliptic fibration @xmath223 has two singular fibers of type @xmath189 , two singular fibers of type @xmath191 and ten sections . hence the assertion follows from the shioda - tate formula ( cf . shioda @xcite , corollary 1.7 ) . incidentally , by the shioda - tate formula , we also see that the order of the group of the sections of @xmath185 is equal to 10 and so the group is isomorphic to @xmath224 . now , we consider a rational vector field @xmath225 with @xmath226 . then , we have @xmath227 , that is , @xmath228 is @xmath17-closed . on the surface @xmath15 , the divisorial part of @xmath228 is given by @xmath229 we set @xmath230 . then , @xmath231 , that is , @xmath34 is also 2-closed and @xmath34 is of additive type if @xmath232 and of multiplicative type otherwise . moreover , we have @xmath233 from here until theorem [ main ] , the argument is parallel to the one in katsura and kondo @xcite , 4 , and so we give just a brief sketch of the proofs for the readers convenience . the quotient surface @xmath234 is nonsingular . since @xmath15 is a @xmath16 surface , we have @xmath235 . using @xmath236 and the equation ( [ euler ] ) , we have @xmath237 therefore , we have @xmath238 . this means that @xmath34 is divisorial , and that @xmath234 is nonsingular . by the result on the canonical divisor formula of rudakov and shafarevich ( see the equation ( [ canonical ] ) ) , we have @xmath239 [ exceptional ] let @xmath63 be an irreducible curve contained in the support of the divisor @xmath54 , and set @xmath64 . then , @xmath240 is an exceptional curve of the first kind . by direct calculation , @xmath63 is integral with respect to @xmath34 . therefore , we have @xmath68 by proposition [ insep ] . by the equation @xmath241 , we have @xmath242 . since @xmath15 is a @xmath16 surface , @xmath243 is linearly equivalent to zero . therefore , we have @xmath244 therefore , we have @xmath245 and the arithmetic genus of @xmath240 is equal to @xmath9 . hence , @xmath240 is an exceptional curve of the first kind . we denote these 12 exceptional curves on @xmath234 by @xmath246 ( @xmath247 ) , which are the images of irreducible components of @xmath248 by @xmath249 . let @xmath250 be the blowing - downs of @xmath246 ( @xmath247 ) . for simplicity , we denote @xmath251 by @xmath10 . now we have the following commutative diagram : @xmath252 here @xmath154 is the frobenius base change . then , we have @xmath253 the canonical divisor @xmath254 of @xmath10 is numerically equivalent to @xmath9 . as mentioned in the proof of lemma [ exceptional ] , all irreducible curves which appear in the divisor @xmath54 are integral with respect to the vector field @xmath34 . for an irreducible component @xmath63 of @xmath54 , we denote by @xmath240 the image @xmath255 of @xmath63 . then , we have @xmath68 by proposition [ insep ] . therefore , we have @xmath256 since @xmath15 is a @xmath16 surface , @xmath257 therefore , @xmath254 is numerically equivalent to zero . the surface @xmath10 has @xmath258 and @xmath259 . since @xmath260 is finite and purely inseparable , the tale cohomology of @xmath15 is isomorphic to the tale cohomology of @xmath234 . therefore , we have @xmath261 , @xmath262 and @xmath263 . since @xmath264 is the blowing - downs of 12 exceptional curves of the first kind , we see @xmath265 , @xmath266 and @xmath258 . therefore , we have @xmath267 [ main ] under the notation above , the following statements hold . * the surface @xmath268 is a supersingular enriques surface if @xmath269 . * the surface @xmath268 is a classical enriques surface if @xmath270 and @xmath271 . since @xmath272 is numerically trivial , @xmath10 is minimal and the kodaira dimension @xmath273 is equal to @xmath9 . since @xmath274 , @xmath10 is an enriques surface . since @xmath15 is a supersingular k3 surface , @xmath10 is either supersingular or classical . in case that @xmath275 , the integral fiber of the elliptic fibration @xmath185 with respect to @xmath34 exists only over the point @xmath167 defined by @xmath192 . hence @xmath276 has only one multiple fiber . therefore , the multiple fiber is wild , and @xmath10 is a supersingular enriques surface . in case that @xmath277 , the integral fibers of the elliptic fibration @xmath185 with respect to @xmath34 exist over the points @xmath278 defined by @xmath279 and @xmath280 defined by @xmath281 . therefore , the multiple fibers are tame , and we conclude that @xmath10 is a classical enriques surface . recall that the elliptic fibration @xmath282 given in ( [ pencil3 ] ) has two singular fibers of type @xmath189 , two singular fibers of type @xmath191 and ten sections . this fibration induces an elliptic fibration @xmath283 which has two singular fibers of type @xmath172 , two singular fibers of type @xmath174 , and ten 2-sections . thus we have twenty nonsingular rational curves on @xmath10 . denote by @xmath284 the set of curves contained in the support of the divisor @xmath54 : @xmath285 the singular points of four singular fibers of @xmath286 consist of twelve points denoted by @xmath287 which are the images of the twelve curves in @xmath284 . we may assume that @xmath288 are the images of @xmath289 respectively . then @xmath288 ( resp . @xmath290 ) are the singular points of the singular fibers of @xmath286 of type @xmath174 ( resp . of type @xmath172 ) . each of the twenty nonsingular rational curves passes through two points from @xmath291 because its preimage on @xmath15 meets exactly two curves from twelve curves in @xmath284 ( see table [ table2 ] ) . let @xmath292 be the set of fifteen nonsingular rational curves which are ten components of two singular fibers of @xmath286 of type @xmath293 and five 2-sections which do not pass through @xmath294 and @xmath295 , that is , the images of @xmath296 . then the dual graph of the curves in @xmath292 is the line graph of the petersen graph . for the petersen graph , see figure [ petersen ] . here the line graph @xmath297 of a graph @xmath143 is the graph whose vertices correspond to the edges in @xmath143 bijectively and two vertices in @xmath297 are joined by an edge iff the corresponding edges meet at a vertex in @xmath143 . in the following figure [ enriques12 ] , we denote by ten dots the ten points @xmath298 . the fifteen lines denote the fifteen nonsingular rational curves in @xmath292 . on the other hand , let @xmath299 be the set of curves which are the images of @xmath300 . then the dual graph of the curves in @xmath299 is the complete graph with five vertices in which each pair of the vertices forms the extended dynkin diagram of type @xmath131 because all of them pass through the two points @xmath294 and @xmath295 . each vertex in @xmath292 meets exactly one vertex in @xmath299 with multiplicity 2 , because any component of the singular fibers of type @xmath189 meets exactly one section from @xmath300 ( see table [ table2 ] ) and @xmath301 meets only @xmath302 ( @xmath303 ( see the equation ( [ int - sections ] ) ) . on the other hand , the vertex in @xmath299 meets three vertices in @xmath292 with multiplicity 2 , because @xmath302 meets one component of each singular fiber of type @xmath189 and @xmath301 . the dual graph @xmath110 of the twenty curves in @xmath292 and @xmath299 forms the same dual graph of nonsingular rational curves of the enriques surfaces of type @xmath28 given in figure [ figure7 - 7 ] ( fig . 7.7 in @xcite ) . the 15 curves in @xmath292 ( resp . five curves in @xmath299 ) correspond to @xmath304 ( resp . @xmath305 ) in figure [ figure7 - 7 ] . it is easy to see that the maximal parabolic subdiagrams in @xmath110 are @xmath306 which are coresponding to elliptic fibrations of type @xmath307 respectively . it follows from vinberg s criterion ( proposition [ vinberg ] ) that @xmath308 is of finite index in @xmath309 . the same argument in @xcite , ( 3.7 ) implies that @xmath10 contains exactly twenty nonsingular rational curves in @xmath310 . [ injectiv ] the map @xmath311 is injective . let @xmath312 . then @xmath264 preserves each nonsingular rational curve on @xmath10 . since each nonsingular rational curve meets other curves at least three points , @xmath264 fixes all 20 nonsingular rational curves pointwisely . now consider the elliptic fibration @xmath313 . since this fibration has has ten 2-sections , @xmath264 fixes a general fiber of @xmath286 and hence @xmath264 is identity . by proposition [ finiteness ] , we now have the following theorem . [ main2 ] the automorphism group @xmath314 is isomorphic to the symmetric group @xmath1 of degree five and @xmath10 contains exactly twenty nonsingular rational curves whose dual graph is of type @xmath19 . we have already showed that @xmath314 is finite and @xmath10 contains exactly twenty nonsigular rational curves whose dual graph @xmath110 is of type @xmath19 . it follows lemma [ injectiv ] that @xmath314 is a subgroup of @xmath315 . then by the same argument in @xcite , ( 3.7 ) , we see that @xmath316 is represented by automorpisms of @xmath10 . [ non - isotrivial ] the one dimensional family @xmath317 is non - isotrivial . denote by @xmath110 the dual graph of all nonsingular rational curves on @xmath10 which is given in figure [ figure7 - 7 ] . @xmath110 contains only finitely many extended dynkin diagrams (= the disjoint union of @xmath318 ) , that is , @xmath319 ( see also kondo @xcite , page 274 , table 2 ) . note that the elliptic fibrations on @xmath10 bijectively correspond to the extended dynkin diagrams in @xmath110 . this implies that @xmath10 has only finitely many elliptic fibrations . the @xmath187-invariant of the elliptic curve which appears as the fiber @xmath320 defined by @xmath279 of the elliptic fibration @xmath321 is equal to @xmath322 ( cf . section 3 ) . consider the multiple fiber @xmath323 on the elliptic fibration on the enriques surface @xmath10 which is the image of @xmath324 . since we have a purely inseparable morphism of degree 2 from @xmath320 to @xmath325 , we see that the @xmath187-invariant of @xmath325 is equal to @xmath326 . this implies the infiniteness of the number of elliptic curves which appear as the multiple fibers of the elliptic fibration on an enriques surface in our family of enriques surfaces with parameter @xmath327 . therefore , in our family of enriques surfaces there are infinitely many non - isomorphic ones ( see also katsura - kond @xcite , remark 4.9 ) . the pullback of an elliptic fibration @xmath328 to the covering @xmath16 surface @xmath15 gives an elliptic fibration @xmath329 . the type of reducible singular fibers of @xmath330 is @xmath331 if @xmath249 is of type @xmath332 , @xmath333 if @xmath249 is of type @xmath334 , @xmath335 if @xmath249 is of type @xmath336 , and type @xmath337 if @xmath249 is of type @xmath338 , respectively . the following theorem is due to m. schtt and h. ito . [ non - existvii ] there are no singular enriques surfaces with the dual graph of type @xmath19 . assume that there exists an enriques surface @xmath21 with the dual graph of type @xmath19 . in the dual graph of type vii there exists a parabolic subdiagram @xmath339 . by proposition [ singular - fiber ] , it corresponds to an elliptic fibration on @xmath21 with singular fibers of type @xmath340 . for example , the linear system @xmath341 defines a such fibration . moreover the dual graph of type @xmath19 tells us that the singular fiber @xmath342 of type @xmath129 is a multiple fiber because @xmath343 is a 2-section of this fibration ( see figure [ figure7 - 7 ] ) . this contradicts to proposition [ multi - fiber ] , ( ii ) . let @xmath345 be a homogeneous coodinate of @xmath346 . consider the nonsingular quadric @xmath347 in @xmath346 defined by @xmath348 which is the image of the map @xmath349 the involution of @xmath350 @xmath351 induces an involution @xmath352 of @xmath347 whose fixed point set on @xmath347 is one point @xmath353 . consider four lines on @xmath347 defined by @xmath354 @xmath355 and a @xmath356-invariant pencil of quadrics @xmath357 passing through the four vertices @xmath358 of the quadrangle @xmath359 . note that two conics @xmath360 tangent to @xmath361 at two vertices of the quadrangle . obviously @xmath362 and @xmath361 @xmath363 is a nonsingular elliptic curve . thus we have the same configuration of curves given in @xcite , figure 1.1 except @xmath364 and @xmath365 tangent at @xmath353 . now we fix @xmath366 @xmath367 and take artin - schreier covering @xmath368 defined by the triple @xmath369 where @xmath370 , @xmath371 and @xmath372 satisfying @xmath373 and @xmath374 . the surface @xmath21 has four singular points over the four vertices of quadrangle given locally by @xmath375 . in the notation in artin s list ( see @xcite , 3 ) , it is of type @xmath376 . let @xmath15 be the minimal nonsingular model of @xmath21 . then the exceptional divisor over a singular point has the dual graph of type @xmath377 . the canonical bundle formula implies that @xmath15 is a @xmath16 surface . the pencil @xmath378 induces an elliptic fibration on @xmath15 . the preimage of @xmath379 is the singular fiber of type @xmath380 and the preimage of @xmath381 is the union of two singular fibers of type @xmath382 . note that the pencil has four sections . thus we have 24 nodal curves on @xmath15 . note that the dual graph of these 24 nodal curves coincide with the one given in @xcite , figure 1.3 . the involution @xmath356 can be lifted to a fixed point free involution @xmath383 of @xmath15 because the branch divisor @xmath384 does not contain the point @xmath353 . by taking the quotient of @xmath15 by @xmath383 , we have a singular enriques surface @xmath385 . the above elliptic fibration induces an elliptic pencil on @xmath10 with singular fibers of type @xmath386 and of type @xmath382 . since the ramification divisor of the covering @xmath387 is the preimage of @xmath379 , the multiple fiber of this pencil is the singular fiber of type @xmath386 . by construction , @xmath10 contains twelve nonsingular rational curves whose dual graph coincides with the one given in @xcite , figure 1.4 . it follows from vinberg s criterion ( proposition [ vinberg ] ) that @xmath308 is of finite index in @xmath309 , and hence the automorphism group @xmath0 is finite ( proposition [ finiteness ] ) . the same argument as in the proof of @xcite , theorem 3.1.1 shows that @xmath0 is isomorphic to the digedral group @xmath377 of order 8 . thus we have the following theorem . [ ithm ] these @xmath10 form a one dimensional family of singular enriques surfaces whose dual graph of nonsingular rational curves is of type @xmath31 . the automorphism group @xmath314 is isomorphic to the dihedral group @xmath377 of order @xmath137 . [ non - existi ] there are no classical and supersingular enriques surfaces with the dual graph of type @xmath31 . from the dual graph of type @xmath344 , we can see that such enriques surface has an elliptic fibration with a multiple fiber of type @xmath386 . the assersion now follows from proposition [ multi - fiber ] . [ typeinumtrivial ] in the above , we consider special quadrics @xmath388 tangent to @xmath389 . if we drop this condition and consider general @xmath356-invariant quadrics through the four vertices of the quadrangle @xmath359 , we have a two dimensional family of singular enriques surfaces @xmath10 . the covering transformation of @xmath390 descends to a numerically trivial involution of @xmath10 , that is , an involution of @xmath10 acting trivially on @xmath391 . in the appendix [ kummer ] , we discuss enriques surfaces with a numerically trivial involution . we use the same notation as in ( [ type1 ] ) . we consider a @xmath356-invariant pencil of quadrics defined by @xmath393 which tangents to the quadrangle @xmath359 at @xmath394 , @xmath395 , @xmath396 , @xmath397 respectively . let @xmath398 be two lines on @xmath347 which passes the tangent points of @xmath361 and the quadrangle @xmath399 . note that @xmath400 and @xmath361 @xmath363 is a nonsingular elliptic curve . thus we have the same configuration of curves given in @xcite , figure 2.1 . now we fix @xmath366 @xmath401 and take artin - schreier covering @xmath368 defined by the triple @xmath369 where @xmath370 , @xmath371 and @xmath372 satisfying @xmath373 and @xmath374 . the surface @xmath21 has four singular points over the four tangent points of @xmath402 with the quadrangle and four singular points over the four vertices of the quadrangle . a local equation of each of the first four singular points is given by @xmath403 and the second one is given by @xmath404 . in the first case , by the change of coordinates @xmath405 ( @xmath406 , @xmath407 ) , then we have @xmath408 which gives a rational double point of type @xmath409 . in the second case , obviously , it is a rational double point of type @xmath410 . let @xmath15 be the minimal nonsingular model of @xmath21 . then the exceptional divisor over a singular point in the first case has the dual graph of type @xmath409 and in the second case the dual graph of type @xmath410 . the canonical bundle formula implies that @xmath15 is a @xmath16 surface . the pencil @xmath378 induces an elliptic fibration on @xmath15 . the preimage of @xmath379 is the singular fiber of type @xmath411 and the preimage of @xmath412 is the union of two singular fibers of type @xmath413 . note that the pencil has four sections . thus we have 24 nodal curves on @xmath15 . note that the dual graph of these 24 nodal curves coincide with the one given in @xcite , figure 2.3 . the involution @xmath356 can be lifted to a fixed point free involution @xmath383 of @xmath15 because the branch divisor @xmath384 does not contain the point @xmath353 . by taking the quotient of @xmath15 by @xmath383 , we have a singular enriques surface @xmath385 . the above elliptic fibration induces an elliptic pencil on @xmath10 with singular fibers of type @xmath414 and of type @xmath413 . since the ramification divisor of the covering @xmath387 is the preimage of @xmath379 , the multiple fiber of this pencil is the singular fiber of type @xmath414 . by construction , @xmath10 contains twelve nonsingular rational curves whose dual graph @xmath110 coincides with the one given in @xcite , figure 2.4 . the same argument as in the proof of @xcite , theorem 3.2.1 shows that @xmath308 is of finite index in @xmath309 and @xmath10 contains only these twelve nonsingular rational curves . it now follows from proposition [ finiteness ] that the automorphism group @xmath0 is finite . by the similar argument as in the proof of lemma [ injectiv ] , we see that the map @xmath415 is injective . moreover , by the same argument as in the proof of @xcite , theorem 3.2.1 , @xmath0 is isomorphic to @xmath416 . thus we have the following theorem . [ iithm ] these @xmath10 form a one dimensional family of singular enriques surfaces whose dual graph of nonsingular rational curves is of type @xmath417 . the automorphism group @xmath314 is isomorphic to the symmetric group @xmath418 of degree four . [ non - existii ] there are no classical and supersingular enriques surfaces with the dual graph of type @xmath417 . from the dual graph of type @xmath392 , we can see that such enriques surface has an elliptic fibration with a multiple fiber of type @xmath414 . the assersion now follows from proposition [ multi - fiber ] . over the field of complex numbers , the following example was studied by dardanelli and van geemen @xcite , remark 2.4 . this surface @xmath10 is isomorphic to the enriques surface of type @xmath419 given in @xcite ( in @xcite , remark 2.4 , they claimed that @xmath10 is of type @xmath420 , but this is a misprint ) . their construction works well in characteristic 2 . let @xmath421 be a homogeneous coodinate of @xmath422 . consider the surface @xmath21 in @xmath422 defined by @xmath423 let @xmath424 @xmath425 the ten lines @xmath426 and ten points @xmath427 lie on @xmath21 . by taking partial derivatives , we see that @xmath21 has ten nodes at @xmath427 . let @xmath15 be the minimal nonsingular model of @xmath21 . then @xmath15 is a @xmath16 surface . denote by @xmath428 the proper transform of @xmath426 and by @xmath429 the exceptional curve over @xmath427 . the cremonat transformation @xmath430 acts on @xmath15 as an automorphism @xmath383 of order 2 . note that the fixed point set of the cremonat transformation is exactly one point @xmath431 . hence @xmath383 is a fixed point free involution of @xmath15 . the quotient surface @xmath385 is a singular enriques surface . obviously the permutation group @xmath1 acts on @xmath21 which commutes with @xmath383 . therefore @xmath1 acts on @xmath10 as automorphisms . the involution @xmath383 changes @xmath428 and @xmath432 , where @xmath433 . the images of twenty nonsingular rational curves @xmath428 , @xmath429 give ten nonsingular rational curves on @xmath10 whose dual graph is given by the following figure [ petersen ] . note that this graph is well known called the petersen graph . here @xmath434 is the image of @xmath428 ( and @xmath432 ) . note that @xmath1 is the automorphism group of the petersen graph . the hyperplane section @xmath435 on @xmath21 is the union of the double line @xmath436 and two lines through @xmath437 defined by @xmath438 . thus we have additional twenty nodal curves on @xmath15 . note that the cremonat transformation changes two lines defined by @xmath438 . thus @xmath10 contains twenty nonsingular rational curves whose dual graph @xmath110 coincides with the one of the enriques surface of type vi ( see fig.6.4 in @xcite ) . it now follows from proposition [ finiteness ] that the automorphism group @xmath0 is finite . the same argument as in the proof of @xcite , theorem 3.1.1 shows that @xmath10 contains only these 20 nonsingular rational curves . by a similar argument to the one in the proof of lemma [ injectiv ] , we see that the map @xmath415 is injective . since the classes of twenty nonsingular rational curves generate @xmath439 , @xmath0 is isomorphic to @xmath440 . thus we have the following theorem . [ vithm ] the surface @xmath10 is a singular enriques surfaces whose dual graph of nonsingular rational curves is of type @xmath441 . the automorphism group @xmath0 is isomorphic to the symmetric group @xmath1 of degree five . [ non - existvi ] there are no classical and supersingular enriques surfaces with the dual graph of type @xmath441 . a pentagon in the figure [ petersen ] , for example , @xmath442 , defines an elliptic fibration on @xmath10 . the multiple fiber of this fibration is nothing but the pentagon , that is , of type @xmath172 . the assertion now follows from proposition [ multi - fiber ] . [ type7 ] over the field of complex numbers , ohashi found that the enriques surface of type @xmath28 in @xcite is isomorphic to the following surface ( see @xcite , 1.2 ) . let @xmath421 be homogeneous coodinates of @xmath422 . consider the surface in @xmath422 defined by @xmath443 which has five nodes at coodinate points and whose minimal resolution is a @xmath16 surface @xmath15 . the standard cremonat transformation @xmath444 acts on @xmath15 as a fixed point free involution @xmath383 . thus the quotient surface @xmath385 is a complex enriques surface . in characteristic 2 , the involution @xmath383 has a fixed point @xmath431 on @xmath15 , and hence the quotient is not an enriques surface . in each case of type @xmath382 , @xmath420 , @xmath446 , from the dual graph ( cf . kondo @xcite , figures 3.5 , 4.4 , 5.5 ) we can find an elliptic fibration which has two reducible multiples fibers . in fact , the parabolic subdiagram of type @xmath447 in case @xmath382 ( of type @xmath448 in case @xmath420 , of type @xmath449 in case @xmath446 ) defines such an elliptic fibration ( see @xcite , table 2 , page 274 ) . hence if an enriques surface with the same dual graph of nodal curves exists in characteristic 2 , then it should be classical ( proposition [ multi - fiber ] ) . on the other hand , in each case of type @xmath382 , @xmath420 , @xmath446 , there exists an elliptic fibration which has a reducible multiple fiber of multiplicative type ( see @xcite , table 2 , page 274 ) . however this is impossible because any multiple fiber of an elliptic fibration on a classical enriques surface is nonsingular or singular of additive type ( proposition [ multi - fiber ] ) . thus we have prove the following theorem . [ non - existiii ] there are no enriques surfaces with the same dual graph as in case of type @xmath128 , @xmath129 or @xmath450 . combining theorems [ main2 ] , [ non - existvii ] , [ ithm ] , [ non - existi ] , [ iithm ] , [ non - existii ] , [ vithm ] , [ non - existvi ] , [ non - existiii ] , we have the table [ table1 ] in the introduction . [ extra ] in characteristic 2 , there exist enriques surfaces with a finite group of automorphisms whose dual graphs of all nonsingular rational curves do not appear in the case of complex surfaces . for example , it is known that there exists an enriques surface @xmath10 which has a genus 1 fibration with a multiple singular fiber of type @xmath451 and with a 2-section ( ekedahl and shepherd - barron@xcite , theorem a , salomonsson@xcite , theorem 1 ) . we have ten nonsingular rational curves on @xmath10 , that is , nine components of the singular fiber and a 2-section , whose dual graph is given in figure [ e10dynkin ] . ( 120,30 ) ( 0 , 20 ) ( 0 , 30)(0 , 0 ) ( 2 , 20)(1 , 0)15 ( 20 , 20 ) ( 20 , 30)(0 , 0 ) ( 22 , 20)(1 , 0)15 ( 40 , 20 ) ( 40 , 30)(0 , 0 ) ( 42 , 20)(1 , 0)15 ( 60 , 20 ) ( 60 , 30)(0 , 0 ) ( 62 , 20)(1 , 0)15 ( 80 , 20 ) ( 80 , 30)(0 , 0 ) ( 82 , 20)(1 , 0)15 ( 100 , 20 ) ( 100 , 30)(0 , 0 ) ( 102 , 20)(1 , 0)15 ( 120 , 20 ) ( 120 , 30)(0 , 0 ) ( 122 , 20)(1 , 0)15 ( 140 , 20 ) ( 140 , 30)(0 , 0 ) ( 142 , 20)(1 , 0)15 ( 160 , 20 ) ( 160 , 30)(0 , 0 ) ( 40 , 17)(0 , -1)15 ( 40 , 0 ) ( 50 , 0)(0 , 0 ) ( 120 , 0)(0 , 0 ) it is easy to see that they generate @xmath452 . moreover it is known that the reflection subgroup generated by reflections associated with these @xmath109-vectors is of finite index in @xmath309 ( vinberg @xcite , table 4 ; also see proposition [ vinberg ] ) and hence @xmath0 is finite ( proposition [ finiteness ] ) . in this section we prove the following theorem . [ height ] in characteristic @xmath17 , if @xmath16 surface @xmath15 has a fixed point free involution , then the height @xmath453 of the formal brauer group of @xmath15 is equal to @xmath30 . [ height2 ] let @xmath15 be the covering @xmath16 surface of a singular enriques surface . then the height @xmath454 . suppose @xmath455 . since @xmath456 is the tangent space of the formal brauer group of @xmath15 ( cf . artin - mazur @xcite , corollary ( 2.4 ) ) , the frobenius map @xmath457 is a zero map . then , we have an isomorphism @xmath458 let @xmath459 be the sheaf of ring of witt vectors of length @xmath460 on @xmath15 . assume @xmath461 is an isomorphism . we have an exact sequence @xmath462 where @xmath463 is the verschiebung and @xmath180 is the restriction . then , we have a diagram @xmath464 by the assumption of induction , the first and the third downarrows are isomorphisms . therefore , by the 5-lemma , we have an isomorphism @xmath465 therefore , taking the projective limit , we have an isomorphism @xmath466 therefore , denoting by @xmath7 the quotient field of the ring of witt vectors @xmath467 of infinite length , we have an isomorphism @xmath468 let @xmath469 be the second 2-adic tale cohomology of @xmath15 . then , we have an exact sequence @xmath470 ( cf . crew @xcite , ( 2.1.2 ) for instance ) . therefore , we have @xmath471 . on the other hand , we consider the quotient surface @xmath10 of @xmath15 by the fixed point free involution . then , @xmath10 is a singular enriques surface and under the assumption crew showed @xmath472 for the @xmath16 covering y of @xmath10 ( crew @xcite , p41 ) , a contradiction . in this section we show that enriques surfaces given in remark [ typeinumtrivial ] are obtained from kummer surfaces associated with the product of two ordinary elliptic curves . let @xmath473 be two ordinary elliptic curves and let @xmath474 be the inversion of the abelian surface @xmath475 . let @xmath476 be the minimal resolution of the quotient surface @xmath477 . it is known that @xmath476 is a @xmath16 surface called kummer surface associated with @xmath475 ( shioda @xcite , proposition 1 , see also katsura @xcite , theorem b ) . the projection from @xmath475 to @xmath147 gives an elliptic fibration which has two singular fibers of type @xmath478 and two sections . let @xmath479 be the unique non - zero 2-torsion points on @xmath147 , @xmath154 respectively . denote by @xmath480 the translation of @xmath475 by the 2-torsion point @xmath481 . the involution @xmath482 induces a fixed point free involution @xmath383 of @xmath476 . thus we have an enriques surface @xmath483 . the involution @xmath484 ( or @xmath480 ) induces a numerically trivial involution @xmath485 of @xmath21 . let @xmath487 be two ordinary elliptic curves . the inversion @xmath488 is then expressed by @xmath489 and the translation by the non - zero 2-torsion on @xmath147 is given by @xmath490 then the function field of @xmath491 is given by @xmath492 with the relation @xmath493 where @xmath494 ( see shioda @xcite , the equation ( 8) ) . the fixed point free involution @xmath383 is expressed by @xmath495 and the involution induced by @xmath484 on @xmath476 is given by @xmath496 on the other hand , we consider the quadric @xmath347 given in ( [ type1quadric ] ) . instead of @xmath356 in ( [ typeiinv ] ) , we consider the involution given by @xmath497 whose fixed point is @xmath498 . the artin - schreier covering is defined by the equation @xmath499 ( in the example given in the subsection [ type1 ] , the term @xmath500 appears in the artin - schreier covering . if @xmath501 , then changing @xmath502 by @xmath503 where @xmath504 , we can delete this term ) . now , by putting here @xmath505 and considering an affine locus @xmath506 , we have @xmath507 which is the same as the equation given in ( [ shiodakummer ] ) . moreover the lifting of @xmath508 and the covering transformation of the artin - schreier covering coincide with the ones given in ( [ enriques - inv ] ) and ( [ num - tri - inv ] ) respectively . all complex enriques surfaces with cohomologically or numerically trivial automorphisms are classified by mukai and namikawa @xcite , main theorem ( 0.1 ) , and mukai @xcite , theorem 3 . there are three types : one of them is an enriques surface associated with @xmath476 and the second one is mentioned in remark [ typeinumtrivial ] . for the third one we refer the reader to mukai @xcite , theorem 3 . in positive characteristic , dolgachev ( @xcite , theorems 4 and 6 ) determined the order of cohomologically or numerically trivial automorphisms . however , the explicit classification is not known . the above theorem [ nt ] implies that two different type of complex enriques surfaces with a numerically trivial involution coincide in characteristic 2 . 99 m. artin , , iwanami shoten , publishers , cambridge univ . press , 1977 . . m. artin and b. mazur , ( 1977 ) , 87131 . w. barth and c. peters , , e. bombieri and d. mumford , , f. cossec and i. dolgachev , r. m. crew , e. dardanelli and b. van geemen , , i. dolgachev , i. dolgachev , i. dolgachev and c. liedtke , t. ekedahl and n. i. shepherd - barron , g. fano , , h. ito , , t. katsura , t. katsura and s. kond , to appear in pure appl . q. ; t. katsura and y. takeda , t. katsura and k. ueno , s. kond , , w. lang , , w. lang , , q. liu , d. lorenzini and m. raynaud , , s. mukai , , s. mukai and y. namikawa , , s. mukai and h. ohashi , , v. nikulin , a. n. rudakov and i. r. shafarevich , p. salomonsson , t. shioda , t. shioda , t. shioda , e. b. vinberg , , in

electronic raman scattering measurements have been performed on single crystals in the superconducting state . pure electronic raman spectra with no phonon structures hindering the analysis of the electronic continuum have been obtained . as a consequence , the spectra in the pure and symmetries are directly and reliably analyzed and the pure contribution can be easily identified . below the critical temperature two electronic structures at and are clearly seen . both are observed simultaneously in pure symmetry , the highest energy one being located at the energy of the maximum . these two maxima disappear at and do not soften significantly as the temperature is raised up to the low energy frequency dependence of the electronic response is strongly linear , for various excitation lines in the 476.5 to 647.1 nm range . such experimental data can not be reconciled with a pure symmetry . instead , they strongly advocate in favor of an anisotropic superconducting gap with two distinct gap maxima and of nodes existing outside the [ 110 ] and [ 1,,0 ] directions in * k*-space . we discuss in detail the simplest order parameter compatible with our experimental findings .

cond-mat9804060

since the last few years , identifying the symmetry of the pairing state has been the expected major step towards an understanding of high @xmath14 superconductivity . the controversy is not yet resolved : while there seems to be an agreement on the existence of nodes in the gap , some experiments appear to advocate in favor of a pure @xmath12 symmetry @xcite , whereas others seem to exhibit a significant s - wave contribution @xcite . with respect to this problem , inelastic light scattering has been shown very early to be a powerful tool because , besides probing the bulk ( in contrast with photoemission and tunneling ) , the polarization selection rules on the incident and scattered light make the spectra sensitive to the wave - vector of the electronic excitations @xcite . a theoretical approach of electronic raman scattering has been developed , in the case of @xmath12 symmetry . the emphasis is put on the low energy spectrum , which is expected in this case to exhibit well defined power laws , according to the symmetry channel , e.g. the polarization of the incoming and outgoing light . more precisely , the so - called @xmath8 and @xmath15 symmetry should display a linear dependence versus frequency , as these channels are sensitive to the nodes in the @xmath16 direction , whereas the @xmath9 symmetry probes mainly directions around the @xmath17 @xmath18 directions and should display a @xmath19 frequency dependence @xcite . such calculations omit other excitations such as phonons . the experimental major difficulty lies then in the fact that in most investigated compounds , e.g. @xmath20 @xmath21 ( y-123 ) , @xmath22 ( bi-2212 ) and @xmath23 ( tl-2201 ) , a large phonon contribution is superimposed to the electronic excitations in particular at low frequency and has therefore to be subtracted , which precludes an accurate determination of the shape of the electronic contribution in the superconducting state @xcite . despite this difficulty , in a number of reports , a satisfactory agreement is claimed to be found between the electronic response obtained after subtraction of the phonon lines , and the @xmath12 model @xcite in this work , we report _ pure _ electronic raman spectra in @xmath0 ( hg-1223 ) single crystals , belonging to the highest @xmath14 cuprate family . this provides a large spectral range below the superconducting gap frequency where the low energy electronic spectrum can be analyzed . moreover , within this range , no subtraction of the phonons is required to discuss the electronic response @xcite . indeed , as shown in our previous work [ 16 - 18 ] , the very low intensity of the raman phonon peaks for electric fields within the @xmath24 planes allows a reliable analysis of the electronic excitations in the superconducting state . the crystallographic structure of hg-1223 is tetragonal ( @xmath25 ) @xcite , which yields an unambiguous comparison with theoretical calculations based on tetragonal symmetry , without the complications due to the orthorhombic distortion as in y-123 @xcite . hg-1223 is therefore particularly well suited for the study of the electronic raman scattering . our most striking results are i ) the existence of two electronic maxima at @xmath26 and @xmath27 , both detected in @xmath28 symmetry . these two maxima disappear at @xmath14 . ii ) the low frequency behavior of the @xmath29 spectrum displays what we believe to be an intrinsic linear term , which is observed for all excitation energies used in this work ( 476.5 - 647.1 nm spectral range ) . such observations imply that the gap is strongly anisotropic with two distinct maxima and that nodes exist outside the [ 110 ] and [ 1,@xmath13,0 ] directions . we argue that our results , which are at odds with a pure @xmath12 gap symmetry , imply an order parameter with two maximum gap values and eight nodes . we briefly recall the standard theoretical framework for electronic raman scattering which we will use in the analysis of our experimental results @xcite . electronic raman scattering is a process of inelastic light scattering where an incident photon is absorbed by the crystal and a scattered one is emitted , with the simultaneous creation ( stokes ) or annihilation ( anti - stokes ) of an electronic excitation . here we will only have to deal with stokes processes . since two photons are coming into play , raman scattering is a second order process in the electromagnetic field . this second order effective interaction with electronic excitations comes both from a direct second order term in the interaction hamiltonian , and from a first order term treated up to second order in perturbation . as introduced by abrikosov and genkin @xcite , one may consider that both terms can be gathered in a single effective second order term in the hamiltonian which can be written as : @xmath30 where e is the electronic charge and m its bare mass . @xmath31 and @xmath32 are the vector potentials of the incoming laser and scattered light , and the bracket is for the proper matrix element over the photons states . the difference between the incident and the scattered photon frequencies is noted @xmath33 , and the difference between the photon momenta is @xmath34 . the operator @xmath35 is given by : @xmath36 it is quite similar to the standard density operator , the only difference being the @xmath37 for the scattering process , known as the raman vertex , which is given explicitely by : @xmath38 & & + { 1 \over \hbar m}\sum\nolimits\limits_{{n}_{m}}^ { } { \left\langle{{\rm n}_{f},\bf k\rm + \bf q\rm \left|{{e}^{-i{\bf k}_{\rm s}.\bf r}\rm { \bf e}_{\rm s}.\bf p}\right|{\rm n}_{m},\bf k\rm + { \bf k}_{l}}\right\rangle\left\langle{{\rm n}_{m},\bf k\rm + { \bf k}_{l}\rm \left|{{e}^{i{\bf k}_{\rm l}.\bf r}\rm { \bf e}_{\rm l}.\bf p}\right|{\rm n}_{i},\bf k}\right\rangle \over { \rm \varepsilon } _ { { n}_{i},\bf k}\rm -{\varepsilon } _ { { n}_{m},\bf k\rm + { \bf k}_{l}}\rm + { \omega } _ { l}+i\eta } + ( l\leftrightarrow s ) \label{eq3}\end{aligned}\ ] ] it is quite important that this raman vertex has a @xmath39-dependence , otherwise electronic raman scattering would essentially vanish , for symmetry reasons in @xmath29 and @xmath40 symmetry , and because of screening in @xmath28 symmetry . the interest of the reduced form in eq.(2 ) is that we have to deal only with electrons in a single band . in particular when the photon energies @xmath41 and @xmath42 are neglected in the denominator of eq.(3 ) , it can be shown that the raman vertex is identical to the contraction of the @xmath43 -dependent inverse effective mass tensor of this band with the polarization vectors of the incident and the scattered light . this latter reduced formula is sometimes used to make explicit calculations of the raman vertex from an assumed knowledge of the band structure . in general a single band is assumed at the fermi level . it is not obvious at all that this is reasonable in high @xmath44 compounds , although one may argue that the raman process is dominated by a single band ( this is strongly suggested by the similarity of electronic raman scattering between various high-@xmath44 superconductors ) . moreover the effective mass formula assumes that the energy of all the states belonging to other bands are far enough from the fermi level in order to neglect the photon energies in the denominator . while this approximation is reasonable for simple metals ( one band is near the fermi level and the other bands are far away ) , it appears much more doubtful in high-@xmath44 superconductors . in particular it is known that the absorption is important at the incident frequency , linked to interband transitions . this invalidates the approximation of neglecting the photon energies and implies instead the existence of resonant terms . this is actually coherent with what we observe experimentally , since the electronic raman scattering varies in an important way as a function of the incident photon energy ( as we will see in section v.2 ) . moreover , in order to deal consistently with these quasi - resonant terms , we have to take into account the finite lifetime of the highly excited intermediate states . we expect , on the basis of optical data , that this relaxation rate is large , of order of a few tenths of ev . all this makes it clear that the effective mass approximation is not valid in our experimental situation . therefore we will rather consider that the raman vertex is actually an effective quantity which depends not only on the wavevector @xmath45 , but also on the incident and scattered photon energies . for simplicity we will not indicate explicitly this dependence in the following , but we will consider it in the analysis of our data . the differential cross section is then obtained in terms of the generalized susceptibility @xmath46 corresponding to the generalized density eq.(2 ) . it is related to the imaginary part @xmath47 of this susceptibility by : @xmath48 where @xmath49 . the standard procedure is then to calculate @xmath50 in the superconducting state within the single bubble approximation . moreover it is appropriate to neglect the momentum transfer @xmath51 because it is small compared to the inverse coherence length and almost perpendicular to the @xmath24 planes . one finds then within bcs theory and in the limit of low temperature @xmath52 , which is relevant for our experiments @xcite : @xmath53 @xmath54 is the density of states for both spin orientations at the fermi level , and the brackets indicate an average over the fermi surface . @xmath55 stands for the superconducting , @xmath45-dependent gap . for our purpose eq.(5 ) has two related major drawbacks . first it leads to a vanishing raman scattering in the limit of large frequency , where we should recover the physical properties of the normal state . it also leads in general to a singular behavior in the vicinity of the thresholds for creation of a pair of excitations , which occurs at twice the maxima of the gap @xmath55 . this is in contrast with our experimental results . as we will see , we obtain smooth spectral shapes , and , in the high frequency regime , we recover a strong raman scattering , which is well known in the normal state . these two discrepancies between experiment and theory can be ascribed to the fact that no collisional damping of any kind is included in the process of deriving eq.(5 ) . nevertheless this is not so important here since our purpose is much more to make a qualitative analysis of our results than to provide a full quantitative account of experiment . however it is clear that inclusion of lifetime effects in the theory is quite desirable to provide a more accurate account of experimental results . work in this direction is in progress and will be reported elsewhere . finally there is a physical effect , essential for a proper analysis of raman scattering , which we have not included up to now . indeed if we had a constant raman vertex , we would merely look for a density - density response . as pointed out already by wolff @xcite , perfect screening in metals at long wavelength implies that in this case raman scattering would be essentially zero . therefore it is necessary to include screening by the coulomb interaction in order to obtain a proper physical result . in the limit of perfect screening , the result is : @xmath56 \label{eq6}\end{aligned}\ ] ] where @xmath57 is defined by : @xmath58 here @xmath59 and @xmath60 are general vertices , 1 being the proper vertex for a true density interaction ( this comes in eq.(6 ) because only this true density is screened by coulomb interaction ) . in @xmath9 and @xmath15 symmetry , @xmath61 is zero for symmetry reasons , the last term in eq.(6 ) drops out and we merely recover eq.(5 ) . in other words , the @xmath9 and @xmath15 symmetry are not screened because the average effective density is zero in these cases . on the other hand this is not so for the @xmath8 symmetry , and screening has to be properly taken into account by making use of eq.(6 ) . nevertheless one sees easily from this equation that any * k*-independent contribution to the raman vertex drops out of the result . physically such a contribution is irrelevant because it is fully screened out . the measurements were carried on two single crystals labelled a and b. these crystals were grown by a single step synthesis as previously described for hg-1223 @xcite . they are parallelepipeds with typical 0.5 x 0.5 @xmath62 cross section and thickness 0.3 mm . the [ 100 ] crystallographic direction lies at @xmath63 of the edge of the square and the [ 001 ] direction is normal to the surface . they were characterized by x - ray diffraction and wave length dispersive spectrometry @xcite . identification of the proper phase requires several experimental characterization techniques , and one may be easily mislead in particular by the critical temperature . the presence of intergrowth phases , corresponding to different stacking numbers of the @xmath64 planes along the _ c_-axis has been established @xcite . this was detected through : i ) the presence of diffusion streaks in the oscillation photographs , ii ) the lattice parameter values obtained from an enraf - nonius cad-4 diffractometer at 293 k , iii ) the chemical analysis carried out using a camebax sx50 electron probe . the electron beam probes 1 @xmath65 of material at each measurement ( the probed depth is approximatively 10 times the penetration depth of the excitation light ) . 20 measurements were performed for the a crystal and 40 for the b crystal . the a crystal exhibits an overall composition of @xmath66 the under - stoechiometry in ca and cu can be assigned to the existence of hg-1212 intergrowth phase . the b crystal has the following composition : @xmath67 which reveals a mixture of hg-1223 and hg-1234 . a detailed x - ray analysis has been performed on the a crystal . the lattice parameters deduced from the diffractometer are a=3.844(2)@xmath68 and c=15.72(3)@xmath68 , hence the c parameter of this crystal is slightly smaller than the perfectly stoechiometric hg-1223 ( c=15.831 ) and definitely much larger than c=12.71@xmath69 for hg-1212 . therefore , according to the composition and the lattice parameters and in agreement with the chemical probe , we may consider that we are dealing essentially with the hg-1223 compound in both a and b crystals . the crystals are slightly under - doped , with @xmath70 less than 0.1 . @xmath70 refers to an excess of oxygen atoms randomly arranged within the hg planes . under - doping is further confirmed by dc magnetization measurements displayed in fig.1 . although there is an onset of diamagnetism around 133 k , it appears that the main transition occurs at @xmath71 optimal doping increases @xmath72 hence introduces disorder in the hg basal plane . therefore an under - doped system is closer to four - fold symmetry than an optimally doped one and is even more appropriate to symmetry considerations . finally we wish to point out that the extensive study of the crystal structure described above had not been completed for all the samples ( coming from a single batch ) by the time of our earlier work @xcite . one crystal was clearly identified as hg-1212 , and we eventually published the better quality spectra from another one , unaware of the fact , which is established by now , that two crystals coming from the same batch may be either hg-1212 or hg-1223 . the latter sample was subsequently analyzed and turned out to be , as explained above , hg-1223 . it is important to quote that actually the raman spectra of the two samples studied in @xcite were similar . this will be shown in the experimental results . raman measurements were performed with a double monochromator using a single channel detection and the @xmath73 and @xmath74 laser lines . the spectral resolution was set at 3 @xmath75 the crystals were mounted in vacuum @xmath76 mbar ) on the cold finger of a liquid helium flow cryostat . the temperature was controlled by a si diode located inside the cold finger . the incident laser spot is less than 100 @xmath77 in diameter and the intensity onto the crystal surface was kept below @xmath78 in order to avoid heating . at our lowest temperature , the difference in temperature between the cooled face and the illuminated face of the crystals was estimated to be @xmath79 1 k , resulting in a temperature of 13 k inside the scattering volume , instead of a nominal 12 k temperature . the incidence angle was @xmath80 . the scattered light was collected along the normal to the crystal surface . the polarization is denoted in the usual way : x[100 ] ( a axis ) , y[010 ] , z[001 ] ( c axis ) , x[110 ] , y[1@xmath130 ] . in order to compare our experimental data with the theoretical calculations [ 6,22 ] , the pure @xmath15 ( xy ) and @xmath9 ( xy ) symmetries are convenient and well suited for the investigation of a @xmath12 order parameter . to make this point clear , we recall briefly the symmetry properties : in the @xmath81 case , the raman vertex @xmath82 is zero by symmetry for @xmath83 , so the electronic scattering is insensitive to the gap structure around @xmath63 ; in contrast , the @xmath84 and @xmath85 regions do contribute , giving weight to the gap @xmath86 in these directions . conversely , in the @xmath2 case , the raman vertex @xmath82 is zero by symmetry for @xmath84 or @xmath85 or and non zero elsewhere , hence provides weight in the @xmath83 direction . in our experimental procedure , we start by measuring the @xmath1 channel . the incident angle being not zero , the crystal is then rotated by @xmath63 in order to get the @xmath2 one , the polarizer and the analyzer remaining unchanged ( see fig . 2-a and 2-b ) . the @xmath87 and @xmath88 symmetries are obtained from the ( xx ) and ( xx ) polarizations respectively . in order to probe the same area for all symmetries , the impact of the laser beam onto the crystal surface was precisely located before turning the crystal . finally , a very weak diffusive spot was carefully selected on each crystal in order to minimize the amount of spurious elastic scattering . another possibility which guarantees that the location of the laser spot is unchanged for @xmath1 and @xmath2 symmetries , is to rotate the polarization instead of the crystal ( see fig . 2-c ) , when starting from pure @xmath9 symmetry . after a @xmath63 rotation , the incident electric field no longer lies within the xy plane , which induces symmetry mixing for @xmath89 we have estimated the amount of mixing with @xmath87 ( xx ) and @xmath90 ( xz ) , taking into account the high refractive index @xmath91 which brings the propagation of the beam inside the crystal closer to normal incidence ; we found @xmath92 admixture of @xmath87 ( xx ) and @xmath93 of @xmath90 ( xz ) . we shall denote this mixed symmetry as @xmath94 . similarly , we can measure the @xmath95 channel ( bringing the analyzer parallel to the polarizer ) , which comprises besides the @xmath87 spectrum , @xmath92 of @xmath2 ( xy ) and @xmath93 of @xmath90 ( xz ) . to get a complete set of reliable data , we have performed raman measurements in both configurations : i ) rotation of the crystal without changing the polarization , so as to get pure @xmath1 and @xmath2 symmetries and ii ) rotation of the polarization without moving the crystal , in order to work with the same spot . various spots have been probed on each crystal . all the resulting data are consistent one with the other . the hg-1223 raw spectra at 13 k , in the three symmetries obtained with the 514.52 nm laser line by rotating the crystal are displayed in fig . these pure @xmath96 @xmath9 and @xmath15 spectra were previously reported in ref @xcite . figure 3-b shows the spectra obtained by rotating the polarization , yielding the @xmath95 and @xmath94 mixed symmetries . the two procedures ( performed in the same a crystal ) yield similar results for each symmetry although there is some extra scattered intensity in the mixed @xmath95 symmetry at high energy . since several spots have been selected , this crossed procedure confirms the reliability of the data . note that due to the high surface quality of the hg-1223 crystal , the residual elastic scattering is confined below 50 @xmath97 and is quite weak . in crossed polarization measurements @xmath98 and @xmath99 this scattering is just not seen , hence no correction has been performed to take it into account . the @xmath88 spectrum obtained from the b crystal , is displayed in fig.4 . for the sake of the comparison , we show in fig.5 preliminary raw spectra from hg-1212 . the surface quality being not so good for this sample , we see clearly , especially in the @xmath8 symmetry , the rayleigh scattering . nevertheless , these results are suggestive of a very strong similarity of the raman response in both compounds . let us focus on fig . 3 and 4 . remarkably and in sharp contrast with y-123 and bi-2212 , we get a large spectral range below the energy gap where no raman active phonon mode hinders the @xmath1 and @xmath2 electronic scattering analysis . weak and narrow peaks appear at 240 , 388 , 480 and 580 @xmath100 in the @xmath87 and @xmath88 but their energy location and their intensity are such that no correction is actually needed to discuss the electronic spectra . the 580 @xmath100 peak shows a phonon asymmetric lineshape with an antiresonance on the high energy side ( see fig.3-a ) . this is characteristic of the interference between an electronic continuum and a single phonon state . this fano lineshape has already been detected in y-123 and bi-2212 for the @xmath1 normal mode in the vicinity of the gap structure maximum [ 7,10,23 ] . the 640 @xmath100 peak detected in fig.4 is assigned to some trace of @xmath101 impurity @xcite . overall , the spectra do not require the delicate handling of phonon subtraction and we are in a position to turn immediately to the analysis of the data . two well marked maxima are observed : one is located around 530 @xmath100 in the @xmath87 ( fig.3-a ) , @xmath102 ( fig.4 ) symmetries and the @xmath95 ( fig.3-b ) mixed symmetry ; the other is observed at 800 @xmath100 in the @xmath9 symmetry . in contrast , no clear maximum appears in the @xmath2 and the mixed @xmath94 symmetries , for which actually the scattered intensity levels out smoothly around 530 @xmath75 in the @xmath87 spectrum ( fig.3-a ) , we observe a weak shoulder near 800 @xmath97 not seen in the mixed @xmath103 symmetry ( fig.3-b ) . in the @xmath88 spectrum ( fig.4),beyond the well defined maximum at 530 @xmath97 a broad maximum is clearly visible at 800 @xmath75 there is in the latter case no possible contamination by the @xmath9 channel , whereas the @xmath15 channel does not display any feature above 500 @xmath75 all this strongly suggests that the 800 @xmath100 maximum is an intrinsic electronic feature in the @xmath8 channel . this important point will be discussed later . the observation of two electronic maxima below @xmath10 is currently mentioned for two distinct channels [ 7 - 9 , 13 ] . it is a common feature for numerous high-@xmath10 cuprates studied by raman spectroscopy near the optimally doped regime . the lower energy peak @xmath104 is usually observed in @xmath87 symmetries and the higher energy one @xmath105 is detected in pure @xmath1 symmetry . the energies of these two peaks for various cuprates are listed in table i. two well defined energy scales appear : ( 5 @xmath106 and @xmath107 remarkably enough , their ratio is approximatively 1.5 , whatever the compound . what is new in our study is that , by looking at a less commonly studied symmetry channel , we find that these two energy scales are present within one single channel , namely @xmath108 finally , the temperature dependence of the imaginary parts of the electronic response for the @xmath109 and @xmath9 symmetry as a function of the temperature from 13 k to 150 k is shown in fig.6-a and 6-b respectively . the imaginary part of the electronic response function @xmath110 $ ] is derived by subtracting the photomultiplier dark current , the rayleigh scattering intensity , when present , and by correcting for the response of the diffraction grating and for the bose - einstein factor @xmath111^{-1}. $ ] in both symmetries , the low energy scattered intensity increases , approaching that of the normal state as the temperature is raised . the maxima at 540 @xmath112 ( for @xmath113 @xmath114 and 800 @xmath100 ( for @xmath115 gradually collapse and disappear above @xmath11 these two maxima do not shift significantly ( at most 10% ) towards low energies as the temperature increases . the latter result is fairly general in raman and tunneling experiments . we wish now to concentrate on the low energy part of the spectra . we collect in fig . 7 the imaginary parts of the electronic response functions @xmath116 $ ] obtained from the raw data after performing the same corrections described just above , associated to the @xmath117 @xmath118 and @xmath2 symmetries , already reported in @xcite . the dotted lines represent @xmath119 , $ ] after smoothing the spectra ( for clarity ) . as could be inferred from the raw spectra , the three response functions seem to extrapolate to zero at zero frequency , within our experimental accuracy . we recall @xcite the main observations which emerge from fig . 7 , namely : \i ) the @xmath1 response function exhibits a clear decrease of the electronic scattering at low energy in the superconducting state with respect to the normal state . in contrast , the @xmath87 and @xmath2 response functions decrease barely at low energy . actually , there is very little change between the normal and the superconducting state in the @xmath15 channel . the change is instead much larger in the @xmath8 channel but occurs around the peak energy . \ii ) the @xmath1 spectrum exhibits a steady quasi - linear increase with energy , in contrast with bi-2212 spectra where , after subtracting the phonon contribution , the spectrum below the peak energy exhibits a strong upward curvature @xcite . \iii ) the electronic response functions in the normal and superconducting state merge in all three symmetries around 1000 @xmath75 to investigate further the low energy behavior of the @xmath9 spectrum in the superconducting state , we have performed raman measurements at t=13 k with various excitations lines . the imaginary parts of the electronic response functions in the @xmath1 symmetry are displayed in fig . 8-a for five different excitation lines , namely 647.1 , 568.2 , 514.52 , 488.0 and 476.5 nm . for clarity , the spectra have been normalized so as to achieve the same intensity around the 800 @xmath100 maximum and shifted vertically . the spectra reported here have been taken from sample b. there is a clear change in the overall shape of electronic response as the excitation energy decreases . a simple visual inspection of the spectra suggests that they can be classified in two categories : for the 647.1 and 568.2 nm lines , the linear part appears to extend up to higher energy than for the 514.52 , 488.0 and 476.5 nm excitation lines . also the maximum for the two former ones seems to develop at a somewhat lower energy . in order to confirm the occurrence of two sets of data displaying a similar spectral dependence , we have firstly checked quantitatively that within the two sets , the spectra are indeed proportional . then , we have used this calculation to normalize the spectra for the two lower energy excitation lines ( 647.1 and 568.2 nm ) on one hand , and for the three higher energy excitation lines ( 514.52 , 488.0 and 476.5 nm ) on the other hand . the result is shown in fig.8-b . indeed , within each set , the higher excitation energy spectra and the lower ones fall exactly one on top of the other , within our experimental resolution . as mentioned in section ii , a source of complication of the analysis may arise from the dependence of the raman vertex on @xmath120 and @xmath121 the dependence on @xmath120 is revealed in our experiments as a change in the strength of the raman scattering with the laser frequency . the dependence on @xmath122 translates into a dependence on the raman shift @xmath123 which could affect the validity of the low frequency analysis . let us first comment on the possible origin of the @xmath124 dependence . it is interesting to note that in the spectral range of interest ( 15000 - 22000 @xmath125 various cuprates ( ybco , lasrcuo , bisrycuo ) exhibit two bands , located typically around 12000 - 14000 @xmath100 ( the so - called charge - transfer band ) , and 22000 @xmath100 [ 25 - 28 ] . since , to the best of our knowledge , similar spectra are not yet available for the hg compounds , we assume that the absorption bands lie in the same range as for other cuprates . the three high energy excitation lines are clearly located in the absorption range of the high energy band . the 568.2 nm line ( 17600 @xmath126 actually lies in - between the two bands , still closer to the charge transfer band , and the raman spectrum exhibits a strong similarity with the 647.1 nm ( 15450 @xmath126 one . the latter is definitely in the vicinity of the charge transfer band . therefore , the change of the raman spectra with the excitation energy bears an actual consistency with the absorption bands in the visible range , supporting the idea that resonant scatttering is effective . it is not very surprising that we find within our experimental resolution , that the data can be separated into two sets , since two bands seem to be involved . the question which is raised once this is recognized is which frequency range is most affected by the change of the raman vertex with energy . when fitting the absorption bands to lorentz oscillators @xcite , we found typical widths of 1ev . a typical energy scale for the variation of the raman vertex with @xmath127 would then be 0.5 - 1ev . a first order expansion of the raman vertex ( we omit the @xmath43 dependence of @xmath128 for simplicity ) with respect to the laser frequency writes : @xmath129 therefore in the limit @xmath130 , which is of interest here , the raman vertex is independent of @xmath131 which implies that the low frequency behavior should not depend on the laser excitation energy . as @xmath132 increases , the frequency dependence of @xmath133 may not be neglected . using this line of reasoning , we have compared the low frequency dependence of the two sets of spectra obtained by varying the laser excitation . a simple way to do it is to select from each set the spectrum exhibiting the best signal to noise ratio , namely the ones associated to the 568.2 and the 514.5 nm lines . 9 shows that indeed , the low frequency dependence ( up to 500 @xmath126 is the same , and that the spectra gradually separate as the frequency increases . we assign this behavior to the change of the raman vertex with frequency , as explained above . therefore , we shall consider henceforward that there is no change of the low frequency behavior when changing the excitation energy . note incidentally the strong linear term of the @xmath9 spectrum , best seen in fig.9 . we may now turn to the comparison of our data with existing theories . we start with the @xmath12 model , as computed by devereaux et al . , which was found to describe well the bi-2212 results @xcite . in a pure @xmath12 model , symmetry considerations imply that the @xmath1 spectrum is insensitive to the nodes at @xmath63 ( hence the @xmath19 dependence ) and displays a maximum at @xmath134 , whereas the @xmath2 symmetry exhibits a linear low frequency dependence ( because it probes the nodes ) and a smeared gap @xcite . for the same reason the @xmath3 symmetry has also a linear low @xmath132 behaviour . we find a fair agreement of this model with our data for the linear dependence of the low energy part of the @xmath2 symmetry ( see fig . our linear @xmath3 spectrum agrees also quite well with the prediction of the @xmath12 model at low frequency , but this is not the case for the higher frequency range [ fig.12-a ] . indeed we note that the @xmath12 model has only a single maximum which gives rise to a single peak , both in @xmath1 and @xmath3 symmetry . naively one expects these two peaks to be essentially at the same frequency . this is already in contrast with experiments on bi-2212 where the @xmath3 peak is found at much lower frequency than the @xmath1 one . nevertheless if one takes into account the strong screening in the @xmath3 symmetry together with a sizeable smearing , one may obtain a large softening of the @xmath3 peak and reconcile theory with experiment . however , we find it eventually difficult in our calculations to get the necessary softening without obtaining simultaneously an unduly wide maximum . we give an example of such a calculation in fig.12-a . anyway our present observation of two distinct peaks in the same @xmath8 channel deduced from the @xmath109 symmetry ( see fig . 4 ) can not be reconciled with a pure @xmath12 model . we note that the second peak , essentially at the same frequency as the @xmath9 one , had not been seen in our earlier experiments @xcite because only the @xmath135 symmetry had been measured . finally we consider the @xmath1 symmetry which , together with the @xmath2 symmetry , is easier to analyze because it is not affected by screening , in contrast with the @xmath3 symmetry . most strikingly the low frequency behavior of the theoretical @xmath12 @xmath1 spectrum is incompatible with our experimental results . in the @xmath12 model , the response function for @xmath1 should increase as @xmath19 @xcite and it is clear , by inspection of fig . 9 and 11 , that our results are essentially linear in frequency . if we analyze our data more quantitatively , we have to recall that such a power law behavior is an appropriate description only at low frequency , and therefore the pending question is the choice of the proper energy scale when trying to fit the results . fitting the spectrum from fig.11 to a power law @xmath136 up to 300 @xmath100 yields @xmath137 inconsistent with an @xmath19 dependence @xcite . however , the choice of the energy range for fitting may be disputable , and a second fit to @xmath138 was attempted in order to provide a more accurate estimate of the relative weight w of the linear part with respect to the cubic part . w was computed over various energy ranges , e.g. up to 300 , 400 , 500 , 600 @xmath97 still using the spectra shown in fig.11 . actually , referring to fig.9 , the low energy range of interest may not exceed 500 @xmath75 we found @xmath139 , meaning that a dominant linear component must be taken into account , inconsistent with a pure @xmath12 model . @xmath9 being linear at low frequency appears to agree with the tl-2201 results , at least at high enough excitation energy ( @xmath140 2.18 ev , or @xmath141 5734 ) @xcite . it has been argued that impurities are responsible for the observed low frequency density of states in the @xmath1 channel . devereaux @xcite has indeed shown that , in the unitary limit , for a @xmath12 gap , impurities induce for @xmath1 a linear rise of the electronic scattering at low frequency , crossing over to the @xmath19 dependence at higher frequency . the cross - over energy @xmath142 found in this case is typically the width @xmath133 of the band of states bound to impurities @xcite . for unitary scatterers this width is of order @xmath143 @xmath144 where @xmath145 @xmath146 @xmath147 @xmath148 with @xmath147 the impurity concentration and @xmath149 the maximum of the d - wave gap . in the born limit it is in general much smaller @xmath150 @xmath151 @xmath152 @xmath153 where @xmath152 is the normal state relaxation time , and impurities bring no qualitative change in the density of states at low energy@xcite except for large relaxation rates . therefore the unitary case is the most favorable one for an important effect of impurities on raman scattering . however this is a very specific case , and experimental support is still lacking . note that in the case of a small or intermediate relaxation rate , one goes rapidly out of the unitary regime as soon as the scattering phase shift goes away from @xmath154 . only for a large relaxation rate @xmath155 does the unitary regime extend over a fair range of scattering phase shifts . in the absence of experimental evidence , a random distribution of scattering phase shifts would seem a reasonable assumption . it is worth recalling that this assumption of unitary scatterers was raised [ 32,33 ] in order to reconcile the robustness of the critical temperature against impurity scattering , which is unexpected in a d - wave superconductor , with the much less robust temperature dependence of the london penetration depth , whose linearity is thought to be affected by scattering . nevertheless , due to the difficulty of measuring the absolute value of the penetration depth , there is no strong experimental evidence for this assumption . actually , published data are somewhat contradictory [ 34,35 ] . let us nevertheless consider this unitary limit . our experimental results can be compared reasonably well with devereaux s calculation @xcite , taking @xmath156 this means a scattering rate @xmath145 of order of the gap @xmath149 itself . such a high scattering rate should strongly reduce the critical temperature or even destroy completely superconductivity . this is in marked contradiction with @xmath157 of our single crystals which does not differ in an appreciable way from the critical temperature of supposedly rather clean crystals . finally we remark that the fairly sharp structure that we observe in @xmath8 is difficult to reconcile with a large scattering rate . all together this makes rather unlikely the explanation of the linear behavior of the low frequency @xmath9 by an impurity effect . accordingly we consider that our results are representative of pure crystals . then an inescapable consequence of the observed linear frequency dependence is the occurrence of a finite density of states exhibiting a similar energy dependence , which is most naturally assigned to nodes in the gap . since these nodes are probed in both @xmath1 and @xmath2 symmetries , they can not be located only in the @xmath63 direction . we are thus left with the conclusion that nodes exist outside the @xmath63 direction . it is worth recalling that although raman scattering probes the electronic excitations along various * k * directions , it can not give accurately the location of the nodes @xcite . we have therefore to explore this point more quantitatively . from the low energy part of our spectra , we know there are nodes outside the @xmath63 direction . it can readily be seen that a small shift of the nodes away from @xmath63 is not enough to account for experiment . indeed if we use a toy model @xmath158 @xmath159 obtained by artificially rotating the @xmath160 order parameter by an angle @xmath161 we find that a rotation by @xmath162 is barely noticeable in the spectra : the linear rise at low @xmath132 produced in @xmath1 by this rotation is quite small and the spectrum is essentially unchanged within experimental accuracy . having at this stage exhausted all information contained in the low energy dependence , in order to build up a specific model more appropriate than @xmath12 , we have to rely now on the high energy part of our spectra . while the @xmath15 symmetry offers only a hint of a maximum around @xmath163 the @xmath9 symmetry and the @xmath8 symmetry are much more informative . indeed we have the occurrence of two maxima at 500 and @xmath164 seen simultaneously in pure @xmath8 symmetry . the last one is seen separately in @xmath9 in a very clear way while the former is compatible with the structure seen in the @xmath15 symmetry . together these observations strongly suggest two different energy amplitudes for the gap . we assume that the gap has the symmetry of one of the one - dimensional representation of the tetragonal group , namely @xmath165 @xmath166 @xmath15 or @xmath167 we need only to consider the range of the fermi surface where the wavevector * k * has an angle @xmath168 with the x axis between 0 and @xmath169 the rest can be deduced by symmetry operations . our model must accommodate a main maximum m , a secondary maximum m - for the absolute value of the order parameter - and at least one node n ( we assume for simplicity that there is only a single secondary maximum and no secondary minimum since there is no experimental evidence for them ) . this does not match with any of the simplest representative for the order parameter , namely @xmath170 @xmath171 symmetry ) , @xmath172 @xmath98 symmetry ) , @xmath173 @xmath174 @xmath175 symmetry ) , or @xmath173 @xmath174 @xmath172 @xmath176 symmetry ) . for instance , as pointed out by leggett , the simple @xmath177 symmetry is compatible with the low energy part of our spectra ; but it does not provide a proper account of the high energy part since it would give only peaks at a single frequency corresponding to the maximum of the gap . therefore we have to deal with a more complicated order parameter . from symmetry we must find at @xmath178 either m or m or n , and similarly at @xmath63 ( we eliminate the possibility of a discontinuous order parameter ) . since the main maximum m is seen in @xmath9 symmetry and not in @xmath15 symmetry , we have to put it at @xmath178 . similarly the secondary maximum m being inferred in @xmath15 symmetry but not seen in @xmath9 symmetry implies a location at @xmath63 . then the only place left for nodes is somewhere between @xmath178 and @xmath63 . since we have assumed a single secondary maximum , we can put only one node there . this leads to an order parameter which has over the whole fermi surface 4 ( equivalent ) m , 4 m and 8 nodes . since this order parameter is continuous and is nonzero at @xmath178 and @xmath63 , it has the @xmath8 symmetry . therefore we come to the conclusion that our raman scattering data imply an order parameter which does not break spontaneously the crystal tetragonal symmetry . however it is clear that any slight shift of m and m away from their symmetric locations will essentially not be felt in the resulting spectra . for example since our @xmath15 is quite featureless , we may worry that the presence of the main maximum m is washed out in it . so m could be away from @xmath178 . since m is at @xmath63 , we can only put a node at @xmath178 . but there is a node away from @xmath178 because @xmath15 is linear at low frequency . the only location for this node is between m and @xmath63 . we end up with an order parameter which has the @xmath15 @xmath179 symmetry , with 8 m , 4 m and 12 nodes over the whole fermi surface . similarly we might worry that a small contribution of m to the @xmath9 spectrum is not seen in our experimental results . however this seems almost incompatible with our data , except if m is quite near @xmath63 , but this would lead to an almost discontinuous order parameter since we have to put a node at @xmath63 . in this way we would have a @xmath9 symmetry order parameter , with 4 m , 8 m and 12 nodes on the fermi surface . finally we could also think of putting nodes at @xmath178 and @xmath63 , with m and m slightly away from these locations . we would need to put a node between them . hence we would have a @xmath177 order parameter , with 8 m , 8 m and 16 nodes . again this is hardly compatible with our @xmath9 data . in conclusion the @xmath8 order parameter is at the same time the simplest and the most compatible with our data . in the following we will concentrate on this symmetry and investigate it more quantitatively . a simple one - parameter model having the @xmath8 symmetry and satisfying the above requirements is @xmath180 @xmath181 @xmath182 $ ] where s = ( @xmath183 - @xmath184 ) / @xmath185 and @xmath149 = ( @xmath183 + @xmath186 ) / 2 . the gap maximum @xmath183 is obtained for @xmath168 = 0 while the secondary maximum @xmath187 occurs for @xmath168 = @xmath188 . the node lies at @xmath189 arccos(-s ) and in this simple model its location is linked to the ratio @xmath186 @xmath190 . naturally by introducing other parameters we can decouple these two quantities , as we will consider briefly below . since we make the choice s = 0.27 according to experiment , we find obviously the peaks for @xmath29 and @xmath40 in agreement with our experimental results ( fig . 10 and 11 ) . we calculate the raman spectra for @xmath29 and @xmath40 symmetry by taking for the raman vertices the simplest form compatible with symmetry , namely @xmath191 , @xmath192 and a density of states independent of @xmath193 naturally the bare result for these spectra display a logarithmic singularity at their respective maximum and we have to invoke some broadening , most likely due to lifetime effects , in order to obtain theoretical results compatible with experiment . in this paper we include this broadening at the most phenomenological level by merely convoluting the bare result with a broadening function . we take a simple normalized lorentzian for this function . however we expect lifetime effects to be quite small at low energy and to grow progressively at higher energy . we account for this by choosing the width @xmath194 of our lorentzian proportional to the frequency at which we are looking : @xmath194 = a @xmath132 . explicitly our final spectrum @xmath195 is obtained from the bare spectrum @xmath196 by @xmath195 = @xmath197 d@xmath198 @xmath199 @xmath200 @xmath201 @xmath202 @xmath203 where @xmath199 @xmath200 @xmath201 @xmath202 = ( 1/ @xmath204 @xmath194 / ( ( @xmath132 @xmath200 @xmath205 @xmath206 . here the integration goes from @xmath207 to @xmath208 and @xmath196 is extended by symmetry @xmath209 = @xmath200 @xmath196 in the whole frequency range . this broadening method insures f(0 ) = 0 , whatever @xmath210 in fig.10 and 11 we have taken our broadening parameter a = 0.15 . note that this parameter is our only way to adjust the peak height with respect to the low energy part of the spectrum . hence the fact that we find a good agreement in @xmath9 with experiment with a reasonably small broadening is satisfactory ( fig.11 ) . this is directly due to the fact that our node location @xmath211 = @xmath212 is far enough from @xmath63 , otherwise we would have trouble obtaining a strong enough low energy spectrum for @xmath166 as we have seen in our toy model . as expected , the part of the experimental spectrum above the peak frequency is not properly described by our model , inasmuch as we go toward higher energy . this is merely due to the fact that our theoretical result goes to zero for very high energies , in contrast with experiment . this discrepancy would be cured , at least partially , by taking proper account of lifetime effects , but this is beyond the scope of the present paper . in particular we ascribe the moderate agreement between theory and experiment in @xmath15 symmetry to strong lifetime effects in this channel ( fig.10 ) . we consider now in detail the @xmath8 symmetry . in our theoretical analysis , we take screening fully into account in the standard way . this feature makes the interpretation of the @xmath8 spectrum less easy than the @xmath9 or @xmath15 spectrum , because the result is not systematically related in an intuitive way to the shape @xmath213 of the order parameter . the most obvious manifestation of this is the fact that a constant @xmath8 raman vertex gives a zero contribution because of full screening . therefore we have to look for the next term in a fourier expansion in order to obtain a nonzero result . this term is proportional to @xmath214 however it goes to zero for @xmath168 @xmath215 that is very near the location of the node in our specific model . this implies that the @xmath8 electronic response will be quite small at small energy , in contradiction with experiment , which shows a strong linear dependence . for a sizeable @xmath8 response at low energy , we need the zero of the @xmath8 raman vertex and the node of the gap to be clearly apart . since we have already chosen our order parameter , hence the node location , we achieve this by considering a more complicated raman vertex , e.g. including the next order fourier component : we take @xmath216 as seen in fig.12-b , we obtain a reasonably good agreement with experiment for a = 0.05 and a small smearing a = 0.03 . we note that we are led to use different smearing for the different raman symmetries , in contrast to what has been done in @xcite . this does not make a problem : the smearing is meant to account for lifetime effects at a phenomenological level . hence when dealing with different channels , there is no reason to have the same effective scattering . we obtain a sizeable contribution at low energy , at @xmath186 and at @xmath183 : this is clearly related to the fact that the raman vertex is not small for @xmath168 = 0 , for @xmath168 @xmath217 and at the node @xmath168 @xmath218 @xmath219 note that , while the size of the low frequency part is directly related to the strength of the raman vertex around the node , the same is not true for the strength of the peaks at @xmath186 and @xmath220 in particular one may obtain a single broad bump between @xmath186 and @xmath183 instead of two peaks . this is naturally due to the complex effect of the screening term . the small value a=0.05 , which implies that the @xmath221 contribution is negligible in the raman vertex compared to the @xmath222 contribution , comes from the strong constraints that the various characteristics of our @xmath28 spectrum put on the order parameter . we had already been led to consider a similar vertex in ref.@xcite to account for a @xmath8 spectrum which did not display a maximum at @xmath183 . while this is not expected beforehand , this does not imply anything pathological about the raman vertex we use and such a situation may very well occur accidentally . as we mentioned we are merely led to this case by our request of a sizeable vertex at 0 , @xmath223 /4 and at the node . clearly we can achieve this goal in other ways if we allow more flexibility in our order parameter . we demonstrate this by considering the following order parameter : @xmath213 @xmath146 @xmath183 for 0 @xmath224 @xmath168 @xmath224 @xmath225 /s , @xmath213 @xmath146 s ( @xmath226 ) for @xmath225 /s @xmath224 @xmath168 @xmath224 @xmath211 + @xmath186 /s , @xmath213 @xmath146 @xmath200 @xmath186 for @xmath211 + @xmath186 /s @xmath224 @xmath168 @xmath224 @xmath188 . its shape is similar to the simple one that we have used above , but in this way we can choose more freely the location of the node @xmath211 of @xmath213 and its slope s. on the other hand we restrict ourselves to the simplest raman vertex @xmath227 . we find a reasonably good agreement with experiment ( see fig.12-b ) for @xmath211 = @xmath228 and s = 3.6 . this shows that we can account for the @xmath8 spectrum by the simplest raman vertex together with an order parameter having two maxima and a node . however the bare @xmath9 spectrum provided by this latter order parameter has a very strong peak at @xmath183 ( inasmuch as we have taken a simple flat maximum model ) and a quite small low energy part , due to the node proximity of @xmath63 . this can not be reconciled with experiment by any reasonable smearing . again the strength of the linear low energy part of the @xmath9 spectrum shows that the node is rather far away from @xmath63 . note however that we could cure somewhat this difficulty by assuming a more complicated @xmath9 raman vertex , which would have a small strength for @xmath168 = 0 ( in addition of being zero by symmetry for @xmath168 = @xmath188 ) . in conclusion we believe that it is possible to account for experiment for all symmetries by an order parameter having the qualitative features that we have considered , together with raman vertices reasonably close to the simplest one . this might require perhaps that we use a varying density of states , while here we have assumed it to be constant for simplicity . however we have not tried to explore this problem more quantitatively . our model gives a natural explanation of the different positions of the @xmath1 and @xmath2 peaks and the existence of these two peaks in @xmath8 symmetry ( even if screening complicates the matter ) . note that peaks with different positions are also found in the other high-@xmath10 superconductors ( see table i ) . at this point , we can notice that the existence of two electronic structures is not only detected in raman scattering ( see table i ) but also in giaever tunneling @xcite and spectroscopy tunneling measurements @xcite in ybco . the energies of these two structures were found at @xmath229 and @xmath230 which corresponds to @xmath231 and 480 @xmath112 respectively ( very close to the values reported for ybco in table i ) . a d+s wave model as developed by m.t.beal-monod et al . @xcite accounts for the @xmath15 spectrum but can not explain the @xmath8 and @xmath9 spectra unless one introduces a very large scattering rate , which is incompatible with our critical temperature as we have discussed above . we note finally that the high values of the maxima ( @xmath232 ) as well as the non observation of a significant softening of the maxima , as a function of the temperature , as reported in section v.1 , are incompatible with a weak coupling theory @xcite . in conclusion , we have presented pure electronic raman spectra of hg-1223 single crystals . in the superconducting state our most significant results are i ) the two electronic maxima usually observed separately at 530 and 800 @xmath100 in @xmath15 and @xmath1 symmetries respectively , are seen simultaneously in @xmath8 symmetry . the existence of such two electronic features at @xmath233 ( 5 - 6 ) and ( 8 - 9 ) , are reported not only in raman but also on tunneling measurements . ii ) these two electronic structures disappear at @xmath10 without showing significant softening as the temperature is raised . iii ) the electronic response function exhibits an intrinsic linear @xmath132 dependence - not only in the @xmath2 spectrum - but also in the @xmath1 spectrum , which has been extensively studied in this work : the linear dependence is robust against the change of the excitation frequency , implying that this is an intrinsic feature arising from the gap symmetry . these experimental results strongly suggest a very anisotropic superconducting gap , involving two characteristic energies and nodes located away from the [ 110 ] and [ 1@xmath130 ] directions . such observations are inconsistent with a pure @xmath12 order parameter symmetry . the simplest order parameter compatible with our experimental data displays two maximum gap values and 8 nodes . it has the @xmath8 symmetry . we have developed a detailed comparison with the experimental spectra and we have shown that such an order parameter can account satisfactorily for our experimental results . we thank p. monod , m. cyrot and g. deutscher for very fruitful discussions . d. a. wollman et al . phys . rev . 74 , 797 ( 1995 ) , c.c . tsuei et al . nature 387 , 481 ( 1997 ) . koutnetzov et al , phys . 79 , 3050 ( 1997 ) ; c. rossel et al . , physica c 282 - 287 , 136 ( 1997 ) . m. v. klein and s. b. dierker , phys.rev . b 29 , 4976 ( 1984 ) . a. a. abrikosov and v. m. genkin , zh.eksp.teor.fiz . 65 , 842 ( 1973 ) [ sov.phys.jetp 38 , 417 ( 1974 ) ] . s. b. dierker , m. v. klein , g. w. webb and z. fisk , phys . 50 , 853 ( 1983 ) . deveraux et al , phys . 72 , 396 , ( 1994 ) ; t. p. devereaux and d. einzel , phys . rev . b 51 , 16336 ( 1995 ) s. l. cooper , m. v. klein , b. g. pazol , j. p. rice and d. m. ginsberg , phys . rev . b 37 , 5920 ( 1988 ) ; s. l. cooper et al , phys . b 38 , 11934 , ( 1988 ) . r. hackl , w. glser , p. mller , d. einzel and k. andres , phys . rev . b 38 , 7133 ( 1988 ) ; r. hackl , in `` gap symmetry and fluctuations in high - tc superconductors '' edited by j. bok , g. deutscher , d. pavuna and s. wolf published in plenum press series nato asi ( 1998 ) . t. staufer , r. nemetschek , r. hackl , p. mller and h. veith , phys.rev.lett . 68 , 1069 ( 1992 ) . d. h. leach , c. thomsen , m. cardona , l. mihaly and c.kendziora solid . stat . 88 , 457 ( 1993 ) . r. nemetschek , o. v. misochko , b. stadlober and r. hackl , phys . b 47 , 3450 ( 1993);. x. k. chen , j.c.irwin , h. j. trodhal , t. kimura and k. kishio , phys . rev 73 , 3290 ( 1994 ) . m. krantz and m. cardona , j. low temp . 99 , 205 ( 1995 ) . l. v. gasparov , p. lemmens , m.brinkmann , n. n. kolesnikov and g. gntherodt , phys . b 55 , 1223 ( 1997 ) t. strohm and m. cardona , solid . stat . 104 , 233 ( 1997 ) . a. sacuto , r. combescot , n. bontemps , p. monod , v. viallet and d. colson , europhys . 39 , 207 ( 1997 ) ; a. sacuto and r. combescot in `` gap symmetry and fluctuations in high - tc superconductors '' edited by j. bok , g. deutscher , d. pavuna and s. wolf published in plenum press series nato asi ( 1998 ) . a. sacuto et al , physica c 259 , 209 ( 1996 ) . m. cantoni , a. schilling , h .- u . nisen and h. r. ott , physica c 215 , 11 ( 1993 ) . p. a. wolff , phys . 171 , 436 ( 1968 ) ; a. kawabata , j.phys.soc.jpn 30 , 68 ( 1971 ) . d. colson , a. bertinotti , j. hammann , j .- f . marucco , a. pinatel , physica c 233 , 231 ( 1994 ) ; a. bertinotti et al , physica c 250 , 213 ( 1995 ) . a. bertinotti , d. colson , j - f . marucco , v. viallet , j. le bras , l.fruchter , c. marssenat , a. carington , j. hammann , in `` studies of high temperature superconductors '' edited by narlikar , nova science publisher ( ny ) ( 1997 ) . t. p. devereaux , journal of superconductivity 8 , 421 ( 1995 ) ; t. p. devereaux et al . 72 , 3291 ( 1994 ) . a. sacuto , c. julien , v. a. shchukin , c. perrin and m. mokhtari , phys . rev . b 52 , 7619 ( 1995 ) . r.m.macfarlane , h.j.rosen , e.m.engler , r.d.jacovitz and v.y.lee , phys.rev . 38 , 284 ( 1988 ) . a. el azrak , phd thesis , universit paris - sud ( 1993 ) s.l . cooper et al , phys . b47 , 8233 ( 1993 ) s. uchida et al , phys . rev . b 43 , 7942 ( 1991 ) i. terasaki et al , physica c 165 , 152 ( 1990 ) moonsoo kang , g. blumberg , m. v. klein and n. n. kolesnikov , phys . 77 , 4434 ( 1996 ) . t. p. devereaux , phys . 74 , 4313 ( 1995 ) . g. preosti , h. kim and p. muzikar , phys . b 50 , 1259 ( 1994 ) j. annett , n. goldenfeld and s.r . renn , phys . b 43,2778 ( 1991 ) . p.j . hirschfeld , w.o . putikka and d.j . scalapino , phys . b 50 , 10250 ( 1994 ) . d.a . bonn et al , phys . b 40,4051 ( 1994 ) . s. hensen , g. mller , c.t . rieck and k. sharnberg , phys . b 56,6237 ( 1997 ) . i . maggio - aprile , phys . 75 , 2754 ( 1995 ) . g. deutscher in `` gap symmetry and fluctuations in high - tc superconductors '' edited by j. bok , g. deutscher , d. pavuna and s. wolf published in plenum press series nato asi ( 1998 ) . m. t. beal - monod , j. b. bieri and k. maki , europhys . lett . , 40 , and erratum ( 1997 ) ; n. nemetschek et al . , submitted to epj r. combescot , cond - mat/9710115 . @xmath234 + fig.1 dc magnetization recorded during field cooling in a 10 oe field for a and b samples + fig.2 experimental configurations chosen to obtain the @xmath9 , @xmath15 and @xmath235 symmetries + fig.3 raw raman spectra of the hg-1223 ( sample a ) at t= 13 k obtained from both configuration described in fig.2 . a)@xmath96 pure @xmath15 and @xmath9 symmetries ( obtained by rotating the crystal ) . b ) mixed @xmath236 and @xmath237 symmetries ( obtained by rotating the polarization ) + fig.4 raw raman spectrum of hg-1223 ( sample b ) at t= 13 k in @xmath109 symmetry . note the shoulder which develops on the high energy side , at 800 @xmath75 + fig.5 raw raman spectra of hg-1212 taken at 13k ( shown for comparison with fig.3 ) + fig.6 imaginary part of the response functions at various temperatures ( sample b ) a)@xmath109 spectrum b)@xmath9 spectrum + fig.7 imaginary parts of the response function @xmath238 $ ] deduced from experimental data of fig . the dashed lines represent @xmath239 $ ] . + fig.8 imaginary part of the response function @xmath238 $ ] in @xmath9 symmetry , obtained from various excitation laser lines ( sample b ) . zero amplitude is indicated by the short solid line . a)after normalization of the amplitude at the peak energy . b)a different normalization ( see text ) shows that the data can be gathered into two sets of spectra displaying the same frequency dependence + fig.9 imaginary part of the electronic response for two excitation lines ( 514.5 nm , circles , and 568.2 nm , triangles ) previously shown in fig.8-b as belonging to the two sets of data . within a scale factor , it appears clearly in this plot that the low frequency response exhibits the same frequency dependence up to 500 @xmath97 for both excitation lines + fig.10 @xmath15 data ( sample a ) and calculations for the @xmath12 model ( dashed line ) and for a model with 8 nodes lying at @xmath240 = @xmath212 ( solid line ) + fig.11 @xmath9 data ( sample a ) and calculation for the @xmath12 model ( dashed line ) and for a model with 8 nodes lying at @xmath241 ( solid line ) . + fig.12 @xmath109 spectrum ( sample b ) and calculations + a ) for the @xmath12 model ( solid line ) + b ) s=0.27 a=0.05 a=0.03 ( solid line ) s=3.6 @xmath242 = @xmath228 a=0.05 ( dashed line ) +

we study the classical anisotropic ferromagnetic spin chain with frustration . the behavior of soliton and kink solutions in the vicinity of the ground state phase transition from the ferromagnetic to the spiral phase is studied . the dependence of the soliton energy on small anisotropy parameter is established using scaling estimates and numerical minimization of the energy functional . conditions of the existence of the solitons are determined . it is shown that solitons survive in the spiral phase though with some restrictions on their size . a comparison of the energies of the classical solitons and the bound magnon complexes in the quantum model shows the functional similarity between them . the influence of the finite - size effects on the soliton states is studied and it is shown that the localized solitons originate from the uniform state when the system size exceeds some critical value depending on the anisotropy .

0908.0068

lately , there has been considerable interest in low - dimensional spin systems that exhibit frustration @xcite . a very interesting class of such systems is chain compounds consisting of edge - sharing @xmath0 units . recently , a variety of these copper oxides were synthesized and found to show unique physical properties @xcite . the frustration in these compounds arises from the competition of exchange interactions between magnetic @xmath1 ions carrying spins @xmath2 . due to a specific geometry of these systems ( @xmath3 angle is close to @xmath4 ) the nearest - neighbor ( nn ) interaction is ferromagnetic , while the next - nearest - neighbor ( nnn ) one is antiferromagnetic and absolute values of these interactions are comparable @xcite . an appropriate model describing the magnetic properties of such copper oxides is so - called f - af spin chain model the hamiltonian of which has a form @xmath5 where @xmath6 and @xmath7 . this model is characterized by a frustration parameter @xmath8 . the quantum f - af @xmath9 model has been intensively studied last years @xcite . most of studies of this model are related to the isotropic case ( @xmath10 ) . it is known that the ground state of the isotropic model is ferromagnetic for @xmath11 . at @xmath12 the phase transition to the incommensurate singlet phase with spiral spin correlations takes place . remarkably , this transition point does not depend on spin value @xmath13 at @xmath14 . it was shown also that the f - af @xmath9 model with anisotropic interactions has a rich phase diagram @xcite . in our papers @xcite we investigated weakly anisotropic ( @xmath15 , @xmath16 ) quantum spin-1/2 model ( [ h ] ) . it was shown that even small anisotropy essentially effects on the properties of the model . in particular , the transition point from the ferromagnetic to the spiral - like ground state shifts from @xmath12 . an interesting feature of the quantum anisotropic model ( [ h ] ) is the existence of the multimagnon bound complexes in the ferromagnetic phase . these complexes govern the low - temperature thermodynamics @xcite . it was noted also that the nnn interaction strongly affects the excitation spectrum especially in the vicinity of the transition point @xmath12 . it is known that there is a close relation between the multimagnon bound complexes in the quantum spin models and soliton and kink excitations in the classical counterparts . in particular , these soliton states have been studied extensively for the classical easy - axis ferromagnetic chain and a connection between them and the bound magnon complexes in the quantum @xmath9 model ( [ h ] ) at @xmath17 was discussed @xcite . on the other hand , the influence of the frustration on the solitons has not been considered before . the f - af model represents a suitable model to study this problem and to investigate the connection between the quantum excitation spectrum and the soliton solutions of the classical frustrated spin model . in @xcite we showed that the ground state phase diagram of the quantum model ( [ h ] ) with both @xmath18 and @xmath19 are qualitatively similar to that for the model with the anisotropy of the nn interaction only . therefore , for simplicity we consider model ( [ h ] ) with @xmath20 and @xmath16 . in this case model ( [ h ] ) takes the form @xmath21 where we put @xmath22 as an energy unit and added constant shifts to secure the energy of the ferromagnetic state to be zero . the phase diagram of the classical f - af model with @xmath23 consists of the ferromagnetic and the spiral phases . the ferromagnetic ground state is simple , i.e. , it has all spins parallel to the z axis . however , the soliton excitations in this phase are not trivial especially near the transition point between the phases . our main goal is to study the behavior of the solitons in the vicinity of the isotropic transition ( it ) point ( @xmath24 , @xmath12 ) and to compare it with the properties of the excitations of the quantum model . as it will be shown the frustration effects strongly modify the soliton states especially for small anisotropy . in particular , the exponent characterizing the power dependence of the gap ( the soliton energy with respect to the ground state ) on the anisotropy is different from that for @xmath25 . the paper is organized as follows . in sec.ii we represent the known results for soliton solutions of the anisotropic ferromagnetic chain ( model ( [ hh ] ) at @xmath25 ) . in sec.iii we consider the classical continuum f - af model in the vicinity of the ground state phase transition from the ferromagnetic to the spiral phase . we deduce the corresponding energy functional and establish the scaling form of the soliton energy as a function of the anisotropy and the frustration parameter . in this section we also obtain asymptotes of the soliton solutions at large distances and determine the necessary conditions of the soliton stability . in sec.iv we present results of the numerical minimization of the energy functional which confirm the scaling estimations . in sec.v we study the finite - size effects on the soliton solution . in sec.vi we show that the behavior of both classic solitons and @xmath26-magnon quantum spin excitations are functionally similar to @xmath26-boson bound complexes of the bose model with the attractive interaction if @xmath26 is not large . in sec.vii we give a summary of results . in the classical approximation the spin operators @xmath27 are replaced by the classical vectors @xmath28 of the fixed length @xmath13 which are parameterized by spherical coordinates @xmath29 in terms of the angles @xmath30 and @xmath31 the discrete classical f - af model ( [ hh ] ) takes the form @xmath32 -\lambda[1-\cos(\theta_{n+2}-\theta_n ) ] + \alpha[1-\cos(\theta_{n+1})\cos(\theta_n)]\right\ } \nonumber \\ & & + s^2\sum\left\ { [ 1-\cos(\varphi_{n+1}-\varphi_n)]\sin\theta_n\sin\theta_{n+1 } -\lambda[1-\cos(\varphi_{n+2}-\varphi_n)]\sin\theta_n\sin\theta_{n+2 } \right\ } \label{edis}\end{aligned}\ ] ] where @xmath33 . in this section we briefly review the known results for the classical heisenberg ferromagnetic chain with an easy - axis anisotropy , i.e. , model ( [ edis ] ) with @xmath25 . in this case the model is exactly solved in the continuum limit @xcite . the continuum limit of the classical model assumes that the vectors @xmath28 can be replaced by the classical vector field @xmath34 with slowly varying orientations , so that @xmath35 where the lattice constant is chosen as unit length . the direction of the vector field @xmath34 is determined by two angular variables @xmath36 and @xmath37 according to eq.([n ] ) . the dynamics of the vector field @xmath34 is governed by the landau - lifshitz equation : @xmath38 where we put @xmath39 . here @xmath40 is the energy as a functional of the vector field @xmath34 . using the continuum approximation ( [ dsdx ] ) , hamiltonian ( [ hh ] ) goes over into the well - known energy functional : @xmath41 \label{exxz}\ ] ] classical equation of motion ( [ ll ] ) for model ( [ exxz ] ) has two constants of the motion : the magnetization ( continuum analog to a number of magnons ) @xmath42 where we subtract a constant to make @xmath43 finite for solitons , and the momentum @xmath44 the ground state of ( [ exxz ] ) is the ferromagnetic configuration with all spins parallel ( or antiparallel ) to the z axis , i.e. @xmath45 ( @xmath46 ) . the soliton configurations are specified by the boundary conditions @xmath47 ( @xmath48 at @xmath49 . fortunately , the classical equations of motion ( [ ll ] ) for model ( [ exxz ] ) are exactly solvable @xcite . in particular , the energy for a given values of @xmath43 and @xmath50 is @xmath51 remarkably , eq.([emp ] ) reproduces the exact result for the energy at @xmath52 of the @xmath43-magnon bound state @xcite for the most quantum case @xmath9 . here we will show that simple scaling arguments are able to establish the correct scaling dependence for the energy . for this aim we rescale the coordinate @xmath53 and introduce the normalized spin vector field @xmath54 in eq.([exxz ] ) , which results in @xmath55 \label{eee}\ ] ] we notice that the integrand in equation ( [ eee ] ) does not depend on any parameters and is expressed through the normalized spin vector field . this means that the energy scales as @xmath56 , which is correct for a kink or for a large soliton excitations ( see eq.([emp ] ) ) . however , we can make one more step . the same procedure for magnetization ( [ m ] ) gives @xmath57 this expression means that the magnetization forms a scaling parameter @xmath58 . in a similar manner one can find that momentum ( [ p ] ) produces a dimensionless parameter @xmath59 independent of @xmath60 . thus , the energy of a soliton of the size @xmath43 having a momentum @xmath50 can be written in a form : @xmath61 where @xmath62 is a scaling function , which can not be found in the framework of this scaling estimate . fortunately , for model ( [ exxz ] ) this function is known exactly ( [ emp ] ) , which validates the above scaling arguments . thus , simple scaling estimates allowed us to establish the scaling parameters and the scaling form for the soliton energy . there is one more important fact which is worth noting here . the exact static solution for a kink in the discrete xxz model with classical spins has been constructed by gochev in ref.@xcite . he found that the kink energy has a form @xmath63 which reproduces the exact result for the case @xmath9 @xcite . this allows us to assume that the energy of kink ( or large soliton , which has double energy of kink ) for xxz chain with any value of @xmath13 has an universal dependence on @xmath64 ( [ ekink ] ) . we will see that this universality is destroyed in case @xmath65 . the nnn term in model ( [ hh ] ) causes the frustration in the system and immediately destroys the integrability of the model . however , the behavior of the system for @xmath66 and weak easy - axis anisotropy remains very similar to the case @xmath25 . in the classical and continuum approximation the hamiltonian reduces to an energy functional ( [ exxz ] ) with renormalized factor @xmath67 at @xmath68 . in particular , this means that the found scaling form for the soliton energy ( [ escxxz ] ) remains valid . the situation drastically changes near the it point ( @xmath69 ) where the ground state phase transition in the isotropic case takes place . let us deduce the energy functional describing the vicinity of this point . the continuum approximation assumes that the differences @xmath70 and @xmath71 are small and eq.([edis ] ) can be expanded in these differences . however , in the vicinity of the it point it is more clear and instructive to derive the continuum approach in terms of a spin vector field . for this aim we rewrite hamiltonian ( [ hh ] ) in the form : @xmath72 where the parameters @xmath73 and @xmath74 are small in the vicinity of the it point . in the classical and the continuum approach the expression containing spin operators in the first term in eq.([h14 ] ) is replaced by @xmath75 and that in the second term in eq.([h14 ] ) according to eq.([dsdx ] ) . thus , hamiltonian ( [ h14 ] ) is mapped to the energy functional @xmath76 \label{etr}\ ] ] the effect of the fourth - order term @xmath77 in the energy functional like ( [ etr ] ) was studied before @xcite , but it was considered as a small correction to the main contribution given by the term @xmath78 . on the contrary for our model ( [ etr ] ) near the it point the fourth - order term @xmath77 becomes the leading one . the appearance of the fourth - order term is related to the fact that the one - magnon spectrum in the it point becomes @xmath79 . the energy functional ( [ etr ] ) in terms of the angular variables @xmath80 and @xmath81 has a rather cumbersome form : @xmath82 \nonumber \\ & & + \frac{s^{2}}{8}\int \mathrm{d}x\left [ ( 4 - 2\sin ^{2}\theta ) \varphi ^{\prime 2}\theta ^{\prime 2}+(2\varphi ^{\prime \prime } \varphi ^{\prime } \theta ^{\prime } -\varphi ^{\prime 2}\theta ^{\prime \prime } ) \sin ( 2\theta ) \right ] \label{elong}\end{aligned}\ ] ] where the prime denotes the space derivatives @xmath83 . one can check that eq.([elong ] ) represents the leading terms in the expansion of eq.([edis ] ) in small differences @xmath70 and @xmath71 , which is actually assumed in the continuum approximation . at first , let us study the phase diagram of the classical continuum model ( [ elong ] ) . numerical calculations ( details will be given below ) confirmed a natural assumption that the ground state spin configuration for the easy - axis anisotropy case ( @xmath84 ) corresponds to the choice @xmath85 . in this case the energy functional ( [ elong ] ) simplifies to @xmath86 variation of the energy functional ( [ etheta ] ) in @xmath81 leads to the euler equation @xmath87 the ground state in the ferromagnetic phase has a trivial solution @xmath45 ( or @xmath46 ) with zero energy . in the isotropic case ( @xmath88 ) the transition from the ferromagnetic to the spiral phase takes place at @xmath12 . for @xmath89 the solution @xmath81 has a pure spiral form @xmath90 which evidently has zero mean spin projection on any axis @xmath91 and describes the spiral in the xz plane . in the anisotropic case the spiral function ( [ sp ] ) does not satisfy eq.([eiler ] ) . however , it represents a good approximation for the solution in the spiral phase . using this function as a variational one for the energy functional ( [ etheta ] ) we find that the transition between the ferromagnetic and the spiral phases occurs on the line @xmath92 . assuming that the correction to the spiral solution @xmath93 is small , we found from eq.([eiler ] ) that the first correction has the oscillating form : @xmath94 this function is not a pure spiral : due to the oscillating correction the spin vectors prefer to direct along the z axis , so that @xmath95 . however the mean values of all total projections remain zero : @xmath96 . the calculation of the energy functional ( [ etheta ] ) with the function ( [ spiral ] ) yields @xmath97 one can see that the correction to the energy is really small in the spiral phase region . energy ( [ espiral ] ) becomes negative for @xmath98 . this means that taking the correction in eq.([spiral ] ) into account shifts the transition line from the ferromagnetic to the spiral phase on the value less than @xmath99 ( the smallness of the correction in eq.([spiral ] ) allows us to leave the name spiral for this phase ) . the phase transition from the ferromagnetic to the spiral phase at @xmath100 is obviously the first order one . the transition line @xmath101 is shown in fig.[phase ] . it is worth noting here that the found behavior of the transition line corresponds to the classical version of spin model ( [ hh ] ) and it is shifted for the quantum spin-@xmath13 model . quantum corrections for @xmath102 change numerical coefficient in eq.([espiral ] ) . but for the @xmath9 case they lead to another form of the transition line : @xmath103 @xcite . now let us study the excitations of model ( [ etr ] ) . similar to the case @xmath25 the lowest configuration in the ferromagnetic region for a given @xmath104 is described by the static soliton - like solutions of eq.([eiler ] ) with @xmath85 over the ferromagnetic configuration @xmath45 ( or @xmath46 ) . unfortunately , we could not find the exact soliton solution near the it point . however , it is possible to determine the scaling dependence of the energy of soliton of size @xmath43 on parameters @xmath60 and @xmath105 using scaling estimates as we did for the case @xmath25 . near the it point we perform the rescaling @xmath106 and @xmath107 which transforms the energy functional ( [ etr ] ) to @xmath108 \label{etrscal}\ ] ] with @xmath109 . since the integrand in eq.([etrscal ] ) depends on the parameter @xmath110 only , we conclude that the energy of a kink or of a large soliton is @xmath111 . as a result of rescaling the magnetization ( [ m ] ) becomes : @xmath112 so , the magnetization produces a scaling parameter @xmath113 . the momentum ( [ p ] ) forms the same parameter @xmath114 as for the case @xmath25 . thus , the energy of a soliton of the size @xmath43 can be represented in a form : @xmath115 where the scaling function @xmath62 can be found numerically . this scaling form is exactly the same as was found for the quantum spin-@xmath2 model near the it point @xcite . we note that similar scaling procedure for the @xmath116-dimensional version of model ( [ etr ] ) gives the scaling form for the energy of the soliton of the size @xmath43 ( even small - amplitude solitons are stable for model ( [ etr ] ) in two- and three - dimensional cases @xcite ) : @xmath117 and the case @xmath118 requires special treatment . though the exact solution of the corresponding equation of motion ( [ ll ] ) is unknown it is possible to identify the necessary conditions for a stability of solitons and establish their asymptotic behavior . it is convenient to introduce the complex function @xmath119 so that @xmath120 and @xmath121 . equation of motion ( [ ll ] ) for @xmath122 near the it point reads @xmath123 where @xmath124 is rescaled time . solitons are localized objects and , therefore , it requires that @xmath125 far from the soliton center . so , at large distance from the center one can linearize eq.([llpsi ] ) in @xmath126 by putting @xmath127 : @xmath128 we seek the asymptotic of @xmath126 in conventional exponential form @xmath129 \label{psiasymp}\ ] ] where @xmath130 and @xmath131 are normalized linear and angular velocities defined as @xmath132 substituting eq.([psiasymp ] ) into eq.([psilinear ] ) we obtain equation for @xmath133 @xmath134 this equation has four roots . the condition of the soliton existence is that all roots have non - zero real part , which secures the decay of solutions at @xmath135 . the region in @xmath136 plane where @xmath126 exponentially vanishes at @xmath135 is defined as @xmath137 . the dependence @xmath138 is parametrically definable function @xmath139 where the parameter @xmath140 runs from @xmath141 to @xmath142 for @xmath143 and @xmath144 for @xmath145 . plane for different @xmath110 . allowable values of @xmath130 and @xmath131 lie below the boundaries.,width=288 ] equations ( [ stable ] ) of the boundary of the soliton existence region coincide with the condition of the spin - wave instability . the existence regions for a few values of the parameter @xmath110 are shown in fig.[vomega ] . for @xmath143 the region has a form similar to the case @xmath25 with the quadratic dependence @xmath138 near the maximum point ( @xmath146 ) . for the case @xmath147 this dependence becomes @xmath148 . when @xmath145 , the dependence @xmath138 shifts down and a cusp appears at the maximal point ( @xmath149 ) . the region of the soliton existence contains allowable values of @xmath130 and @xmath131 for a given @xmath110 . as it can be seen in fig.[vomega ] the soliton can exist in the spiral phase ( @xmath150 ) . but , for example , the kink is unstable for @xmath151 since the point ( @xmath152 ) corresponding to the kink does not belong to the existence region . it is worth noting that one should be careful in treating of the part of the existence region with large negative @xmath131 , where the derivatives of @xmath81 and @xmath80 near the center of soliton become large and the continuum approximations eqs.([dsdx ] ) , ( [ dsdx2 ] ) become doubtful . within the existence region the roots of eq.([eqkappa ] ) have the forms @xmath153 where @xmath154 , @xmath155 and @xmath156 are real nonnegative quantities and @xmath157 . certainly , one should take into account only the roots providing the decay of @xmath158 at the corresponding limit . for determinacy we will consider the limit @xmath159 and , therefore , analyze the roots @xmath160 and @xmath161 . at infinity only the root @xmath162 with the smaller real part is vital . so , the asymptotes of the angles at @xmath159 are @xmath163 $ ] and @xmath164 . the case of the static soliton ( @xmath165 ) requires special treatment , because in this case both roots @xmath162 and @xmath166 have equal real parts @xmath167 ( they immediately split if @xmath168 ) . fortunately , for the static soliton the solution of eq.([eqkappa ] ) can be obtained explicitly and this allows to analyze in detail the behavior of the soliton asymptotes for given values of @xmath110 and @xmath131 . the decaying roots at @xmath159 are @xmath169 and @xmath170 . simple analysis of eq.([kappa ] ) shows that there are three regimes of the asymptotic ( [ psiasymp ] ) . the exponent @xmath162 is real for @xmath171 , which means that @xmath172 smoothly tends to zero at infinity . this region lies totally in the ferromagnetic phase . in the region @xmath173 the parameter @xmath162 contains both real and imaginary parts . here the superposition of asymptotes ( [ psiasymp ] ) with @xmath162 and @xmath170 results in the decay with oscillation at infinity : @xmath174 and @xmath175 . finally , for @xmath176 the parameter @xmath162 is pure imaginary , which implies that the solution oscillates but does not decay at infinity . therefore , in this region there are no static soliton - like excitations . as was noted above only one asymptotic with @xmath162 having smaller real part survives at infinity for @xmath177 . this implies that the asymptotic of the function @xmath172 has no zeros . however , in the region close to the soliton center the function @xmath172 can have a finite number of zeros and it is really observed numerically . generally , the relation between the parameters @xmath178 and @xmath179 is unknown analytically . however , it can be found for large values of @xmath180 . for large solitons ( @xmath181 ) the soliton energy and , therefore , the scaling function @xmath182 saturates to some finite value . this means that for @xmath181 both @xmath183 and @xmath184 . the boundaries between the regions corresponding to different asymptotic regimes for large solitons are obtained from the above analysis and they are shown in fig.[phase ] . as one can see in fig.[phase ] the oscillating large solitons exist not only in the ferromagnetic phase but , partly , in the spiral phase region @xmath185 . the amplitude of the soliton oscillations grow as @xmath110 increases and the oscillations are most strong in the spiral phase . however , one should remember that the soliton excitations in the spiral phase lie in the high - energy part of the spectrum , because the energy of the soliton - like solutions differs from the ferromagnetic one on some finite value ( [ escal ] ) . therefore , the soliton energy is higher than the spiral one ( [ espiral ] ) on the value proportional to the system size . nevertheless , these states can be important in strong magnetic field close to the saturation value . if we expand @xmath186 and take into account the terms @xmath187 in eq.([llpsi ] ) , then eq.([llpsi ] ) near the border of the soliton existence reduces to the modified non - linear schrdinger equation containing the fourth - order derivative term @xmath188 unfortunately , in spite of substantial simplification of the initial equation of motion ( [ llpsi ] ) , the exact solution of this equation is unknown and we have to use numerical calculation . we note that eqs.([psiasymp]),([eqkappa ] ) are valid for small @xmath52 , but for any @xmath105 . in particular , for the case @xmath25 ( @xmath189 ) they correctly reproduce the known asymptotic for @xmath190 and @xmath191 . we have carried out a numerical analysis of the discrete and the continuum versions of the classical spin model ( [ hh ] ) in the vicinity of the it point . both energy functionals ( [ edis ] ) and ( [ elong ] ) were minimized numerically over angles @xmath192 and @xmath193 with fixed values of magnetization ( [ m ] ) and momentum ( [ p ] ) . the periodic boundary conditions for soliton and the open boundary conditions for kink excitations were imposed . the numerical calculations showed that for small @xmath60 the difference between the discrete and the continuum models is negligible as expected and it increases as @xmath60 grows . on the investigation of the dynamics of solitons we restrict ourself to small anisotropy @xmath52 , when the continuum approximation is valid , because in the discrete model with finite @xmath60 a so - called pinning potential appears @xcite and the momentum becomes undefined . so , on default we will present numerical results for the discrete classical model ( [ edis ] ) , keeping in mind that the continuum approach corresponds to the limit @xmath194 . we studied the finite size effects for both discrete and continuum models and found that the system size @xmath195 should be taken so that the parameter @xmath196 . the finite size effects will be discussed in detail in sec.v . here we only notice that when the relation @xmath196 is fulfilled the convergence of a solution accelerates exponentially with @xmath195 . at first we present and discuss the results for small values of @xmath60 and @xmath105 , when the continuum approximation is justified and the soliton energy takes the scaling form ( [ escal ] ) . in order to verify the found scaling equation we plotted the results for a fixed @xmath197 and different values of @xmath60 , @xmath105 , @xmath43 as @xmath198 vs. scaling parameters @xmath113 and @xmath109 . for small @xmath60 and @xmath105 all data must lie on one curve which actually represents the scaling function @xmath182 . for example , the numerical data for the static solitons ( @xmath199 ) and for @xmath147 is shown in fig.[emmu0 ] . here the calculated dependencies of the soliton energy on the soliton size @xmath43 for three values of @xmath200 , @xmath201 , @xmath202 are demonstrated in axes @xmath203 vs. @xmath113 . as we see all three solid curves lie very close to each other and they rapidly converges in the limit @xmath194 , so that the curve corresponding to @xmath204 perfectly describes the scaling function @xmath205 . versus scaled soliton size @xmath206 for @xmath147 and @xmath199 . solid lines are soliton energies for three values of @xmath207 from top to bottom . circles ( joined by dashed line ) are energies of magnon complexes of quantum @xmath9 f - af chain for @xmath208 and @xmath209 @xcite . dot - dashed line is the energy of boson bound states of model ( [ hbose2]).,width=288 ] the function @xmath205 , shown in fig.[emmu0 ] , reaches a maximum at @xmath210 and then rapidly saturates , yielding the energy of large solitons @xmath211 . for comparison we put also in fig.[emmu0 ] similar data for @xmath199 and @xmath12 obtained in ref.@xcite for the quantum spin-@xmath2 chain of length @xmath208 and @xmath209 , which represents a rough estimate for the spin-@xmath2 scaling function @xmath212 . one can see that the behavior of the scaling functions for the quantum @xmath9 and the classical @xmath213 models are very similar , though they have different limits at @xmath214 . thus , we believe that the scaling law eq.([escal ] ) is valid for quantum model ( [ hh ] ) with general @xmath13 and the corresponding scaling functions @xmath215 behave similar to @xmath216 and @xmath62 . it is known @xcite that for @xmath25 there is an equivalence between the energies of the classic static solitons of small size and the bound boson complexes of the bose - hamiltonian which is mapped from the anisotropic heisenberg model . we expect that this mapping is valid for @xmath65 as well and it allows to determine the functional form of the energy for small solitons near the it point . as it will be shown in sec.vi the energy of the @xmath43-boson bound state of the corresponding bose - model is @xmath217 where @xmath218 and the dependence @xmath219 is shown in fig.[gbeta ] . for the energy of small static solitons ( triangles ) and @xmath219 for the energy of boson complexes eq.([eboson ] ) ( solid line).,width=288 ] we found that the numerical data for @xmath220 at @xmath221 is perfectly fitted by eq.([egb ] ) and the function @xmath222 for the classical solitons perfectly coincides with @xmath219 ( see fig.[gbeta ] ) . therefore , we believe that the scaling functions for solitons and boson complexes are the same at @xmath223 and are given by eq.([egb ] ) . however , they are different generally ( see fig.[emmu0 ] ) . in particular , at @xmath181 the energy of the soliton saturates while the boson complex energy diverges . for small solitons with @xmath199 according to eq.([omegav ] ) @xmath131 is given by the derivative of the right - hand side of eq.([egb ] ) . therefore , the boundary between different regimes of asymptotes given by equation @xmath224 ( see eq.([kappa ] ) ) transforms for small solitons to the relation @xmath225 for @xmath226 . numerical calculation of the function @xmath222 ( fig.[gbeta ] ) indicates that there is only one solution @xmath227 of the above relation . this means that the oscillation behavior of the solitons of size @xmath43 exists if @xmath228 and that the first two terms in the expansion in small @xmath180 ( [ egb ] ) is not sufficient to determine the boundary where solitons disappear . versus scaled soliton size @xmath206 for @xmath229 and for different @xmath197.,width=288 ] the scaling function @xmath182 for @xmath230 at @xmath229 is demonstrated in fig.[emmu1 ] . as can be seen in fig.[emmu1 ] the scaling function @xmath231 converges at @xmath214 to the same finite value for all @xmath197 , which is natural , because the spectrum of large solitons is flat . another observation followed from fig.[emmu1 ] is that the scaling function for @xmath229 oscillates substantially stronger than that in the case @xmath147 ( fig.[emmu0 ] ) . as we have stated above the necessary condition for the soliton existence is @xmath232 which means that @xmath233 for @xmath234 and @xmath235 for @xmath236 for any velocity . this inequality imposes certain limitations on the soliton size ( or the allowable values of @xmath180 ) with given values @xmath110 and @xmath197 . however , the existence regions in the ( @xmath179 ) plane can not be obtained directly from that in ( @xmath237 ) plane , because the analytical relations between @xmath238 and @xmath239 near the it point are unknown in contrast with the integrable case @xmath25 . therefore , the stability of the soliton for different values of parameters @xmath110 , @xmath180 and @xmath197 can be established by numerical minimization of the energy functional only . our calculations show that for @xmath143 the solitons of any size and for all @xmath197 exist and inequality ( [ stab ] ) is satisfied . however , the situation changes for @xmath145 . for example , the solitons with @xmath229 and @xmath199 are stable for @xmath240 only as it is shown on fig.[emmu1 ] . as a matter of fact for @xmath145 only part of a whole phase space ( a half - strip @xmath241 ) is allowable for solitons . in general , the allowable region in ( @xmath179 ) half - strip has a complicated form and can consist of many disconnected parts . for instance , at @xmath242 the static solitons with @xmath199 are stable in the range @xmath243 though for @xmath244 small solitons exist . though a detailed phase picture can be find only numerically , nevertheless we can make some statements about it . since soliton is a localized object , its energy inevitably saturates at @xmath214 to some finite value independent of @xmath197 . this means that for large solitons @xmath245 . therefore , for @xmath246 all large enough solitons with @xmath247 are allowable . on the other hand deeply in the spiral phase when @xmath151 the allowable region is restricted by @xmath248 and large solitons do not exist . in fact , we did not find large solitons for @xmath151 in our calculations . it is interesting to study the constraints on the soliton size in the isotropic limit of model ( [ hh ] ) at @xmath89 , when @xmath194 and @xmath249 . it can be shown that the function @xmath182 at @xmath221 and any @xmath244 behaves as ( @xmath250 ) , where @xmath251 does not depend on @xmath60 . in this case condition ( [ stab ] ) reduces to @xmath252 . it means that the maximal possible size of soliton grows at @xmath253 . this fact qualitatively agrees with the observed behavior @xcite of the stability of magnon bound complexes with @xmath254 in the isotropic quantum @xmath9 f - af model at @xmath12 . this is one more indication of the resemblance of the classical solitons and the quantum bound complexes in the f - af model . for @xmath255 ( dot - dashed line ) , @xmath256 ( dashed line ) and @xmath257 ( solid line).,width=288 ] in fig.[ep ] the dependence of the soliton energy on the momentum for @xmath229 is plotted for three different soliton sizes corresponding to @xmath255 , @xmath256 and @xmath257 . it is interesting to note that similarly to the case @xmath25 , there is a periodical dependence of the energy on the momentum in the _ continuum _ model near the it point . we see that the dependence of the energy on the momentum rapidly flattens with the increase of the soliton size . in other words , the soliton mass exponentially grows with the soliton size . in this respect the behavior of the soliton spectrum is similar to the exactly solvable case @xmath25 ( see eq.([emp ] ) ) . however , the dependence @xmath258 has a very complex form in contrast to the case @xmath25 . as it is shown on fig.[ep ] @xmath259 can be @xmath199 , @xmath254 or even intermediate value between them . for @xmath256 the minimum of @xmath258 is obtained at @xmath254 . with the subsequent increase of the soliton size @xmath180 the form of the spectrum alternates : it has the minimum either at @xmath199 or at @xmath254 . however , for @xmath260 the spectrum becomes so flat that it is difficult to distinguish these cases numerically . the soliton velocity @xmath130 is zero in both points @xmath199 and @xmath254 . ) and large ( @xmath261 ) solitons for @xmath147 and @xmath199.,width=288 ] ) and large ( @xmath261 ) solitons for @xmath229 and @xmath199.,width=288 ] ) and large ( @xmath261 ) solitons for @xmath229 and @xmath254.,width=288 ] we have obtained numerically the soliton shapes for different values of the parameters @xmath110 , @xmath180 and @xmath197 . the shapes of static ( @xmath199 ) solitons at @xmath147 for the cases @xmath262 and @xmath261 are demonstrated in fig.[solitonmu0p0 ] . the soliton solutions oscillate around @xmath45 ( and @xmath46 for large soliton ) and these oscillations exponentially decay as moving off the domain walls . numerical calculations show that at approaching to the spiral phase the amplitude of the oscillations grows and its damping decreases with the distance from the soliton center . the shape of solitons of two different sizes with @xmath262 and @xmath261 at @xmath229 , which is very close to the transition line , is shown in fig.[solitonmu1p0 ] . here one can see the increase of oscillations in comparison with the case @xmath147 . the observed behavior of the soliton shape is in full accord with conclusions followed from the analysis of eq.([kappa ] ) . the shape of solitons changes with the momentum @xmath197 . this change is more pronounced for small solitons : they become narrower and higher . at the same time the shapes of large solitons remain almost the same . as an example the shapes of small and large solitons for the momentum @xmath254 and @xmath229 are demonstrated in fig.[solitonmu1ppi ] . as shown in fig.[solitonmu1ppi ] the function @xmath81 has a cusp at the maximum , which resembles the case @xmath25 . however , the behavior of the function @xmath263 drastically differs from the case @xmath25 @xcite . in particular , @xmath263 is discontinuous at @xmath264 in contrast with that for @xmath25 , where this function is linear near @xmath264 . one more illustration of the oscillation behavior of localized excitations in the f - af model is the kink solution for the particular case @xmath147 ( @xmath12 ) . though we did not find the analytic solution of the non - linear equation ( [ eiler ] ) , we found for this particular case some signs that it can admit a solution in a closed form . namely , we studied numerically the kink excitation described by equation ( [ eiler ] ) with @xmath265 . the boundary conditions are @xmath266 , @xmath267 and @xmath268 . numerical solution showed with very high accuracy that the asymptotic of the kink solution centered at @xmath264 has a form @xmath269 the oscillation behavior of the kink asymptotes is in full accord with eq.([kappa ] ) . at @xmath270 @xmath271 besides , the calculated kink energy has a surprisingly simple form @xmath272 where the numerical factor @xmath273 was found with high precision . moreover , the contributions to this kink energy from different terms in eq.([etheta ] ) have simple rational ratios . all these facts together give a hope to obtain the exact solution of eq.([eiler ] ) for @xmath265 in closed form . the continuum approximation is valid for @xmath52 . if the parameter @xmath60 is not very small there are corrections to the soliton energy due to a discreteness . the energy of large soliton ( @xmath181 ) at @xmath12 obtained numerically is shown in fig.[ealpha ] . as one can see the energy is perfectly described by equation @xmath274 which is valid up to @xmath275 . here the first term is described by the continuum approximation , while the second term represents the correction coming from the discreteness of the lattice . the found correction to the soliton energy ( @xmath276 ) differs from the correction to the energy of multimagnon complex ( @xmath277 ) found for the case @xmath9 in ref.@xcite . for @xmath278.,width=288 ] we found that similar behavior of the energy of the large soliton takes place for any @xmath110 . so , the energy for @xmath181 can be written as @xmath279 where @xmath280 is the saturated values of @xmath182 at @xmath214 . ( solid line ) and @xmath281 ( dashed line ) describing two leading terms for the energy of large solitons ( eq.([e3454])).,width=288 ] the behaviors of the functions @xmath280 and @xmath281 are shown in fig.[fgmu ] . as one can see @xmath280 goes down with the increase of @xmath110 and becomes negative in the spiral phase at @xmath282 . at the same time , the correction in eq.([e3454 ] ) increases at approaching to the spiral phase . in the opposed limit of large anisotropy @xmath283 , the asymptotic of the soliton energy is found using a simple perturbation theory , which gives @xmath284 as follows from this equation the energy is not simply proportional to @xmath285 in contrast with the case @xmath25 ( eq.([ekink ] ) ) . this means that the dependence of the energy of kink or large soliton is not a universal function of @xmath60 for general @xmath13 as it is for @xmath25 . in the preceding sections we investigated the soliton and the kink excitations of model ( [ edis ] ) in the infinite system . now we study the finite - size effects of these states . an interesting property of the soliton solutions of the classical model ( [ edis ] ) on finite ring is the existence of the critical value @xmath286 below which the uniform solution ( @xmath287 ) is realized . at @xmath288 this state develops into localized soliton . this transition from the uniform to the localized solution is similar to the well - known gross - pitaevskii transition in the bose - systems @xcite . we consider the static soliton on the ring of size @xmath289 with periodic boundary conditions . we use the continuous approach because the finite - size effects are essential at small values of @xmath60 . in the beginning we study the case @xmath25 . in this case the function @xmath81 providing the extremum of the energy functional ( [ exxz ] ) satisfies the lagrange - euler equation @xmath290 with boundary conditions @xmath291 the lagrange multiplier @xmath131 ensures the condition of a given normalized magnetization @xmath292 simple analysis shows that for any given values of @xmath293 and @xmath289 there is a critical value of parameter @xmath294 so that at @xmath295 eq.([eqxxz ] ) has a single and uniform solution @xmath296 . the energy of this solution is @xmath297 at @xmath288 another , non - trivial solution of eq.([eqxxz ] ) appears , which is a precursor of localized soliton solution . in particular , for the case @xmath298 the critical value is @xmath299 and the uniform solution at @xmath295 is @xmath300 . the non - trivial solution at @xmath301 can be expanded in small parameter @xmath302 : @xmath303 where @xmath304 . the energy for this case can be written as @xmath305 where @xmath306 and @xmath307 . therefore , the second derivative of @xmath40 with respect to @xmath308 is discontinuous at @xmath309 ( @xmath310 ) . at @xmath311 the solution for the case @xmath298 becomes @xmath312\ ] ] and @xmath40 tends to the known results @xmath313 . a similar behavior of the soliton solution takes place for other values of @xmath314 . it is straightforward to obtain the value @xmath315 and the soliton solution near @xmath315 for the general case in the same manner as for the case @xmath25 . for example , for @xmath12 and @xmath298 the critical value is @xmath316 and the solution at @xmath317 is @xmath318 where @xmath319 . the energy in this case is @xmath320 where @xmath321 and @xmath322 . for @xmath12 and @xmath199 . circle is a critical point of soliton creation.,width=288 ] the calculated dependence @xmath323 for @xmath12 and @xmath298 is shown in fig.[gp ] . as one can see the energy rapidly saturates to its asymptotic value at @xmath324 . therefore , in order to reduce finite - size effects in numerical calculations near the transition point @xmath12 one should choose the system size @xmath195 so that the parameter @xmath325 . the critical @xmath315 for a given magnetization @xmath293 can be restored from the case @xmath298 by the relation @xmath326 the process of the formation of the kink ( domain - wall ) occurs in a somewhat different way . the kink is realized in open chain with the boundary conditions @xmath327 , @xmath328 . in this case the lowest state at small @xmath60 is the spiral one @xmath329 and its energy remains finite at @xmath88 in contrast with the solitons . for example , for @xmath12 it is @xmath330 when @xmath60 increases the spiral configuration continuously transforms to the kink . in this respect the situation is different from the definite and sharp transition from the uniform to soliton - like state studied before . the crossover between the spiral and the kink states takes place at @xmath331 . at @xmath332 the energy of the kink is a half energy of the soliton . it is worth noting that the transition of such kind coming from the finite - size effects occurs in the quantum spin model ( [ hh ] ) as well . for very small @xmath60 , when the corresponding parameter @xmath333 , the system is in a uniform state , which is the state with a given total projection @xmath334 and the maximal total spin @xmath335 . the energy of this state is given by eq.([elin ] ) with @xmath336 as in the classical model . this uniform state transforms continuously to the localized state of bound magnon complex when @xmath60 increases . the crossover between the uniform and the localized states takes place at @xmath337 for @xmath25 and at @xmath338 for @xmath12 . it is known that there is a resemblance between the classical solitons and the quantum magnon complexes at @xmath25 @xcite . in the preceding sections we observed such a resemblance for the classical f - af model and its quantum counterpart with @xmath9 as well . here we consider the excitations in the quantum model with general @xmath13 and compare them with the classical solitons . a standard method to treat the quantum spin models is a mapping of the spin - hamiltonian to the bose - one . though this method is approximate it gives important indications about the behavior of the spin systems . using the dyson - maleev transformation @xmath340 we represent the hamiltonian ( [ hh ] ) in the form @xmath341 where @xmath342 and @xmath343 are conventional bose - operators . in the long - wavelength and weakly - coupling limit @xmath344 and the one - magnon spectrum @xmath345 in the vicinity of the it point is @xmath346 in this limit hamiltonian ( [ hbose ] ) is equivalent to the bose - model with the attractive @xmath339 -function interaction @xmath347 where @xmath26 denotes number of the bose particles . for the case @xmath25 one can neglect the fourth - order derivative term and such model becomes exactly solvable one @xcite . however , in the vicinity of the it point the fourth - order derivative term in eq.([hbose2 ] ) is important and it destroys the exact solution found in ref.@xcite . therefore , we have to use approximations . first , we consider the two - boson bound state near the it point and compare it with the two - magnon bound state found in ref.@xcite . the wave function of two bosons with the total momentum @xmath348 and the energy @xmath349 is @xmath350 . the function @xmath126 is determined from the schrdinger equation @xmath351 where @xmath352 and @xmath353 ( @xmath354 ) is the binding energy . here we see that the fourth - order derivative term introduces the dependence of the function @xmath355 and the binding energy @xmath356 on the total momentum . the function @xmath355 as well as its first- and second - order derivatives are continuous at @xmath357 , while the third - order derivative satisfies the condition @xmath358 the wave function of the bound state of two particles with @xmath339 -attraction has a form : latexmath:[\[\psi ( x)= e^{-\kappa |x|}\left [ \eta \cos ( \eta x)+\kappa \sin ( \eta @xmath133 and @xmath360 are determined by the above stated conditions for the function @xmath355 and its derivatives at @xmath357 . the solution has a form @xmath361 where @xmath362 the binding energy is @xmath363 in particular , for @xmath265 ( @xmath12 ) and @xmath199 it is @xmath364 the wave function ( [ psi2boson ] ) has one interesting feature . it oscillates when @xmath360 is real and has a form of superposition of two decaying exponents otherwise : @xmath365 . the boundary between these regions is defined by equation @xmath366 , which has a solution @xmath367 . thus , the wave function of two coupled bosons oscillates for @xmath368 and does not oscillate for @xmath369 . the oscillation region diminishes with the increase of the total momentum . certainly , the oscillatory behavior of the wave function is the effect of the frustration similarly to that observed for the solitons in the spin model . we can compare eq.([e2boson ] ) with the binding energy of two magnons for @xmath9 found by us in @xcite @xmath370 it follows from eqs.([e2boson ] ) and ( [ e2magnon ] ) that the leading term in @xmath371 at @xmath194 coincides up to numerical factor with the two - boson binding energy , but subsequent terms in the expansion of @xmath371 are absent in ( [ e2boson ] ) . similar difference between @xmath371 and @xmath372 takes place for any values of @xmath314 . for example , for @xmath25 it is @xmath373 as will be shown below , this difference plays a key role at comparison of the multimagnon and multiboson bound states . the exact wave function of the @xmath26-boson state at @xmath25 is known for any @xmath26 @xcite . as was shown in ref.@xcite at @xmath25 the hartree approximation correctly reproduces the exact energy of the multiboson states ( @xmath374 ) with zero total momentum . therefore , we expect that this approach gives reliable results for @xmath374 when @xmath65 . here we consider bose model ( [ hbose2 ] ) in a parametric regime corresponding to the vicinity of the it point of model ( [ hh ] ) . the energy functional for zero total momentum in the hartree approximation is @xmath375 \label{ehartree}\ ] ] rescaling @xmath376 and @xmath377 transforms the energy functional ( [ ehartree ] ) to @xmath378 \label{eharnor}\ ] ] where @xmath379 . the function @xmath380 satisfies the conventional normalization condition , which after rescaling gives the same normalization condition for @xmath381 @xmath382 the hartree equation comes from the minimization of the energy over @xmath381 . for @xmath374 it has a form : @xmath383 where @xmath384 is the lagrange multiplier secured the norma condition ( [ norma ] ) . unfortunately , the solution of this equation at present is unknown . however , if we assume the existence of a localized solution of eq.([eqhartree ] ) , then the integral in eq.([eharnor ] ) converges yielding some function of parameter @xmath226 . so , the energy of @xmath26-boson bound complex takes the form @xmath385 the behavior of the function @xmath219 obtained by the numerical minimization of the functional ( [ eharnor ] ) is demonstrated in fig.[gbeta ] . numerical calculations also showed that there is a critical value of the parameter @xmath386 separating the regions with ( @xmath387 ) and without ( @xmath388 ) oscillations of the function @xmath380 . the value @xmath227 is the same as was found for the classical spin model , which certificates the distinct correspondence between the boson and the classical spin models for small solitons . however , the bound complex of bosons exists in the whole region of parameter @xmath226 , which means that the instability of solitons studied in sec.iii comes from the effects neglecting in the boson approach . it is interesting to compare eq.([eboson ] ) with the energy of the @xmath26-magnon bound state of the quantum model ( [ hh ] ) with @xmath9 at @xmath12 . the latter has been obtained in ref.@xcite and has a form @xmath389 where @xmath390 and @xmath391 are numerical coefficients . similarly to the case @xmath392 the leading terms of the expansions of these two energies in small @xmath60 coincide with eq.([eboson ] ) up to numerical factors . however , energy ( [ eboson ] ) tends to @xmath141 at @xmath393 ( collapse phenomenon ) , while the energy of @xmath26-magnon bound complex in the spin model are finite at @xmath393 and behaves as @xmath394 @xcite . therefore , subsequent terms in the energy expansion in @xmath60 of the spin model are responsible for a short distance repulsion of magnons preventing the collapse . a comparison of the expression for the energy of the classical soliton of size @xmath43 ( [ escal ] ) with a formula for the energy of @xmath26-boson complex ( [ eboson ] ) indicates that the obtained expression for the energy of @xmath26-boson complex ( [ eboson ] ) is a particular case of a more general scaling relation ( [ escal ] ) for @xmath199 and @xmath395 as we discussed in sec.iv the scaling function of the boson model @xmath396 at @xmath221 coincides with the scaling function of the classical spin model . the comparison of this equation with eq.([escal ] ) allows us to assume that eq.([fboson ] ) represents two first terms of the scaling function of the quantum spin-@xmath13 model ( including the classical limit ) in small parameter @xmath180 @xmath397 where functions @xmath398 depend on @xmath13 . eq.([fboson ] ) reproduces ( probably exactly for any @xmath13 ) two first terms in this expansion . but the expansion ( [ fseries ] ) contains infinite number of terms in contrast with eq.([fboson ] ) . it leads to the finite energy of both the classic solitons and the quantum bound magnon complexes while the energy of @xmath26-boson complex diverges at @xmath393 . summarizing all the above facts , we believe that the bound energy of @xmath26 -magnon complex for a quantum spin-@xmath13 model ( [ hh ] ) near the it point is correctly described by the classical scaling formula ( [ escal ] ) @xmath399 the scaling functions @xmath215 for quantum spin-@xmath13 case does not coincide with the function @xmath62 obtained in the classical continuum approach , though all of them have very similar behavior and @xmath400 . we studied the soliton excitations in the classical f - af model with the easy - axis anisotropy . the f - af model has two parameters : the frustration parameter @xmath401 and the anisotropy @xmath60 . we found that in a weakly anisotropic limit ( @xmath402 ) the behavior of the soliton solutions for small frustration parameter @xmath11 is qualitatively similar to that for the exactly solvable easy - axis xxz chain ( @xmath25 case ) . however , the situation drastically changes near the it point ( @xmath88 , @xmath12 ) , where the transition from the ferromagnetic to the spiral ground state takes place in the isotropic case . in the vicinity of the it point the corresponding energy functional in the continuum approximation qualitatively changes and does not admit the exact solution . the analysis of the derived energy functional allowed us to estimate the behavior of the transition line between the ferromagnetic to the spiral ground state in ( @xmath403 ) plane near the it point . we mainly interested in the behavior of the solitons in the vicinity of the it point . we showed that these localized states are separated from the ferromagnetic state by a finite gap . the dependence of the soliton energy ( the gap ) on model parameters near the it point was established on a base of the scaling arguments . as a result we found that the soliton energy is proportional to @xmath404 and is expressed by the scaling function ( [ escal ] ) depending on three scaling parameters : @xmath405 , the scaled soliton size @xmath180 and the momentum @xmath197 . the analysis of the asymptotic solutions of the corresponding equation of motion provided us with the necessary conditions of the soliton stability . it was shown that solitons of all sizes and any momentum exist when @xmath234 , while in the region @xmath145 there are definite restrictions on the soliton size @xmath180 and the momentum @xmath197 . the distribution of the allowable values of the soliton parameters in ( @xmath179 ) plane for @xmath145 has very complicated form , which can be determined numerically only . for example , the static solitons ( @xmath199 ) only of the middle size @xmath243 are stable at @xmath242 . nevertheless , some facts about the soliton existence region was ascertained analytically and then confirmed by numerical calculations . in particular , small static ( @xmath199 ) solitons are unstable for @xmath145 , while small solitons with non - zero momentum exist for any @xmath110 . large solitons ( @xmath181 ) exist for @xmath246 and , therefore , they survive in the part of the spiral phase @xmath406 . though the soliton excitations in the spiral phase lie in the high - energy part of the spectrum , they can play an essential role in the magnetization processes . in the isotropic limit ( @xmath249 ) only small solitons with non - zero momentum survive and the maximal allowable size of these solitons increases when the frustration parameter tends to the critical value @xmath407 . such a dependence of the soliton size on the frustration parameter is qualitatively similar to the observation that the size of the multimagnon bound complexes with @xmath254 in the quantum f - af model grows when @xmath407 . we found that the frustration reveals itself in the oscillating shape of solitons . the amplitude of the oscillations grows at approaching to the spiral phase and further inside of it . generally , at some value of @xmath110 the solution starts to oscillate without the decay at infinity , i.e. , soliton - like solution with given values of @xmath180 and @xmath197 disappears . we studied the finite - size effects on the soliton solutions . it was shown that the soliton solution on finite ring originates in the uniform ( non - localized ) state . the transition from the uniform to the soliton - like state occurs at the critical value of the anisotropy @xmath408 , below which the uniform solution is realized . this finite - size effect is similar to the well - known gross - pitaevskii transition in the bose - systems . in order to establish a connection between the properties of the solitons and the multi - magnon complexes of the quantum counterpart of this model we used the dyson - maleev mapping of the spin model to the bose one . it turned out that the dependence of the energy of boson bound complexes on model parameters represents a particular case of the found scaling expression for the classical soliton energy . moreover , the energy of the bound magnon complexes for quantum spin-@xmath2 model found by us before perfectly coincides with the scaling equation for soliton energy . therefore , we believe that the bound energy of multi - magnon complex for a quantum spin-@xmath13 model near the it point is characterized by the same critical exponents and by the identical scaling parameters as the soliton energy , though the corresponding scaling functions are different for different @xmath13 . it is known that such a resemblance takes place for the heisenberg ferromagnetic chain with an easy - axis anisotropy . our study shows that the multi - magnon complexes behave substantially as the classical objects for the frustrated model as well . for future it would be interesting to study the behavior of the classical spin model deeply in the spiral phase . we believe that it can help in understanding of unusual magnetization processes and shed light on the behavior of the multi - magnon complexes at high magnetic fields . mikeska and a. k. kolezhuk , in _ quantum magnetism _ , lecture notes in physics vol . * 645 * , edited by u. schollwck , j. richter , d. j. j. farnell , and r. f. bishop , eds . ( springer - verlag , berlin , 2004 ) , p. 1 . r. d. somma and a. a. aligia , phys . b * 64 * , 024410 ( 2001 ) ; r. jafari and a. langari , phys . rev . b * 76 * , 014412 ( 2007 ) ; a. avella , f. mancini , and e. plekhanov , eur . j. : cond . matter , * 66 * , 295 ( 2008 ) . r. kanamoto , h. saito , and m. ueda , phys . a * 67 * , 013608 ( 2003 ) ; g. m. kavoulakis , phys . rev . a * 67 * , 011601(r ) ( 2003 ) ; k. sakmann , a. i. streltsov , o. e. alon , and l. s. cederbaum , phys . a * 72 * , 033613 ( 2003 ) .

we introduce the hofer - zehnder-semicapacity of a symplectic manifold ( or-sensitive hofer - zehnder capacity ) with respect to a subgroup ( ) and prove that given a geometrically bounded symplectic manifold and an open subset admitting a hamiltonian free circle action with order greater than two then has bounded hofer - zehnder-semicapacity , where is the subgroup generated by the orbits of the action , provided that the index of rationality of is sufficiently great ( for instance , if|_{\pi_2(m)}=0 $ ] ) . we give several applications of this result . using p. biran s decomposition theorem , we prove the following : let be a closed khler manifold ( ) with \in h^2(m,\z)$ ] and a complex hypersurface representing the poincar dual of $ ] , for some . suppose either that vanishes on or that . then there exists a decomposition of into an open dense subset such that has finite hofer - zehnder-semicapacity and an isotropic cw - complex , where is the subgroup generated by the obvious circle action on the normal bundle of . moreover , we prove that if is subcritical then has finite hofer - zehnder-semicapacity . we also show that given a hyperbolic surface and endowed with the twisted symplectic form , where is the khler form on , then the hofer - zehnder-semicapacity of the domain bounded by the hypersurface of kinetic energy minus the zero section is finite if , where is the subgroup generated by the fibers of . finally , we will consider the problem of the existence of periodic orbits on prescribed energy levels for magnetic flows . we prove that given any weakly exact magnetic field on any compact riemannian manifold then there exists a sequence of contractible periodic orbits of energy arbitrarily small , extending a previous result of l. polterovich .

math0205030

the problem of the existence of periodic orbits for hamiltonian systems has a very far and rich history in the development of mathematics and physics . in fact , some of the major contributions on this question include the celebrated poincar - birkhoff theorem on the existence of periodic orbits for area preserving twist maps on the annulus and the morse - lyusternik - schnirelman theory on the existence of closed geodesics on compact riemannian manifolds . recently , a special class of symplectic invariants was introduced by h. hofer and e. zehnder @xcite as a link between rigidity phenomena of symplectic mappings and the existence of periodic orbits provided by classical variational methods . more precisely , they introduced an axiomatic definition of symplectic capacities in the following way : consider the class of all symplectic manifolds @xmath33 of fixed dimension equal to 2n . a symplectic capacity is a map @xmath34 that associates to @xmath2 a non negative number or @xmath35 and that satisfies the following axioms : * * monotonicity : * if there exists a symplectic embedding @xmath36 then @xmath37 * * conformality : * @xmath38 where @xmath39 is the ball of radius @xmath40 and @xmath41 is the cylinder over the @xmath42-plane . a distinguished symplectic capacity was introduced by hofer and zehnder and is directly related to the question of the existence of periodic orbits on prescribed energy levels of hamiltonian flows : [ defchz ] given a symplectic manifold @xmath2 , define the hofer - zehnder capacity of @xmath18 by @xmath43 where @xmath44 is the set of _ admissible hamiltonians _ @xmath45 defined on @xmath18 , that is , * @xmath46 , where @xmath47 is the set of _ pre - admissible hamiltonians _ defined on @xmath18 , that is , @xmath45 satisfies the following properties : @xmath48 , there exist an open set @xmath49 such that @xmath50 and a compact set @xmath51 satisfying @xmath52 ; * every non - constant periodic orbit of @xmath53 has period greater than 1 . intuitively , this capacity measure the sufficient oscillation for a pre - admissible hamiltonian to have fast periodic orbits , that is , of period less than one . it is easy to see that it satisfies the monotonicity and conformality properties . remarkably , by making use of classical variational methods , hofer and zehnder @xcite showed that it also satisfies the non - triviality axiom , giving , in particular , a dynamical new proof of gromov s nonsqueezing theorem @xcite . moreover , this symplectic invariant has important consequences on the question of the existence of periodic orbits on prescribed energy levels of hamiltonian systems . in fact , it is easy to prove that if a symplectic manifold @xmath2 has _ bounded hofer - zehnder capacity _ , that is , if @xmath54 for every open subset @xmath55 with compact closure , then given any hamiltonian @xmath56 with compact energy levels , there exists a dense subset @xmath57 such that for every @xmath58 the energy hypersurface @xmath59 has a periodic solution . this result was improved by m. struwe in @xcite showing that if the energy hypersurfaces bound a compact submanifold then @xmath22 has total lebesgue measure . this condition is not necessary as was proved recently in @xcite . note that as an immediate consequence we have a proof of the weinstein conjecture @xcite for manifolds with bounded hofer - zehnder capacity , since we can construct in a neighborhood of a contact hypersurface @xmath60 a distinguished parametrized family of hypersurfaces with equivalent flows . unfortunately , it is , in general , a difficult task to determine when a symplectic manifold has bounded hofer - zehnder capacity . for @xmath61 , where @xmath62 denotes the canonical symplectic form , it follows by hofer - zehnder theorem @xcite . for surfaces , it was proved by k. siburg @xcite , that the hofer - zehnder capacity of an open subset @xmath63 coincides with the area of @xmath63 . however , the situation is dramatically different for higher dimensional manifolds . actually , hofer and viterbo showed in @xcite that @xmath64 where @xmath2 is any closed symplectic manifold and @xmath65 is the two dimensional disk of radius @xmath66 , with @xmath40 satisfying the condition @xmath67 , where @xmath68 is the _ index of rationality _ of @xmath2 , that is , @xmath69}\om;\ \ [ u ] \in \pi_2(m ) \text { satisfies } \int_{[u]}\om > 0\bigg\}.\ ] ] if the above set is empty , we define @xmath70 . recall that a 2-form @xmath71 on @xmath18 is _ weakly exact _ if it is closed and @xmath70 . this result was extended by g. lu @xcite to noncompact symplectic manifolds satisfying certain conditions at the infinity . more precisely , he showed that @xmath72 provided that @xmath2 is geometrically bounded and @xmath73 . a symplectic manifold @xmath2 is called _ geometrically bounded _ if there exists an almost complex structure @xmath74 on @xmath18 and a riemannian metric @xmath75 such that * @xmath74 is uniformly @xmath71-tame , that is , there exist positive constants @xmath76 and @xmath77 such that @xmath78 for all tangent vectors @xmath79 and @xmath80 to @xmath18 ; * the sectional curvature of @xmath75 is bounded from above and the injectivity radius is bounded away from zero . the closed symplectic manifolds are clearly geometrically bounded ; a product of two geometrically bounded symplectic manifolds is also such a manifold . moreover , it was proved in @xcite that the cotangent bundle of a compact manifold endowed with any twisted symplectic form is geometrically bounded ( a proof of this fact can also be found in @xcite ) . a _ twisted symplectic structure _ on a cotangent bundle @xmath81 is a symplectic form given by @xmath26 , where @xmath62 is the canonical symplectic form , @xmath82 is the canonical projection and @xmath15 is a closed 2-form on @xmath18 . thus , using the monotonicity property , we can conclude that @xmath83 has bounded hofer - zehnder capacity , where @xmath6 is any compact manifold and @xmath62 is the canonical symplectic form ( because we can embed symplectically any bounded subset of @xmath84 into @xmath85 ) . actually , by the same reason , we can prove that @xmath86 has bounded hofer - zehnder capacity for any geometrically bounded symplectic manifold @xmath18 . it should be noted that it is necessary to consider product symplectic forms @xmath87 . in fact , there is an example due to zehnder @xcite of a symplectic form on @xmath88 with unbounded hofer - zehnder capacity . note that any symplectic form on @xmath89 is weakly exact . other known example of cotangent bundle with bounded hofer - zehnder capacity is @xmath90 endowed with any twisted symplectic form @xmath91 , where @xmath15 is a closed 2-form on @xmath92 and @xmath93 is the canonical projection . it was proved by j. jiang @xcite for @xmath94 and by v. ginzburg and e. kerman @xcite for the general case . we will consider here a refinement of the hofer - zehnder capacity by considering periodic orbits whose homotopy class is contained in a given subgroup @xmath0 of @xmath95 . more precisely , we define the hofer - zehnder semicapacity as follows ; it can be compared with the @xmath96-sensitive hofer - zehnder capacity defined by m. schwarz @xcite and lu @xcite and the relative symplectic capacity introduced by p. biran , l. polterovich and d. salamon @xcite : given a symplectic manifold @xmath2 and a subgroup @xmath3 define the _ hofer - zehnder g - semicapacity _ of @xmath18 ( or _ @xmath0-sensitive hofer - zehnder capacity _ ) by @xmath97 where @xmath98 is the set of _ @xmath0-admissible hamiltonians _ defined on @xmath18 , that is , * @xmath99 , that is , @xmath45 is pre - admissible ( see definition [ defchz ] ) ; * every nonconstant periodic orbit of @xmath53 whose homotopy class belongs to @xmath0 has period greater than 1 . we call it a _ semicapacity _ because given a symplectic embedding @xmath100 such that @xmath101 it can not be expected , in general , that @xmath102 , where @xmath103 is the induced homomorphism on the fundamental group . however , we can state the following _ weak monotonicity property _ ( for a proof see section [ proofthma ] ) : given a symplectic embedding @xmath104 such that @xmath101 then @xmath105 for every subgroup @xmath106 . in particular , if @xmath107 is injective we have that @xmath108 . note that obviously , @xmath109 for every subgroup @xmath3 . moreover , it can be show that if the hofer - zehnder @xmath0-semicapacity is bounded then there are periodic orbits with homotopy class in @xmath0 on almost all energy levels for every proper hamiltonian @xcite . a symplectic manifold @xmath2 has _ bounded hofer - zehnder @xmath0-semicapacity _ if for every open subset with compact closure @xmath110 , we have that @xmath111 we remark that the periodic orbits in @xmath112 given by the theorems of hofer - viterbo and lu are _ contractible_. thus , in terms of symplectic semicapacities , their theorems state the stronger result that @xmath113 for any geometrically bounded symplectic manifold @xmath2 and @xmath114 . by the weak monotonicity property , we can conclude then that @xmath115 is bounded , where @xmath116 is the obvious subgroup generated by the circle in the cylinder @xmath117 . in fact , the kernel of the induced homomorphism on the fundamental group of the symplectic embedding of @xmath118 into @xmath119 is given by @xmath120 . the aim of the first part of this paper is to generalize this result in the following way : note that @xmath121 has a natural hamiltonian free circle action with hamiltonian given by the angle form @xmath122 viewed here as a function on @xmath123 . we will consider the following : is the existence of a hamiltonian free circle action sufficient to ensure the boundedness of the hofer - zehnder semicapacity ( with respect to the subgroup of @xmath95 generated by the orbits of the action ) ? before we state our first result we need a previous topological definition related to free circle actions : let @xmath124 be a free circle action on a connected open subset @xmath6 of a manifold @xmath18 . we define the _ order of the action _ @xmath125 as the order of the cyclic subgroup of @xmath95 generated by the homotopy class of the orbits of @xmath125 . now , we are able to state our first theorem . in what follows we will denote the oscillation of a hamiltonian @xmath45 by @xmath126 . [ thma ] let @xmath2 be a geometrically bounded symplectic manifold and @xmath5 a connected open subset that admits a free hamiltonian circle action @xmath124 ( with period equal to @xmath127 ) given by the hamiltonian @xmath128 . suppose that the order of this action @xmath129 ( considered as an action on @xmath18 ) satisfies @xmath130 . given an open subset @xmath131 with compact closure , suppose that @xmath132 then , we have that @xmath133 where @xmath134 is the induced homomorphism on the fundamental group and @xmath135 is the subgroup generated by the orbits of @xmath125 . * note that the previous theorem can not be stated directly for @xmath6 because we need to assume that @xmath18 is geometrically bounded and it not clear that an open subset of a geometrically bounded symplectic manifold is geometrically bounded . * the circle action above needs to be hamiltonian . in fact , the example of herman - zehnder @xcite has unbounded hofer - zehnder capacity and it is easy to see that it admits a free symplectic circle action with infinite order . before we explain the main ideas involved in the proof of theorem [ thma ] , let us state some corollaries and applications . [ coreulerclass ] let @xmath2 be a geometrically bounded symplectic manifold . suppose that @xmath18 admits a free hamiltonian circle action generated by @xmath136 whose euler class @xmath137 \in h^2(m / s^1,\r)$ ] satisfies @xmath137|_{\pi_2(m / s^1 ) } = 0 $ ] . then , for any open subset @xmath110 with compact closure , we have that @xmath138 provided that @xmath139 . actually , it can be proved that given a free circle action on a manifold @xmath18 such that the euler class @xmath137 \in h^2(m / s^1,\r)$ ] satisfies @xmath137|_{\pi_2(m / s^1 ) } = 0 $ ] then the order of the action is infinite ( see section [ eulerclass ] ) . now , consider the finite subgroup @xmath140 . note that , since the circle action is hamiltonian , the induced action of @xmath141 on @xmath18 is symplectic . since the action is free , the quotient manifold @xmath142 has a induced symplectic structure @xmath143 defined uniquely by the property that the pullback of @xmath143 by the quotient projection is @xmath71 . moreover , @xmath142 has an induced free hamiltonian circle action ( with period @xmath144 ) with order greater than or equal to @xmath145 . consequently , we have the following immediate corollary : [ cor1 ] let @xmath2 be a geometrically bounded symplectic manifold . suppose that @xmath18 admits a free hamiltonian circle action generated by @xmath136 and let @xmath55 be an open subset with compact closure . if @xmath146 and @xmath147 , then @xmath148 where @xmath149 is the quotient projection , @xmath150 is the inclusion , @xmath71 is the induced symplectic form on @xmath142 and @xmath125 is the induced circle action . there are a lot of non - trivial examples of symplectic manifolds admitting such free hamiltonian circle actions . the first one that we will consider is the following : let @xmath18 be a compact manifold admitting a free circle action . then the lift of this action to the cotangent bundle of @xmath18 is hamiltonian with respect to the canonical symplectic form . in fact , it is hamiltonian with respect to any twisted symplectic form @xmath151 , where @xmath15 is a closed two - form on @xmath18 given by the pullback of a closed two - form @xmath152 on @xmath153 . thus , we have the following corollaries : [ cor2 ] let @xmath18 be a compact manifold admitting a free circle action @xmath125 whose euler class @xmath137 \in h^2(m / s^1,\r)$ ] satisfies @xmath137|_{\pi_2(m / s^1 ) } = 0 $ ] and @xmath15 a weakly exact two - form on @xmath18 given by the pullback of a two - form @xmath152 on @xmath153 . then the cotangent bundle @xmath154 has bounded hofer - zehnder @xmath155-semicapacity . let @xmath18 be a compact manifold such that @xmath156 and @xmath157 the total space of a circle bundle @xmath158 . then @xmath159 has bounded hofer - zehnder @xmath155-semicapacity . [ cor3 ] let @xmath18 be a compact manifold admitting a free circle action @xmath125 . then @xmath160 has bounded hofer - zehnder @xmath155-semicapacity for any @xmath161 and any weakly exact 2-form @xmath162 , where @xmath163 is the quotient projection . it is easy to construct specific non - trivial examples of such circle bundles . for instance , the lens spaces @xmath164 @xmath165 over @xmath166 , the heisenberg manifold over the 2-torus and , more generally , any 2-step nilmanifold @xcite . in fact , it can be proved that the group of equivalence classes of circle bundles over a manifold @xmath18 is isomorphic to @xmath167 @xcite . thus , any non - trivial element of @xmath167 for a manifold @xmath18 such that @xmath156 corresponds to a non - trivial circle bundle with infinite order . the heisenberg manifold , for example , corresponds to the cohomology class of the area form on @xmath89 . in the section [ proofthmb ] , we show some applications of the theorem [ thma ] where the circle action is not a lifted action . we will consider circle actions given by the magnetic flow associated to monopoles on surfaces . more precisely , let @xmath168 be a closed riemannian manifold , @xmath15 a closed 2-form on @xmath18 and @xmath25 endowed with the twisted symplectic form @xmath91 , where @xmath62 is the pullback of the canonical symplectic form via the riemannian metric . when @xmath18 is a hyperbolic surface and @xmath15 is the area form , this manifold has remarkable properties and is related to mcduff s example of a symplectic manifold with disconnected contact - type boundary @xcite ( see section [ proofthmb ] , where we give a simple construction of such a manifold ) . the hamiltonian flow @xmath169 given by the kinetic energy @xmath170 with respect to @xmath26 is called the magnetic flow generated by the riemannian metric @xmath75 and the magnetic field @xmath15 . it can be show that , when @xmath18 is a surface of constant negative curvature and @xmath15 is the area form , the magnetic flow defines a free circle action with infinite order on certain subsets of @xmath25 . using this fact and the expanding completion of convex manifolds developed by eliashberg and gromov @xcite , we can prove the following theorem : [ thmb ] let @xmath18 be a surface of genus @xmath171 endowed with the hyperbolic metric @xmath172 and the khler form @xmath173 . let @xmath174 be the usual hamiltonian given by the kinetic energy @xmath175 and @xmath176 a positive real number . consider the open subset @xmath177 . then , @xmath178 as long as @xmath179 , where @xmath31 is the subgroup generated by the fibers of the unitary bundle . in particular , the periodic orbits have contractible projection on @xmath18 . on the other hand , we have that @xmath180 for every @xmath181 . by moser s theorem , we have the following corollary : let @xmath18 be a surface of genus @xmath182 and @xmath15 a symplectic form on @xmath18 . then there exists a riemannian metric @xmath75 on @xmath18 such that @xmath183 as long as @xmath184 , where @xmath185 and @xmath186 is the subgroup generated by the fibers of the unitary bundle as above . when @xmath18 is the 2-torus and @xmath15 is a symplectic form , it is well know that @xmath187 for every open subset @xmath188 with compact closure . in fact , it is easy to prove that @xmath189 is symplectomorphic to @xmath190 , where @xmath191 is the canonical symplectic form on @xmath192 @xcite . this result can also be obtained by the theorem [ thma ] . now , we will use theorem [ thma ] to show that for closed khler manifolds @xmath193 the possible obstruction to the boundedness of the hofer - zehnder capacity is a very thin subset of @xmath18 which , in many situations , can be explicitly described . to prove it , we will need the following nice decomposition result of p. biran @xcite which enable us to represent a khler manifold as a disjoint union of two basic components whose symplectic nature is very standard : [ bir ] let @xmath9 be a closed khler manifold with @xmath11 \in h^2(m,\z)$ ] and @xmath194 a complex hypersurface whose homology class @xmath195 \in h_{2n-2}(m)$ ] is the poincar dual to @xmath13 $ ] for some @xmath196 . then , there exists an isotropic cw - complex @xmath197 whose complement - the open dense subset @xmath198 - is symplectomorphic to a standard symplectic disc bundle @xmath199 modeled on the normal bundle @xmath200 of @xmath12 in @xmath18 and whose fibers have area @xmath201 . the symplectic form @xmath202 is given by @xmath203 where @xmath204 is the bundle projection , @xmath40 is the radial coordinate using a hermitian metric @xmath205 and @xmath206 is a connection form on @xmath19 such that @xmath207 . the form @xmath208 is uniquely characterized by the requirements that its restriction to the zero section @xmath12 equals @xmath209 , the fibers of @xmath204 are symplectic and have area @xmath201 and @xmath202 is invariant under the obvious circle action along the fibers . it is called standard because the symplectic type of @xmath210 depends only on the symplectic type of @xmath211 and the topological type of the normal bundle @xmath200 @xcite . let us recall that the pair @xmath23 is called _ subcritical _ @xcite if @xmath24 is a subcritical stein manifold , that is , if there exists a plurisubharmonic morse function @xmath125 on @xmath24 such that @xmath212 for every critical point @xmath213 of @xmath125 . it is equivalent to the condition that the dimension of @xmath214 ( that is , the maximal dimension of the cells of @xmath214 ) is strictly less than @xmath145 . [ thmc ] let @xmath9 be a closed khler manifold ( @xmath10 ) with @xmath11 \in h^2(m,\z)$ ] and @xmath194 a complex hypersurface whose homology class @xmath195 \in h_{2n-2}(m)$ ] is the poincar dual to @xmath13 $ ] for some @xmath196 . then there exists an open dense subset @xmath19 of @xmath18 symplectomorphic to a standard symplectic disc bundle over @xmath22 whose complement is an isotropic cw - complex @xmath215 such that if @xmath11|_{\pi_2(m)}=0 $ ] then @xmath216 where @xmath217 is the subgroup generated by the orbits of the obvious @xmath218-action on @xmath20 . if @xmath11|_{\pi_2(m ) } \neq 0 $ ] and @xmath17 , we have that @xmath219 moreover , if @xmath23 is subcritical and @xmath11|_{\pi_2(m)}=0 $ ] , then @xmath220 where @xmath221 . * in @xcite it is proved that the hypothesis that @xmath10 and that either @xmath11|_{\pi_2(m ) } = 0 $ ] or @xmath17 are not necessary to ensure that @xmath222 is finite . * the result above can not be stated directly for symplectic disk bundles and concluded as a property of symplectic disk bundles plus biran s decomposition result because symplectic disk bundles are not geometrically bounded . actually , a fundamental ingredient in the proof of theorem [ thmc ] is that , by the lefschetz theorem , the circle action on @xmath20 has the same order considered both as an action on @xmath20 itself and as an action on @xmath24 ( this construction is not necessary in @xcite since we do not need there the topological assumption on the order of the action ) . * under the hypothesis that either @xmath223 or @xmath11|_{\pi_2(m)}=0 $ ] , it was proved by p. biran @xcite that the _ gromov capacity _ of @xmath19 satisfies @xmath224 . these assumptions are not necessary as was proved recently by g. lu @xcite . * the result above for _ subcritical _ manifolds was proved by c. viterbo in @xcite . * when @xmath23 is subcritical we do not consider the case @xmath17 because , as was proved in @xcite , every subcritical polarization has degree 1 . we remark that the cw - complex @xmath214 above is given by the union of the stable manifolds of the gradient flow of a plurisubharmonic function @xmath125 defined on @xmath24 such that it can be explicitly computed in many examples @xcite . the next theorem shows that we can get a similar result in the neighborhood of a symplectic hypersurface of donaldson type @xcite : [ thmd ] let @xmath9 be a closed symplectic manifold ( @xmath10 ) with @xmath11 \in h^2(m,\z)$ ] and @xmath194 a symplectic hypersurface whose homology class @xmath195 \in h_{2n-2}(m)$ ] is the poincar dual to @xmath13 $ ] for some @xmath225 . then there exists @xmath226 such that if @xmath227 there exists an open neighborhood @xmath228 of @xmath22 such that if @xmath11|_{\pi_2(m)}=0 $ ] then @xmath229 where @xmath230 , as above , is the subgroup generated by the orbits of the obvious @xmath218-action on @xmath231 . if @xmath11|_{\pi_2(m ) } \neq 0 $ ] , we have that @xmath232 it is important to remark that an extension of biran s result for symplectic hypersurfaces considered above ( which is generally expected to remains valid ) will enable us to extend theorem [ thmc ] for general symplectic manifolds . the essential idea in the proof of theorem [ thma ] is to relate the hofer - zehnder capacity of a symplectic manifold endowed with a hamiltonian circle action with the hofer - zehnder capacity of its reduced symplectic manifold ( in the sense of marsden - weinstein ) with respect to this action . more precisely , we consider the diagonal hamiltonian circle action on @xmath233 whose reduced symplectic manifold @xmath234 is given by @xmath2 . then , given a pre - admissible hamiltonian @xmath235 we construct a @xmath218-invariant pre - admissible hamiltonian @xmath236 on @xmath157 whose reduced dynamics is given by a reparametrization of the hamiltonian vector field of @xmath45 on @xmath18 . thus , we can apply the results of hofer - viterbo and lu to get a periodic orbit for @xmath237 and so , by reduction , a periodic orbit for @xmath53 . the essential step is to show the non - triviality of the projected closed orbit and it is here that the hypothesis on the order of the action plays an essential role . the idea is to use the condition on the homotopy of the periodic orbit together with an upper bound on the period of the orbit to ensure that the orbit can not be tangent to the trajectories of the diagonal action . in the second part of this paper , we will consider the problem of the existence of periodic orbits on prescribed energy levels for a special class of hamiltonian dynamical systems given by the magnetic flows . this problem was first considered by v. arnold @xcite and s. novikov @xcite and for a discussion of the results in this area we refer to @xcite . it was proved by l. polterovich @xcite that for every nontrivial weakly exact magnetic field on a manifold whose euler characteristic vanishes , there exist non - trivial contractible closed orbits of the magnetic flow in a sequence of arbitrarily small energy levels . the proof uses the geometry of the hofer s metric in the group of hamiltonian diffeomorphisms and the fundamental fact that the displacement energy of the zero section of @xmath25 , with respect to the twisted symplectic form given by a non - vanishing magnetic field , is equal to zero . recently , e. kerman @xcite gives the same result for magnetic fields given by symplectic forms . in @xcite we proved the existence of contractible closed orbits for almost all low energy levels , provided that the magnetic field is also symplectic . we will prove here an extension of polterovich s theorem for any manifold , without the assumption on the euler characteristic . [ thme ] let @xmath18 be any closed riemannian manifold and @xmath15 a non - trivial weakly exact magnetic field . then there exists a sequence of arbitrarily small energy levels containing non - trivial contractible periodic orbits . the main idea in the proof is similar to that of theorem [ thma ] . in fact , we consider a lift of the magnetic flow to @xmath84 and use the topological condition on the periodic orbit to ensure the non - triviality of the projected periodic orbit by symplectic reduction . the details are given in the section [ proofthme ] . i am very grateful to gabriel paternain for many helpful comments and suggestions and to the centro de investigacin en matemtica , guanajuato , mexico for the hospitality while part of this work was in progress . i am also grateful to kai cieliebak , dusa mcduff and felix schlenk for very useful comments . before we begin the proof of theorem [ thma ] we need some auxiliary results . [ hzin2 ] let @xmath2 be a geometrically bounded symplectic manifold and @xmath238 endowed with the symplectic form @xmath239 , where @xmath240 is the canonical symplectic form on @xmath117 . let @xmath241 be an open subset with compact closure and suppose that @xmath242 , where @xmath243 is the projection onto the second factor . then given a hamiltonian @xmath244 , there exists a nonconstant periodic orbit @xmath245 of @xmath53 with period @xmath246 moreover , the homotopy class @xmath247 $ ] of @xmath248 belongs to the subgroup @xmath249 . since @xmath250 is compact , there exists a positive constant @xmath251 such that @xmath252 \subset t^*s^1 $ ] . let @xmath62 be the canonical symplectic form on @xmath192 and consider the symplectomorphism @xmath253 \to a : = \{(x , y ) \in \r^2 ; a \leq x^2+y^2 \leq 5a \}$ ] given by @xmath254 note that @xmath255 , where @xmath256 ) \simeq \pi_1(t^*p ) \to \pi_1(t^*m \times \r^2 ) \simeq \pi_1(t^*m)$ ] is the homomorphism induced on the fundamental group by the transformation given by the identity and @xmath169 in the first and second factors respectively . thus , we can suppose , without loss of generality , that @xmath261 is a connected two - dimensional compact submanifold with boundary , such that there exists an open disk of radius @xmath262 with @xmath263 distinct points ( @xmath264 ) @xmath265 and an orientation preserving diffeomorphism @xmath266 such that @xmath267 from a theorem of dacorogna and moser ( see lemma 2.2 in @xcite ) this @xmath268 can be required to be symplectic . thus , consider the hamiltonian @xmath269 given by @xmath270 which is obviously @xmath271 , since @xmath272 , where @xmath273 is a compact subset such that @xmath274 . then it is sufficient to prove that @xmath288 . to do it , suppose , by contradiction , that there exists @xmath289 such that @xmath290 has a nonconstant periodic orbit @xmath245 such that @xmath291 \in \psi_*g$ ] and the period of @xmath245 is less than 1 . then the image of @xmath245 by @xmath292 ( note that the image of @xmath245 is contained in @xmath293 , since @xmath294 outside @xmath293 ) is a periodic orbit of @xmath295 of the same period and @xmath296 \in \psi_*^{-1}(\psi_*g),\ ] ] a contradiction . [ [ construction - of - a - reparametrized - diagonal - circle - action - rho - on - n - times - ts1-and - its - symplectic - reduction . ] ] construction of a reparametrized diagonal circle action @xmath297 on @xmath298 and its symplectic reduction . ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ let @xmath299 , where @xmath300 is endowed with the symplectic form @xmath301 , where @xmath240 is the canonical symplectic form on @xmath117 . note that @xmath117 has a natural hamiltonian circle action given by the hamiltonian @xmath302 . let @xmath303 be such that @xmath304 and @xmath305 . consider the diagonal action @xmath306 on @xmath157 given by the hamiltonian @xmath307 and let @xmath308 be the diffeomorphism given by @xmath309 where @xmath124 is the flow generated by @xmath310 . the factor @xmath145 is not important in this section and it will be used only in the next section in the construction of a @xmath297-invariant pre - admissible hamiltonian from @xmath45 . to show that @xmath315 , it is more convenient to write @xmath268 as @xmath316 where @xmath317 is the projection onto the circle and @xmath318 is the unit vector field tangent to the fibers of @xmath117 such that @xmath319 . consequently , @xmath320 where @xmath321 is the hamiltonian vector field generated by @xmath310 and @xmath322 is the pullback of the angle form @xmath323 on @xmath218 to @xmath157 . thus , @xmath324 but note that @xmath325 and @xmath326 since the vectors are colinear . on the other hand , @xmath327 and @xmath328 because the vectors are orthogonal with respect to the product decomposition @xmath329 and by the definition of @xmath311 . we have then that , @xmath330 where the last equality follows from the fact that @xmath331 . now , note that @xmath332 because @xmath333 and @xmath334 . on the other hand , we have that @xmath335 because @xmath336 and @xmath337 ( note that @xmath322 here is the pullback of the angle form @xmath323 on @xmath218 to @xmath157 ) . consequently , @xmath338 as desired . we claim that the quotient projection @xmath342 is given by @xmath343 , where @xmath344 is the projection onto the first factor . in effect , note that , by the previous lemma , @xmath268 sends the orbits of @xmath345 to the orbits of @xmath346 . on the other hand , @xmath96 is the quotient projection at @xmath347 with respect to the trivial circle bundle given by the orbits of @xmath348 . [ [ construction - of - a - rho - invariant - pre - admissible - hamiltonian - hh - on - n - times - ts1-from - h . ] ] construction of a @xmath297-invariant pre - admissible hamiltonian @xmath236 on @xmath298 from @xmath45 . ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ in view of the last lemma , consider the map @xmath351 defined by @xmath352 such that @xmath353 is the quotient projection . now , let @xmath45 be a pre - admissible hamiltonian on @xmath63 . we will construct from @xmath45 a @xmath297-invariant hamiltonian defined on @xmath157 . firstly , fix a sufficiently small constant @xmath354 satisfying @xmath355 define the new hamiltonian by @xmath356 where @xmath357 is a @xmath271 function such that @xmath358 , @xmath359 , @xmath360 and @xmath361 for every @xmath362 $ ] ( see figure [ graph1 ] ) . thus , @xmath363 is the lift of @xmath364 by the quotient projection @xmath365 . it is easy to see that @xmath368 where @xmath369 is given by ( see figure [ graph2 ] ) @xmath370 in effect , it is clear that @xmath371 . moreover , @xmath372 where @xmath373 is the subset of @xmath157 given by @xmath374 and @xmath49 is an open set such that @xmath50 ( that exists because @xmath375 ) since @xmath376 . note that the interior of @xmath373 is not empty . [ [ existence - of - a - non - trivial - periodic - orbit - of - x_h - given - by - the - reduction - of - a - periodic - orbit - of - x_hh . ] ] existence of a non - trivial periodic orbit of @xmath53 given by the reduction of a periodic orbit of @xmath237 . ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ let us now prove that there exists a periodic orbit @xmath248 of @xmath53 of period less than @xmath380 and homotopy class @xmath247 \in g_\vr$ ] . note firstly that @xmath381 in effect , the image of @xmath382 under the projection @xmath383 is the graph of @xmath384 restricted to @xmath63 and the projection of @xmath385 is the graph of @xmath386 : note that @xmath387 has the hamiltonian vector field with respect to @xmath71 given by @xmath388 . thus , the reduced dynamics of @xmath237 at @xmath339 is a reparametrization of the dynamics of @xmath53 . since @xmath389 , the nonconstant periodic orbits of @xmath390 have period greater than or equal to those of @xmath53 . consequently , by the lemma [ mwreduction ] , it is sufficient to show that the periodic orbit of @xmath236 given by the proposition [ hzin2 ] is not tangent to a fiber given by the action of @xmath125 ( such that it is projected onto a singularity of @xmath53 ) . suppose that there exists a periodic orbit @xmath391 \to \u$ ] of @xmath237 tangent to an orbit of @xmath297 and whose homotopy class belongs to @xmath279 . we will show that the period @xmath392 of @xmath248 is strictly greater than @xmath393 ( see proposition [ hzin2 ] ) . now , let @xmath395 be an almost complex structure on @xmath157 compatible with @xmath311 and @xmath396 be the corresponding riemannian metric . we have that along @xmath248 , @xmath397 on the other hand , the period of the orbits of @xmath345 is equal to @xmath398 because the action on @xmath18 is free . consequently , since @xmath247 \in \pi_1(s^1)$ ] , the projection @xmath399 must wrap around itself at least @xmath129 times . hence , @xmath245 also must wrap around itself at least @xmath129 times and so the period of @xmath248 is given by @xmath400 but , by the definition of @xmath145 , @xmath401 actually , note that the inequality @xmath402 is equivalent to @xmath403 that is , @xmath404 which is exactly the condition on @xmath145 and @xmath405 . thus , it is sufficient to prove that @xmath406 which implies @xmath407 to conclude this inequality , note that @xmath408 where in the second equation we used the fact that @xmath395 defines an isometry . on the other hand , we have that for every @xmath409 , @xmath410 but note that , since @xmath248 is tangent to a fiber , it is projected onto a singularity of @xmath53 , that is , @xmath411 for every @xmath412 . consequently , we have that @xmath413 as desired . finally , note that the periodic orbit @xmath276 of @xmath237 given by the proposition [ hzin2 ] satisfies @xmath414 \in \pi_1(s^1)$ ] . consequently , the corresponding reduced periodic orbit @xmath248 satisfies @xmath247 \in g_\vr$ ] since @xmath415 . let @xmath18 be a surface of genus @xmath182 endowed with a riemannian metric @xmath75 of constant negative curvature . let @xmath25 be the tangent bundle of @xmath18 with the symplectic form @xmath62 given by the pullback of the canonical symplectic form of @xmath81 by the bundle isomorphism induced by @xmath75 . consider on @xmath18 the almost complex structure @xmath74 such that @xmath443 is the khler form . as defined in the introduction , the magnetic flow associated to @xmath75 and @xmath15 is the hamiltonian flow given by the kinetic energy @xmath170 with respect to the twisted symplectic form @xmath444 . it is easy to see that a curve @xmath445 defines an orbit of this flow if and only if @xmath446 . in other words , the trajectories are the curves with constant geodesic curvature . consequently , the dynamics of this magnetic flow can be explicitly described : on the energy levels greater than one half , the magnetic flow is topologically equivalent to the geodesic flow ( which has periodic orbits in every nontrivial homotopy class and does not have projected contractible periodic orbits , because there is no conjugate points ) ; on the sphere bundle , it coincides with the horocycle flow and hence is minimal , that is , all the orbits are dense ; and on the energy levels below than one half , all the orbits are closed and homotopic to the fibers of the sphere bundle . in particular , it defines ( after a suitable reparametrization ) a hamiltonian circle action with infinite order on @xmath447 ( since @xmath156 ) such that we can apply theorem [ thma ] . the problem is the condition on @xmath448 to be geometrically bounded . to circumvent this problem we need the following proposition : to fix the notation , let @xmath456 , where @xmath457 is the curvature map , be the pullback of the canonical symplectic form of @xmath81 by the riemannian metric and @xmath458 . define @xmath459 and @xmath460 . consider the liouville form @xmath461 and the 1-form @xmath462 note that @xmath463 and @xmath464 , where the last equality follows from the fact that @xmath465 defines a connection form on @xmath466 and that @xmath18 is a hyperbolic surface . let us compute the lie derivative of @xmath467 with respect to @xmath468 . initially , we have that , @xmath469 such that @xmath470 but , we have that @xmath471 consequently , @xmath472 thus , @xmath473 and so @xmath474 on the other hand , the lie derivative of @xmath452 with respect to @xmath475 is given by @xmath476 since @xmath477 consequently , we have that @xmath478 it was proved in @xcite that the tangent bundle of a compact manifold endowed with any twisted symplectic form is geometrically bounded . thus , consider the almost complex structure @xmath74 on @xmath25 such that @xmath481 defines a riemannian metric with bounded sectional curvature and positive injectivity radius . by the previous lemma , given any @xmath482 sufficiently small , the subset @xmath483 has contact type boundary , that is , its boundary is convex in the sense of eliashberg - gromov @xcite . thus , we can consider the set @xmath484 ( that we can identify with @xmath448 ) and a symplectic form @xmath450 on @xmath485 such that @xmath450 coincides with @xmath452 on @xmath486 and the vector field @xmath487 is an extension of @xmath453 to @xmath488 ( that is , @xmath489 coincides with @xmath490 on @xmath491 ) such that @xmath492 . note that @xmath450 is exact , since @xmath452 on @xmath486 is exact and @xmath450 restricted to @xmath493 is obviously exact . consider now the almost complex structure @xmath494 on @xmath485 given by @xmath495 on @xmath486 and by the pushforward @xmath496 of @xmath74 by the flow @xmath125 of @xmath487 on @xmath497 . by the construction , @xmath494 is invariant by @xmath125 on @xmath498 . define @xmath499 . by our previous discussion , we have that @xmath500 defines a geometrically bounded riemannian metric on @xmath486 . we need to prove that it also defines a geometrically bounded riemannian metric on @xmath493 . but , we have that @xmath501 , that is , @xmath502 consequently , since @xmath494 is invariant by @xmath125 on @xmath497 , we obtain that @xmath503 thus , the sectional curvature @xmath504 of @xmath500 on @xmath505 is given by @xmath506 and the injectivity radius @xmath507 is greater than or equal to @xmath508 but , @xmath500 has bounded sectional curvature and positive injectivity radius on @xmath486 , proving the proposition . by our previous discussion , the magnetic flow restricted to an energy level @xmath181 is topologically conjugated to the geodesic flow such that it has no projected contractible periodic orbits . consequently , we have that @xmath509 for every @xmath181 . on the other hand , we have that for energies @xmath510 the ( reparametrized ) magnetic flow defines a free hamiltonian circle action ( of period 1 ) on @xmath27 of infinite order on @xmath448 . by the previous proposition , given @xmath482 sufficiently small , there exists a geometrically bounded symplectic form @xmath450 on @xmath511 that coincides with @xmath452 on @xmath512 . thus , by theorem [ thma ] we have that for every @xmath354 there exists a finite constant @xmath513 such that , for every @xmath482 sufficiently small , @xmath514 but , @xmath515 before we give the proof of theorem [ thmc ] , we need some previous lemmas . the first one can be derived by the arguments in the proof of proposition [ tame ] and the fact @xmath24 has a convex boundary . however , for the reader s convenience we present an independent proof using the structure of stein manifolds : [ tame1 ] given a neighborhood @xmath63 of @xmath12 , there exists an exact geometrically bounded symplectic form @xmath297 on @xmath24 such that @xmath297 coincides with @xmath15 on the complement of @xmath63 . note that @xmath12 can be obtained as the zero set of a holomorphic section @xmath516 of a complex line bundle defined by the divisor @xmath12 . it can be show that the function @xmath517 given by @xmath518 is plurisubharmonic @xcite . the next lemma proved in @xcite shows that we can construct a new plurisubharmonic function on @xmath24 that coincides with @xmath125 on @xmath519 and whose gradient vector field is complete . since @xmath18 is closed , we have that @xmath533 defines a geometrically bounded riemannian metric on @xmath534 , that is , the sectional curvature of @xmath533 restricted to @xmath534 is bounded from above and the injectivity radius is positive . we need to prove that it also defines a geometrically bounded riemannian metric on all @xmath24 . given @xmath535 , there exists @xmath536 such that @xmath537 for some @xmath538 , where @xmath539 is the gradient flow of @xmath525 with respect to @xmath533 . but , we have that @xmath540 , that is , @xmath541 consequently , since the complex structure @xmath74 is invariant by @xmath539 , we obtain that @xmath542 thus , the sectional curvature @xmath504 of @xmath533 on @xmath505 is given by @xmath506 and the injectivity radius @xmath507 is greater than or equal to @xmath508 but , @xmath533 has bounded sectional curvature and positive injectivity radius on @xmath534 , proving that @xmath543 is geometrically bounded . [ nullset ] suppose that @xmath18 is subcritical and let @xmath545 be an open subset with compact closure . let @xmath297 be a geometrically bounded symplectic form on @xmath24 given by the lemma [ tame1 ] . then there exists a symplectomorphism isotopic to the identity @xmath546 such that @xmath547 . remember that the isotropic cw - complex @xmath214 is given by the union of the stable manifolds of the gradient vector field @xmath548 of a certain plurisubharmonic function @xmath125 with a complete gradient flow such that @xmath549 @xcite . since @xmath18 is subcritical , we have that @xmath550 and so there exists a hamiltonian isotopy @xmath551 compactly supported in an arbitrarily small neighborhood of @xmath214 such that @xmath552 . since @xmath214 is compact , there exists a neighborhood @xmath228 of @xmath214 such that @xmath553 . fix @xmath392 so large that @xmath554 . since @xmath555 moves @xmath228 away from itself , we have that @xmath556 because @xmath557 leaves @xmath558 invariant . since @xmath548 is conformally symplectic , we have that @xmath559 is the desired symplectomorphism . by theorem [ bir ] , there exists an isotropic cw - complex @xmath215 ( given by the union of the stale manifolds of the gradient vector field associated to a plurisubharmonic function defined on @xmath560 ) whose complement @xmath561 is symplectomorphic to a standard symplectic disc bundle @xmath199 over @xmath22 modeled on the normal bundle @xmath562 of @xmath22 in @xmath18 , where @xmath563 moreover , this symplectomorphism @xmath564 sends @xmath22 to the zero section of @xmath565 . note that @xmath566 has an obvious hamiltonian free circle action ( of period @xmath127 ) generated by the hamiltonian @xmath567 such that @xmath568 also generates a hamiltonian free circle action on @xmath569 . by the lefschetz theorem @xcite , we have that the inclusion @xmath570 induces an injective map on the fundamental group such that the order of the circle action on @xmath571 is equal to the order of the circle action @xmath572 generated by @xmath573 on @xmath574 . since the chern class of @xmath565 is given by @xmath575 $ ] , we have that if @xmath11_{\pi_2(m)}=0 $ ] then , by the proposition [ propeulerclass ] and the lefschetz theorem , the order of the action generated by @xmath310 is infinite . on the other hand , the order of the action is always grater than or equal to @xmath28 . in effect , note that the @xmath218-bundle @xmath576 is equivalent to the quotient by @xmath577 of a @xmath218-bundle @xmath578 with euler class @xmath579 $ ] such that the subgroup of @xmath580 generated by the orbits of the action has order greater than or equal to @xmath28 . let @xmath450 be the exact geometrically bounded symplectic form on @xmath24 given by the lemma [ tame1 ] that coincides with @xmath15 on the complement @xmath581 of a tubular neighborhood of radius @xmath582 of @xmath12 . note that @xmath583 also has a hamiltonian free circle action that coincides with @xmath584 on @xmath581 . now , let @xmath585 be an open subset with compact closure and @xmath482 sufficiently small such that @xmath586 . then , by the theorem [ thma],we have that if @xmath11_{\pi_2(m)}=0 $ ] then , @xmath587 where @xmath588 is the subgroup generated by the homotopy class of the orbits of @xmath584 . on the other hand , if @xmath11_{\pi_2(m ) } \neq 0 $ ] and @xmath17 we have the inequality @xmath589 finally , suppose that @xmath23 is subcritical and let @xmath590 be an open subset with compact closure and @xmath482 sufficiently small such that @xmath591 . then , by the lemma [ nullset ] , there exists a symplectomorphism isotopic to the identity @xmath592 such that @xmath593 . take @xmath482 sufficiently small such that @xmath594 . thus , by the weak monotonicity property , we conclude that @xmath595 in fact , by the symplectic neighborhood theorem , there exists a neighborhood @xmath228 of @xmath12 and a symplectomorphism @xmath596 between @xmath597 and a neighborhood @xmath395 of the zero section of the canonical disc bundle @xmath199 , where @xmath563 moreover , @xmath598 sends @xmath22 to the zero section of @xmath565 . now , notice that @xmath453 point towards the interior of the compact domains bounded by the hypersurfaces @xmath601 , that is , the subset @xmath602 has a convex boundary in the sense of eliashberg - gromov @xcite , where @xmath603 and @xmath604 is the radial coordinate of @xmath605 . thus , we can consider the set @xmath606 and a symplectic form @xmath450 on @xmath485 such that @xmath450 coincides with @xmath202 on @xmath607 and the vector field @xmath608 is a complete extension of @xmath453 to @xmath609 ( that is , @xmath610 coincides with @xmath490 on @xmath611 ) such that @xmath612 . note that @xmath614 has an obvious hamiltonian free circle action ( of period @xmath127 ) generated by the hamiltonian @xmath615 such that @xmath568 also generates a hamiltonian free circle action on @xmath616 . by the lefschetz theorem for symplectic hypersurfaces of donaldson type @xcite , there exists @xmath617 such that if @xmath227 then the inclusion @xmath618 induces an injective map on the fundamental group and so the order of the circle action on @xmath619 is equal to the order of the circle action @xmath572 generated by @xmath573 on @xmath620 . let @xmath621 . since the chern class of @xmath565 is given by @xmath575 $ ] , we have that , as in the proof of theorem [ thmc ] , if @xmath11_{\pi_2(m)}=0 $ ] then the order of the action generated by @xmath310 is infinite . on the other hand , the order of the action is always grater than or equal to @xmath28 . let @xmath450 be the geometrically bounded symplectic form on @xmath24 given by the lemma [ tame2 ] that coincides with @xmath15 on the complement @xmath581 of a tubular neighborhood of radius @xmath582 of @xmath12 . now , let @xmath626 be a relative compact open subset and @xmath482 sufficiently small such that @xmath627 . then , by the theorem [ thma ] , we have that if @xmath11_{\pi_2(m)}=0 $ ] then , @xmath587 where @xmath628 is the subgroup generated by the homotopy class of the orbits of @xmath584 . analogously , if @xmath11_{\pi_2(m ) } \neq 0 $ ] and @xmath629 we have the inequality @xmath589 a diffeomorphism of a symplectic manifold @xmath2 is called hamiltonian if it is given by the time one map of a time - dependent hamiltonian vector field . let @xmath630 be the group of hamiltonian diffeomorphisms of @xmath2 . it was discovered by hofer @xmath631 that this group carries a natural biinvariant finsler metric with a non - degenerate distance function . this distance is defined as follows . the lie algebra @xmath632 of @xmath630 consists of all smooth functions on @xmath18 satisfying a normalization condition which ensures that , when @xmath18 is open , @xmath633 iff @xmath598 has compact support , and , when @xmath18 is closed , @xmath633 iff @xmath598 has zero mean with respect to the volume form induced by @xmath71 . the adjoint action of @xmath630 on @xmath632 is the standard action of diffeomorphisms on functions . consider the @xmath634-norm in @xmath632 , @xmath635 this norm is invariant under the adjoint action , and thus defines a biinvariant finsler metric on @xmath630 . now , let @xmath636 , @xmath637 $ ] , a path of hamiltonian diffeomorphisms , that is , a curve @xmath638 \to ham(m,\om)$ ] . let @xmath639 be its normalized hamiltonian function , that is , @xmath640 for all @xmath641 . we define then the length of this curve by @xmath642 the hofer s metric is then defined by @xmath643 where the infimum is taken over all paths @xmath248 which join @xmath169 and @xmath268 . for a subset @xmath63 of @xmath18 define the displacement energy @xmath644 as the measure of the distance between the identity map and the set of @xmath645 which displaces @xmath63 from itself , in the sense that @xmath646 . using the geometry of the hofer s metric and that , for a manifold whose euler characteristic vanishes , the displacement energy of the zero section of @xmath81 , with the twisted symplectic form given by a non - vanishing magnetic field , is equal to zero , polterovich @xcite proved the following theorem : [ polterovich @xcite ] let @xmath18 be a closed manifold whose euler characteristic vanishes and @xmath15 a non - vanishing weakly exact magnetic field . then there exist contractible magnetic closed orbits on a sequence of arbitrarily small energy levels . the idea of the proof is the following @xcite : take a smooth function @xmath647 , @xmath648 , which equals @xmath649 on @xmath650 $ ] , vanishes on @xmath651 and is strictly increasing on @xmath652 . thus , the closed orbits of the ( normalized ) hamiltonian @xmath653 corresponds to reparametrized magnetic periodic orbits whose energy is less than @xmath654 . the key point is that the minimum set of @xmath598 coincides with the zero section of @xmath81 and so its displacement energy vanishes @xcite . it implies that the asymptotic non - minimality of @xmath598 is strictly less than 1 and thus that @xmath598 ( as a vector in the lie algebra @xmath632 ) does not generate a minimal geodesic in @xmath630 @xcite . then , using a result of lalonde and mcduff @xcite , we conclude that the hamiltonian @xmath598 has a non - constant contractible closed orbit , since @xmath15 is weakly exact . it should be noted that the contractibility of the periodic orbits given by lalonde and mcduff follows by the fact that the proof of theorem 5.4 of @xcite relies on the hofer - zehnder capacity - area inequality ( theorem 1.17 of @xcite ) proved by floer , hofer and viterbo @xcite , where the periodic orbits are contractible ( see @xcite ) . consider the trivial @xmath218-bundle @xmath655 , the flat connection @xmath656 and the product metric @xmath657 . let @xmath658 be the pullback of the magnetic field @xmath15 to @xmath157 . the lifted circle action to @xmath659 is hamiltonian with respect to the twisted symplectic form @xmath660 , where @xmath661 is the canonical projection . in fact , let @xmath662 the vector field on @xmath659 that generates the circle action . then , @xmath663 where @xmath656 above is viewed as a function in @xmath659 . moreover , the form @xmath656 is invariant by the magnetic flow of @xmath659 , because taking a magnetic orbit @xmath248 in @xmath157 , we have that @xmath664 where @xmath665 denotes the lorentz force of the magnetic field @xmath658 ( defined as the bundle map @xmath666 uniquely characterized by the property that @xmath667 ) and in the third equation we used the fact that @xmath453 is parallel . thus , the magnetic flow on @xmath659 is equivariant with respect to the hamiltonian lifted action of @xmath218 to @xmath659 . on the other hand , it is easy to see that the marsden - weinstein reduced symplectic manifold with respect to the circle action on @xmath668 is given by @xmath669 . moreover , the pullback of the hamiltonian metric @xmath670 in @xmath25 by @xmath671 is given by @xmath672 . consequently , the reduced dynamics of the magnetic flow @xmath673 of @xmath157 on @xmath674 is given by the magnetic flow @xmath675 on @xmath18 . now , suppose that @xmath15 is weakly exact . since @xmath157 is the product , the fibers of @xmath157 are non - contractible and , in fact , the subgroup generated by the fibers in @xmath676 is isomorphic to @xmath677 . since the euler characteristic of @xmath157 vanishes and @xmath678 is weakly exact ( because @xmath679 ) , we can apply polterovich s theorem to conclude that @xmath673 has contractible closed orbits @xmath680 of arbitrarily small energy . since the subgroup in @xmath676 generated by the fibers of @xmath157 is cyclic infinite , these orbits can not be given by the fibers . this ensures that they are projected on non - constant contractible periodic orbits of @xmath675 . finally , note that the energy of these projected orbits is equal to @xmath681 where @xmath682 , because @xmath683 as @xmath145 grows to infinite . y. eliashberg , m. gromov , _ convex symplectic manifolds _ several complex variables and complex geometry , part 2 ( santa cruz , ca , 1989 ) , 135 - 162 , proc . sympos . pure math . , 52 , part 2 , amer . soc . , providence , ri , 1991 . m. herman , _ exemples de flots hamiltoniens do nt aucune perturbation en topologie @xmath271 na dorbites periodiques sur un ouvert de surfaces denergies _ , comptes rendus de lacadmie des sciences de paris * 412 * ( 1991 ) , 989994 . initially , note that @xmath11|_{\pi_2(m)}=0 $ ] iff @xmath420 = 0 $ ] , where @xmath421 is the universal cover . in effect , it is clear that if @xmath422 is exact , then @xmath11|_{\pi_2(m)}=0 $ ] . to prove the converse , note that , by the hurewicz s theorem , we have that @xmath423 thus , the condition that @xmath15 is weakly exact is equivalent that the integral of @xmath420 $ ] over any 2-cycle is equal to zero . but , by de rham s theorem , it is equivalent to @xmath422 be exact . to prove the claim , note that , since the fiber is path connected , @xmath427 is simply connected thus , it is sufficient to prove that @xmath427 covers @xmath18 . now , observe that @xmath429 where @xmath430 is the group of deck transformations on @xmath425 . on the other hand , @xmath424 commutes with the action of @xmath430 , since it is a lifted action from @xmath157 . consequently , it induces an action of a discrete group @xmath431 on @xmath427 such that @xmath432 proving the claim .

with the developments of the last decade on complete constant mean curvature ( - ) surfaces in the hyperbolic-space , many examples of such surfaces are now known . however , most of the known examples have regular ends . ( an end is irregular , resp . regular , if the hyperbolic gauss map of the surface has an essential singularity , resp . at most a pole , there . ) there are some known surfaces with irregular ends , but they are all either reducible or of infinite total curvature . ( the surface is reducible if and only if the monodromy of the secondary gauss map can be simultaneously diagonalized . ) up to now there have been no known complete irreducible - surfaces in with finite total curvature and irregular ends . the purpose of this paper is to construct countably many-parameter families of genus zero - surfaces with irregular ends and finite total curvature , which have either dihedral or platonic symmetries . for all the examples we produce , we show that they have finite total curvature and irregular ends . for the examples with dihedral symmetry and the simplest example with tetrahedral symmetry , we show irreducibility . moreover , we construct a genus one - surface with four irregular ends , which is the first known example with positive genus whose ends are all irregular .

0805.3751

let @xmath2 denote the unique simply connected complete @xmath1-dimensional riemannian manifold with constant sectional curvature @xmath3 , which we call the hyperbolic @xmath1-space . associated to a complete finite - total - curvature -@xmath0 ( constant mean curvature one ) conformal immersion @xmath4 of a riemann surface @xmath5 are two meromorphic maps called the hyperbolic gauss map and the secondary gauss map , which we denote by @xmath6 and @xmath7 respectively ( to be defined in the next section ) . using these two gauss maps , we can define two characteristics of the surface @xmath8 : 1 . it is known that @xmath5 is biholomorphic to a compact riemann surface with a finite number of points removed , and hence each end is conformally a punctured disk . therefore we may consider the order of the hyperbolic gauss map @xmath6 at each end , and an end is called _ regular _ if @xmath6 has at most a pole singularity at this end . if @xmath6 has an essential singularity , the end is called _ irregular_. 2 . although @xmath6 is single - valued on @xmath5 , the secondary gauss map @xmath7 might be multi - valued on @xmath5 , so we can have a nontrivial monodromy representation defined on the first fundamental group of @xmath5 . this monodromy group is a subgroup of @xmath9 , and if all members of this group can be diagonalized by the same conjugation , we say that the surface @xmath8 is _ reducible_. otherwise , we say that @xmath8 is _ irreducible_. ( irreducibility depends on a global behavior of the surface but not on individual ends . ) if a -@xmath0 immersion is reducible , the surface can be deformed preserving its hyperbolic gauss map @xmath6 and hopf differential ( @xmath10-part of the second fundamental form , see section [ sec : prelim ] ) . on the other hand , an irreducible surface is the only surface with given hyperbolic gauss map and hopf differential . recent progress in the theory of -@xmath0 surfaces in @xmath2 has led to the discovery of many new examples of these surfaces . many examples with regular ends are now known , and various properties of these surfaces are understood . bryant @xcite found a local description for these surfaces in terms of holomorphic data that initiated this recent progress . the last two authors @xcite@xcite developed the theory using bryant s description to find many examples and properties , and work in this direction has been continued by small @xcite , the authors @xcite@xcite , costa - sousa neto @xcite , earp - toubiana @xcite@xcite , yu @xcite@xcite , levi - rossman @xcite , barbosa - berard @xcite , do carmo and gomes and lawson and thorbergsson and silveira @xcite , and others . regarding properties of the ends of embedded examples , collin , hauswirth and rosenberg @xcite have recently shown that any embedded -@xmath0 surface of finite total curvature is either a horosphere or all of its ends are asymptotic to catenoid cousin ends . in @xcite it is further shown that any irregular end can not be embedded , and the limit points of such an end are dense at infinity . recently , pacard and pimentel @xcite established a method for attaching small handles between tangent horospheres and deforming to produce -@xmath0 surfaces , and this construction produces many embedded -@xmath0 surfaces of any genus . also , karcher @xcite has recently constructed periodic -@xmath0 surfaces with fundamental domains in several different types of compact quotients of @xmath2 . a typical example of an irregular end is the end of the enneper cousin , a surface first constructed by bryant @xcite . after that the last two authors @xcite constructed examples of genus zero and two irregular ends , and also many reducible -@xmath0 surfaces of genus zero whose ends are all irregular , using deformations from minimal surfaces . ( the conclusion of remark 4.4 in @xcite contains an error . the number of ends should be @xmath11 , and hence the genus of @xmath12 is zero . ) recently , daniel @xcite has investigated irregular ends from the viewpoint of nevanlinna theory . after @xcite , no further surfaces with irregular ends and finite total curvature had been constructed . ( however , such an example with infinite total curvature can be found in @xcite . ) in particular , until now no irreducible -@xmath0 surfaces with irregular ends and finite total curvature had been known . the purpose of the paper is to construct countably many @xmath0-parameter families of genus zero -@xmath0 surfaces with irregular ends and finite total curvature , which have either dihedral or platonic symmetries . we further show that examples with dihedral symmetries , and the simplest example with tetrahedral symmetry , must be irreducible . all of our examples have irregular ends of finite type in the sense of daniel @xcite . to do the construction , we start with the meromorphic data for the genus zero irreducible -@xmath0 surfaces with regular ends found in @xcite and @xcite and modify this data to make surfaces with irregular ends . the spirit of the construction is similar to the construction of trinoids in @xcite , where -@xmath0 surfaces with prescribed gauss maps are constructed by reflecting spherical triangles , and we use monodromy killing arguments like in @xcite and @xcite , but the techniques are brought to bear more intricately here . in section [ sec : prelim ] we give necessary preliminaries . as our construction is done by reflecting abstract spherical triangles , we discuss this in section [ sec : reflection ] , and introduce a method to construct -@xmath0 surfaces with irregular ends ( theorem [ thm : gen ] ) , which is proved in section [ sec : proof ] . as an application of the theorem , we construct examples of genus zero with either dihedral or platonic symmetries in section [ sec : platonic ] . finally , in section [ sec : torus ] , we construct a -@xmath0 surface of genus @xmath0 with four irregular ends , which is the first known example with positive genus whose ends are all irregular . here we recall from @xcite some fundamental properties of null meromorphic curves in @xmath13 . let @xmath14 be a meromorphic map defined on a riemann surface @xmath5 with a local complex coordinate @xmath15 . then @xmath16 is called _ null _ if @xmath17 . ( this condition does not depend on the choice of coordinate @xmath15 . ) let @xmath14 be a null meromorphic map . we define a matrix @xmath18 by @xmath19 and set @xmath20 then the pair @xmath21 is a meromorphic function @xmath7 and a meromorphic @xmath0-form @xmath22 on @xmath5 satisfying @xmath23 conversely , let @xmath7 be a meromorphic function and @xmath22 a holomorphic @xmath0-form on @xmath5 . then the ordinary differential equation is integrable and the solution @xmath16 is a null map into @xmath13 ( since we will always choose the initial condition to be in @xmath13 ) defined on the universal covering of @xmath24 . in general , @xmath16 might not be single - valued on @xmath5 itself , and @xmath16 may have essential singularities at poles of @xmath18 . we call the pair @xmath21 the _ weierstrass data _ of @xmath16 . let @xmath25 be a null meromorphic map of @xmath5 into @xmath13 with weierstrass data @xmath21 . we call @xmath26 the _ hyperbolic gauss map _ of @xmath16 . furthermore , we call @xmath7 in the _ secondary gauss map _ and @xmath27 the _ hopf differential _ of @xmath16 . we remark that the secondary gauss map @xmath7 satisfies @xmath28 let @xmath29 be a null meromorphic map . then for @xmath30 , @xmath31 is also a null meromorphic map . then the associated two gauss maps @xmath32 , @xmath33 , and the hopf differential @xmath34 of @xmath35 are @xmath36 where , for any matrix @xmath37 and any function @xmath6 , @xmath38 is the mbius transformation of @xmath6 by @xmath39 : @xmath40 let @xmath15 be a complex coordinate on a neighborhood @xmath41 of @xmath5 . now we consider the schwarzian derivatives @xmath42 and @xmath43 on @xmath41 of @xmath6 and @xmath7 , where @xmath44\,dz^2 \qquad \left ( ' = \frac{d}{dz}\right).\ ] ] the description of the schwarzian derivative depends on the choice of complex coordinate @xmath15 . however , any difference of two schwarzian derivatives , as a meromorphic @xmath45-differential , is independent of the choice of complex coordinate . note that the schwarzian derivative is invariant under mbius transformations : @xmath46 the following identity can be checked : @xmath47 conversely , the following lemma holds : [ lem : unique ] let @xmath48 be a pair of meromorphic functions on @xmath5 such that @xmath49 is not identically zero . then there exists a unique ( up to sign ) null meromorphic map @xmath50 such that @xmath6 and @xmath7 are the hyperbolic gauss map and the secondary gauss map of @xmath16 . now , for later use , we point out the following elementary fact from linear algebra : [ fact ] a matrix @xmath51 satisfies @xmath52 if and only if it is of the form @xmath53 we identify the minkowski 4-space @xmath54 , which has the canonical lorentzian metric @xmath55 of signature @xmath56 , with the space of @xmath57 hermitian matrices @xmath58 . more explicitly , @xmath59 is identified with the matrix @xmath60 the hyperbolic @xmath1-space can be defined as the upper component @xmath61 of the hyperboloid in @xmath54 with the induced metric . in @xmath58 this is represented as @xmath62 where @xmath63 . the complex lie group @xmath13 acts isometrically on @xmath2 by @xmath64 , where @xmath51 and @xmath65 . let @xmath5 be a riemann surface and @xmath50 a null holomorphic immersion . then @xmath66 is a conformal -@xmath0 immersion . conversely , for any conformal -@xmath0 immersion @xmath67 , there exists a null holomorphic immersion @xmath68 such that @xmath69 , where @xmath70 denotes the universal covering of @xmath5 . we call @xmath16 a _ lift _ of the conformal -@xmath0 immersion @xmath8 . let @xmath35 be another lift of @xmath8 . then , we have the expression @xmath71 for some matrix @xmath72 . let @xmath21 be the weierstrass data of the null map of the lift @xmath16 . then the first fundamental form @xmath73 and the second fundamental form @xmath74 are given by ( see @xcite , for example ) @xmath75 where @xmath27 is the hopf differential of @xmath16 . let @xmath66 be a complete conformal -@xmath0 immersion whose total gaussian curvature is finite . since the gaussian curvature @xmath76 of -@xmath0 surface is non - negative , finiteness of the total gaussian curvature is equivalent to @xmath77 where @xmath78 is the area element with respect to the first fundamental form . then there is a compact riemann surface @xmath79 and a finite number of points @xmath80 @xmath81 @xmath82 @xmath83 such that @xmath84 . we call each @xmath85 an _ end _ of @xmath8 . the hyperbolic gauss map @xmath6 on @xmath5 does not necessarily extend meromorphically on @xmath79 . the end @xmath85 is called a _ regular _ end if @xmath85 is at most a pole singularity of @xmath6 , and otherwise is called an _ irregular _ end . namely , an irregular end is an end at which the hyperbolic gauss map has an essential singularity . on the other hand , the hopf differential @xmath86 can always be extended as a meromorphic @xmath45-differential on @xmath79 . we denote by @xmath87 the order of the first non - vanishing term of the laurent expansion of the hopf differential @xmath86 at @xmath88 . ( by definition , @xmath89 at zeros of @xmath86 and @xmath90 at poles of @xmath86 . ) the following lemma is well - known ( cf . @xcite , lemma 2.3 of @xcite ) . [ lem : irreg ] an end @xmath85 is regular if and only if @xmath91 . now we set @xmath92 . then @xmath93 is a pseudometric of constant curvature @xmath0 with conical singularities ( see the appendix of this paper , and also proposition 4 of @xcite ) . it follows from and the gauss equation that @xmath94 hence @xmath95 is the pull - back of the fubini - study metric @xmath96 on @xmath97 induced by the secondary gauss map @xmath98 . by and we have @xmath99 in addition to having conical singularities at the ends @xmath85 , the pseudometric @xmath100 also has a conical singularity at each umbilic point @xmath101 of @xmath8 . the conical order of @xmath100 at each point is defined in the appendix of this paper . since @xmath73 is positive definite at @xmath102 , we have the following : at umbilic points , the conical order of @xmath93 equals the order of @xmath86 . in this section , we introduce a method to construct -@xmath0 surface with irregular ends . in @xcite , examples with regular ends are constructed from the holomorphic data @xmath6 ( the hyperbolic gauss map ) and @xmath86 ( the hopf differential ) . however , since the hyperbolic gauss map has essential singularities at irregular ends , it is hard to find an explicit expression of @xmath6 in our case . thus , our construction is based on the secondary gauss map @xmath7 and the hopf differential @xmath86 . though @xmath7 is not a well - defined meromorphic function on the surface , the pseudometric @xmath93 as in is a spherical metric with conical singularities which is well - defined on the surface . so , we start by constructing a spherical metric @xmath100 with conical singularities , using reflections of spherical triangles ( see @xcite ) . in this section , we consider abstract spherical triangles and their extensions by reflection . first , we shall define abstract triangles . we set @xmath103 and label each vertex @xmath104 , @xmath105 , @xmath106 of this closed triangular region @xmath107 as in figure [ fig : triangle ] . cc ( 11,11)(-5,-5 ) ( -1,0)(1,0)6 ( 0,-1)(0,1)6 ( 0,4)(1,-1)4 ( 0.2,-0.2)(0,0)[lt]@xmath106 ( 4,-0.2)(0,0)[lt]@xmath105 ( -0.2,4)(0,0)[rb]@xmath104 ( 0,0 ) ( 4,0 ) ( 0,4 ) ( 0.8,0.8)(0,0)[cc]@xmath108 ( 3.0,0.4)(0,0)[cc]@xmath109 ( 0.4,3)(0,0)[cc]@xmath110 ( 5.2,-0.4)(0,0)[cc]@xmath111 ( -0.1,5.3)(0,0)[cc]@xmath112 ( 2,0)(0,1)0.5 ( 2,0)(0,-1)0.5 ( 0,2)(1,0)0.5 ( 0,2)(-1,0)0.5 ( 2,2)(1,1)0.35355339 ( 2,2)(-1,-1)0.35355339 ( 2.2,-0.3)(0,0)[lt]@xmath113 ( -0.3,1.8)(0,0)[rt]@xmath114 ( 2.6,2)(0,0)[lt]@xmath115 & ( 11,11)(-5,-5 ) ( -5,0)(1,0)10 ( 0,-5)(0,1)10 ( 0,4)(1,-1)4 ( 0,0)(1,0)4 ( 0,0)(0,1)4 ( 4,0)(0,1)4 ( 0,4)(1,0)4 ( -4,0)(1,1)4 ( 0,-4)(1,1)4 ( 0.2,-0.2)(0,0)[lt]@xmath106 ( 4,-0.2)(0,0)[lt]@xmath105 ( -0.2,4)(0,0)[rb]@xmath104 ( 2,0)(0,1)0.5 ( 2,0)(0,-1)0.5 ( 0,2)(1,0)0.5 ( 0,2)(-1,0)0.5 ( 2,2)(1,1)0.35355339 ( 2,2)(-1,-1)0.35355339 ( 2.2,-0.3)(0,0)[lt]@xmath113 ( -0.3,1.8)(0,0)[rt]@xmath114 ( 2.6,2)(0,0)[lt]@xmath115 ( 1.2,1.2)(0,0)[cc]@xmath107 ( -1.6,1.2)(0,0)[cc]@xmath116 ( 1.2,-1.6)(0,0)[cc]@xmath117 ( 3,3)(0,0)[cc]@xmath118 an _ abstract spherical triangle _ is a pair @xmath119 , where @xmath120 is a riemannian metric defined on @xmath107 with constant curvature @xmath0 such that the three edges forming the boundary @xmath121 are geodesics . let @xmath110 , @xmath109 and @xmath108 be the interior angles of @xmath107 at the vertices @xmath104 , @xmath105 , and @xmath106 with respect to the metric @xmath120 respectively . the angles @xmath110 , @xmath109 and @xmath108 are positive , but may take any positive values , including those greater than or equal to @xmath122 . moreover , if @xmath123 , then we call the abstract spherical triangle @xmath119 _ non - degenerate_. the following fact is known : [ lem : trig ] let @xmath124 be a non - degenerate abstract spherical triangle , then the three angles @xmath110 , @xmath109 , @xmath108 satisfy the inequality @xmath125 conversely , if a triple of positive real numbers @xmath126 satisfies , then there exists a unique non - degenerate abstract spherical triangle such that the angles at @xmath104 , @xmath105 and @xmath106 are @xmath110 , @xmath109 and @xmath108 respectively . we take a double ( two identical copies ) of @xmath124 and glue them along their corresponding vertices and edges . then we get a conformal pseudometric on @xmath127 with three conical singularities with conical angles @xmath128 , @xmath129 , @xmath130 . consequently , the conical orders are @xmath131 respectively . if @xmath124 is non - degenerate , then @xmath132 . by corollary 3.2 of @xcite , the metric on @xmath127 is irreducible . then follows from ( 2.19 ) of @xcite . conversely , suppose that holds . by theorem 2.4 of @xcite , there exists a unique conformal pseudometric on @xmath127 with three conical singularities with conical angles @xmath133 . the uniqueness of this pseudometric implies that it has a symmetry and can be considered as a gluing of two identical non - degenerate spherical triangles . the above lemma implies that a non - degenerate abstract spherical triangle is uniquely determined by its angles @xmath134 . so we denote it by @xmath135 now we fix a non - degenerate abstract spherical triangle @xmath136 . since @xmath107 is simply connected , there exists a meromorphic function @xmath137 such that the pull - back of the fubini - study metric on @xmath138 by @xmath7 is @xmath120 . however , such a choice of the developing map has an ambiguity up to an @xmath9-matrix action @xmath139 ( @xmath140 ) . we shall now remove this ambiguity , using a normalization : there exists a unique ( up to sign ) developing map @xmath141 of @xmath120 satisfying ( see figure [ fig : triangle ] ) @xmath142 we call the developing map @xmath143 the _ normalized developing map _ of the triangle @xmath136 . let @xmath144 ( @xmath145 ) be the reflections of the triangle @xmath136 across the three edges , as in figure [ fig : triangle ] . let @xmath146 be the closed domain obtained by reflecting @xmath107 with respect to @xmath144 ( see figure [ fig : triangle ] ) . then each reflection @xmath144 can be regarded as an involution on the domain @xmath147 . [ lem : p - i ] let @xmath136 be a non - degenerate abstract spherical triangle and @xmath148 @xmath149 be the normalized developing map of @xmath136 . then the following identities hold : @xmath150 where @xmath151 to simplify the notation , we set @xmath143 . let @xmath152 , @xmath153 and @xmath154 be the three edges of @xmath107 which are stabilized by the reflections @xmath113 , @xmath114 and @xmath115 , respectively . since the edge @xmath152 is a geodesic , implies that @xmath155 lies on the real axis . then by the reflection principle , @xmath156 holds . similarly , by , @xmath157 is real . hence @xmath158 holds . then we have the second assertion . there exists a rotation @xmath39 centered at @xmath159 of the unit 2-sphere @xmath160 such that the image of @xmath161 is real . such an isometry @xmath39 of @xmath127 can be represented as a matrix @xmath162 . then we have @xmath163 lies on the real axis . hence by the reflection principle , @xmath164 holds , and then we have @xmath165 in particular , we have @xmath166 now we glue @xmath136 to a double of @xmath136 along corresponding edges and vertices , giving us a constant curvature one metric of three conical singularities on @xmath127 with conical angles @xmath128 , @xmath129 , and @xmath130 just as in the proof of lemma [ lem : trig ] . the open domains @xmath117 , @xmath116 and @xmath118 can be regarded as the interior of the second triangle in the double of @xmath136 . the monodromy of reflections for such metrics on @xmath127 are determined in @xcite . then , as shown at the bottom of page 82 of @xcite , we have @xmath167 with @xmath102 as in . let @xmath79 be a closed riemann surface and @xmath168 a simply connected closed domain bounded by three real analytic curves @xmath152 , @xmath153 and @xmath154 . we label the vertices @xmath104 , @xmath105 , @xmath106 of this closed triangular region @xmath169 such that @xmath152 , @xmath153 , and @xmath154 correspond to the three edges @xmath170 , @xmath171 and @xmath172 , respectively . the riemann surface @xmath79 is generated by @xmath169 if there are three anti - holomorphic reflections @xmath113 , @xmath114 , @xmath115 of @xmath79 stabilizing the three edges @xmath152 , @xmath153 , and @xmath154 of @xmath169 such that any point of @xmath79 can be contained in the image of @xmath169 by a suitable finite composition of these three reflections ( see figure [ fig : domain ] ) . , and is obtained from the triangle @xmath169 by reflections.,width=151 ] in this case , @xmath169 is called a _ fundamental domain _ of @xmath79 . a meromorphic @xmath45-differential @xmath86 on @xmath79 is said to be symmetric with respect to @xmath169 if @xmath173 holds . we let @xmath174 denote the set of conformal pseudometrics with conical singularities on @xmath79 . a metric @xmath175 is called _ symmetric _ with respect to the fundamental domain @xmath169 if it is invariant under the three reflections @xmath113 , @xmath114 and @xmath115 . moreover , if the restriction @xmath176 is isometric to @xmath136 , we denote the metric by @xmath177 a meromorphic function @xmath7 on @xmath79 is called _ @xmath9-symmetric _ with respect to @xmath169 if @xmath178 is symmetric with respect to @xmath169 . the following is the main theorem in this paper : [ thm : gen ] let @xmath79 be a closed riemann surface which is generated by a triangular fundamental domain @xmath179 by reflections , and label the vertices of @xmath169 as @xmath104 , @xmath105 and @xmath106 . let @xmath7 be an @xmath9-symmetric meromorphic function on @xmath79 with respect to @xmath169 , and let @xmath86 be a symmetric meromorphic @xmath45-differential on @xmath79 with respect to @xmath169 . suppose that : 1 . [ item : gen : angle ] there exist @xmath180 such that @xmath181 , with @xmath182 as in . 2 . [ item : gen : ord ] @xmath86 is holomorphic on @xmath183 and has a pole at @xmath105 with @xmath184 . [ item : gen : branch ] the branch point set of @xmath7 outside the poles of @xmath86 equals to the zero set of @xmath86 , and at each zero of @xmath86 , the order of @xmath86 is equal to the conical order of @xmath182 . let @xmath185 be the set of poles of @xmath86 . then , for some @xmath186 , there exist a smooth function @xmath187 satisfying @xmath188 and a @xmath0-parameter family of conformal -@xmath0 immersions @xmath189 for @xmath190 with the following properties : [ item : gen : concl-1 ] the hopf differential of @xmath191 is @xmath192 and @xmath193 . [ item : gen : concl-3 ] @xmath191 has irregular ends at @xmath194 . [ item : gen - concl-4 ] @xmath191 is symmetric with respect to @xmath169 . that is , the image of @xmath191 is generated by the reflections across the edges of @xmath195 . the construction method for proving theorem [ thm : gen ] will still work if @xmath196 . the stronger assumption @xmath184 in [ item : gen : ord ] is required only to make the ends irregular ( see lemma [ lem : irreg ] ) . the next theorem gives conditions which imply the surfaces @xmath191 in theorem [ thm : gen ] are irreducible . [ thm : irred ] under the assumptions in theorem [ thm : gen ] , if @xmath197 and @xmath198 for a local loop @xmath199 about @xmath105 , then @xmath191 is irreducible for @xmath200 sufficiently close to zero . the proofs of these theorems are given in section [ sec : proof ] . in this section , we construct examples of finite total curvature -@xmath0 surfaces with irregular ends and either dihedral or platonic symmetries . examples with dihedral symmetries and the simplest example with tetrahedral symmetry ( the case @xmath201 in and ) are irreducible . ( though we expect all the other examples to be irreducible , we have not checked them yet . ) let @xmath202 be an integer and @xmath203 the jorge - meeks @xmath204-noid is the complete minimal immersion @xmath205 given by the weierstrass representation as @xmath206 the @xmath207 extension @xmath208 of the dihedral group @xmath209 acts isometrically on the image of @xmath210 . there exists a one - parameter family of corresponding -@xmath0 immersions of @xmath211 to @xmath2 such that the hyperbolic gauss map is @xmath212 and the hopf differential is proportional to @xmath213 ( see @xcite , and figure [ fig : trinoid ] for the @xmath214 case ) . since @xmath215 @xmath216 , the ends of these corresponding surfaces are regular . ( 3,3)(-1.5,-1.5 ) ( -1.5,0)(1,0)3 ( 0,-1.5)(0,1)3 ( 0,0 ) ( 1,0 ) ( 0.30901699,0.95105652 ) ( 0.30901699,-0.95105652 ) ( -0.80901699,0.58778525 ) ( -0.80901699,-0.58778525 ) ( 0,0)(1.2,0 ) ( 0,0)(0.30901699,0.95105652 ) ( 0,0)(0.80901699,0.58778525 ) ( 1.1,-0.1)(0,0)@xmath0 ( 0.4,1.1)(0,0)@xmath217 ( -0.1,-0.1)(0,0)@xmath218 ( 0.6,0.2)(0,0)@xmath219 ( 1.18,0.65)(0,0)@xmath220 ( 0.1,-0.1)(0,0)@xmath104 ( 0.9,-0.1)(0,0)@xmath105 however , as we wish to produce -@xmath0 surfaces in @xmath2 whose ends are _ not _ regular , we now modify our choices for @xmath86 and @xmath7 to accomplish this : let @xmath221 be an integer and set @xmath222 then @xmath223 and @xmath224 are a meromorphic function and a meromorphic @xmath45-differential on @xmath225 respectively , which are symmetric with respect to the fundamental domain @xmath219 as in figure [ fig : fund - jm ] . moreover , @xmath226 satisfies the assumptions [ item : gen : angle][item : gen : branch ] of theorem [ thm : gen ] for @xmath227 and @xmath228 . then , for each @xmath204 and @xmath229 , there exists a one - parameter family of -@xmath0 immersions @xmath230 ( @xmath231 ) with symmetry group @xmath208 whose hopf differential is @xmath232 ( see figure [ fig : trinoid ] ) . the total gaussian curvature of these surfaces will be approximately @xmath233 . [ cols="^,^ " , ] let @xmath199 be a loop surrounding the end @xmath234 . since @xmath235 we have @xmath236 thus by theorem [ thm : irred ] , the surfaces are irreducible for sufficient small @xmath200 . it is well - known that there exists a minimal immersion @xmath237 with @xmath238 catenoid ends and tetrahedral symmetry @xcite and corresponding -@xmath0 surfaces in @xmath2 @xcite with regular ends . we denote by @xmath239 and @xmath240 the hopf differential and the gauss map of @xmath241 respectively . since each end is asymptotic to a catenoid , @xmath242 ( @xmath243 ) . then there exists @xmath238 umbilic points ( zeros of @xmath239 ) @xmath244 such that @xmath245 ( @xmath243 ) . the gauss map @xmath240 is a meromorphic function on @xmath246 whose branch points are @xmath247 each with branch order @xmath0 . ( 7,7 ) ( 4,6)(4.5,1)(1,2)(4,6 ) ( 4,6)(6,3)(4.5,1 ) ( 1,2)(6,3 ) ( 4,6)(2.75,1.5 ) ( 4.5,1)(2.5,4 ) ( 1,2)(4.25,3.5 ) ( 4,6.4)(0,0)@xmath248 ( 4.5,0.6)(0,0)[cc]@xmath249 ( 0.6,2)(0,0)[cc]@xmath250 ( 6.4,3)(0,0)[cc]@xmath251 ( 2.7,3.15)(0,0)[cc]@xmath252 ( 3.8,4)(0,0)[cc]@xmath169 moreover , @xmath5 is obtained by reflections across the edges of the fundamental domain @xmath169 , which is a triangle with vertices @xmath253 , @xmath254 , @xmath106 . ( see figure [ fig : fund - tet ] . see also the construction in @xcite . ) the hopf differential @xmath239 and the gauss map @xmath240 are symmetric with respect to the fundamental domain @xmath169 . consider the schwarzian derivative @xmath255 of @xmath240 , as in , where @xmath15 is the usual complex coordinate of @xmath246 . then @xmath256 is a meromorphic @xmath45-differential on @xmath246 . moreover , since @xmath240 is symmetric , @xmath256 is invariant under reflections about the edges of @xmath169 . the branch points of @xmath257 are poles of @xmath256 , and each pole of @xmath255 has order @xmath258 . so @xmath256 has @xmath238 poles of order @xmath45 at the @xmath259 ( @xmath243 ) and is holomorphic on @xmath260 . since the total order of a meromorphic @xmath45-differential on @xmath246 is @xmath261 , @xmath256 has @xmath238 zeros ( counting multiplicity ) . if there exists a zero of @xmath256 on the interior of @xmath169 , @xmath256 has at least @xmath262 zeros because @xmath246 consists of @xmath262 copies of the fundamental region @xmath169 , which is impossible . similarly , if a zero of @xmath256 lies on the interior of a edge of @xmath169 , @xmath256 has at least @xmath263 zeros , which is also impossible . if the vertex @xmath106 of @xmath169 is a zero of @xmath256 , there exist @xmath264 zeros , which is again impossible . since @xmath253 is a pole of @xmath256 , the set of zeros of @xmath256 must be @xmath265 , and @xmath266 for @xmath267 . for an integer @xmath221 , we define a meromorphic @xmath45-differential @xmath268 as @xmath269 where @xmath270 ( resp . @xmath271 ) is the symmetric product of @xmath272 copies of @xmath239 ( resp . @xmath229-copies of @xmath256 ) . since the poles of @xmath239 and the zeros of @xmath256 are @xmath265 , @xmath268 has the poles @xmath265 and is holomorphic on @xmath5 . more precisely , @xmath273 hold . since @xmath239 and @xmath256 are invariant under the reflections , so is @xmath268 . consider an abstract spherical triangle @xmath274 with @xmath275 and identify it with the fundamental domain @xmath169 . then we have a pseudometric @xmath276 . since neighborhoods of @xmath104 , @xmath105 and @xmath106 are generated by @xmath264 , @xmath264 and @xmath238 copies of the fundamental domain @xmath169 , respectively , @xmath277 is a pseudometric whose conical orders are all integers . hence by proposition [ prop : a ] in appendix , the developing map @xmath278 of @xmath277 is meromorphic on @xmath246 . then one can easily check that @xmath279 satisfies the assumptions [ item : gen : angle][item : gen : branch ] of theorem [ thm : gen ] . hence for each @xmath229 , there exists a one - parameter family of -@xmath0 immersions @xmath280 of @xmath5 into @xmath2 with irregular ends and tetrahedral symmetry . finally , we check irreducibility for @xmath201 . we may set @xmath281 and @xmath282 ( see page 221 of @xcite ) . hence the umbilic points are @xmath283 . by direct calculation , we have @xmath284 on the other hand , @xmath285 is a meromorphic function which branches at the umbilic points with branch order @xmath238 . then we have @xmath286 by the riemann - hurwitz formula . choose a rotation @xmath140 such that @xmath287 , where we set @xmath288 . then @xmath259 @xmath289 are not poles of @xmath290 , because the multiplicity of @xmath7 at @xmath259 is @xmath291 for each @xmath292 and @xmath293 . moreover , @xmath294 has a conical singularity at @xmath218 with conical order @xmath238 . hence , by symmetricity , we have @xmath295 , and we may write @xmath296 for some nonzero constants @xmath39 and @xmath297 . such a function @xmath7 exists if and only if all residues of the right - hand side vanish , which is equivalent to @xmath298 . then one can check irreducibility by direct calculation and theorem [ thm : irred ] . there are genus zero minimal surfaces in @xmath299 with catenoidal ends and the symmetry of any platonic solid ( @xcite ) . by similar arguments to the tetrahedral case above , one can obtain -@xmath0 immersions with irregular ends and any platonic symmetry . table [ tab : platonic ] shows the data for such surfaces . @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath110 & @xmath304 + dihedral & @xmath204 & @xmath45 & @xmath305 & @xmath306 & @xmath307 & @xmath308 & @xmath309 ' '' '' + tetrahedral & @xmath238 & @xmath238 & @xmath0 & @xmath310 & @xmath311 & @xmath312 & @xmath313 ' '' '' + octahedral & @xmath314 & @xmath264 & @xmath45 & @xmath310 & @xmath315 & @xmath316 & @xmath313 ' '' '' + octahedral & @xmath264 & @xmath314 & @xmath0 & @xmath317 & @xmath311 & @xmath312 & @xmath318 ' '' '' + icosahedral & @xmath319 & @xmath263 & @xmath1 & @xmath310 & @xmath320 & @xmath321 & @xmath313 ' '' '' + icosahedral & @xmath263 & @xmath319 & @xmath0 & @xmath322 & @xmath311 & @xmath312 & @xmath323 ' '' '' + take a real number @xmath324 and let @xmath325 where @xmath326 then we have @xmath327 since @xmath328 is non - degenerate ( by the assumption [ item : gen : angle ] ) , @xmath110 , @xmath304 and @xmath329 satisfy the relation . then for each @xmath109 sufficiently close to @xmath304 , there exists an abstract spherical triangle with angles @xmath110 , @xmath109 and @xmath330 . we identify the domain @xmath179 with this triangle . then by reflecting the metric , we have @xmath331 . let @xmath332 and @xmath333 the universal covering . by proposition [ prop : a ] in the appendix of this paper , the developing map @xmath334 of @xmath335 is defined on @xmath70 . to simplify the notation , we set @xmath336 then by the monodromy principle ( lemma [ lem : p - i ] ) , we have @xmath337 one may regard the triangle @xmath183 as generating @xmath70 by the three reflections across its edges . we denote these reflections by @xmath338 , @xmath339 , and @xmath340 ; that is , each @xmath341 ( @xmath145 ) is an antiholomorphic transformation on @xmath70 which preserves the @xmath292th edge of the triangle @xmath183 . we set @xmath342 let @xmath343 be a solution of the following ordinary differential equation on @xmath70 : @xmath344 such a solution @xmath345 is uniquely determined on @xmath70 . by , the right - hand side of the ordinary differential equation is traceless and so @xmath345 takes values in @xmath13 . then @xmath346 ( @xmath145 ) has the hopf differential @xmath347 and the secondary gauss map satisfies @xmath348 . however , @xmath349 also has the hopf differential @xmath350 and secondary gauss map @xmath351 , by . thus , by , we have @xmath352 this implies that @xmath346 and @xmath349 are both solutions of the same ordinary differential equation , and thus they differ only by the choice of initial values at the base point @xmath106 . so there exists a matrix @xmath353 such that @xmath354 since @xmath355 for @xmath356 , @xmath357 holds . thus we have @xmath358 in particular , the matrices @xmath359 and @xmath360 do not depend on @xmath200 nor the angle @xmath109 . since @xmath361 , implies that @xmath362 next we shall describe the matrix @xmath363 . we have @xmath364 where we used the fact @xmath365 is the identity . thus we have @xmath366 by lemma [ fact ] , the matrix @xmath363 can be written in the following form @xmath367 we also have @xmath368 we may assume that the small disk centered at @xmath104 consists of @xmath369-copies of @xmath169 . let @xmath370 be the branching order of @xmath7 at @xmath104 . by the condition [ item : gen : branch ] , we have @xmath371 since @xmath372 is the rotation of angle @xmath128 at @xmath104 , we have @xmath373 : @xmath374 on the other hand , one can easily check that the eigenvalues of @xmath375 are @xmath376 . then by , the eigenvalues of @xmath377 are @xmath378 , that is @xmath379 so we have @xmath380 which implies that @xmath381 this implies that the eigenvalues of @xmath382 are of the form @xmath383 for some integer @xmath384 . since @xmath382 is continuous with respect to the parameter @xmath200 , we can conclude that the eigenvalues of @xmath382 do not change by @xmath200 . in particular , @xmath385 on the other hand , since @xmath386 and @xmath361 , we have @xmath387 by , and , we have @xmath388 since @xmath389 , we have @xmath390 now we would like to find a real valued smooth function @xmath391 such that @xmath392 by the implicit function theorem , a sufficient condition for the existence of such a @xmath391 is @xmath393 and by , , and the assumption [ item : gen : angle ] , we have @xmath394 this proves the existence of such a @xmath391 @xmath395 for a sufficiently small @xmath186 . when @xmath396 , it holds that @xmath397 , so @xmath398 for sufficiently small @xmath200 ( @xmath399 ) , by continuity . now we set @xmath400{\frac{\nu_2(t , b(t))}{\nu_1(t , b(t ) ) } } \;.\ ] ] obviously , @xmath401 satisfies the ordinary differential equation for @xmath402 . in particular , @xmath401 has the hopf differential @xmath403 and the secondary gauss map @xmath404 . then by and , we get the following relations @xmath405 where @xmath406 since @xmath407 and @xmath408 , we have @xmath409 thus we have @xmath410 for @xmath145 . we now set @xmath411 by , the first fundamental form of @xmath191 is given by @xmath412 by the condition [ item : gen : branch ] of the theorem , it is positive definite , and thus @xmath191 is a conformal -@xmath0 immersion whose hopf differential is @xmath403 and the secondary gauss map @xmath404 . we shall now prove that the conformal -@xmath0 immersion @xmath191 is single - valued on @xmath413 . let @xmath414 be a non - negative integer and @xmath415 are sequences of three reflections @xmath416 on @xmath5 such that @xmath417 where @xmath333 be the universal covering . to show the single - valued property of @xmath191 , it is sufficient to show that @xmath418 . in fact , we have @xmath419 since @xmath7 is single - valued on @xmath5 , @xmath420 satisfies @xmath421 on the other hand , since @xmath422 is the secondary gauss map of @xmath423 , we have by that @xmath424 thus we can conclude that @xmath425 since @xmath426 ( @xmath427 ) , this implies that @xmath428 is single - valued on @xmath5 . moreover , by the assumption [ item : gen : ord ] and lemma [ lem : irreg ] , each end is irregular . the hopf differential of @xmath191 is @xmath429 , and the pseudometric @xmath430 defined in is @xmath431 . since they are symmetric with respect to @xmath169 , and imply that the first and second fundamental forms of @xmath191 are invariant under the reflections @xmath144 @xmath149 . then by the fundamental theorem of surfaces , @xmath191 is symmetric with respect to @xmath169 . let @xmath199 be a loop surrounding the point @xmath105 with the base point @xmath106 and @xmath432 the covering transformation of @xmath70 corresponding to @xmath199 . suppose that a neighborhood of @xmath105 is generated by @xmath433-copies of @xmath169 . then @xmath434 holds . thus we have @xmath435 here , by the argument of step 6 of the proof of theorem [ thm : gen ] , we have @xmath436 . hence at the base point @xmath106 , @xmath437 since @xmath438 . since @xmath345 is a solution of , it holds that @xmath439\,dz & = d \left[\left.\frac{\partial f_{t , b}}{\partial t } \right|_{(t , b)=(0,b_0)}\right]\\ & = \left.\frac{\partial } { \partial t } df_{t , b}\right|_{(t , b)=(0,b_0)}\\ & = \left.\frac{\partial}{\partial t } \left [ t f_{t , b } \begin{pmatrix } { \hat g}_b & -{\hat g}_b^2 \\ 1 & -{\hat g}_b \end{pmatrix}\frac{\hat q}{d\hat g_b } \right]\right|_{(t , b)=(0,b_0 ) } \\ & = \begin{pmatrix } \hat g & -\hat g^2 \\ 1 & -\hat g\hphantom{^2 } \end{pmatrix}\frac{\hat q}{d\hat g } \ ; , \end{aligned}\ ] ] because @xmath440 . then by the assumption and , we have @xmath441 this implies that @xmath442 is not constant on @xmath443 for sufficiently small @xmath186 , and hence we have @xmath444 hence the eigenvalues of @xmath445 are different from those of @xmath446 . the secondary gauss map @xmath447 of @xmath401 changes by the covering transformation @xmath432 as @xmath448 now , let @xmath449 ( see figure [ fig : endreflect ] ) . ( 11,7)(-5,-1 ) ( 0,4)(1,-1)4 ( 0,0)(1,0)4 ( 0,0)(0,1)4 ( -4,0)(1,1)4 ( -4,0)(1,0)4 ( 0.2,-0.2)(0,0)[lt]@xmath106 ( 4,-0.2)(0,0)[lt]@xmath105 ( -0.2,4)(0,0)[rb]@xmath104 ( -4,-0.2)(0,0)[lt]@xmath450 ( 2,0)(0,1)0.5 ( 2,0)(0,-1)0.5 ( 0,2)(1,0)0.5 ( 0,2)(-1,0)0.5 ( 2,2)(1,1)0.35355339 ( 2,2)(-1,-1)0.35355339 ( -2,2)(-1,1)0.35355339 ( -2,2)(1,-1)0.35355339 ( 2.2,-0.3)(0,0)[lt]@xmath451 ( -0.3,1.8)(0,0)[rt]@xmath452 ( 2.6,2)(0,0)[lt]@xmath453 ( -3.2,2)(0,0)[lt]@xmath454 we denote by @xmath454 the reflection about the edge @xmath455 . then we have @xmath456 let @xmath457 be a loop surrounding @xmath450 with base point @xmath106 , and let @xmath458 be the covering transformation corresponding to @xmath457 . then we have @xmath459 and @xmath460 so , to show irreducibility , it is sufficient to show that the matrices @xmath461 do not commute . by , @xmath462 holds . then by lemma [ fact ] , the off - diagonal components of @xmath39 are coincide and pure imaginary . set @xmath463 then we have @xmath464 and @xmath465 = a b - b a = \begin{pmatrix } 0 & -2\beta{\operatorname{im}}r\\ 2\beta{\operatorname{im}}r & 0 \end{pmatrix}\;,\ ] ] that is , @xmath39 and @xmath370 commute if and only if @xmath466 or @xmath467 is a real number . first , we consider the case @xmath466 . then @xmath468 is a diagonal matrix whose eigenvalues are different from @xmath469 for sufficiently small @xmath200 , because @xmath391 is not constant . in particular , the two eigenvalues of @xmath39 are distinct . this implies that the eigenspaces of @xmath39 coincide of those of @xmath470 . since @xmath39 is diagonal , this implies that @xmath471 is also a diagonal matrix , a contradiction . next , assume @xmath467 is real . then there exists a real number @xmath472 such that @xmath473 in this case , @xmath474 is a diagonal matrix whose eigenvalues differ from @xmath469 , for sufficiently small @xmath475 . then , by a similar argument to the previous case , we have @xmath476 is diagonal . if @xmath477 , this contradicts to and . hence @xmath39 and @xmath370 do not commute . in the final section , we construct an example of a -@xmath0 surface of genus one with four irregular ends . let @xmath478 be the lattice of gaussian integers of @xmath479 and let @xmath480 we consider @xmath79 as the square @xmath481\times[-\frac{1}{2},\frac{3}{2}]$ ] in @xmath482 , with opposite edges identified . take a triangle @xmath169 on @xmath79 as in figure [ fig : fund - torus ] . then @xmath79 is obtained from @xmath169 by reflecting across the edges of @xmath169 . ( 3,3)(-1,-1 ) ( -1,0)(1,0)2.8 ( 0,-1)(0,1)2.8 ( -0.5,-0.5)(-0.5,1.5)(1.5,1.5)(1.5,-0.5)(-0.5,-0.5 ) ( -0.5,1)(1.5,1 ) ( 1,-0.5)(1,1.5 ) ( -0.5,0.5)(1.5,0.5 ) ( 0.5,-0.5)(0.5,1.5 ) ( 0,0)(0.5,0)(0.5,0.5)(0,0 ) ( 0,0 ) ( 1,0 ) ( 0,1 ) ( 1,1 ) ( 0.3,0.15)(0,0)@xmath169 ( 0.55,-0.1)(0,0)[lc]@xmath483 ( 0.55,0.55)(0,0)[lb]@xmath484 ( -0.05,-0.1)(0,0)[rc]@xmath485 ( 0.1,-0.1)(0,0)[lc]@xmath248 ( 1.1,-0.1)(0,0)[lc]@xmath249 ( 1.1,1.1)(0,0)[lc]@xmath250 ( 0.1,1.1)(0,0)[lc]@xmath251 using the weierstrass @xmath486 function with respect to @xmath487 ( not with respect to @xmath488 ) , we set @xmath489 then @xmath86 has poles at @xmath490 , each with order @xmath264 . the @xmath486-function with respect to the square lattice has the following properties @xmath491 hence @xmath86 is symmetric with respect to @xmath169 . consider an abstract spherical triangle @xmath274 with @xmath492 and identify the triangle with the fundamental region @xmath169 . then the metric of @xmath274 can be extended to @xmath493 by reflections . since @xmath110 , @xmath304 and @xmath108 satisfy , @xmath277 is non - degenerate . let @xmath7 be the developing map of @xmath277 . since the conical orders of @xmath277 are integers , @xmath7 is well - defined on @xmath479 . now , we prove that @xmath7 is well - defined on @xmath79 . by the monodromy principle ( lemma [ lem : p - i ] ) , one can choose @xmath7 such that @xmath494 ( @xmath145 ) , where @xmath495 and @xmath451 , @xmath452 and @xmath453 are reflections along the edges @xmath171 , @xmath172 and @xmath170 , respectively . we denote by @xmath496 and @xmath497 the translations @xmath498 and @xmath499 respectively . then @xmath500 holds . so we have @xmath501 thus @xmath502 hold . this shows that @xmath7 is invariant under the action of the lattice @xmath488 . hence @xmath7 is a meromorphic function on @xmath79 . one can easily see that the same result as theorem [ thm : gen ] holds when @xmath503 , instead of @xmath330 . hence we have a one - parameter family @xmath504 of -@xmath0 immersions of @xmath505 into @xmath2 with irregular ends . for a compact riemann surface @xmath79 and points @xmath506 , a conformal metric @xmath120 of constant curvature @xmath0 on @xmath332 is an element of @xmath174 if there exist real numbers @xmath507 so that each @xmath85 is a conical singularity of conical order @xmath508 , that is , if @xmath120 is asymptotic to @xmath509 at @xmath85 , for @xmath510 and @xmath15 a local complex coordinate around @xmath85 . we call the formal sum @xmath511 the _ divisor _ corresponding to @xmath120 . for a pseudometric @xmath512 with divisor @xmath169 , there is a holomorphic map @xmath513 such that @xmath120 is the pull - back of the fubini - study metric of @xmath97 . this map , called the _ developing map _ of @xmath120 , is uniquely determined up to mbius transformations @xmath514 for @xmath515 . we have the following expression @xmath516 where @xmath517 is a covering projection . consider @xmath512 with divisor @xmath169 as in and with the developing map @xmath7 . since the pull - back of the fubini - study metric of @xmath97 by @xmath7 is invariant under the deck transformation group @xmath518 of @xmath332 , there is a representation @xmath519 such that @xmath520 the metric @xmath120 is called _ reducible _ if the image of @xmath521 is a commutative subgroup in @xmath9 , and is called _ irreducible _ otherwise . since the maximal abelian subgroup of @xmath9 is @xmath522 , the image of @xmath521 for a reducible @xmath120 lies in a subgroup conjugate to @xmath522 , and this image might be simply the identity . we call a reducible metric @xmath120 _ @xmath523-reducible _ if the image of @xmath521 is the identity , and _ @xmath524-reducible _ otherwise ( for more on this , see ( * ? ? ? * section 3 ) ) . [ prop : a ] let @xmath120 be a metric of constant curvature @xmath0 defined on @xmath525 whose conical order at each @xmath85 is an integer . then the developing map @xmath7 of @xmath120 is single - valued on the universal covering of @xmath5 . [ cor : a ] let @xmath120 be a metric of constant curvature @xmath0 defined on @xmath530 @xmath531 whose conical order at each @xmath85 is an integer . then the developing map @xmath7 of @xmath120 is single - valued on @xmath530 and extends as a meromorphic function on @xmath246 . moreover , the divisor of @xmath120 coincides with the ramification divisor of the meromorphic function @xmath7 . ruy5 j. l. barbosa and p. berard , _ eigenvalue and `` twisted '' eigenvalue problems , applications to cmc surfaces _ , j. math . pures appl . * 79 * ( 2000 ) , 427450 . r. bryant , _ surfaces of mean curvature one in hyperbolic space _ , astrisque * 154155 * ( 1987 ) , 321347 . j. berglund and w. rossman , _ minimal surfaces with catenoid ends _ , pacific j. math . * 171 * ( 1995 ) , 353371 . m. p. do carmo , j. de m. gomes and g. thorbergsson , _ the influence of the boundary behaviour on hypersurfaces with constant mean curvature in @xmath532 _ , comment . helvetici * 61 * ( 1986 ) , 429441 . m. p. do carmo and h. b. lawson , _ on alexandrov - berstein theorems in hyperbolic space _ , duke math . j. * 50(4 ) * ( 1983 ) , 9951003 . do carmo and a. m. da silveira , _ index and total curvature of surfaces with constant mean curvature _ , proc . * 110(4 ) * ( 1990 ) , 10091015 . p. collin , l. hauswirth and h. rosenberg , _ the geometry of finite topology bryant surfaces _ , ann . of math . * 153 * ( 2001 ) , 623659 . , _ the gaussian image of mean curvature one surfaces in @xmath2 of finite total curvature _ , advanced studies in pure mathematics 29 , minimal surfaces , geometric analysis and symplectic geometry 2000 . c. j. costa and v. f. de sousa neto , _ mean curvature @xmath0 surfaces of costa type in hyperbolic 3-space _ , thoku math . j. * 53 * ( 2001 ) , 617628 . b. daniel , _ surfaces de bryant dans @xmath2 de type fini _ , bull . * 126 * ( 2002 ) , 581594 . r. sa earp and e. toubiana , _ on the geometry of constant mean curvature one surfaces in hyperbolic space _ , illinois j. math . * 45 * ( 2001 ) , 371401 . , _ meromorphic data for mean curvature one surfaces in hyperbolic space _ , tohoku math . j. ( 2 ) 56 ( 2004 ) , no . 1 , 2764 . m. furuta , _ gluing spherical triangles _ ( kymen sankaku kei wo hariawaseru to ) , in japanese , sgaku tsshin * 6 * ( 2001 ) , 427 . m. furuta and y. hattori , _ two - dimensional spherical space forms _ , preprint . c. c. ges , m. e. e. l. galvo , b. nelli , _ a type weierstrass representation for surfaces in hyperbolic space with mean curvature equal to one _ , an . brasil cienc . 70 ( 1998 ) , no . 1 , 1 - 6 . h. karcher , _ hyperbolic constant mean curvature one surfaces with compact fundamental domains _ , global theory of minimal surfaces , 311323 , clay math . proc . 2 , amer . math . soc . , 2005 . s. kato , _ construction of @xmath204-end catenoids with prescribed flux _ , kodai math . j * 18 * ( 1995 ) , 8698 . m. kokubu , m. umehara and k. yamada , _ an elementary proof of small s formula for null curves in @xmath533 and an analogue for legendrian curves in @xmath533 _ , osaka j. math . 40 ( 2003 ) , no . 3 , 697715 l. l. de lima and w. rossman , _ on the index of mean curvature @xmath0 surfaces in @xmath2 _ , indiana univ . math . j. * 47(2 ) * ( 1998 ) , 685723 . c. mccune and m. umehara , _ an analogue of the up - iteration for constant mean curvature one surfaces in hyperbolic 3-space _ , diff . ( 2004 ) , no . 2 , 197207 . f. pacard and f. a. pimentel , _ attaching handles to bryant surfaces _ , j. inst . jussieu 3 ( 2004 ) , no . 3 , 421459 . l. rodriguez and h. rosenberg , _ half - space theorems for mean curvature one surfaces in hyperbolic space _ math . soc . * 126 * ( 1998 ) , 27632771 . h. rosenberg , _ bryant surfaces _ , the global theory of minimal surfaces in flat spaces ( martina franca , 1999 ) , pp . 67111 , lecture notes in math . 1775 , springer - verlag , 2002 . w. rossman and k. sato , _ constant mean curvature surfaces with two ends in hyperbolic space _ , experimental math . * 7(2 ) * ( 1998 ) , 101119 . w. rossman , m. umehara and k. yamada , _ irreducible constant mean curvature 1 surfaces in hyperbolic space with positive genus _ , thoku math . j. * 49 * ( 1997 ) , 449484 . , _ a new flux for mean curvature @xmath0 surfaces in hyperbolic @xmath1-space , and applications _ , proc . soc . * 127 * ( 1999 ) , 21472154 . , _ mean curvature @xmath0 surfaces with low total curvature in hyperbolic @xmath1-space i _ , hiroshima math . j. 34 ( 2004 ) , 2156 . , _ mean curvature @xmath0 surfaces with low total curvature in hyperbolic @xmath1-space ii _ , thoku j. math . 55 ( 2003 ) , 375395 . , _ period problems for mean curvature one surfaces in @xmath2 ( with application to surfaces of low total curvature ) _ , adv . pure math . 51 ( 2008 ) , 335387 . a. j. small , _ surfaces of constant mean curvature @xmath0 in @xmath2 and algebraic curves on a quadric _ , proc . amer . math . * 122 * ( 1994 ) , 12111220 . m. troyanov , _ metric of constant curvature on a sphere with two conical singularities _ , in `` differential geometry '' , lect . notes in math . vol . 1410 , springer - verlag , ( 1989 ) , 296306 . m. umehara and k. yamada , _ complete surfaces of constant mean curvature-@xmath0 in the hyperbolic @xmath1-space _ , ann . of math . * 137 * ( 1993 ) , 611638 . , _ a parameterization of weierstrass formulae and perturbation of some complete minimal surfaces of @xmath299 into the hyperbolic @xmath1-space _ , j. reine u. angew . * 432 * ( 1992 ) , 93116 . , _ surfaces of constant mean curvature-@xmath534 in @xmath535 with prescribed hyperbolic gauss map _ , math * 304 * ( 1996 ) , 203224 . , _ another construction of a cmc @xmath0 surface in @xmath2 _ , kyungpook math . j. * 35 * ( 1996 ) , 831849 . , _ a duality on cmc @xmath0 surfaces in hyperbolic @xmath1-space and a hyperbolic analogue of the osserman inequality _ , tsukuba j. math . * 21 * ( 1997 ) , 229 - 237 . , _ geometry of surfaces of constant mean curvature @xmath0 in the hyperbolic @xmath1-space _ , suugaku expositions * 10(1 ) * ( 1997 ) , 4155 . , _ metrics of constant curvature @xmath0 with three conical singularities on the @xmath45-sphere _ , illinois j. math . * 44(1 ) * ( 2000 ) , 7294 . y. xu , _ symmetric minimal surface in @xmath299 _ , pacific j. math . * 171 * ( 1995 ) , 203224 . z. yu , _ value distribution of hyperbolic gauss maps _ , * 125 * ( 1997 ) , 29973001 . , _ the inverse surface and the osserman inequality _ , tsukuba j. math . * 22 * ( 1998 ) , 575588 , _ surfaces of constant mean curvature one in the hyperbolic three - space with irregular ends _ , thoku math . j. * 53 * ( 2001 ) 305318 .

we investigate a resonantly modulated harmonic mode , dispersively coupled to a nonequilibrium few - level quantum system . we focus on the regime where the relaxation rate of the system greatly exceeds that of the mode , and develop a quantum adiabatic approach for analyzing the dynamics . semiclassically , the dispersive coupling leads to a mutual tuning of the mode and system into and out of resonance with their modulating fields , leading to multistability . in the important case where the system has two energy levels and is excited near resonance , the compound system can have up to three metastable states . nonadiabatic quantum fluctuations associated with spontaneous transitions in the few - level system lead to switching between the metastable states . we provide parameter estimates for currently available systems .

1401.4702

dispersive coupling of a quantum system to a mechanical or electromagnetic cavity mode has been attracting much attention recently . the coupling provides a means for quantum nondemolition measurement of the occupation number of the mode or of the populations in the energy levels of the system @xcite . the underlying read - out mechanism is the shift of the mode frequency or the system transition frequency , which depends on the state populations of the system or the mode , respectively . in the dispersive regime , a measurement erases information about the quantum phase , but does not cause transitions between energy levels . however , such transitions can happen due to coupling to a thermal reservoir , and also if the mode and/or the system are modulated by external fields . it is well understood that , through dispersive coupling , thermal interstate transitions cause decoherence @xcite . much less is known about the effects of periodic modulation and the interplay of the modulation and dephasing due to the coupling to a thermal reservoir . in this paper we address these problems . we consider a mode @xmath0 ( a harmonic oscillator ) coupled to a dynamical system @xmath1 . the mode and the system are also coupled to separate thermal reservoirs and can be modulated by periodic fields . the couplings and the modulation are assumed weak in the sense that the coupling energy is small compared to the interlevel energy spacing . in other words , the widths of the energy levels and the rabi energies are small compared to the level spacing . the modulation is assumed to be nearly resonant and will be described in the rotating wave approximation ( rwa ) . in distinction from the celebrated jaynes - cummings model @xcite , here the level spacings of the mode and the system are significantly different . for a dispersive @xmath2 coupling the major effect is not energy exchange , but rather it is the change of the level spacing depending on the state population , which occurs already in the first order in the coupling constant . semiclassically , this situation can give rise to multistability in the response to a modulating field as follows . for given modulating field parameters , the combined system may self - consistently support either large amplitude forced vibrations of mode @xmath3 , with the effective mode frequency tuned into good resonance with the driving field via the dispersive coupling , or small amplitude vibrations with an effective mode frequency far from resonance with the driving field . in each case , the vibration amplitude of mode @xmath3 sets the transition frequencies of the system @xmath1 . if system @xmath1 is modulated itself , this determines its quasi - steady - state level occupations . through the dispersive coupling , these level occupations tune the oscillator frequency into or out of resonance with the driving field , leading to the self - consistent mean - field multi - stability , see fig . [ fig : hysteresis ] . of the mode ( the squared mode amplitude is @xmath4 , where @xmath5 and @xmath6 are the mode mass and frequency , respectively ) . the abscissa shows the reduced squared modulation amplitude @xmath7 . both the mode and the system are resonantly modulated . in the rotating wave approximation , the model is described by eqs . ( [ eq : h_rwa1 ] ) , ( [ eq : liouville_osc ] ) and ( [ eq : bloch ] ) . the ratio of the relaxation rate of the two - level system @xmath8 to the relaxation rate of the mode @xmath9 is 30 ; the reduced amplitude of the field modulating the two - level system is @xmath10 . the reduced detunings of the modulating fields from the transitions frequencies of the mode and the system are , respectively , @xmath11 and @xmath12 . the reduced strength of the dispersive coupling is @xmath13 . the inset refers to @xmath14 , in which case the system does not show tristability . the unstable states are shown by dashed lines . the black vertical dashed line shows the modulation field used in fig . [ fig : graphic_solution ] below . ] the mean - field theory describes the semiclassical multistability of the system , but does not account for the role of _ fluctuations_. classical and quantum fluctuations unavoidably come along with relaxation as a consequence of coupling to a bath . in multistable systems , fluctuations cause interstate switching , _ even at zero temperature_. interestingly , where the dispersive coupling is weak , there is obviously no multistability ; however , where it is strong there is also no multistability , because the switching rate becomes comparable to the relaxation rate , and then the very notion of multistability becomes meaningless . in what follows we find the appropriate range of the coupling strength . we provide a general formulation of a mean - field theory and a theory of the switching rates in the important case where the typical relaxation time @xmath15 of the system @xmath1 is much smaller than the relaxation time @xmath16 of the mode . formally , the dispersive coupling hamiltonian @xmath17 is a function of a mode operator @xmath18 and a system operator @xmath19 , which commute with the isolated mode and system hamiltonians , respectively , in the absence of modulation . for example , @xmath18 can be the occupation number of the mode , @xmath20 , where @xmath21 and @xmath22 are the lowering and raising operators , while , if the dynamical system is a spin in a static magnetic field @xmath23 , @xmath24 can be the spin operator @xmath25 . this form of coupling assures that @xmath26 is independent of time in the interaction representation . for illustration , we consider a mode coupled to a two level system ( tls ) , each modulated by its own nearly - resonant field ( with @xmath27 ) : @xmath28 here @xmath29 , where @xmath30 are pauli operators which act on the tls . for nearly - resonant modulations , the detunings @xmath31 and @xmath32 of the modulation frequencies from the transition frequencies @xmath6 and @xmath33 are small compared to the transition frequencies themselves , and to their difference : @xmath34 . the condition on @xmath35 in particular justifies the approximation where only dispersive coupling is taken into consideration . the simplest form of the dispersive coupling of a mode and a tls is @xmath36 , which we now consider . we switch to the interaction representation using the unitary transformation @xmath37 . disregarding the fast - oscillating ( counter - rotating ) terms proportional to the modulation amplitudes @xmath38 , in the spirit of the rwa , we write the transformed hamiltonian @xmath39_{rwa}$ ] as @xmath40 with @xmath41 model ( [ eq : h_rwa1 ] ) describes , in particular , the dispersive coupling of a cavity mode to a two - level atom in cavity qed or to an effectively two - level josephson junction in circuit qed , which has been studied in many experiments , see e.g. refs . and references therein . more generally , @xmath26 may take on a more complicated form . in particular , the coupling does not have to be linear in @xmath20 . similarly , when system @xmath1 has more than two levels , the coupling hamiltonian may involve more complicated combinations of system operators as well . we will generally characterize the energy of dispersive coupling by a parameter @xmath42 , even where the coupling has a form different from @xmath43 in eq . ( [ eq : h_rwa1 ] ) ; we assume @xmath44 . in order to describe the dynamics in the presence of dissipation , we consider the density matrix @xmath45 of the coupled mode and system . assuming markovian dynamics in slow time , i.e. on times long compared to @xmath46 , we can write the equation of motion for @xmath45 in the interaction representation in the form : @xmath47.\ ] ] here , @xmath48 and @xmath49 are liouville operators , or superoperators , cf . ref . ; they describe , respectively , the dynamics of the mode and the system coupled to their thermal reservoirs but isolated from each other . below we will calculate the density matrix in the basis where operators @xmath50 and @xmath24 are diagonal . importantly , @xmath45 must remain hermitian through its evolution via eq . ( [ eq : rho__schematic ] ) . as a consequence , for any operator of the mode and the system @xmath51 , @xmath52 this condition applies also to @xmath48 and @xmath49 taken separately . in the frequently used model of dissipation where coupling of the mode to a thermal reservoir is taken to be linear in the operators @xmath53 , to the leading order in this coupling we have@xcite : @xmath54 + i[\rho,{\tilde h}_{_m}],\end{aligned}\ ] ] where @xmath55^{-1}$ ] is the mode planck number and @xmath56 is the decay rate . we note that eq . ( [ eq : liouville_osc ] ) is not limited to describing ohmic dissipation ; in the microscopic derivation it is assumed that @xmath57 and @xmath58 , and that the time is slow , cf . ref . . we assume that the renormalization of the parameters of the mode due to the coupling to the thermal reservoir has been incorporated into the parameter values . a simple form of relaxation for the two - level system is described via bloch equations . in this case operator @xmath59 in eq . ( [ eq : rho__schematic ] ) has the same form as @xmath60 , except that ( i ) the friction coefficient @xmath56 should be replaced by the parameter @xmath61 that gives the reciprocal lifetime of the two - level system , @xmath62 $ ] , where @xmath63 is the planck number ; ( ii ) operators @xmath21 and @xmath22 in the dissipation term should be replaced by @xmath64 and @xmath65 , respectively , with @xmath66 , and ( iii ) hamiltonian @xmath67 should be replaced with @xmath68 . further , we incorporate additional trasverse relaxation through a term @xmath69 in @xmath70 . to build intuition before the more technical discussion , we now provide a heuristic semi - quantitative picture of the adiabatic mean - field multistability for dispersive coupling to a tls ; the justification and the applicability conditions follow from the general analysis in sec . [ sec : mean_field ] below . suppose that the mode is in a state @xmath71 with @xmath72 . for the mode in this state , the detuning of the effective tls transition frequency from the driving field frequency is given by @xmath73 , as seen from eqs . ( [ eq : hamiltonian_rotating_frame ] ) and ( [ eq : h_rwa1 ] ) . in the adiabatic approximation we solve for the dynamics of the two - level system assuming that this frequency detuning is independent of time . using the well - known result for this problem , see e.g. ref . , we obtain the mean value of @xmath25 for a given value of @xmath74 : @xmath75^{-1}\nonumber\\ & & \gamma = \gamma_{_s}(2\bar n_{_s}+1 ) + \gamma_{\perp } , \quad \delta\omega_{_s}(m)=\delta\omega_{_s } -vm,\end{aligned}\ ] ] where @xmath76 is the decay rate of the spin components @xmath77 . through the interaction term @xmath78 in eqs . ( [ eq : hamiltonian_rotating_frame ] ) and ( [ eq : h_rwa1 ] ) , the average tls population difference @xmath79 acts back on the mode , changing its frequency by @xmath80 . importantly , the mode frequency depends on its degree of excitation , @xmath74 . such dependence is characteristic for nonlinear modes . here it comes from the resonant pumping of the tls . in turn , the typical values of @xmath74 in the stable mode state determine the detuning of the tls from the forcing @xmath81 that modulates it , @xmath82 , thus determining @xmath79 . the mutual tuning of the mode and the two - level system to resonance leads to multistability of the compound system . indeed , the stationary state mean occupation number of a resonantly modulated harmonic oscillator is given by the familiar expression @xmath83 $ ] . given the dependence of the mode frequency on its degree of excitation , @xmath74 , one might expect to find a self - consistency relation for the stationary state of the form : @xmath84 ^ 2\}.\ ] ] the resulting system of nonlinear equations ( [ eq : bloch ] ) and ( [ eq : naive_m ] ) can have multiple solutions . an example is the dependence of the squared mode vibration amplitude ( equal to @xmath85 , for a unit mode mass ) on the modulation strength , which is shown in fig . [ fig : hysteresis ] . in fact , the quantity plotted is the mean - field value of the `` center of mass '' variable @xmath86 of the quasi - stationary wigner distribution over the occupation numbers @xmath74 of the mode ; it is given by eq . ( [ eq : stationary_states ] ) , which for the considered model coincides with eq . ( [ eq : naive_m ] ) and justifies the above qualitative arguments . for the chosen parameters the mode can have up to three stable states at a time . in the mean - field picture where quantum and classical fluctuations are neglected ( see sec . [ sec : fluctuations ] for the role of fluctuations ) , this tristability is revealed by a hysteresis pattern with multiple switching between stable branches with the varying control parameter ( here , the driving strength ) . the onset of multistability can be understood from the graphical solution of eqs . ( [ eq : bloch ] ) and ( [ eq : naive_m ] ) , illustrated in fig . [ fig : graphic_solution ] . the solid lines on this figure show the resonant dependence of the reduced population difference of the tls ( see caption ) on the `` center of mass '' occupation number of the mode @xmath87 . it is given by eq . ( [ eq : bloch ] ) with @xmath74 replaced by @xmath87 . the resonance is a consequence of the tls frequency detuning @xmath88 being linear in @xmath87 . the dashed line shows the resonant dependence of the scaled squared amplitude of the modulated mode @xmath87 on the mode frequency . note that there is always an odd number of intersections ; for the blue and green curves it is equal to 1 and the intersection occurs for small @xmath87 outside the range shown in the figure . the corresponding regime corresponds to the single stable state of the modulated compound system . the case of 3 intersections corresponds to bistability , whereas 5 intersections correspond to tristability . the understanding of this pattern comes from the analysis of the bifurcation curves in sec.[subsec : multistability ] . as a function of the average occupation number of the mode @xmath89 ; for convenience , instead of @xmath79 we plot the reduced frequency shift @xmath90 counted off from @xmath91 . the dependence of @xmath79 on @xmath87 is resonant , which corresponds to the tuning of the two - level system in resonance with the modulating field by varying the mode occupation number . the blue , green , and red solid curves correspond to @xmath92 and 24 in eq . ( [ eq : bloch ] ) . other parameters are the same as in fig . [ fig : hysteresis ] . the dashed line shows the resonant response @xmath93 of a _ linear _ mode with eigenfrequency @xmath94 , cf . ( [ eq : naive_m ] ) ; here we use @xmath95 as a variable and find @xmath87 for a given @xmath95 and for @xmath96 ; such @xmath97 corresponds to the long - dash vertical line in fig . [ fig : hysteresis ] . the points show the solutions of eqs . ( [ eq : bloch ] ) and ( [ eq : naive_m ] ) ; the small-@xmath87 intersections of the dashed line with the blue and green solid lines occur for smaller @xmath98 and are not shown . ] the possibility of bistability of the response of a mechanical mode to resonant modulation in the situation where the mode is coupled to another system ( a massive classical particle diffusing along the mechanical resonator ) was considered earlier @xcite . such coupling is similar to dispersive coupling , as the diffusion changes the mode frequency and is in turn affected by the vibrations . however , in contrast to ref . , the analysis below is fully quantum , it is general as it is not limited to a specific coupling mechanism , the mean - field predictions are different ( for example , tristability ) , and most importantly , the class of systems to which the results refer is much broader . the rest of the paper is organized as follows . in sec . [ sec : adiabatic ] we develop equations of motion for the mode density matrix , and introduce the adiabatic approximation which allows a dominant and tractable part of the coupled set of equations to be isolated . then in sec . [ sec : mean_field ] we consider the semiclassical limit of large mode vibration amplitude , and derive mean field equations which govern the stationary state vibration amplitudes and phases of the mode . the mean field equations capture the multistability of the system and its critical slowing - down near bifurcation points in parameter space . in sec . [ sec : fluctuations ] we study fluctuations and switching between the mean - field metastable states , induced by random transitions of system @xmath1 through the @xmath99 coupling . finally , in sec . [ sec : conclusions ] we summarize our main conclusions and discuss the relevance for various experimental systems of current interest . the central assumption of our analysis is that the relaxation time @xmath15 of system @xmath1 is much smaller than the relaxation time @xmath16 of the mode [ which is given by @xmath100 for the model ( [ eq : liouville_osc ] ) ] . we exploit this separation of timescales to solve eq . ( [ eq : rho__schematic ] ) in an adiabatic approximation : first we solve for the evolution of system @xmath1 for a fixed state of the mode , and then examine how the resulting quasi - stationary states of @xmath1 feed back into the mode dynamics through the coupling @xmath78 . later we will see how quantum fluctuations of @xmath1 lead to switching between metastable states of the mode . we will formally assume that the energy of the dispersive coupling satisfies @xmath101 , although the actual condition of relevance is @xmath102 , as will be seen below . to begin , consider the case where the mode is set to be in an eigenstate @xmath71 of the operator @xmath18 : @xmath103 . the joint system - mode density matrix is given by the tensor product @xmath104 . if we neglect the slow mode dynamics generated by the liouvillian @xmath105 in eq . ( [ eq : rho__schematic ] ) , the reduced density matrix @xmath106 of @xmath1 obeys @xmath107,\ ] ] where @xmath108 and @xmath109 are operators acting only on @xmath1 . because the dispersive coupling @xmath78 commutes with @xmath18 , here it acts on the spin variables through its projection onto the selected mode state @xmath71 , @xmath110 . we will use the solutions of eq . ( [ eq : rho_s_dot ] ) as a basis to build up the solution to the full problem , eq . ( [ eq : rho__schematic ] ) . we solve eq . ( [ eq : rho_s_dot ] ) in terms of the eigenoperators @xmath111 of the superoperator @xmath112 , which appears on its right hand side : @xmath113 note that , from eq . ( [ eq : hermitivity ] ) , @xmath114 . therefore , if @xmath115 is an eigenoperator of @xmath112 with eigenvalue @xmath116 , then @xmath117 is also an eigenoperator with the eigenvalue @xmath118 . the eigenoperators @xmath115 with real eigenvalues can be chosen to be hermitian . if system @xmath119 has @xmath120 states ( @xmath121 for a tls ) , operators @xmath122 are @xmath123 matrices . because the superoperator @xmath124 in eq . ( [ eq : lambda__m ] ) does not commute with its adjoint , there is no guaranty that the set of eigenvectors ( operators ) @xmath125 forms a complete basis for system @xmath1 . specifically , under fine - tuned conditions , eq . ( [ eq : lambda__m ] ) may have less than @xmath126 linearly independent solutions and additional steps are needed to solve the dynamical problem ( [ eq : rho_s_dot ] ) . here we will not treat such secular cases , assuming that the operator @xmath124 is diagonalizable . this condition is generically satisfied for problems of physical interest including the specific examples considered below . furthermore , we will not consider the other structurally unstable case where some of the eigenvalues @xmath127 coincide , as such degeneracy is lifted by an infinitesimally small change of the parameters of @xmath128 . since eq . ( [ eq : rho_s_dot ] ) describes relaxation of the system @xmath129 , the eigenvalues @xmath130 have non - negative real parts . one of these eigenvalues ( with @xmath131 , for concreteness ) is equal to zero , which corresponds to the stationary state of system @xmath119 for the mode in state @xmath71 . the minimal value of re @xmath132 is the relaxation rate of system @xmath1 for a given @xmath71 . the relaxation time @xmath15 is given by the maximal value of @xmath133^{-1}$ ] calculated for the characteristic @xmath74 . we define the inner product of system-@xmath1 operators @xmath134 as @xmath135 $ ] , where tr@xmath136 is taken over the states of system @xmath119 . the expression @xmath137 $ ] then defines how the superoperator @xmath128 acts to the left ( in this case , on the operator @xmath138 ) ; it also defines the adjoint superoperator @xmath139 through @xmath140 . `` left '' eigenoperators @xmath141 of @xmath128 , which we denote with lowered indices , are defined through the equation @xmath142 the left and right _ eigenvalues _ of @xmath128 coincide : from eqs . ( [ eq : lambda__m ] ) and ( [ eq : left_eigenoperators ] ) , @xmath143 . however , the left and right _ eigenoperators _ are not hermitian conjugate . the non - degeneracy of the spectrum implies the orthonormality relation @xmath144 = \delta_{\alpha\beta},\ ] ] where we have imposed an additional normalization condition @xmath145 = 1 $ ] . the orthogonality relation ( [ eq : orthonorm ] ) holds only for the eigenoperators corresponding to the _ same _ mode state @xmath71 . this will be important below when we consider evolution with general mode states which are not diagonal in @xmath74 . over its relaxation time , system @xmath1 reaches a quasi - stationary state for the given mode - state @xmath71 . the reduced density matrix of @xmath1 in the stationary state is given by the right eigenoperator @xmath146 , corresponding to the zero - eigenvalue @xmath147 . note that the trace - preserving property of evolution dictates that @xmath148 = 0 $ ] , for any @xmath108 . by inserting the identity operator @xmath149 to the left of @xmath112 , we see that @xmath149 is a left eigenvector of @xmath112 with eigenvalue 0 . hence we set @xmath150 , such that the orthonormality condition ( [ eq : orthonorm ] ) gives tr@xmath151 . this is a very useful property , which we will employ below . we now use the solutions of the previous section to build up the solution to the full problem of coupled dynamics . to begin , for each @xmath152 we collect the set of eigenoperators @xmath111 , together with corresponding projectors onto the mode states , @xmath153 , to form a single operator @xmath154 which acts on the variables of both @xmath1 and @xmath3 . similarly , we define @xmath155 . we write the full density operator as @xmath156 where the operator @xmath157 acts only on the mode @xmath3 . note that @xmath158 in eq . ( [ eq : chi_alpha ] ) and @xmath159 in eq . ( [ eq : p_alpha ] ) do not commute . therefore the ordering of operators in the definition of @xmath160 in eq . ( [ eq : expansion ] ) is important . in explicit form , we have @xmath161 the set of complex parameters @xmath162 , for @xmath163 and @xmath164 , completely specifies the density matrix of the compound system . the asymmetry , that @xmath115 appears in eq . ( [ eq : rho_1 ] ) while @xmath165 does not , is accounted for once @xmath160 is added to its conjugate in forming the full density matrix @xmath45 . for @xmath160 of the form ( [ eq : rho_1 ] ) , the density matrix depends on time through the coefficients @xmath166 . this parametrization proves to be convenient for the analysis of the slow dynamics of the mode . thus , below we recast the master equation ( [ eq : rho__schematic ] ) as a coupled set of equations for the operators @xmath167 . the dynamical equation for @xmath159 is obtained by substituting @xmath160 in the form ( [ eq : rho_1 ] ) into the equation @xmath168 . we then project out the @xmath159 part by multiplying from the left by @xmath169 and taking the trace over the variables of system @xmath1 . using the relation @xmath170{\vert m ' \rangle}\nonumber\\ & & = ( \hat\lambda_m\chi^\alpha_m)p_\alpha^{mm ' } + i\chi^\alpha_m{\langle m \vert}[p_\alpha,\tilde h_i]{\vert m ' \rangle } , \end{aligned}\ ] ] which results from eq . ( [ eq : rho_s_dot ] ) , along with eq . ( [ eq : lambda__m ] ) for @xmath171 , we obtain : @xmath172 with @xmath173\right),\\ \hat k_{\alpha\beta}{\bf } p_\beta & = & \delta_{\alpha\beta}\hat { l}_{_m } { \bf } p_\alpha- { \rm tr}_{_s}\left[{\bm{}{\chi}}_\alpha^\dagger \hat { l}_{_m}(\bm{}{\chi}^\beta { \bf } p_\beta)\right].\end{aligned}\ ] ] here @xmath174 and @xmath175 are superoperators . both of them result from the dispersive @xmath176 coupling . for @xmath174 this is obvious , as this term explicitly contains the coupling hamiltonian @xmath43 and goes to zero where the coupling energy @xmath177 . the term @xmath178arises because the eigenoperators @xmath179 characterizing the dynamics of system @xmath1 depend on the mode state @xmath74 , via the coupling . as a consequence , the superoperator @xmath48 that describes dissipation of mode @xmath180 does not commute with @xmath181 , i.e. , @xmath182 . we note that , since @xmath183 , we have @xmath184 . also , since @xmath175 comes from the mode dissipation , its typical size is of order @xmath185 . we are interested in the effective nonlinear dynamics of the mode , described by the evolution of its reduced density matrix @xmath186 . since @xmath187 for all @xmath188 , we have @xmath189 however , as seen in eq . ( [ eq : p_equation ] ) , the evolution of @xmath190 is coupled to the behavior of all @xmath191 . thus to find @xmath192 we must examine the full set of coupled dynamical equations . if , as we assume , the relaxation rate @xmath193 of system @xmath119 is large compared to the mode relaxation rate @xmath194 and to the coupling parameter in @xmath43 ( divided by @xmath195 ) , the time evolution of @xmath196 , described by eq . ( [ eq : p_equation ] ) , is qualitatively different from the evolution of operators @xmath197 . the evolution of @xmath190 is governed by the mode liouvillian @xmath105 and @xmath78 , and therefore relaxation of @xmath196 is characterized by time @xmath16 . in contrast , the relaxation rate of @xmath159 for @xmath188 is determined by the values of @xmath198 . therefore , over time @xmath15 , all operators @xmath199 approach quasi - stationary solutions of eq . ( [ eq : p_equation ] ) for @xmath200 , calculated for the instantaneous @xmath201 . moreover , the matrix elements @xmath202 become small compared to the matrix elements of @xmath196 . this is because @xmath203 and @xmath204 , and as we assume @xmath205 $ ] is large compared to @xmath206 ; as we will see below , the actual constraint on @xmath207 is significantly weaker . operator @xmath196 is of primary interest , as it determines the density matrix of the mode ( [ eq : rho__m ] ) . from the arguments of the previous section , for times @xmath208 , to the leading order in @xmath209 , its time evolution is determined by equations @xmath210 the physical picture behind eq . ( [ eq : zero_order ] ) is that system @xmath211 reaches quasi - equilibrium , with distribution @xmath212 , for a given state @xmath74 of the mode , and then the mode ( and the system ) slowly evolve to the self - consistent stationary state given by equation @xmath213 . the superoperator @xmath214 , which describes the effect of the coupling to @xmath211 on the mode dynamics , has a simple form . indeed , @xmath215 , whereas @xmath216 gives the stationary density matrix of system @xmath211 for the mode being in state @xmath71 . in particular , for two nearby mode states @xmath71 and @xmath217 , with @xmath218 and @xmath219 , eqs . ( [ eq : chi_alpha ] ) and ( [ eq : nu_and_k ] ) give to leading order in @xmath220 : @xmath221 where @xmath222 $ ] is the average over the stationary state of system @xmath211 performed for the mode in a given state @xmath74 and @xmath223 . the quantity @xmath224 characterizes the change of level spacing of the mode due to its coupling to system @xmath1 . this change affects the distribution over mode states @xmath71 in a driving field by tuning the mode closer or further away from resonance . in turn , this affects the distribution of the system @xmath216 , which itself determines @xmath224 . it is this mechanism that leads to the multistability of the response in the mean - field approximation . we will assume that modulation of the mode is sufficiently strong that the mode is excited to states with @xmath225 . as we will check a posteriori , the characteristic width of the distribution over @xmath74 is then small compared to the characteristic @xmath74 . it is convenient to change from @xmath226 and @xmath227 to @xmath228 in the considered case the `` center of mass '' parameter @xmath229 is large , @xmath230 , and the major contribution to @xmath231 comes from terms with @xmath232 . in eq . ( [ eq : zero_order ] ) for the matrix elements @xmath226 one can change to @xmath231 , with account taken of eq . ( [ eq : semiclass_p_0 ] ) , and use the semiclassical approximation in which @xmath87 is quasicontinuous . a similar analysis can be done for the operator @xmath233 . this allows calculating the mode density matrix @xmath234 . from the normalization condition , in the semiclassical limit the equation for @xmath192 should have a form of the continuity equation @xmath235 here , @xmath236 is the probability current in variables @xmath237 . in the approximation ( [ eq : nu_m ] ) it is determined by the operator @xmath48 and @xmath95 , with @xmath238 . generally it has a drift part , which is proportional to @xmath234 but does not contain derivatives of @xmath192 , a diffusion part that contains first derivatives , and higher - order derivatives . the expansion in the order of the derivatives is an expansion in @xmath239 , and moreover , in @xmath240 , as will be also seen from the example below . we note that this is not the classical limit . the diffusion coefficient has a contribution from quantum fluctuations , which will be dominating in the example below . clearly , the dynamics of system @xmath211 is purely quantum . to the leading order in @xmath239 , one should keep in @xmath241 only terms @xmath242 , i.e. , @xmath243 . vector @xmath244 has the meaning of the force that drives the mode , in the rotating frame . with account taken of the explicit form of @xmath48 and @xmath214 , from eqs . ( [ eq : zero_order ] ) - ( [ eq : semiclass_p_0 ] ) @xmath245 function @xmath246 describes the effect of dissipation of the mode . for dissipation given by the standard linear friction operator ( [ eq : liouville_osc ] ) , @xmath247 . in a more general case the dependence on @xmath87 can be more complicated . it is important , however , that , since in our model the dissipation operator is independent of the modulation , the thermal reservoir on its own does not have a preferred vibration phase , and thus @xmath248 is independent of @xmath249 . the reservoir coupling lead to diffusion over phase , however , in this section we do not consider the diffusion ( it will be discussed later ) and the corresponding terms are absent in eq . ( [ eq : current_0 ] ) . the approximation @xmath250 corresponds to the mean - field approximation . the mean - field equations of motion for variables @xmath87 and @xmath249 are @xmath251 they can have stationary solutions @xmath252 which describe the stationary states of forced vibrations of the mode . in principle , equations of motion of the type ( [ eq : mean_field_explicit ] ) could also have periodic solutions that correspond to periodic vibration in the rotating frame . however , such solutions require that @xmath253 at least somewhere in phase space . from eq . ( [ eq : current_0 ] ) , @xmath254 has the same form as in the absence of modulation , where the only stationary state is @xmath255 , and therefore @xmath256 . the positions of the stationary states @xmath86 in the presence of modulation are given by @xmath257 ^ 2.\end{aligned}\ ] ] from eq . ( [ eq : mean_field_explicit ] ) , the stationary state ( [ eq : stationary_states ] ) is stable provided @xmath258 . in the absence of modulation , the state @xmath255 ( i.e. , the zero - amplitude state ) is stable on physical grounds and thus the condition @xmath258 is always satisfied for small @xmath87 ; quite generally @xmath259 for @xmath260 . if function @xmath261 is monotonic , eq . ( [ eq : stationary_states ] ) has one solution and the mode has only one stable state of forced vibrations for all modulation amplitudes @xmath97 . for nonmonotonic @xmath261 , the mode can have several stable vibrational states for a given @xmath97 , i.e. , it can display bi- or multi - stability . for large @xmath87 , the function @xmath261 is increasing with @xmath87 , except for the nongeneric case where @xmath262 decreases at least as fast as @xmath263 . then , since @xmath264 both for small and large @xmath87 , it can have only an even number of zeros . these zeros give the positions of the saddle - node bifurcation points @xmath265 , @xmath266 as seen from eqs . ( [ eq : stationary_states ] ) and ( [ eq : bif_point ] ) , if the modulation amplitude @xmath97 is tuned to the bifurcational value @xmath267 , for @xmath268 ( and for the corresponding @xmath269 given by equation @xmath252 ) stable and unstable stationary states @xmath270 merge . thus , the number of coexisting stable states changes by one once @xmath97 goes through @xmath271 . on the frequency detuning of this field @xmath272 as given by eqs . ( [ eq : stationary_states ] ) and ( [ eq : bif_point ] ) . the data refers to a mode dispersively coupled to a two - level system , with hamiltonian ( [ eq : h_rwa1 ] ) and with dissipation described by linear friction , see eq . ( [ eq : liouville_osc ] ) . the right dashed line shows the value of @xmath91 where the system displays tristability with varying @xmath97 ; this @xmath91 corresponds to the main part of fig . [ fig : hysteresis ] . the left dashed line shows @xmath91 used in the inset of fig . [ fig : hysteresis ] . the other parameters are the same as in fig . [ fig : hysteresis ] . ] the values of @xmath86 in the stable states depend on the mode driving strength @xmath97 . thus we consider the branches of stable states as functions of @xmath97 . these branches merge with the branches of unstable states at the bifurcation points ( [ eq : bif_point ] ) . the case where there are two stable - state branches and one branch of unstable states corresponds to vibration bistability and to the familiar @xmath129-shape dependence of the vibration amplitude of the mode on the modulation amplitude , cf . @xcite . for the model of a dispersively coupled mode and tls discussed in sec . [ subsec : intro_multistability ] , this dependence is shown in the inset of fig . [ fig : hysteresis ] . in the region of bistability @xmath261 has two extrema . if @xmath261 has four extrema for a given set of modulation field parameters , the mode has three stable states , as seen in the main panel of fig . [ fig : hysteresis ] . the bifurcational values @xmath271 themselves depend on other parameters of the system , and in particular on the detuning of the modulation frequency @xmath91 . the corresponding bifurcation curves are shown in fig . [ fig : bifurcation_diagram ] . each time any of these curves is crossed by varying parameters ( @xmath97 or @xmath91 ) , the number of the stable and unstable states changes by one . the bifurcation curves form pairs , which emanate from cusp point where the curves meet @xcite . such cusp points are analogous to the critical points on lines of first - order phase transitions . if we are close to a cusp point and go around it in the @xmath273 plane , without crossing the bifurcation curves , the number of stationary states does not change . if on the other hand , we move between the same initial and final values of @xmath273 but cross the bifurcation curves that merge at the cusp point , we go through a region where there is an extra stable and an extra unstable state . on the bifurcation curves this unstable state must merge with two different stable states . at the cusp point all three states merge together . the understanding of this topology makes the plot fig . [ fig : bifurcation_diagram ] convenient . in particular , if we move up along the right dashed line , we start from one stable state for small @xmath97 , then there are added a stable and an unstable state once the lowest bifurcation curve @xmath274 is crossed . when the next bifurcation curve is crossed , since it emanates from another cusp point , there is added another stable and unstable state . there are now three stable and two unstable states . as we cross the still higher curve @xmath274 , the first unstable state merges with one of the stable states and disappears , so that the system now has two stable and one unstable state . when the highest curve @xmath274 is crossed , there remains only one stable state . this behavior precisely corresponds to fig . [ fig : hysteresis ] . the `` beaks '' formed by the bifurcation curves in fig . [ fig : bifurcation_diagram ] open toward opposite sides . as follows from the above analysis , the tristability exists only in the range where the beaks overlap . in fact , the beaks do not go to infinity , they close up , but this occurs too far out to show on the figure . we now consider the vicinity of a bifurcation point , i.e. , we assume that @xmath97 is close to @xmath271 , and expand the right - hand sides of the equations of motion ( [ eq : current_0 ] ) and ( [ eq : mean_field_explicit ] ) about @xmath275 . here , the value of @xmath269 is given by the relation @xmath276 , in which @xmath268 and @xmath277 . if we limit the expansion of @xmath278 to linear terms in @xmath279 , we find that one of the eigenvalues of equations ( [ eq : mean_field_explicit ] ) for @xmath280 is equal to zero at the bifurcation point . correspondingly , for the parameters close to the bifurcation point , a combination of the dynamical variables @xmath281 becomes `` slow , '' i.e. there is a soft mode @xcite . the other eigenvalue of eqs . ( [ eq : mean_field_explicit ] ) remains of order @xmath194 at the bifurcation point ( note that we previously defined @xmath16 as the relaxation time of mode @xmath282 far from the bifurcation point ) . over a time of order @xmath16 , the linear combination of @xmath283 corresponding to the large ( negative ) eigenvalue of eq . ( [ eq : mean_field_explicit ] ) decays . after this decay , a relation between @xmath284 and @xmath285 is established , which to leading order can be obtained by linearizing equations ( [ eq : mean_field_explicit ] ) in @xmath286 and setting @xmath287 , with @xmath277 . this gives @xmath288 , where @xmath289\}$ ] with @xmath268 . the slow dynamics near the bifurcation point are controlled by the quantity @xmath290 , the soft mode , which happens to be the quadrature ( out of phase ) component of forced vibrations . at the bifurcation point , the deviation @xmath291 of @xmath292 from its bifurcational value @xmath293 is static , to linear order in @xmath283 : @xmath294 . close to the bifurcation point , the soft mode dynamics are governed by the nonlinear equation @xmath295 where the derivatives in the expression for @xmath296 are calculated for @xmath268 . as seen from eq . ( [ eq : slow_dynamics ] ) , if @xmath297 , the mode has a stable and an unstable stationary state . these states merge for @xmath277 and disappear for @xmath97 on the opposite side of @xmath271 . one of the best - known nonadiabatic effects in quantum systems is nonadiabatic transitions between stable states @xcite . in the case we study here , nonadiabatic corrections to the mean - field theory also lead to transitions between the stable mode states . in the conventional picture , nonadiabatic transitions usually involve tunneling , for low temperature . in contrast , in our case nonadiabatic transitions are induced by fluctuations that come along with the relaxation @xcite . specifically , these are fluctuations due to the randomness of emission and absorption of excitations of the thermal reservoir by system @xmath211 . these fluctuations lead to fluctuations of the level spacing of the mode through the mode - system coupling . classically , they correspond therefore to noise of the mode frequency . even though the noise is of quantum origin , it causes activated - like interstate transitions over an effective barrier in phase space , see eq . ( [ eq : switching_rate ] ) . the nonadiabaticity parameter is the ratio of the relaxation times @xmath209 . our analysis will be based on a perturbation theory . we will express functions @xmath197 in the equation for the mode density matrix ( [ eq : p_equation ] ) in terms of @xmath196 and then substitute them into equation ( [ eq : zero_order ] ) for @xmath196 . the major nonadiabatic corrections come from the term @xmath298 in eq . ( [ eq : p_equation ] ) ; the terms @xmath299 are proportional to @xmath194 and thus lead to small corrections to the parameters of the operator @xmath48 . to the leading order in @xmath209 , for time @xmath300 one can set @xmath301 for @xmath200 , which gives a slowly varying in time solution @xmath302 . in turn , this gives an extra term in eq . ( [ eq : zero_order ] ) for @xmath196 , which now reads @xmath303 with @xmath304\end{aligned}\ ] ] ( here the superoperator @xmath305 acts on the operator inside the bracket ) . we will be interested in the matrix elements @xmath306 between the mode states @xmath71 and @xmath217 . the calculation is simplified by the fact that operators @xmath43 , @xmath307 , and @xmath308 which appear in @xmath309 are all diagonal in @xmath74 . we will consider the semiclassical region of large @xmath310 and @xmath311 . as used above , in this region one can assume that @xmath87 , @xmath74 are quasicontinuous variables and expand the coupling hamiltonian , @xmath312 . then to the leading order in @xmath313 we have @xmath314 in addition to the leading order term displayed in eq . ( [ eq : nu_m ] ) , the function @xmath315 in the equation for the matrix elements of @xmath316 also has a term @xmath317 , i.e. , @xmath318\,p_0^{mm'}.\end{aligned}\ ] ] it is helpful to further process the @xmath319-term in this expression , by evaluating the quantity @xmath320 which appears inside the trace . this can be done by formally differentiating the equation @xmath321 over @xmath74 , using eq . ( [ eq : rho_s_dot ] ) , and by expanding @xmath322 in @xmath323 . the result is similar to the right - hand side of eq . ( [ eq : d_explicit ] ) , except for the extra factor @xmath324 and the fact that in the second trace one should replace the product of the operators @xmath325 and @xmath326 with their commutator . we note that an operator @xmath327 with respect to the variables of system @xmath211 can be written as @xmath328 $ ] . further , it is convenient to consider system-@xmath211 operators in the heisenberg representation in the rotating frame . from eqs . ( [ eq : rho_s_dot ] ) and ( [ eq : lambda__m ] ) , in this representation @xmath329 for @xmath330 ; similarly , @xmath331 . one can then define @xmath332.\ ] ] using this definition , one obtains @xmath333p_0^{mm'}\ ] ] with @xmath334\right.\nonumber\\ & \left.\times\left[\partial_\mu \hat h_i(\mu;t ) - \langle \partial_\mu\hat h_i(\mu)\rangle_{_s}\right]\right\rangle_{_s}.\end{aligned}\ ] ] here we used that , for real eigenvalues @xmath127 , operators @xmath335 are hermitian , whereas for the pairs of complex conjugate @xmath127 there are corresponding pairs of the hermitian conjugate operators @xmath335 . by its construction as the average of a correlator of the same operator over the states of system @xmath211 , the coefficient @xmath336 . clearly , @xmath337 is quadratic in the dispersive coupling constant @xmath42 contained in @xmath43 , and @xmath338 , i.e. , @xmath339 . note that @xmath337 , which we will see below plays the role of a diffusion constant , is small when the relaxation time of system @xmath1 is very short . this dependence captures the motional narrowing that occurs when system @xmath1 rapidly switches between its states . with account taken of the terms @xmath317 in equation for @xmath226 , the equation for the density matrix @xmath340 in @xmath237-variables takes the form of the fokker - planck equation @xmath341 where the vector @xmath342 has components @xmath343 and the drift vector @xmath344 is given by eq . ( [ eq : current_0 ] ) . function @xmath192 satisfies the semiclassical normalization condition @xmath345 . we do not consider the diffusion term that comes from the direct coupling of the mode to the thermal reservoir , as described by the operator @xmath48 . this term adds a contribution to the phase diffusion coefficient @xmath337 proportional to @xmath194 ; in addition , and importantly , this contribution is @xmath346 . operator @xmath48 also introduces diffusion along the @xmath87-variable , with a diffusion coefficient that scales as @xmath347 . taking this diffusion into account will not change the analysis below , and in particular will just renormalize the coefficient @xmath348 in eqs . ( [ eq:1d_diffusion ] ) and ( [ eq : switching_rate ] ) below . we assume that the diffusion is weak . this means that the distribution @xmath192 in the stationary state has narrow peaks at the stable states of forced vibrations , which are given by the condition @xmath349 . from eq . ( [ eq : rho_diffusion ] ) , for @xmath350 , the peaks are gaussian near the maximum . their typical width is @xmath351 , and the peaks at different stable states are well separated from each other . further away from the stable states the stationary solution of eq . ( [ eq : rho_diffusion ] ) can be sought in the eikonal form @xmath352.\ ] ] to the leading order in @xmath337 , function @xmath353 is independent of @xmath337 and can be found from a nonlinear equation of the form of the hamilton - jacobi equation @xcite . one can see from the full nonadiabatic equation for the mode operators @xmath354 , eq . ( [ eq : p_equation ] ) , that the condition that the ratio @xmath355 be small requires smallness of the parameter @xmath356 , where @xmath357 is the typical width of the distribution over @xmath74 , or equivalently , @xmath358 . this estimate of @xmath355 takes into account only the leading terms , which are described by the operator @xmath298 , and applies in the time range @xmath208 where all @xmath197 have reached stationary values for a given @xmath196 . from eq . ( [ eq : eikonal ] ) , @xmath359 . near peaks of @xmath192 , where @xmath360 , the condition @xmath361 reduces to @xmath362 , which has been our major assumption all along . on the far tail of the distribution we have @xmath363 , and therefore the ratio @xmath364 is not necessarily small . if this is the case , the adiabatic perturbation theory breaks down and the far tail of the distribution is not described by eq . ( [ eq : rho_diffusion ] ) . however , as we will see , in the most interesting regime for studying the switching between metastable states , where the system is close to a bifurcation point , @xmath365 . equation ( [ eq : rho_diffusion ] ) allows one to find , in a simple explicit form , the rate of switching from a metastable state near the saddle - node bifurcation point where this state disappears . near this point the dynamics is controlled by the slow variable @xmath366 , see eq . ( [ eq : slow_dynamics ] ) . the distribution @xmath192 is a gaussian peak with width @xmath367 in the direction transverse to the slow variable @xmath292 , whereas in the @xmath292-direction it is much broader @xcite . the distribution over the @xmath292-variable @xmath368 can be found following the arguments of ref . . to the leading order in @xmath369 from eqs . ( [ eq : slow_dynamics ] ) and ( [ eq : rho_diffusion ] ) one obtains @xmath370 + d_{\mu b}\partial_y^2\rho_{_m}(y),\ ] ] where @xmath371 is the coefficient of diffusion along the @xmath292-axis , and @xmath372 is the derivative of @xmath373 calculated at the bifurcation point . equation ( [ eq:1d_diffusion ] ) allows one to find the rate of escape , @xmath374 , from a metastable state near a bifurcation point . this rate is described by the kramers theory @xcite , @xmath375^{1/2}\ ] ] with @xmath376^{1/2}$ ] . as seen from eq . ( [ eq : switching_rate ] ) , the activation energy of switching near a saddle - node bifurcation point scales as the distance to the bifurcation point @xmath377 to the power 3/2 . this is typical in the case where fluctuations are induced by gaussian noise . in the present case this noise comes from quantum fluctuations of system @xmath211 which modulate the frequency of the mode @xmath282 . the analysis of secs . [ sec : adiabatic ] - [ sec : fluctuations ] can be applied , for example , to the problem of a mode coupled to a two - level system . a qualitative description of the mean - field dynamics of this model was given in sec . [ subsec : intro_multistability ] . the consistent mean - field analysis outlined above leads to eqs . ( [ eq : bloch ] ) and ( [ eq : naive_m ] ) , with @xmath378 replaced by the stationary value of the wigner disitribution center - of - mass variable @xmath86 of eq . ( [ eq : stationary_states ] ) . with this replacement , ( [ eq : naive_m ] ) and ( [ eq : stationary_states ] ) coincide . this justifies the results on the multistability of a mode coupled to a tls presented in sec . [ sec : introduction ] . the mean - field picture disregarded the effect of quantum fluctuations of the tls . when relaxation of the mode is slow , the diffusion caused by these fluctuations is described by eq . ( [ eq : rho_diffusion ] ) . using the bloch equations for the tls dynamics , one can show that the effective diffusion coefficient of the mode s vibrational phase , which is defined by eq . ( [ eq : d_time_correlator ] ) , has the form @xmath379 ^ 2}\right)\right],\end{aligned}\ ] ] where @xmath79 is given by eq . ( [ eq : bloch ] ) with @xmath74 replaced by @xmath87 . one can show from eq . ( [ eq : bloch ] ) that @xmath380 . it is clear that @xmath381 ( see also discussion below eq . [ eq : d_time_correlator ] ) . the condition of the applicability of the approach is @xmath382 . in fig . [ fig : d_mu ] we show the scaled values of @xmath337 along the mean - field response curve of fig . [ fig : hysteresis ] that displays tristability . as seen from this figure , @xmath383 remains small for the considered example . we emphasize that for the tls planck number @xmath384 , the noise described by the parameter @xmath337 is purely quantum . the noise is due to the randomness of spontaneous transitions between the states of the tls , with corresponding emission of excitations of the thermal bath . on average , the transitions lead to relaxation of the tls , but because they happen at random , they also cause fluctuations . for a mode coupled to a tls , scaled by the mode decay rate @xmath56 . the data refer to the mean - field characteristic in fig . [ fig : hysteresis ] , which displays tristability . the value of @xmath337 on the stable and unstable branches is shown by solid and dashed lines , respectively . diffusion is caused by quantum fluctuations of the tls , where we set the planck number @xmath385 . the point where the uppermost dashed line joins the lowermost solid line ( @xmath386 ) accidentally lies very close to the solid line that starts from @xmath387 and corresponds to the lowest branch @xmath388 in fig . [ fig : hysteresis ] . ] we have developed a theory of a vibrational mode ( which may be of mechanical or electromagnetic origin ) dispersively coupled to a quantum system , where both the mode and the system are driven far from thermal equilibrium . the results reveal new aspects of dispersive coupling . a profound consequence of the coupling is multistability of the nonlinear response , where the compound system can have multiple stable states in the mean - field approximation this situation is very different from the familiar bistability due to intrinsic nonlinearity of a mode . the multistability happens because , as a result of the interaction , the resonance frequency of the mode depends on the state of the system , while the state of the system depends on the degree of excitation of the mode . effectively , the mode becomes nonlinear , with the transition frequency depending on the distribution over the states of the mode . in the simple but highly relevant case of the coupling to a two - level system , we found a regime where the compound system can have up to three stable states . our analysis refers to the case where the relaxation rate of the system is large compared to the mode relaxation rate . this case is of utmost interest for the broad range of currently studied compound systems . here we discuss a few examples from the literature , which may naturally satisfy the conditions under which our description applies . * _ example 1 _ : double quantum dot charge qubit coupled to superconducting cavity mode . such a system was studied in ref . , with mode frequency @xmath389 ghz , mode lifetime @xmath390 s , qubit transition frequency @xmath391 ghz , qubit lifetime @xmath392 s , and jaynes - cummings coupling strength @xmath393 mhz . here the cavity mode quality factor was only @xmath394 , already giving @xmath395 . with expected device improvements the condition @xmath396 will be reached . the dispersive coupling @xmath42 in our theory can be tuned via the qubit - cavity detuning @xmath397 , giving e.g. @xmath398 , which easily satisfies the weak noise condition @xmath399 . * _ example 2 _ : superconducting `` transmon '' qubit coupled to a superconducting stripline cavity mode . in the recent experiment described in ref . , the parameter values are : @xmath400 ghz , @xmath401 ns , @xmath402 ghz , @xmath403 ns , and jaynes - cummings coupling @xmath404 mhz . the condition @xmath405 is weakly satisfied . the value of the dispersive coupling @xmath42 in our model is approximately equal to the `` qubit - qubit coupling '' @xmath406 mhz . this gives @xmath407 , though the lifetimes and @xmath42 can be presumably decreased . interestingly , this setup features two qubits coupled to the same cavity mode , allowing the possibility for studying the case where the system @xmath1 has more than just two levels . * _ example 3 _ : cooper pair box qubit coupled to a nanomechanical resonator . a device of this type was used in ref . @xcite to perform a nanomechanical measurement of the qubit state , with parameters : @xmath408 mhz , @xmath409s , @xmath410 ghz , @xmath411 khz . the qubit relaxation time @xmath412 was not measured , but from similar devices it can be expected to be of order @xmath413 ns . thus both the adiabaticity condition @xmath414 and the weak noise limit @xmath415 are easily satisfied . we have shown that quantum fluctuations in the system dispersively coupled to the mode cause switching between coexisting stable states . the switching rates are explicitly calculated in the most interesting region , i.e. , near bifurcation point where metastable states disappear . we find that the effective switching activation energy displays power - law scaling with the distance to the bifurcation point , with exponent 3/2 . this analysis holds in the regime of weak quantum noise , quantified by the parameter combination @xmath399 . going beyond the limit described above , we also expect that the system displays interesting behavior where the condition that the quantum noise is weak is violated . in this case quantum noise leads to unusually large fluctuations between the areas centered near the mean - field stable states . such behavior may be manifested in some of the systems described above . in other words , the system becomes an amplifier of the nonequilibrium quantum noise . a detailed analysis of this effect is beyond the scope of the present paper , but is a worthwhile direction for future study . 27ifxundefined [ 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 @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1038/nature11559 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1088/1367 - 2630/15/4/045017 [ * * , ( ) ] @noop _ _ ( , ) @noop _ _ ( , ) @noop _ _ ( , ) @noop * * , ( ) in @noop _ _ , ( , ) pp . @noop _ _ ( , ) in @noop _ _ , vol . , ( , ) pp . @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . ( , ) @noop _ _ ( , ) @noop _ _ , ed . ( , ) @noop * * , ( ) @noop _ _ , ed . ( , , ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.110.173603 [ * * , ( ) ]

gravitational waves are one of the most important diagnostic tools in the analysis of strong - gravity dynamics and have been turned into an observational channel with ligo s detection of gw150914 . aside from their importance in astrophysics , black holes and compact matter distributions have also assumed a central role in many other branches of physics . these applications often involve spacetimes with dimensions where the calculation of gravitational waves is more involved than in the four dimensional case , but has now become possible thanks to substantial progress in the theoretical study of general relativity in . here , we develop a numerical implementation of the formalism by godazgar & reall based on projections of the weyl tensor analogous to the newman - penrose scalars that allows for the calculation of gravitational waves in higher dimensional spacetimes with rotational symmetry . we apply and test this method in black - hole head - on collisions from rest in spacetime dimensions and find that a fraction of the arnowitt - deser - misner mass is radiated away from the system , in excellent agreement with literature results based on the kodama - ishibashi perturbation technique . the method presented here complements the perturbative approach by automatically including contributions from all multipoles rather than computing the energy content of individual multipoles .

1609.01292

gravitational waves ( gws ) entered the limelight with the recent detection of gw150914 @xcite soon followed by a second detection gw151226 @xcite which not only constitutes the first observation of a black - hole ( bh ) binary system , but also marks a true milestone in gravitational physics . this breakthrough has opened a quantitatively new path for measuring bh parameters @xcite , testing einstein s theory of general relativity @xcite and probing extreme astrophysical objects and their formation history @xcite , and substantially broadens the scope of multi - messenger astronomy @xcite . gw modelling , however , finds important applications beyond the revolutionary field of gw astronomy . many fundamental questions in general relativity in @xmath3 and @xmath0 spacetime dimensions concern the stability of strong - gravity sources ( see @xcite and references therein ) in the context of cosmic censorship violation , the solutions significance as physical objects or expanding our understanding of the strong - field regime of general relativity . for instance , gw extraction from numerical simulations of rapidly spinning myers - perry bhs demonstrates how excess angular momentum is shed in order to allow the bh to settle down into a more slowly rotating configuration @xcite . gw emission also represents a channel for mass - energy loss in ultra - relativistic collisions that are studied in the context of the so - called tev gravity scenarios that may explain the hierarchy problem in physics ; for reviews see e.g. @xcite . the calculation of gw signals in the theoretical modelling of @xmath3 dimensional sources in the framework of general relativity has been increasingly well understood following seminal work by pirani , bondi , sachs and others in the 1950s and 1960s ; see e.g. @xcite and @xcite for a review . applications are now routinely found in numerical and ( semi-)analytic calculations @xcite even though care needs to be taken when applied to numerical simulations on finite domains @xcite . the numerical study of solutions to einstein s equations has proven incredibly useful for understanding the behaviour of black holes and other compact objects . most recently , the application of numerical relativity in the generation of gravitational waveform templates for gw data analysis @xcite contributed to the above mentioned detection of gw150914 . the wave extraction techniques presently used in numerical simulations of astrophysical gw sources can be classified as follows : perturbative methods based on the formalism developed by regge , wheeler , zerilli and moncrief @xcite ; application of the quadrupole formula @xcite in matter simulations @xcite ; a method using the landau - lifshitz pseudo - tensor @xcite ; cauchy characteristic extraction @xcite ; and , probably the most popular technique , using the weyl scalars from the newman - penrose formalism @xcite . for a comparison of various of these techniques , see @xcite . the calculation of gws in higher dimensional relativity requires generalization of these techniques to @xmath0 . the extraction of the gw energy flux from the landau - lifshitz pseudotensor has been generalized straightforwardly to higher dimensions in @xcite . an extension of the regge - wheeler - zerilli - moncrief formalism for perturbations of spherically symmetric background spacetimes is available in the form of the kodama and ishibashi ( ki ) formalism @xcite and forms the basis of the wave extraction techniques developed in @xcite . the main assumption there is that far away from the strong - field regime , the spacetime is perturbatively close to the tangherlini @xcite spacetime . the deviations from this background facilitate the construction of master functions according to the ki formalism which in turn provide the energy flux in the different @xmath4 radiation multipoles . even though both methods are in practice applied at finite extraction radius , their predictions have been found to agree within a @xmath5 error tolerance when applied to bh head - on collisions starting from rest in @xmath6 @xcite . recent years have also seen considerable progress in the understanding of the peeling properties of the weyl tensor ; see @xcite and references therein . in particular , godazgar & reall @xcite have performed a decomposition of the weyl tensor in higher dimensions , and derived a generalisation of the newman - penrose formalism for wave extraction to @xmath0 . the numerical implementation of this formalism and probing the accuracy for a concrete example application is the subject of this paper . for this purpose , we require a formulation of the higher dimensional einstein equations suitable for numerical integration . for computational practicality , all higher dimensional time evolutions in numerical relativity have employed symmetry assumptions to reduce the problem to an effective `` @xmath7 + 1 '' dimensional computation where typically @xmath8 . this has been achieved in practice by either ( i ) imposing the symmetries directly on the metric line element @xcite , ( ii ) dimensional reduction by isometry of the einstein field equations @xcite or ( iii ) use of so - called _ cartoon _ methods @xcite . in our implementation , we use the latter method combined with the baumgarte - shapiro - shibata - nakamura @xcite ( bssn ) formulation of the einstein equations as expanded in detail in @xcite . the relevant expressions for the gw computation , however , will be expressed in terms of the arnowitt - deser - misner @xcite ( adm ) variables , and the formalism as presented here can be straightforwardly applied in other common evolution systems used in numerical relativity . the paper is structured as follows . in section [ sec : notation ] we describe the notation used in the remainder of this work . in section [ sec : theory ] we recapitulate the key results of @xcite which sets up the formalism . in section [ sec : mc ] we put the formalism into a form compatible with the modified cartoon dimensional reduction of our simulations . in section [ sec : numerical ] we describe the numerical set up used in our simulations , analyse the energy radiated in bh collisions in @xmath1 and compare the predictions with literature results based on alternative wave extraction techniques . the key goal of our work is to extract the gw signal from dynamical , asymptotically flat @xmath9 dimensional spacetimes , i.e. manifolds @xmath10 equipped with a metric @xmath11 , @xmath12 , of signature @xmath13 that obeys the einstein equations with vanishing cosmological constant @xmath14 here , @xmath15 is the energy momentum tensor which we assume to vanish in the wave zone , @xmath16 and @xmath17 denote the ricci tensor and scalar , respectively , and we are using units where the gravitational constant and the speed of light @xmath18 . furthermore , we define the riemann tensor and christoffel symbols according to the conventions of misner , thorne & wheeler @xcite . the adm space - time decomposition as reformulated by york @xcite writes the metric line element in the form @xmath19 where @xmath20 and @xmath21 , @xmath22 denote the lapse function and shift vector , respectively , and @xmath23 is the spatial metric that determines the geometry of hypersurfaces @xmath24 . for this choice of coordinates and variables , the einstein equations ( [ eq : einsteind ] ) result in one _ hamiltonian _ and @xmath25 _ momentum _ constraints as well as @xmath26 evolution equations cast into first - order - in - time form by introducing the extrinsic curvature @xmath27 through @xmath28 for a detailed review of this decomposition see @xcite . in the remainder of this work we assume that the adm variables are available . the bssn formulation , for example , employs a conformally rescaled spatial metric @xmath29 , the rescaled traceless extrinsic curvature @xmath30 , a conformal factor @xmath31 , the trace of the extrinsic curvature @xmath32 and contracted christoffel symbols @xmath33 associated with the conformal metric . these are related to the adm variables by @xmath34 \tilde{\gamma}^{{{\scriptscriptstyle{i } } } } = \tilde{\gamma}^{{{\scriptscriptstyle{m}}}{{\scriptscriptstyle{n } } } } \tilde{\gamma}^{{{\scriptscriptstyle{i}}}}_{{{\scriptscriptstyle{m}}}{{\scriptscriptstyle{n}}}}\ , , & \label{eq : bssnvars}\end{aligned}\ ] ] so that the adm variables can be reconstructed straightforwardly at every time in the evolution . for other popular evolution systems such as the conformal z4 system @xcite or the generalized harmonic formulation @xcite the adm variables are obtained in similar fashion or directly from the spacetime metric components through eqs . ( [ eq : admlineelement ] ) , ( [ eq : ek ] ) . many practical applications of higher dimensional numerical relativity employ symmetry assumptions that reduce the computational domain to an effective three dimensional spatial grid using the aforementioned techniques . the reasons are two - fold : ( i ) the computational cost of simulations in @xmath35 dimensions massively increase with any dimension added and ( ii ) the @xmath36 rotational symmetry accomodates many applications of interest that can therefore be handled by relatively straightforward extensions of numerical codes originally developed for astrophysical systems in four spacetime dimensions . in general , the number of effective ( spatial ) dimensions can range inside @xmath37 , where @xmath38 corresponds to spherical symmetry , i.e. @xmath39 isometry , and the other extreme @xmath40 corresponds to the axisymmetric case @xmath41 with rotational symmetry about one axis . as already detailed in @xcite , axisymmetry , i.e. @xmath40 , represents a special case that imposes fewer constraints on the components of tensors and their derivatives and is therefore most conveniently handled numerically in a manner similar but not identical to the general case . we will discuss first in detail the case @xmath42 and then describe the specific recipe for dealing with @xmath40 . probably the most important situation encountered in practical applications is that of an effective @xmath43 dimensional spatial computational domain . whenever the expressions developed in the remainder of this work for general @xmath7 are not trivially obtained for the special choice @xmath43 , we shall therefore explicitly write down the @xmath43 version in addition to the general case . let us start by considering a spacetime with at least two rotational killing vectors which is the scenario discussed in secs . 2 and 3 of ref . @xcite . for convenience , we will employ two specific coordinate systems adapted to this situation . the first is a set of cartesian coordinates @xmath44 where middle latin indices without ( with ) a caret range from @xmath45 ( @xmath46 ) , early latin indices run from @xmath47 , and rotational killing vector fields are presumed to exist in all planes spanned by any two of the coordinates @xmath48 . the second is a system of spherical coordinates @xmath49 where greek indices run from @xmath50 . we use middle , upper case latin indices to denote all spatial coordinates of either of these systems , so that @xmath51 and @xmath52 with @xmath53 . in the special case @xmath43 , we use the notation @xmath54 , so that eq . ( [ eq : coords_cart ] ) becomes @xmath55 we orient the cartesian coordinates ( [ eq : coords_cart ] ) such that they are related to the spherical coordinates ( [ eq : coords_spher ] ) by @xmath56 ( w^{2}\equiv)~~x^{2 } & = r\sin \phi^2\,\cos \phi^3\ , , \nonumber \\ \vdots & \nonumber \\[0pt ] ( w^{d-1}\equiv)~~x^{d-1 } & = r\sin \phi^2\,\ldots\,\sin \phi^{d-1}\,\cos\phi^d\ , , \nonumber \\[10pt ] ( w^d\equiv)~~z & = r\sin \phi^2\,\ldots\,\sin \phi^{d-1}\ , \sin \phi^d\,\cos \phi^{d+1}\ , , \nonumber \\[10pt ] w^{d+1 } & = r\sin \phi^2\,\ldots\,\sin \phi^{d-1}\ , \sin \phi^d\,\sin \phi^{d+1}\,\cos \phi^{d+2}\ , , \nonumber \\ \vdots \nonumber \\[10pt ] w^{d-3 } & = r\sin \phi^2\,\ldots\,\sin \phi^{d-3}\,\cos \phi^{d-2 } \ , , \nonumber \\[10pt ] w^{d-2 } & = r\sin \phi^2\,\ldots\,\sin \phi^{d-3}\,\sin \phi^{d-2 } \,\cos \phi^{d-1}\ , , \nonumber \\[10pt ] w^{d-1 } & = r\sin \phi^2\,\ldots\,\sin \phi^{d-3}\,\sin \phi^{d-2 } \,\sin \phi^{d-1}\ , . \label{eq : coordinates}\end{aligned}\ ] ] here @xmath57 $ ] , and all other @xmath58 $ ] , and we have formally extended the @xmath59 coordinates to also include ( in parentheses in the equation ) @xmath60 which turns out to be convenient in the notation below in sec . [ sec : mc ] . note that for the orientation chosen here , the @xmath61 axis denotes the reference axis for the first polar angle rather than the @xmath62 axis which more commonly plays this role for spherical coordinates in three spatial dimensions . for orientation among the different sets of indices , we conclude this section with a glossary listing index variables and their ranges as employed throughout this work . * upper case early latin indices @xmath63 range over the full spacetime from 0 to @xmath25 . * upper case middle latin indices @xmath64 denote all spatial indices , inside and outside the effective three dimensional computational domain , running from 1 to @xmath25 . * lower case middle latin indices @xmath65 denote indices in the spatial computational domain and run from @xmath66 to @xmath7 . for @xmath43 , we have @xmath67 . * lower case middle latin indices with a caret @xmath68 run from @xmath66 to @xmath69 and represent the @xmath70 ( without caret ) excluding @xmath62 . in @xmath43 , we write @xmath71 . * lower case early latin indices @xmath72 denote spatial indices outside the computational domain , ranging from @xmath73 to @xmath25 . * greek indices @xmath74 denote all angular directions and range from @xmath75 to @xmath25 . * greek indices with a caret @xmath76 denote the subset of angular coordinates in the computational domain , i.e. range from @xmath77 . as before , a caret thus indicates a truncation of the index range . * put inside parentheses , indices cover the same range but merely denote labels rather than tensor indices . * an index 0 denotes a contraction with the timelike normal to the foliation , rather than the time component , as detailed below in section [ d-1 + 1split ] . * @xmath78 denotes the covariant derivative in the full @xmath9 dimensional spacetime , whereas @xmath79 denotes the covariant derivative on a spatial hypersurface and is calculated from the spatial metric @xmath23 . * we denote by @xmath17 with appropriate indices the riemann tensor ( or ricci tensor / scalar ) of the full @xmath9 dimensional spacetime , and by @xmath80 the spatial riemann tensor ( or ricci tensor / scalar ) calculated from @xmath23 . our wave extraction from numerical bh simulations in @xmath0 dimensions is based on the formalism developed by godazgar & reall @xcite . in this section , we summarize the key findings and expressions from their work . the derivation @xcite is based on the definition of asymptotic flatness using bondi coordinates @xcite @xmath81 where @xmath82 is retarded time , @xmath83 the radius and @xmath84 are @xmath13 angular coordinates covering the unit @xmath13 sphere . a spacetime is asymptotically flat at future null infinity @xcite if the metric outside a cylindrical world tube of finite radius can be written in terms of functions @xmath85 , @xmath86 , @xmath87 as @xmath88 with @xmath89 where @xmath90 is the unit metric on the @xmath13 sphere . for an asymptotically flat spacetime @xmath91 can be expanded as @xcite @xmath92 and the bondi news function is obtained from this expansion as the leading - order correction @xmath93 . in analogy with the @xmath3 case , a null frame of vectors is constructed which is asymptotically given by ( see e.g. @xcite ) . in practice , the error arising from the use of an asymptotic form of the tetrad at finite extraction radii is mitigated by extracting at various radii and extrapolating to infinity @xcite and we pursue this approach , too , in this work . ] @xmath94 note that all the tetrad vectors are real in contrast to the @xmath3 dimensional case where the two vectors @xmath95 and @xmath96 are often written as two complex null vectors . next , the components of the weyl tensor are projected onto the frame ( [ eq : tetrad ] ) and the leading order term in the radial coordinate is extracted . following @xcite , we denote this quantity by @xmath97 and its components are given by @xmath98 here @xmath99 denote a set of vectors forming an orthonormal basis for the unit metric @xmath90 on the @xmath13 sphere . in practice , this basis is constructed using gram - schmidt orthonormalisation starting with the radial unit vector . as with the newman - penrose scalar @xmath100 in the four dimensional case , we note that this is the contraction of the weyl tensor with the ingoing null vector twice and two spatial vectors . whereas in @xmath3 the two polarisations of the gravitational waves are encoded in the real and imaginary parts of @xmath100 , here @xmath101 is purely real , with the @xmath102 labels providing the different polarisations . the final ingredient for extracting the energy radiated in gws is the rate of change of the bondi mass given by @xcite @xmath103 by substituting in for @xmath104 from the definition of @xmath105 we obtain an expression for the mass loss . @xmath106 where the notation @xmath107 implies summation over the @xmath108 labels inside the parentheses , and @xmath109 denotes the area element of the @xmath13 sphere . in practice , we will apply eq . ( [ mdot ] ) at constant radius @xmath83 , therefore replace retarded time @xmath82 with `` normal '' time @xmath110 and start the integration at @xmath111 rather than @xmath112 , assuming that gws generated prior to the start of the simulation can be neglected . the formalism summarized in the previous section is valid in generic @xmath9 dimensional spacetimes with or without symmetries . we now assume that the spacetime under consideration obeys @xmath36 isometry with @xmath42 , and will derive the expressions required for applying the gw extraction formalism of sec . [ sec : theory ] to numerical simulations employing the modified cartoon method . throughout this derivation , we will make use of the expressions for scalars , vectors and tensors in spacetimes with @xmath36 symmetry and the regularisation of their components at @xmath113 as listed in appendices a and b of ref . the key result of these relations for our purposes is that the adm variables @xmath21 , @xmath22 , @xmath23 , @xmath27 for a spacetime with @xmath36 isometry can be expressed completely in terms of their @xmath7 dimensional components @xmath114 , @xmath115 and @xmath116 as well as two additional functions @xmath117 and @xmath118 according to @xmath119 while the scalar @xmath21 remains unchanged . from the viewpoint of numerical applications , the key relations of the procedure reviewed in sec . [ sec : theory ] are eqs . ( [ eq : weylcomps ] ) and ( [ mdot ] ) . the first provides @xmath101 in terms of the weyl tensor and the normal frame , and the second tells us how to calculate the mass loss from @xmath101 . the latter is a straightforward integration conveniently applied as a post processing operation , so that we can focus here on the former equation . for this purpose , we first note that in practice wave extraction is performed in the wavezone far away from the sources . even if the sources are made up of non - trivial energy matter fields , the gw signal is calculated in vacuum where the weyl and riemann tensors are the same . our task at hand is then twofold : ( i ) calculate the riemann tensor from the adm variables and ( ii ) to construct a null frame . these two tasks are the subject of the remainder of this section . [ [ section ] ] the adm formalism is based on a space - time decomposition of the @xmath9 dimensional spacetime manifold into a one - parameter family of spacelike hypersurfaces which are characterized by a future - pointing , unit normal timelike field @xmath121 . this normal field together with the projection operator @xmath122 allows us to split tensor fields into components tangential or orthogonal to the spatial hypersurfaces by contracting each tensor index either with @xmath123 or with @xmath124 . for a symmetric rank ( 0,2 ) tensor , for example , we thus obtain the following three contributions @xmath125 the most important projections for our study are those of the riemann tensor which are given by the gauss - codazzi relations used in the standard adm splitting of the einstein equations ( see e.g. @xcite ) @xmath126 where in the last line we used the fact that in vacuum @xmath127 and , hence , its projection vanishes ( note , however , that in general @xmath128 even in vacuum ) . furthermore @xmath129 is the covariant derivative of the extrinsic curvature defined on the spatial hypersurface , with christoffel symbols calculated from the induced metric @xmath130 . equations ( [ gauss])-([contrgauss ] ) tell us how to reconstruct the full @xmath9 dimensional riemann tensor from @xmath25 dimensional quantities defined on the spatial hypersurfaces which foliate our spacetime . from this point on , we will use coordinates adapted to the @xmath120 split . in such coordinates , we can replace in eqs . ( [ gauss])-([contrgauss ] ) the spacetime indices @xmath131 on the left and right - hand side by spatial indices @xmath132 while the time components of the spacetime riemann tensor are taken into account through the contractions with the unit timelike normal @xmath121 and which we denote with the suffix @xmath133 in ( [ codazzi]),([contrgauss ] ) . note that more than two contractions of the riemann tensor with the timelike unit normal @xmath121 vanish by symmetry of the riemann tensor . [ [ section-1 ] ] the expressions given in the previous subsection for the components of the riemann tensor are valid for general spacetimes with or without symmetries . in this section , we will work out the form of the components of the riemann tensor in spacetimes with @xmath36 isometry for @xmath42 . for this purpose we recall the cartesian coordinate system @xmath135 of eq . ( [ eq : coords_cart ] ) , adapted to a spacetime that is symmetric under rotations in any plane spanned by two of the @xmath136 . we discuss in turn how the terms appearing on the right - hand sides of eqs . ( [ gauss])-([contrgauss ] ) simplify under this symmetry . we begin with the components of the spatial riemann tensor , given in terms of the spatial metric and christoffel symbols by @xmath137 the rotational symmetry imposes conditions on the derivatives of the metric , the christoffel symbols and the components of the riemann tensor that are obtained in complete analogy to the derivation in sec . 2.2 and appendix a of @xcite . we thus calculate all components of the riemann tensor , where its indices can vary over the coordinates inside and outside the computational domain , and obtain @xmath138 for the right - hand side of eq . ( [ contrgauss ] ) , we also need the spatial ricci tensor which is obtained from contraction of the riemann tensor over the first and third index . in @xmath36 symmetry , its non - vanishing components are @xmath139 note that the last expression , @xmath140 , does _ not _ involve a summation over @xmath59 , but merely stands for the product of @xmath141 with the expression ( [ eq : r3wuwu ] ) . the components of the extrinsic curvature are given by eq . ( [ eq : admso3 ] ) . its derivative is directly obtained from the expressions ( a.1)-(a.12 ) in appendix a of @xcite and can be written as @xmath142 next , we plug the expressions assembled in eqs . ( [ eq : spatialriemann])-([eq : ek3 ] ) into the gauss - codazzi equations ( [ gauss])-([contrgauss ] ) where , we recall , early latin indices @xmath131 are now replaced by @xmath132 following our switch to adapted coordinates . splitting the index range @xmath143 into @xmath144 for components inside and outside the computational domain , and recalling that an index 0 denotes contraction with @xmath145 , we can write the resulting components of the spacetime riemann tensor as @xmath146 with these expressions , we are able to calculate all components of the spacetime riemann tensor directly from the adm variables @xmath115 , @xmath117 , @xmath116 and @xmath118 and their spatial derivatives . there remains , however , one subtlety arising from the presence of terms containing explicit division by @xmath62 . numerical codes employing vertex centered grids need to evaluate these terms at @xmath113 . as described in detail in [ sec : regularization ] , all the above terms involving division by @xmath62 are indeed regular and can be rewritten in a form where this is manifest with no divisions by zero . the null frame we need for the projections of the weyl tensor consists of @xmath9 unit vectors as given in eq . ( [ eq : tetrad ] ) : ( i ) the ingoing null vector @xmath147 , ( ii ) the outgoing null vector @xmath148 which , however , does not explicitly appear in the scalars ( [ eq : weylcomps ] ) for the outgoing gravitational radiation , and ( iii ) the @xmath149 vectors @xmath150 pointing in the angular directions on the sphere . we begin this construction with the @xmath13 unit basis vectors on the @xmath13 sphere , @xmath151 , and recall for this purpose eq . ( [ eq : coordinates ] ) that relates our spherical coordinates @xmath152 to the cartesian @xmath153 . the set of spatial vectors , although not yet in orthonormalised form , then consists of a radial vector denoted by @xmath154 and @xmath13 angular vectors @xmath155 whose components in cartesian coordinates @xmath156 on the computational domain @xmath157 are obtained through chain rule @xmath158 we can ignore time components here , because our coordinates are adapted to the space - time split , so that all spatial vectors have vanishing time components and this feature is preserved under the eventual gram - schmidt orthonormalisation . plugging eq . ( [ eq : coordinates ] ) into ( [ eq : cart2sphere ] ) , we obtain for @xmath155 ( after rescaling by @xmath159 ) @xmath160 w^{1 } w^{2 } \\ \vdots \\ \vdots \\ \vdots \\ \vdots \\ w^{1 } w^{d-1 } \end{matrix } \right)}_{=\tilde{m}^{{{\scriptscriptstyle{i}}}}_{(2 ) } } \,,\ldots\ , , \underbrace { \left ( \begin{matrix } \left . \begin{matrix } 0~ \\ \vdots~ \\ 0~ \\ \end{matrix } \right\ } ( \alpha-2 ) \times \\[15pt ] - { \displaystyle \sum_{s=\alpha}^{d-1 } ( w^{s})^2 } \\[15pt ] w^{\alpha-1 } w^{\alpha } \\ \vdots \\ w^{\alpha-1 } w^{d-2 } \\ w^{\alpha-1 } w^{d-1 } \end{matrix } \right)}_{=\tilde{m}^{{{\scriptscriptstyle{i}}}}_{(\alpha ) } } \,,\ldots\ , , \underbrace { \left ( \begin{matrix } 0 \\ \vdots \\ \vdots \\ \vdots \\ 0 \\[5pt ] -(w^{d-2})^2 - ( w^{d-1})^2 \\[5pt ] w^{d-3 } w^{d-2 } \\[5pt ] w^{d-3 } w^{d-1 } \end{matrix } \right)}_{=\tilde{m}^{{{\scriptscriptstyle{i}}}}_{(d-2 ) } } \ , , ~~ \underbrace { \left ( \begin{matrix } 0 \\ \vdots \\ \vdots \\ \vdots \\ 0 \\[5pt ] 0 \\[5pt ] -(w^{d-1})^2 \\[5pt ] w^{d-2 } w^{d-1 } \end{matrix } \right)}_{=\tilde{m}^{{{\scriptscriptstyle{i}}}}_{(d-1 ) } } \ , . \label{eq : mtilde}\ ] ] in @xmath3 dimensions , these vectors , together with @xmath154 of eq . ( [ eq : radial ] ) would form the starting point for gram - schmidt orthonormalisation ; see e.g. appendix c in @xcite . in @xmath161 dimensional spacetimes with @xmath36 symmetry , however , we face an additional difficulty : on the computational domain @xmath157 , all components of the vectors @xmath162 vanish and their normalisation would result in divisions of zero by zero . this difficulty is overcome by rewriting the cartesian components of the vectors in terms of spherical coordinates and then exploiting the freedom we have in suitably orienting the frame . the details of this procedure are given in [ coordtrans ] where we derive a manifestly regular set of spatial vectors given by @xmath163 where @xmath164 , we have restored , for completeness , the time component and the vertical bars highlight the three component sectors : time , spatial on - domain , and spatial off - domain . equations ( [ eq : tildemdp1gen])-([eq : tildemdm1gen ] ) can , of course , be conveniently written in short - hand notation as @xmath165 . for the special case @xmath43 , the vectors are given by @xmath166 the next step is to orthonormalise these vectors . clearly the vectors @xmath167 with components in the @xmath168 dimensions are normalised by : @xmath169 for the remaining @xmath7 vectors given by eqs . ( [ eq : tildem1gen])-([eq : tildemdgen ] ) or , for @xmath43 , the spatial triad consisting of the three vectors ( [ eq : tildem1])-([eq : tildem3 ] ) , we use standard gram - schmidt orthonormalisation . note that under this procedure the components outside the computational domain of these vectors remain zero and can therefore be ignored . the final element of the null frame we need is the ingoing null vector , which we call @xmath147 . given in @xcite as @xmath170 asymptotically , we transform out of bondi coordinates , sending @xmath171 and furthermore use the freedom of rescaling this null vector by applying a constant factor of @xmath172 @xmath173 expressing the timelike unit normal field @xmath121 in terms of our gauge variables @xmath174 we find @xmath175 where @xmath176 . this result provides the ingoing null vector for any choice of @xmath7 and is the version implemented in the code . finally , we calculate the projections of the weyl tensor that encode the outgoing gravitational radiation @xmath177 [ cf . ( [ eq : weylcomps ] ) ] where @xmath147 is given by eq . ( [ eq : ka ] ) and the normal frame vectors @xmath178 are those obtained from gram - schmidt orthonormalising the right - hand sides of eqs . ( [ eq : tildem1gen])-([eq : tildemdm1gen ] ) . we first note that @xmath105 is symmetric in @xmath179 , so contractions solely with @xmath180 will result in @xmath181 components @xmath182 . for the special case @xmath43 , we obtain the three components @xmath183 . the null vector @xmath184 has vanishing @xmath59 components and from eqs . ( [ eq : rijkl])-([eq : ra0i0 ] ) we see that all components of the riemann tensor where an odd number of indices is in the range @xmath185 are zero . the only non - vanishing terms involving the riemann tensor with off - domain indices @xmath185 , therefore , have either four such indices or two and contain a kronecker delta @xmath186 ; cf . ( [ eq : ribkd ] ) , ( [ eq : rabcd ] ) , ( [ eq : ra0ck ] ) , ( [ eq : ra0b0 ] ) . in consequence , the mixed components @xmath187 and the purely off - domain components @xmath188 . the list of all non - vanishing components @xmath105 is then given by @xmath189\ , , \label{eq : omegaij } \\[10pt ] \omega'_{(a)(b ) } & = \delta_{ab } \ , \omega'_{(w)(w)}\,,\\[10pt ] \omega'_{(w)(w ) } & = \frac{1}{4\gamma_{ww}}\left[r_{w0w0 } - r_{w0wk}m_{(1)}^k - r_{w0wl}m_{(1)}^l + r_{wkwl}m_{(1)}^k m_{(1)}^l\right]\ , . \label{eq : omegaww}\end{aligned}\ ] ] where @xmath190 and all components of the riemann tensor on the right - hand sides are listed in the set of eqs . ( [ eq : rijkl])-([eq : ra0i0 ] ) . it should be noted here that @xmath105 is trace free , and so @xmath191 can be calculated from the diagonal terms @xmath192 . in a numerical simulation , the components of @xmath105 are calculated as functions of time and then can be integrated according to eq . ( [ mdot ] ) to extract the amount of energy radiated in gravitational waves . in the axisymmetric case @xmath40 there exists only one @xmath59 direction ( off domain ) . as discussed in section 4 of @xcite , we keep all tensor components as we would in the absence of symmetry , and the modified cartoon method and , thus , the rotational symmetry only enters in the calculation of spatial derivatives in the @xmath59 direction . for @xmath41 symmetry , the extraction of gravitational waves therefore proceeds as follows . * all components of the adm metric and extrinsic curvature are extracted on the @xmath13 dimensional computational domain . * the spatial riemann tensor and its contractions are directly evaluated using eq . ( [ eq : spatialriemann ] ) using the relations of appendix c in @xcite for off - domain derivatives . * the necessary components of the spacetime riemann tensor and its projections onto the timelike unit normal are evaluated through eqs . ( [ gauss])-([contrgauss ] ) . * the null frame is constructed as detailed in sec . [ mvecs ] , simply setting @xmath40 . * all the projections of the weyl tensor onto the null frame vectors are obtained from eq . ( [ eq : omegaij ] ) , but now covering the entire range of spatial indices @xmath193\,.\end{aligned}\ ] ] note that with the existence of more components of the riemann tensor , more projections of the weyl tensor now exist , specifically cross - terms such as @xmath194 . this can be seen straightforwardly by using @xmath41 modified cartoon terms from appendix c of @xcite and the expressions for the full and spatial riemann tensor given in eqs . ( [ gauss ] ) and ( [ eq : spatialriemann ] ) . for example , we can see that a component such as @xmath195 is non - zero . this will contribute to terms of the form @xmath196 . as already emphasized in @xcite , the key gain in employing the modified cartoon method for simulating axisymmetric spacetimes does _ not _ lie in the elimination of tensor components , but in the dimensional reduction of the computational domain . in the remainder of this work , we will implement the specific version of the wave extraction for the case @xmath43 and @xmath1 and simulate head - on collisions of equal - mass , non - spinning bhs starting from rest . we will calibrate the numerical uncertainties arising from the numerical discretisation of the equations ( fourth order in space and time and second order at the outer and refinement boundaries ) , the use of large but finite extraction radii and also consider the dependency of the results on the initial separation of the bhs . this type of collisions has already been studied by witek _ @xcite who calculate the gw energy using the kodama - ishibashi formalism , which enables us to compare our findings with their values . we perform evolutions using the lean code @xcite which is based on cactus @xcite and uses carpet @xcite for mesh refinement . the einstein equations are implemented in the bssn formulation with the modified cartoon method employed to reduce computational cost . for the explicit equations under the @xmath134 symmetry that we use , see section 3.2 of @xcite with parameter @xmath43 . without loss of generality , we perform collisions along the @xmath61-axis , such that the centre - of mass is located at the origin of the grid , and impose octant symmetry . we specify the gauge in terms of the `` 1+log '' and `` @xmath197 driver '' conditions for the lapse function and shift vector ( see e.g. @xcite ) according to @xmath198 with initial values @xmath199 , @xmath200 . the bh initial data is calculated using the higher dimensional generalization of brill - lindquist data @xcite given in terms of the adm variables by @xmath201^{(d-3)/2 } } \,,\label{blid}\ ] ] where the summation over @xmath202 and @xmath32 extends over the multiple bhs and spatial coordinates , respectively , and @xmath203 denotes the position of the @xmath204 bh . as mentioned above , we place the bhs on the @xmath61 axis in the centre - of - mass frame , so that in the equal - mass case , we have @xmath205 . our initial configuration is therefore completely specified by the initial separation which we measure in units of the horizon radius @xmath206 of a single bh . the bh mass and the radius @xmath206 are related through the mass parameter @xmath207 by @xmath208 where @xmath209 is the surface area of the unit @xmath149 sphere . the computational domain used for these simulations consists of a set of eight nested refinement levels which we characterize in terms of the following parameters : ( i ) the resolution @xmath210 on the innermost level which gets coarser by a factor of two on each consecutive outer level , ( ii ) the size @xmath211 of the domain which describes the distance of the outermost edge from the origin , and ( iii ) the resolution @xmath212 on the refinement level where the gravitational waves are extracted . for each simulation , we calculate the @xmath105 on our three dimensional computational grid and project them onto a two dimensional array representing a spherical grid at fixed coordinate radius . the data thus obtained on the extraction sphere are inserted into eq . ( [ mdot ] ) . the @xmath105 are scalars and so in our angular coordinate system do not depend on @xmath213 , so the integral over the sphere in ( [ mdot ] ) can be simplified : @xmath214~\sin^{d-3}(\phi^2 ) ~\sin^{d-4}(\phi^3)~d\phi^3d\phi^2\,,\ ] ] where @xmath215\equiv \left ( \int_{-\infty}^{u } \omega'_{(\alpha)(\beta ) } d\tilde{u}\right)^2 $ ] . a final integration over time of the variable @xmath216 then gives the total radiated energy . we begin our numerical study with an estimate of the uncertainty in our gw estimates arising from the discretisation of the equations . for this purpose , we have evolved two bhs initially located at at @xmath217 using a computational grid of size @xmath218 and three resolutions @xmath219 , @xmath220 and @xmath221 which corresponds to @xmath222 , @xmath223 and @xmath224 in the extraction zone . we measure the radiated energy in units of the total adm mass of the spacetime , which for brill - lindquist data is given by eq . ( [ eq : rhmmu ] ) with @xmath225 , the mass parameters of the initial bhs . the radiated energy as a function of time is shown in the upper panel of fig . [ fig : convergence ] . the radiation is almost exclusively concentrated within a window of @xmath226 around merger . during the infall and the post - merger period , in contrast , @xmath227 remains nearly constant . by comparing the high - resolution result with that obtained for the coarser grids , we can test the order of convergence . to leading order , the numerical result @xmath228 for some variable obtained at finite resolution @xmath210 is related to the continuum limit solution @xmath229 by @xmath230 , where @xmath231 denotes the order of convergence . by evaluating the quotient @xmath232 we can then plot the two differences @xmath233 and @xmath234 and test whether their ratio is consistent with a given value @xmath231 . the results for our study are shown in the lower panel of fig . [ fig : convergence ] which demonstrates that our numerical results converge at fourth order . ( solid curve ) and richardson extrapolated to infinite resolution assuming fourth - order convergence ( dashed curve ) . lower panel : convergence plot for the radiated energy @xmath227 extracted at @xmath235 from an equal - mass collision of two non - spinning bhs in @xmath1 starting from a separation @xmath236 . the results shown have been obtained using resolutions @xmath219 , @xmath220 and @xmath221 . the difference in radiated energy between the medium and high - resolution simulations has been rescaled by a factor @xmath237 expected for fourth - order convergence.,title="fig:",scaledwidth=80.0% ] ( solid curve ) and richardson extrapolated to infinite resolution assuming fourth - order convergence ( dashed curve ) . lower panel : convergence plot for the radiated energy @xmath227 extracted at @xmath235 from an equal - mass collision of two non - spinning bhs in @xmath1 starting from a separation @xmath236 . the results shown have been obtained using resolutions @xmath219 , @xmath220 and @xmath221 . the difference in radiated energy between the medium and high - resolution simulations has been rescaled by a factor @xmath237 expected for fourth - order convergence.,title="fig:",scaledwidth=80.0% ] the discretisation error of the total radiated energy is then obtained as the difference between the finite resolution result and that predicted by richardson extrapolation ( see upper panel in the figure ) . we obtain for the high - resolution case @xmath238 with a discretisation error of @xmath239 . a second source of error arises from the extraction at finite radius . following standard practice ( see e.g. @xcite ) , we estimate this uncertainty by extracting the gw energy at a set of seven or eight finite radii in the range @xmath240 to @xmath241 and extrapolating these values assuming a functional dependency @xmath242 where @xmath243 is a coefficient obtained through the fitting of the numerical data . by applying this procedure , we estimate the uncertainty due to the extraction radius at @xmath244 at @xmath245 and @xmath246 at @xmath247 . finally , we have measured the dependency of the total radiated energy on the initial separation of the bhs . in addition to the simulations discussed so far , we have performed high - resolution simulations placing the bhs at @xmath248 and at @xmath249 . we have found very small variations at a level of @xmath250 in the radiated energy for these cases , well below the combined error budget of @xmath251 obtained above . compared with collisions in @xmath3 dimensions ( see e.g. table ii in @xcite ) , @xmath227 shows significantly weaker variation with initial separation in @xmath1 . we attribute this to the more rapid fall - off of the force of gravity in higher dimensions leading to a prolonged but dynamically slow infall phase which generates barely any gws . in summary , we find the total energy radiated in gravitational waves in a head - on collision of two equal - mass , non - spinning bhs to be @xmath252 in excellent agreement with the value @xmath253 reported in @xcite using the kodama - ishibashi formalism . the extraction of gravitational waves from numerical simulations is one of the most important diagnostic tools in studying the strong - field dynamics of compact objects in four as well as higher dimensional spacetimes . in this work we have formulated the weyl tensor based wave extraction technique of godazgar & reall @xcite a higher dimensional generalization of the newman - penrose scalars in a form suitable for numerical simulations of @xmath0 dimensional spacetimes with @xmath36 , @xmath254 , symmetry employing the modified cartoon method . the only prerequisite for implementing our formalism is the availability on each timelike hypersurface of the effective computational domain of the adm variables . these are constructed straightforwardly from all commonly used numerical evolution systems such as bssn , generalized harmonic or conformal z4 . the recipe for extracting the gw signal then consists of the following steps . * ( ) * 0.5 cm 0.7 cm 0.2 cm 0.5ex plus0.2ex minus0.1ex 0ex plus0.2ex computation of the on and off - domain components of the spatial riemann tensor ( which equals the weyl tensor in the vacuum extraction region ) and the derivative of the extrinsic curvature according to eqs . ( [ eq : r3ijkl])-([eq : ek3 ] ) . reconstruction of the components of the spacetime riemann tensor as well as its contractions with the unit timelike normal from the quantities of the previous step according to eqs . ( [ eq : rijkl])-([eq : ra0i0 ] ) . construction of the null - frame vectors through gram - schmidt orthonormalising the expressions of eqs . ( [ eq : tildem1gen])-([eq : tildemdm1gen ] ) and then using ( [ eq : ka ] ) for the ingoing null vector . calculation of the projections @xmath105 of the weyl tensor onto the null frame vectors using eqs . ( [ eq : omegaij])-([eq : omegaww ] ) . calculation of the energy flux in gws through eq . ( [ mdot ] ) and integration in time of the flux to obtain the total radiated energy . the most common case of modeling higher dimensional spacetimes with rotational symmetries is the case of @xmath43 effective spatial dimensions which allows for straightforward generalization of existing codes ( typically developed for 3 + 1 spacetimes ) and also accommodates sufficiently complex dynamics to cover most of the important applications of higher dimensional numerical relativity . we have , for this purpose , explicitly given the specific expressions of some of our relations for @xmath43 where these are not trivially derived from their general counterparts . for testing the efficacy and accuracy of this method , we have applied the wave extraction to the study of equal - mass , non - spinning headon collisions of bhs starting from rest in @xmath1 using @xmath43 . we find these collisions to radiate a fraction @xmath255 of the adm mass in gws , in excellent agreement with a previous study @xcite employing a perturbative extraction technique based on the kodama - ishibashi formalism . we find this energy to be essentially independent of the initial separation which we have varied from @xmath256 to @xmath257 and @xmath258 times the horizon radius of a single bh . we attribute this result to the higher fall - off rate of the gravitational attraction in higher dimensions and the correspondingly slow dynamics during the infall stage . we finally note that the weyl tensor based wave extraction ideally complements the perturbative extraction technique of the kodama - ishibashi formalism . the latter provides the energy contained in individual @xmath4 radiation multipoles but inevitably requires cutoff at some finite @xmath259 . in contrast , the @xmath105 facilitate calculation of the total radiation , but without multipolar decomposition . it is by putting both extraction techniques together , that we obtain a comprehensive description of the entire wave signal . future applications include the stability of highly spinning bhs and their transition from unstable to stable configurations , the wave emission in evolutions of black rings and an extended study of higher dimensional bh collisions over a wider range of dimensionality @xmath9 , initial boosts and with non - zero impact parameter . these studies require particularly high resolution to accurately model the rapid fall - off of gravity , especially for @xmath260 , and are therefore beyond the scope of the present study . however , the foundation for analysing in detail the gw emission in these and many more scenarios is now available in as convenient a form as in the more traditional 3 + 1 explorations of numerical relativity . we thank pau figueras , mahdi godazgar , markus kunesch , harvey reall , saran tunyasuvunakool and helvi witek for highly fruitful discussions of this topic . this work has received funding from the european union s horizon 2020 research and innovation programme under the marie skodowska - curie grant agreement no 690904 , from h2020-erc-2014-cog grant no . magrath " 646597 , from stfc consolidator grant no . st / l000636/1 , the sdsc comet and tacc stampede clusters through nsf - xsede award nos . phy-090003 , the cambridge high performance computing service supercomputer darwin using strategic research infrastructure funding from the hefce and the stfc , and dirac s cosmos shared memory system through bis grant no . st / j005673/1 and stfc grant nos . st / h008586/1 , st / k00333x/1 . w.g.c . is supported by a stfc studentship . [ [ section-2 ] ] for the axisymmetric case @xmath40 , we only need to regularise terms appearing in the calculation of derivatives in the off - domain @xmath59 direction . all these terms are given explicitly in appendix c of @xcite , so that in the following we can focus exclusively on the additional terms appearing for @xmath261 , i.e. for spacetimes admitting two or more rotational killing vector fields . the treatment of these terms proceeds in close analogy to that of the bssn equations in the modified cartoon approach as described in detail in appendix b of @xcite . in contrast to that work , however , we will not be using the conformally rescaled metric of the bssn equations , which satisfies the simplifying condition @xmath262 , and so certain regularised terms involving the inversion of the metric will differ from the expressions obtained for the bssn equations . we start with a brief summary of the techniques and the main assumptions we will use to regularise expressions : * _ 1 . regularity : _ * we require all tensor components and their derivatives to be regular when expressed in cartesian coordinates . under transformation to spherical coordinates this implies that tensors containing an odd ( even ) number of radial indices , i.e. @xmath62 indices in our notation , contain exclusively odd ( even ) powers of @xmath62 in a series expansion around @xmath113 . using such a series expansion enables us to trade divisions by @xmath62 for derivatives with respect to @xmath62 . for example , for the @xmath62 component of a vector field @xmath263 , we obtain @xmath264 where we have introduced the symbol @xmath265 to denote equality in the limit @xmath266 . * _ 2 . absence of conical singularities : _ * we require that the spacetime contain no conical singularity at the origin @xmath113 . for the implications of this condition , we consider the coordinate transformation from @xmath267 to @xmath268 . as no other @xmath269 , coordinates will enter into this discussion we shall refer to @xmath168 as @xmath59 . in these coordinates we have that @xmath270 and the line element for vanishing @xmath271 and @xmath272 , @xmath273 , is given by @xmath274 requiring the circumference to be the radius times @xmath275 , we have that @xmath276 . substituting the above expressions and taking the limit @xmath277 , we obtain @xmath278 taking the time derivative of this relation and using the definition of the extrinsic curvature , we find that likewise @xmath279 * _ 3 . inverse metric _ * : various terms that we need to address contain factors of the inverse metric @xmath280 . in the practical regularisation procedure , these terms are conveniently handled by writing expressing @xmath280 in terms of the downstairs metric components @xmath115 and @xmath117 which are the fields we evolve numerically . we know the metric takes the following form : @xmath281 and we shall denote the upper left quadrant by the matrix @xmath282 . for simplicity , we will use the index @xmath283 to denote @xmath284 in this section , so e.g. cofactors @xmath285 and @xmath286 , and similarly indices @xmath287 will denote the same range , but including the @xmath62 component . we can now denote the cofactor of an element in the top left quadrant of @xmath23 as @xmath288 where @xmath289 and the notation @xmath290 denotes the determinant of the matrix @xmath291 obtained by crossing out the @xmath292 row and @xmath293 column . likewise , we may add further inequalities inside the braces to denote matrices obtained by crossing out more than one row and column . we can then insert this expression for @xmath294 and the determinant of the right hand side of eq . ( [ eq : gammamatrix ] ) , @xmath295 to obtain expressions for inverse metric components according to @xmath296 for @xmath43 , this procedure starts from the spatial metric @xmath297 the components @xmath294 of the cofactor matrix ( which is symmetric ) are given by @xmath298 \cdots & c_{yy } = \gamma_{ww}^n ( \gamma_{xx } \gamma_{zz } - \gamma_{zx}^2)\ , , & c_{yz } = -\gamma_{ww}^n ( \gamma_{xx } \gamma_{zy } - \gamma_{zx } \gamma_{xy})\ , , \\[10pt ] \cdots & \cdots & c_{zz } = \gamma_{ww}^n ( \gamma_{xx } \gamma_{yy } - \gamma_{xy}^2)\ , , \end{array } } \nonumber \\[10pt ] \label{eq : cofactor}\end{aligned}\ ] ] the determinant becomes @xmath299 and the inverse metric follows by inserting these into eq . ( [ eq : invg ] ) . using these techniques , we can regularise all terms in eqs . ( [ eq : r3iwjw ] ) , ( [ eq : gamma3mww ] ) , ( [ eq : r3wuwu ] ) , ( [ eq : ek3 ] ) and ( [ eq : rw0wk ] ) that contain divisions by @xmath62 . it turns out to be convenient to combine the individual terms into the following six expressions . * ( ) * 0.5 cm 0.7 cm 0.2 cm 0.5ex plus0.2ex minus0.1ex 0ex plus0.2ex @xmath308 we use @xmath309 and trade a division by @xmath62 for a @xmath62 derivative . the result is @xmath310 0 & \text{if } ( i , j)=(\hat{i},z ) \text { or } ( z,\hat{j } ) \\[10pt ] -\frac{1}{2 } \partial_z \partial_z \gamma_{ww } & \text{if } i = j = z \end{array } \right.\,.\\\nonumber \end{aligned}\ ] ] @xmath311 + using eqs . ( [ eq : gammamatrix])-([eq : invg ] ) , we express the inverse metric components @xmath312 in terms of the downstairs metric and trade the division by @xmath62 for a @xmath62 derivative . we thus obtain @xmath313 which in the case @xmath43 reduces to @xmath314 @xmath315 the regularisation of this term proceeds in analogy to that of term ( 9 ) in appendix b of @xcite , except we do not set @xmath316 . by rewriting @xmath317 , trading divisions by @xmath62 for @xmath62 derivatives and using @xmath318 , we obtain @xmath319 which in the case @xmath43 reduces to @xmath320 @xmath321 the division by @xmath62 is again traded for a derivative if @xmath322 and for @xmath323 , we use @xmath324 , so that @xmath325 0 & \text{if } i = z \end{array } \right.\,.\end{aligned}\ ] ] in this section , we discuss how the set of spatial normal frame vectors , eq . ( [ eq : mtilde ] ) , can be recast in a form suitable for applying gram - schmidt orthonormalisation . it turns out to be convenient to first rescale the @xmath155 such that they would acquire unit length in a flat spacetime with spatial metric @xmath326 . denoting these rescaled vectors with a caret , we have @xmath327 -\sum_{s=\alpha}^{d-1 } ( w^{s})^2 \\[15pt ] \left . \begin{matrix } w^{\alpha-1}w^{\alpha } \\ \vdots \\ w^{\alpha-1 } w^{d-2 } \\ w^{\alpha-1 } w^{d-1 } \end{matrix } ~~\right\ } ( d-\alpha)\times \end{matrix } \right)\,,~~\alpha=2,\,\ldots,\,d-1\ , . \label{eq : ma_v1}\ ] ] recall that we formally set @xmath328 , @xmath329 . as a convenient shorthand , we define @xmath330 so that , for instance , @xmath331 , @xmath332 , @xmath333 . this definition allows us to write @xmath334 -\rho_{\alpha}^2 \\[10pt ] \left . \begin{matrix } w^{\alpha-1}w^{\alpha } \\ \vdots \\ w^{\alpha-1 } w^{d-2 } \\ w^{\alpha-1 } w^{d-1 } \end{matrix } ~~\right\ } ( d-\alpha)\times \end{matrix } \right)\ , . \label{eq : ma_v2}\ ] ] we can now express the angles @xmath84 in terms of the radial variables @xmath335 , @xmath336 using these relations in ( [ eq : ma_v2 ] ) , we obtain @xmath337 -\sin \phi^{\alpha } \\[10pt ] \left . \begin{matrix } \cos \phi^{\alpha } \,\cos \phi^{\alpha+1 } \\ \cos \phi^{\alpha } \,\sin \phi^{\alpha+1 } \,\cos \phi^{\alpha+2 } \\ \vdots \\ \cos \phi^{\alpha } \,\sin \phi^{\alpha+1 } \,\ldots\ , \sin \phi^{d-2 } \,\cos \phi^{d-1 } \\ \cos \phi^{\alpha } \,\sin \phi^{\alpha+1 } \,\ldots\ , \sin \phi^{d-2 } \,\sin \phi^{d-1 } \\ \end{matrix } ~~\right . \end{matrix } \right ) = \left ( \begin{matrix } \left . \begin{matrix } 0~~ \\ \vdots~~ \\ 0~~ \\ \end{matrix } \right\ } ( \alpha-2 ) \times \\[25pt ] -\sin \phi^{\alpha } \\[10pt ] \begin{matrix } \vdots\\ \cos \phi^{\alpha } \left ( { \displaystyle \prod_{s=\alpha+1}^{\alpha+n-1 } } \sin \phi^{s } \right ) \cos \phi^{\alpha+n } \\ \vdots \end{matrix } \end{matrix } \right)\ , , \label{eq : ma_v3}\ ] ] where @xmath338 , and we formally set @xmath339 and @xmath340 . now , in our computational domain @xmath341 , which , from the definition of our coordinate system in eq . ( [ eq : coordinates ] ) gives @xmath342 since @xmath343 are arbitrary in our computational domain , we must have either @xmath344 . without loss of generality , we choose @xmath345 , which fixes the @xmath69 vectors @xmath346 which , up to rescaling by @xmath347 , are equal to the vectors in eqs . ( [ eq : tildem2gen])-([eq : tildemdgen ] ) . for the remaining vectors , we can use the rotational freedom in the angles @xmath348 . any choice for these values will satisfy @xmath349 as required on our computational domain and we merely need to ensure that we choose these angles such that the resulting set of vectors is orthogonal . this is most conveniently achieved by setting @xmath350 which , inserted into eq . ( [ eq : ma_v3 ] ) , implies @xmath351 combined with eqs . ( [ eq : hatm2])-([eq : hatm3 ] ) and restoring the tilde in place of the caret on the @xmath352 , we have recovered eqs . ( [ eq : tildemdp1gen])-([eq : tildemdm1gen ] ) in section [ mvecs ] for the angular vectors . for the case @xmath43 we have just two non - trivial vectors : @xmath353 recovering eqs . ( [ eq : tildem2])-([eq : tildemdm1 ] ) 10 m. dafermos and i. rodnianski . . in _ on recent developments in theoretical and experimental general relativity , astrophysics and relativistic field theories . proceedings , 12th marcel grossmann meeting on general relativity , paris , france , july 12 - 18 , 2009 . . 1 - 3 _ , pages 132189 , 2010 . arxiv:1010.5137 [ gr - qc ] . william g. cook , pau figueras , markus kunesch , ulrich sperhake , and saran tunyasuvunakool . . 3rd amazonian symposium on physics and 5th nrhep network meeting : celebrating 100 years of general relativity belem , brazil , september 28-october 2 , 2015 _ , 2016 . arxiv:1603.00362 [ gr - qc ] . j. r. van meter , j. g. baker , m. koppitz , and d .- i . choi . how to move a black hole without excision : gauge conditions for the numerical evolution of a moving puncture . , 73:124011 , 2006 . gr - qc/0605030 .

we derive the continuum equations and boundary conditions governing phonon - mediated heat transfer in the limit of small but finite mean free path from asymptotic solution of the linearized boltzmann equation in the relaxation time approximation . our approach uses the ratio of the mean free path to the characteristic system lengthscale , also known as the knudsen number , as the expansion parameter to study the effects of boundaries on the breakdown of the fourier descrition . we show that , in the bulk , the traditional heat conduction equation using fourier s law as a constitutive relation is valid at least up to second order in the knudsen number for steady problems and first order for time - dependent problems . however , this description does not hold within distances on the order of a few mean free paths from the boundary ; this breakdown is a result of kinetic effects that are always present in the boundary vicinity and require solution of a boltzmann boundary - layer problem to be determined . matching the inner , boundary layer , solution to the outer , bulk , solution yields boundary conditions for the fourier description as well as additive corrections in the form of _ universal _ kinetic boundary layers ; both are found to be proportional to the bulk - solution gradients at the boundary and parametrized by the material model and the phonon - boundary interaction model ( boltzmann boundary condition ) . our derivation shows that the traditional no - jump boundary condition for prescribed temperature boundaries and no - flux boundary condition for diffusely reflecting boundaries are appropriate only to zeroth order in the knudsen number ; at higher order , boundary conditions are of the jump type . we illustrate the utility of the asymptotic solution procedure by demonstrating that it can be used to predict the kapitza resistance ( and temperature jump ) associated with an interface between two materials . all results are validated via comparisons with low - variance deviational monte carlo simulations .

1508.04694

microscale and nanoscale solid state heat transfer as mediated by phonon transport has received considerable attention in connection with a number of diverse practical applications , such as heat management in microelectronic devices , passive cooling and thermoelectric energy conversion @xcite , but also due to the number of scientific challenges it poses . particularly notable is the wide range of scales present in these problems , typically starting from the atomistic ( including quantum ) and extending to the macroscopic ( device ) . kinetic - theory approaches based on the boltzmann transport equation ( bte ) @xcite , especially if informed by ab - initio information on the material properties @xcite , can be quite effective in bridging this range of scales . one limitation of such approaches appears in the small mean free path limit , @xmath0 , where kinetic descriptions become stiff . here , @xmath1 denotes the knudsen number defined as the ratio of the mean free path to the characteristic system lengthscale ; a more precise definition will be given in section [ background ] . as is well known , in the limit @xmath2 , the stiff boltzmann description need not be used because it can be replaced by the heat conduction equation ; derivation of the bulk thermal conductivity from the boltzmann equation in the relaxation approximation via a chapman - enskog type of expansion @xcite is well established , thus providing a `` pathway '' for recording the effect of molecular structure on the constitutive behavior in that limit . however , the chapman - enskog expansion is only applicable in the bulk and provides no information on the boundary conditions that need to supplement the heat conduction description in order to obtain solutions that are consistent with the ( more fundamental ) boltzmann solution . moreover , a rather large gap exists between lengthscales that truly satisfy @xmath3 and the regime where boltzmann equation solution is no longer problematic ( @xmath4 ) . in this paper , we use an asymptotic expansion procedure using @xmath1 as a small parameter to derive , from the bte , the `` continuum '' equations governing phonon - mediated heat transfer in the small mean free path limit . this procedure recovers the classic heat conduction equation ( including fourier s law as a constitutive relation ) as the equation governing the temperature field that is consistent with solution of the boltzmann equation to order @xmath5 , as expected . however , in contrast to chapman - enskog - type procedures , this procedure , also _ derives _ the boundary conditions that the heat equation is to be solved subject to . specifically , for fixed temperature boundaries , the fourier boundary conditions are found to be of the dirichlet type at the boundary temperature ; for diffusely specular walls , the fourier boundary conditions are shown to be the neumann no - flux boundary condition . although these results have been _ empirically _ established centuries ago , this is the first time they are shown to arise , rigorously , from a solution of the boltzmann equation . more importantly , by extending the asymptotic expansion to first and second order in @xmath1 , we derive the governing `` continuum - level '' equation _ and boundary conditions _ for finite but small values of the knudsen number ( @xmath0 ) . specifically , for steady problems , the governing equation is shown to be the steady heat conduction equation up to order @xmath6 , while the corresponding boundary conditions are shown to be of the temperature - jump type , with jump coefficients that , in general , depend on the material and boundary properties . for unsteady problems , we show that the governing equation is the unsteady heat conduction equation up to first order in @xmath1 with boundary conditions remaining the same as in the steady case up to that order for the case of prescribed - temperature boundaries . jump boundary conditions have been observed before in solutions of the boltzmann equation @xcite and attempts were made @xcite to explain these invoking differences in local equilibrium conditions across interfaces . the present work shows how temperature jumps arise as a result of the incompatibility between the isotropic distributions associated with boundary conditions and the anisotropic distribution associated with non - equilibrium resulting from transport ( temperature gradients ) . a well - known manifestation of this physical behavior are the temperature jumps associated with the kapitza interface problem . in section [ kapitza ] we show how our asymptotic approach can be used to calculate the interface conductance ( and associated temperature jump ) from first principles ( at the kinetic level , that is , given the interface transmission and reflection coefficient ) . the temperature jump relations derived in this work are manifestations of what is known in the kinetic theory community as `` slip '' , which gives its name to the slip regime , @xmath7 . it is generally known @xcite that in this regime the material constitutive law may still be used _ unmodified _ and kinetic effects are accounted for by modified boundary conditions . in the field of rarefied gas dynamics , cercignani @xcite and sone with co - workers @xcite were the first to provide systematic asymptotic solutions up to second order in @xmath1 , demonstrating the possibility of using the traditional `` continuum '' fluid dynamics , albeit with modified boundary conditions , beyond the slip regime and into the early transition regime . the transition regime is typically defined by @xmath8 and represents the regime in which transport transitions from diffusive ( @xmath0 ) to ballistic ( @xmath9 ) . discussions of the use of asymptotic solutions of the boltzmann equation in rarefied gas dynamics can be found in @xcite . the practical implications of the present work are twofold : first , solution of the heat equation is significantly easier ( analytically or numerically ) compared to the boltzmann equation , especially in the regime @xmath0 where the latter becomes stiff . in addition to ease of solution , centuries of investment in continumm formulations such as the heat equation , either in the form of education , mathematical solution techniques or numerical solution software , make this by far the preferred approach . this can be easily seen from the considerable efforts expended in developing approximate `` effective thermal conductivity '' concepts that enable the use of fourier s law in the transition regime . the present work provides _ rigorous _ methods for obtaining solutions _ consistent with the boltzmann equation _ in the slip and early transition regime . studies in rarefied gas dynamics show that , depending on the problem and the amount of error that can be tolerated , slip / jump formulations could be used up to @xmath10 and sometimes beyond @xcite . second , by using the asymptotic solution as a control in deviational monte carlo schemes , one can overcome the stiffness associated with the @xmath0 regime . this happens because @xcite the asymptotic solution becomes increasingly more accurate as @xmath3 , thus requiring increasingly less computational resources to describe the deviation therefrom as this limit is approached . this yields computational methods that are able to efficiently simulate problems characterized by @xmath11 locally or globally , in contrast to traditional boltzmann solution methods . the present paper is organized as follows : in section [ background ] we introduce the governing ( boltzmann ) equation and the notation used in this paper ; in section [ asymptotic_analysis ] , we present the asymptotic analysis leading to derivation of the governing equation in the bulk up to second order in the knudsen number . associated boundary conditions and boundary layer corrections up to first order in the knudsen number are derived in section [ order1bl ] . in section [ order2bl ] we present results obtained from extending the boundary layer analysis to second order in knudsen number . in section [ summary ] we summarize and discuss our results and provide example applications to one - dimensional and two - dimensional problems . in section [ time - dep ] we discuss the applicability of the asymptotic theory and its results ( governing equations , boundary conditions and corrective boundary layers ) to time - dependent problems . in section [ kapitza ] we show how the asymptotic theory can be used to calculate the kapitza conductance ( and temperature jump ) associated with the interface between two materials . we conclude with some final remarks in section [ final ] . we consider the boltzmann equation for phonon transport in the relaxation time approximation @xmath12 where @xmath13 is the occupation number of the phonon states , @xmath14 the position vector in physical space , @xmath15 the group velocity , @xmath16 the phonon frequency , @xmath17 the phonon polarization , @xmath18 the unit vector denoting phonon traveling direction , @xmath19 the temperature and @xmath20 an equilibrium distribution at the `` pseudotemperature '' @xmath21 defined by energy conservation considerations ( refer for instance to @xcite for details on the definition of @xmath20 ) . in this work we primarily consider steady problems . extension to time - dependent problems directly follows by extending the methodology presented here . scaling analysis in section [ time - dep ] shows that , assuming diffusive time scaling , time dependence may modify the results presented here at order @xmath22 . in other words , the results obtained for steady state in this paper may be applied directly to order @xmath5 and @xmath23 with very few modifications , explained in section [ time - dep ] . assuming small deviations from equilibrium at temperature @xmath24 , the linearized steady - state boltzmann equation reads @xmath25 where @xmath26 , with @xmath27^{-1}$ ] . by noting that @xmath28 and writing energy conservation @xcite in the form @xmath29 where @xmath30 denotes the density of states , we obtain the expression @xmath31 here , and in what follows , unless otherwise stated , @xmath32 . in the above expression , @xmath33 also , @xmath34 and @xmath35 respectively refer to the unit vector defining the direction of propagation and to the differential solid angle , expressed as @xmath36 in spherical coordinates . in the interest of simplicity , in the above expressions and in what follows , we use a single integral symbol to denote both integrals over multiple variables and sum over polarization . in this study , relaxation times and group velocities may depend on frequency and polarization . for this reason , the knudsen number is defined in an average sense . we choose the following ( somewhat arbitrary ) definition @xmath37 where @xmath38 and @xmath39 , which we will denote by kn . in the expression for @xmath40 , @xmath41 is the magnitude of the group velocity introducing the dimensionless coordinate @xmath42 as well as the normalization @xmath43 we write the boltzmann equation in the form @xmath44 where @xmath45 the usual macroscopic quantities of interest such as temperature , energy density and heat flux can be calculated from @xmath46 we will refer to @xmath47 as the deviational temperature , since it represents the deviation from the equilibrium temperature @xmath24 . the asymptotic solution relies on a `` hilbert - type '' @xcite expansion of the solution @xmath48 in the form @xmath49 given the nature of the proposed solution , similar expansions can be written for the temperature and the heat flux fields @xmath50 in this section , we only consider solutions far from any boundary . as will be shown below , close to the boundary , kinetic effects become important due to the incompatibility of the bulk solution with the kinetic ( boltzmann ) boundary condition and a separate , boundary layer analysis is required . therefore , we let @xmath51 be the bulk solution , anticipating that @xmath52 , where @xmath53 represents kinetic boundary layer corrections that are zero in the bulk and will be similarly expanded later . when the expansion for @xmath54 is inserted in the boltzmann equation we obtain @xmath55}{\text{kn } } \label{expanded_bte}\ ] ] by equating terms of the same order ( @xmath23 and higher powers ) and assuming that @xmath56 , we obtain the following relationship for all @xmath57 @xmath58 . \label{eq : bulk_relationship}\ ] ] in addition , considering the two terms of order 0 in the right hand side of , we find that @xmath59 is determined by the solution of the equation @xmath60 the assumption @xmath56 is easily satisfied when the range of free paths is relatively small ( and is exactly satisfied in the single free path case @xmath61 ) , but becomes harder to justify in materials with wide range of free paths . in the latter cases , it has the effect of reducing the value of @xmath1 for which the theory presented here is valid . this is further discussed and quantified in section [ bl_results ] . from equation ( [ phig0 ] ) we deduce that @xmath59 is a function that depends on @xmath62 only , since this is the case for @xmath63 . we note here that any function that only depends on @xmath62 is a solution . additionally , since @xmath64 , we find that the zeroth order deviational bulk temperature is given by @xmath65 and that @xmath66 at this stage , the spatial dependence of @xmath59 is undetermined . the additional information needed will be inferred from the application of a solvability condition to @xmath67 . using we find the following expression for the order 1 solution @xmath68 this equation states that a necessary condition for @xmath67 to be the order 1 solution is that it is equal to the sum of @xmath69 and a function that only depends on @xmath62 . since the temperature associated with @xmath70 is zero , we can write @xmath71 finally , order 2 may be derived following the same procedure for eq ( [ eq : bulk_relationship ] ) for @xmath72 , which yields @xmath73 in the following section , while deriving the governing equation for @xmath74 , we also show that the temperature associated with @xmath75 is @xmath76 . the solvability condition required to determine @xmath77 is the statement of energy conservation which , applied to @xmath78 , becomes @xmath79 using results in the following condition @xmath80 that needs to be satisfied for all @xmath57 . applying this relationship to @xmath67 , we obtain @xmath81 in the above expression , the integral over the solid angle is zero in all terms where a component of the traveling direction appears with an odd exponent . this implies @xmath82 this concludes the proof that the 0-th order temperature field obeys the steady state heat equation . moreover , from it follows that @xmath83 in appendix [ laplace_demo ] we show that higher - order ( in fact , possibly all order ) terms similarly obey the heat equation . in other words , @xmath84 and @xmath85 , are determined by solution of @xmath86 before we close this section , we note that although in the laplace - type equations derived above for the temperature the thermal conductivity does not appear , the above asymptotic analysis still clearly predicts that _ in the bulk , the material constitutive relation ( thermal conductivity ) is equal to the `` traditional '' bulk value_. this can be seen from first - principles by inserting into to obtain @xmath87 where the second equality follows from recognizing the well known expression @xmath88 in this section , we extend the asymptotic analysis of the previous section to the vicinity of boundaries , where as will be shown below , a boundary layer analysis is required for matching the bulk solution of the previous section to the kinetic ( bte ) boundary conditions of interest . here we will consider two kinetic boundary conditions , namely , those of prescribed temperature and diffuse adiabatic reflection . in this work we assume that boundaries are flat ; boundary curvature will be considered in a future publication . without loss of generality we assume that the boundary is located at @xmath89 and with an inward normal pointing in the positive @xmath90 direction ; @xmath91 and @xmath92 will denote cartesian coordinates in the plane of the boundary . moreover , we will use @xmath93 , @xmath94 and @xmath95 to refer to the components of the unit vector @xmath34 in the coordinate system @xmath96 . in other words , @xmath97 , @xmath98 and @xmath99 . we now derive the general equation governing the boundary layer correction required in the boundary vicinity for matching the bulk solution to the kinetic ( bte ) boundary conditions . we introduce the boundary layer function @xmath100 , written as a hilbert expansion ( @xmath101 ) with @xmath102 and insert it in the boltzmann equation , obtaining @xmath103 in the vicinity of the boundary , a new characteristic lengthscale , namely the distance from the boundary , becomes important . similarly to @xcite , we introduce a `` stretched '' variable defined by @xmath104 . equation can thus be written in the form @xmath105 by equating terms of the same order , we find that each boundary layer term is solution to a 1d ( in physical space ) boltzmann - type equation . for @xmath106 , this equation is @xmath107 the equations for @xmath108 , @xmath109 include `` volumetric source '' terms resulting from the derivatives of the lower order boundary layers in the boundary tangential directions ( @xmath91 and @xmath92 ) . specifically , for each order @xmath110 : @xmath111 the case @xmath112 will be considered in the following section , where second - order boundary layer analysis is carried out . the term `` prescribed temperature boundary '' is typically used to describe a boundary approximating a black - body , absorbing incoming phonons and emitting phonons from an equilibrium ( isotropic ) distribution at a given temperature . in other words , the boltzmann boundary condition associated with such a boundary at deviational temperature @xmath113 is a bose - einstein ( equilibrium ) distribution at the wall temperature , denoted here by @xmath114 . in the linearized case , the incoming distribution of deviational particles is therefore @xmath115 or simply , in terms of quantity @xmath48 defined in , @xmath116 . we note that @xmath59 is isotropic and is thus able to match @xmath117 provided we set @xmath118 at the boundary . therefore , at order 0 , the solution to the boltzmann equation with prescribed temperature boundaries is given by the heat equation complemented by the traditional dirichlet boundary conditions and no boundary layer correction is required ( @xmath119 , which also implies that @xmath120 ) . this situation changes at order 1 . the order 1 distribution @xmath121 is not isotropic due to the gradient of @xmath74 . as a consequence , there is a mismatch between the order 1 solution and the boundary condition ( which has been satisfied by @xmath59 and is thus zero for all subsequent orders ) . this mismatch can be corrected by introducing a boundary layer term @xmath106 governed by equation ( [ eq : k1_boltz ] ) and subject to boundary condition @xmath122 , which translates into the following relation @xmath123 the term @xmath124 is known from the order 0 solution . the term @xmath125 is unknown and determined by the fact that there exists only one value for @xmath125 such that @xmath106 tends to 0 for @xmath126 @xcite . this determination proceeds by writing @xmath127 where each of @xmath128 is the solution to an equation of the form with the associated boundary condition : @xmath129 anticipating the values of @xmath130 to scale with @xmath131 in the above equations we have set @xmath132 . the constants @xmath133 are uniquely determined by the condition that @xmath134 , @xmath135 and @xmath136 individually tend to zero for @xmath126 . one can easily verify that for @xmath137 , @xmath138 , with @xmath139 is a solution to with boundary condition . the temperature field associated with these functions is zero . here we note that the above solutions have the property @xmath140 and thus are also solutions of with the term @xmath141 removed . we will use this observation throughout this paper for obtaining analytical solutions to a number of boundary layer problems . the problem for @xmath134 must be solved numerically . given the boundary condition it needs to satisfy , we write @xmath142 and solve for @xmath143 . the numerical method developed and used for this purpose is explained in detail in ref @xcite . in the case of a debye and gray material referred to here as the single free path case ( @xmath144 for all @xmath145 ) , it yields @xmath146 , while the resulting @xmath147 is plotted in figure [ fig : order1_smfp ] . we note that refs . @xcite also report the value 0.7104 in the context of other kinetic particle transport , and develop other efficient methods for solving this problem . , for three relaxation time models . the @xmath148-axis is scaled by the maximum free path @xmath149 of each model . ] ( 1,2.2 ) node[left ] ; ( -3.2,5.5 ) node[left ] ; [ fig : order1_smfp ] in summary , the boundary condition for the order 1 bulk temperature field is @xmath150 or more generally @xmath151 where @xmath152 refers to the derivative in the direction of the normal to the boundary pointing into the material , @xmath153 , and @xmath154 the boundary location . in other words , the boundary condition is of the jump type and the associated temperature jump is proportional to the derivative of the 0th order solution in the direction normal to the boundary . the amplitude of the corrective boundary layer that is added near the wall is also proportional to the normal derivative : @xmath155 note that although a non - zero temperature field is associated with @xmath134 , the corresponding heat flux is zero . this is explained by the fact that @xmath134 , by construction , tends to 0 at infinity . since the boundary layer problem is one - dimensional in space , by energy conservation , the heat flux has to be constant in @xmath90 and is therefore zero everywhere . we also note that although @xmath135 and @xmath136 do not contribute to the temperature field , they do contribute in the heat flux @xmath156 in the direction parallel to the boundary . their contribution can be obtained by substituting into ; the result is summarized in section [ summary ] . in section [ boundary_conditions ] we reported the value of the coefficient @xmath157 and boundary - layer function @xmath134 in the single free path case . in this section we report results for two more realistic material models . specifically , we consider a material with realistic dispersion relation and a single relaxation time , as well as a material with realistic dispersion relation and frequency - dependent relaxation times . the dispersion relation in both cases is taken to be that of the [ 100 ] direction in silicon . the single relaxation time is taken to be 40ps . in the case of a variable relaxation time we use a slightly modified born - von karman - slack ( mbvks ) model @xcite with parameters from @xcite and @xcite , where the grain size used for boundary scattering is 0.27 mm instead of 2.7 mm . the reason for this approximation is that it facilitates the verification of the order 1 behavior with monte carlo simulation . we do not consider optical phonons in this work , but the method can be straightforwardly extended to this case . we find @xmath158 in the single relaxation time model and @xmath159 in the mbvks model . the associated boundary layers are plotted in figure [ fig : order1_smfp ] . it is important to note that : * the values of coefficient @xmath157 and the function @xmath160 depends on the definition of @xmath1 or , equivalently , @xmath161 , which is rather arbitrary . this , however , does not influence the final result because the asymptotic temperature field , ultimately ( see ( [ t_expansion ] ) ) depends on the products @xmath162 and @xmath163 ( see for instance solution ) . * the boundary layer in the mbvks model is particularly wide ( on the order of millimeters ) . this observation , as well as the large value of @xmath157 , is a manifestation of the stiffness ( multiscale nature ) of this problem , resulting from the wide range of free paths present in this material ; mathematically , it is due to the factor @xmath164 that appears in and which tends to give more weight to modes with very large free paths and makes the assumption @xmath56 hard to satisfy . since , by assumption , the sum of all @xmath165 should exist which requires @xmath166 this has the overall effect of limiting the range of applicability of the asymptotic model to knudsen numbers that are lower than the nominal @xmath167 . it is important to note , however , that this limitation is a result of the fundamental physics of the problem : even at `` low '' knudsen numbers given by @xmath168 , there exist modes with long free paths ( i.e. @xmath169 ) introducing kinetic effects and making the zeroth order solution ( @xmath170 ) inadequate . we validate our result using a one - dimensional problem , in which a mbvks material is placed between two boundaries at prescribed temperatures and located at @xmath171 and @xmath172 , respectively . the order 0 ( traditional fourier ) solution to this problem is a linear temperature profile @xmath173 which yields a heat flux @xmath174 , where @xmath175 is the temperature difference between the boundaries ; here , @xmath176 denotes the bulk thermal conductivity associated with the mbvks material . the temperature profile @xmath177 is obtained by solving the laplace equation with jump conditions @xmath178 and yields the modified heat flux @xmath179 . we note that when calculated from an order @xmath180 temperature field , the heat flux is inherently an order @xmath181 quantity ; in other words , the above result is correct to order 2 . in figure [ fig:1d_prescribed ] , we plot the difference between the actual heat flux ( @xmath182 , obtained using deviational monte carlo simulation @xcite ) and the asymptotic approximation , both normalized by @xmath183 , namely , @xmath184 . the observed asymptotic behavior is order 2 which validates the order 1 accuracy of the asymptotic solution . ( 1,2.7 ) node[left ] ; ( -3,5.6 ) node[left ] ; [ fig:1d_prescribed ] the case of diffuse adiabatic boundaries can be treated through a similar approach , where the mismatch between the bulk asymptotic solution and the boundary condition is analyzed and corrected . the boundary condition at the kinetic level is given by @xcite @xmath185 a major difference from the prescribed temperature boundary is that applying this condition to the 0th order bulk solution gives no information , because @xmath59 satisfies regardless of its value at the wall . the boundary condition for @xmath74 is obtained by analyzing the order 1 mismatch . the order 1 boundary layer problem may be defined by applying the boundary condition to @xmath186 . it results in the following condition : @xmath187 the isotropic term @xmath177 readily cancels from both sides of the equality . similarly to section [ boundary_conditions ] , we define @xmath188 where each @xmath130 is associated with the temperature gradient in direction @xmath189 ( as given by a right - handed set with @xmath90 being the direction normal to the boundary ) and is a solution to the boltzmann - type equation with boundary condition : @xmath190 we find that solutions satisfy the above conditions for @xmath112 and @xmath191 respectively , and do not impose any condition over the tangential derivatives of @xmath74 . for @xmath192 , results in @xmath193 the only solution possible with this boundary condition is @xmath194 . this can be seen by noting that if @xmath195 , multiplying the above equation by @xmath93 and integrating over @xmath196 yields @xmath197 , which is impossible ( this can be seen by starting from the equation governing @xmath134of the type and integrating over @xmath198 and @xmath199 and using the condition @xmath200 ) . we thus conclude that @xmath74 must satisfy the boundary condition @xmath201 which is agrees with the neumann boundary conditions associated with adiabatic boundaries . the second order correction @xmath202 must be solution of for @xmath112 , namely @xmath203 with the boundary conditions @xmath204 here we note that the derivatives of the first order boundary layer which appear in the rhs of introduce four volumetric source terms in the governing equation . the boundary condition includes three terms with first order partial derivatives of @xmath177 and nine terms with second order derivatives . taking into account the four source terms on the rhs of , we introduce sixteen constants such that the order 2 `` temperature jump '' , @xmath205 , may be written as @xmath206 we accordingly introduce sixteen boundary layer functions such that the total order 2 boundary layer may be written as : @xmath207 the 16 unknown coefficients and boundary layer functions can be determined using a combination of numerical and analytical techniques ; these are discussed in appendix [ second order prescribed ] . here we summarize the final result , which , conveniently , is quite compact . the second order temperature jump is given by the condition @xmath208 due to its simplicity and compactness , this result lends itself particularly well to implicit application of boundary conditions ; this is discussed in section [ implicit ] . the analogy to the order one temperature jump extends to the temperature boundary layer that is given by @xmath209 in addition to this temperature boundary layer , the analysis yields a second order heat flux boundary layer . it may be calculated analytically by inserting expression for @xmath202 into @xmath210 which can be written in terms of incomplete gamma functions . validation of these results can be found in @xcite . in section [ diffuse_order0 ] , we resorted to an analysis of the order 1 boundary layers to obtain the order 0 boundary condition , and showed the latter amounts to the well - known neumann boundary condition . similarly , we here proceed with the order 2 analysis in order to find the boundary condition for the order 1 temperature field . inserting in and introducing a boundary layer term yields , for @xmath211 and for all frequency / polarization modes : @xmath212 moving to the coordinate system @xmath96 and the stretched coordinate @xmath213 , we first note that in , the derivatives @xmath214 are zero for @xmath137 because @xmath215 . boundary layer @xmath202 may be decomposed into 4 components , @xmath216 , @xmath217 , @xmath218 and @xmath219 . components @xmath217 and @xmath218 are similar to the order 1 boundary layers @xmath135 and @xmath136 ( see expression ) , with the only difference being that @xmath74 is replaced by @xmath177 . component @xmath219 corrects the anisotropic mismatch associated with the bulk term @xmath220 . it is a solution to the 1d boltzmann equation with boundary condition @xmath221 for @xmath211 , and 0 at infinity , and is therefore given by @xmath222 components @xmath217 , @xmath218 and @xmath219 do not contribute to a temperature jump or ( temperature ) corrective layer , but they do contribute to the heat flux boundary layer . the last component is solution to the following problem : @xmath223 although we could solve problem using the numerical method described in @xcite , we will here directly find the value of @xmath224 without specifically calculating @xmath216 . we first proceed by multiplying the boundary condition ( second equation of problem ) by @xmath93 and integrating over the half sphere described by @xmath211 to obtain @xmath225 we also multiply the first equation of problem by @xmath226 and integrate it over all frequencies and solid angles and @xmath227 to obtain @xmath228 \\= -\frac{\pi}{4 } \int_{\omega , p } v_g \frac{\text{kn}^2}{\langle \text{kn } \rangle^2 } c_{\omega , p } d \omega \sum_{i=2}^3 \frac{\partial^2 t_{g0}}{\partial x_i^2}\end{gathered}\ ] ] since @xmath216 tends to 0 at infinity and @xmath170 , we deduce the jump relation @xmath229 with @xmath230 which can be rewritten in the form @xmath231 in the single free path model , @xmath232 . validation of this result can be found in @xcite . note also that the approach that we used for finding @xmath224 may be used for finding the heat flux associated with the boundary layer @xmath216 . _ a note on the physical interpretation of _ at first glance , the boundary condition seems to suggest that energy is not conserved since the net heat flux into the ( diffusely reflective ) boundary is not zero . in fact , contrary to appearances , this form _ ensures _ energy conservation at the boundary . this can be seen by considering that @xmath233 ( only possible in two or three dimensions ) implies variations in the temperature gradient _ along _ the boundary , which in turn implies variations in the heat flux along the boundary due to first - order kinetic boundary layers ( see ) . imposing energy conservation at the boundary reveals that exactly balances the terms resulting from gradients along the boundary @xcite . we have derived the continuum equations and associated boundary conditions that provide solutions equivalent to those of the boltzmann equation up to second - order in knudsen number for steady problems . this derivation shows that the governing equation in the bulk , up to at least second order in knudsen number , is the steady heat conduction equation _ with the bulk thermal conductivity_. kinetic effects , always present at the boundaries due to the inhomogeneity introduced by the boundary and the concomitant mismatch between the distribution introduced by the kinetic ( boltzmann ) boundary condition and the distribution function in the bulk , become increasingly important ( can be observed in larger parts of the physical domain ) as the knudsen number increases . fortunately , these kinetic effects can be systematically described and incorporated into the continuum solution relatively straightforwardly via the addition of kinetic boundary layer functions that are universal for a given material and material - boundary interaction model . we have studied two types of kinetic boundary conditions : prescribed wall temperature and diffuse reflection . we now summarize the procedure for obtaining the temperature and heat flux fields for an arbitrary problem of interest . _ prescribed wall temperature boundary condition _ : let @xmath234 denote the prescribed temperature along the system boundary denoted by @xmath154 with boundary normal @xmath153 . according to the asymptotic theory , the temperature and heat flux fields can be calculated from @xmath235 where * @xmath236 is solution to @xmath237 subject to @xmath238 * @xmath84 is solution to @xmath239 subject to @xmath240 * @xmath85 is solution to @xmath241 subject to @xmath242 * @xmath243 * @xmath244 * @xmath245 , @xmath246 * @xmath247 with @xmath248 given by . * @xmath249 with @xmath202 given by . we recall here that the coordinate @xmath213 is a stretched ( by @xmath250 ) version of the local normal to the boundary . the boundary layer functions @xmath251 , @xmath252 and @xmath253 are unique ( universal ) for each material and material - boundary interaction model . figure [ fig : order1_smfp ] shows results for @xmath251 for three material models . the method for calculating this function is described in detail in @xcite . the boundary layer functions @xmath252 and @xmath253 are known analytically . we also note that due to the absence of kinetic boundary layer corrections , at order zero @xmath254 . _ diffusely reflecting boundary _ : in the case of a diffusely reflecting boundary located at @xmath154 with normal vector @xmath153 , the temperature and heat flux fields can be calculated from @xmath255 where * @xmath236 is solution to @xmath237 subject to @xmath256 * @xmath84 is solution to @xmath239 subject to @xmath257 with @xmath224 given by . * @xmath258 , @xmath246 * @xmath259 with @xmath248 given by . * @xmath249 with the components of @xmath202 given in section [ diffuse_order1 ] . we note here that @xmath260 is identical to the corresponding boundary layer function that appeared in the prescribed - temperature boundary condition case . we also note that due to the structure of the boundary - layer problem for the diffusely reflecting boundary , the first - order analysis yields a zeroth order boundary condition , while a second - order analysis yields a first order boundary condition ; as a result the asymptotic solution for the temperature terminates at first order in @xmath1 . we see that , in both cases , the `` traditional '' fourier description corresponds to the zeroth order solution . in this section we consider a simple 1d problem as a means of illustrating the application of the asymptotic theory to problems of interest . we consider a silicon slab of thickness @xmath261 confined between two boundaries at different prescribed temperatures . using dimensionless coordinates , the boundaries are located at @xmath262 and @xmath263 and have deviational temperatures @xmath264 and @xmath265 , respectively . we recall that under the asymptotic analysis , the temperature field is given by @xmath266 the order 0 solution straightforwardly reads @xmath267 since it is the solution of the heat conduction equation subject to no - jump boundary conditions . therefore , the boundary conditions for the order 1 field are @xmath268 which results in @xmath269 the boundary layer @xmath270 contributes to the solution near the boundary at @xmath262 , while the function @xmath271 contributes close to the boundary at @xmath263 . the resulting solution correct to order 1 ( eq ) is plotted in figure [ fig : order1 ] for @xmath272 in the single relaxation time model and compared to our benchmark ( adjoint monte carlo @xcite ) result . the agreement is excellent ; we note in particular that even though the boundary layer correction is small at this knudsen number , the temperature jumps are considerable and are accurately captured by the asymptotic solution . in contrast , the zeroth order solution ( which neglects the temperature jumps ) is clearly inadequate . . ] ( 1,2.4 ) node[left ] ; ( -4.3,6.1 ) node[left ] ; [ fig : order1 ] if desired , calculation of @xmath273 to second order in @xmath1 proceeds by solving the heat conduction equation for @xmath274 subject to the second order boundary conditions . applying to this problem yields @xmath275 with the solution @xmath276 the order 2 solution including kinetic boundary layers is also shown in figure [ fig : order1 ] and clearly exhibits improved accuracy with respect to the order 1 solution . in fact , in this particular problem where only first derivatives are non zero , the process by which was derived can be repeated for all orders without knowledge of the higher order jump coefficients , leading to an asymptotic solution that is , in principle , correct to all orders . in other words , for @xmath277 : @xmath278 summing all orders ( provided @xmath279 ) , we obtain : @xmath280 the boundary layer corrections of all orders can also be obtained ( and summed ) using the same process . for example , for the boundary at @xmath262 , we obtain @xmath281 the second boundary layer ( at @xmath263 ) is obtained in an analogous fashion . this solution is asymptotically accurate to all orders , meaning that the error converges to 0 faster than any power of @xmath1 ; for a discussion on the error associated with the asymptotic expansion see @xcite . figure [ fig : order_inf ] , compares the order 1 , infinite order and `` exact '' ( monte carlo ) solution for @xmath282 . the infinite order solution is in very good agreement with the exact solution , while the order 1 solution is clearly inadequate at this knudsen number . in the rarefied gas dynamics literature @xcite jump boundary conditions are frequently imposed in an `` implicit '' fashion ( in the sense that the unknown is on both sides of the equation , resulting to what is referred to in the mathematical literature as mixed boundary conditions ) thus avoiding the `` stagerred '' solution procedure shown above where the governing equation needs to be solved for each order . for example , a set of boundary conditions up to second order given by @xmath283 @xmath284 and @xmath285 may be imposed by solving @xmath286 subject to @xmath287 one can show that these two approaches are equivalent ( to order @xmath288 ) by expanding @xmath289 and similarly for @xmath290 and substituting into ( [ eq : implicit ] ) . equating terms of the same orders of @xmath1 we obtain equations , and , at order zero , one and two , respectively . clearly the implicit form relies on the jump coefficients ( @xmath291 , @xmath292 , etc ) remaining the same at each order ( e.g. in and ) . if the above condition is satisfied , in addition to requiring less solutions of the governing equation , the implicit form has one more advantage : provided higher order derivatives ( not included in ) do not appear at higher order , the solution will be correct to all orders , since it is easy to verify that then implies that @xmath293 for all @xmath294 . this property can be illustrated with the example of section [ couette - example ] , where @xmath295 and @xmath296 : solution can be obtained directly by solving @xmath297 subject to @xmath298 although an infinite order solution is always welcome , we also need to keep in mind that some fortuity was involved in this problem in which all higher derivatives of the solution are zero . in the general case , given that @xmath296 , we expect the implicit condition to provide solutions that are accurate at least to second order and at most up to order @xmath299 where @xmath300 denotes the order of derivative featuring a non - zero jump coefficient . we close by noting that the implicit approach sometimes results in boundary conditions which feature derivatives of the same order as the governing equation which may raise questions about the well - posedness of the mathematical problem . as a resolution to this paradox , we recall that the derivation process followed here ( sections [ order1bl ] and [ order2bl ] ) produces the staggered forms of the general type - , which do not present posedness problems . in other words , the implicit form is used merely for convenience and should be discarded if any mathematical / numerical issues arise . . at this knudsen number the boundary layer contribution is clearly visible ( the solution is no longer a straight line ) . ] ( 1,2.7 ) node[left ] ; ( -4.3,6.4 ) node[left ] ; [ fig : order_inf ] in this section we use a two - dimensional example to illustrate the application as well as convergence properties of the asymptotic solution theory . specifically , we consider a slab of material that is infinite but subject to a periodic temperature variation in direction @xmath90 ; the slab has thickness @xmath301 in the transverse direction , with the associated dimensionless coordinate @xmath91 defined such that @xmath302 describes the median plane of the slab . the material boundaries at @xmath303 and @xmath304 are at the prescribed ( deviational ) temperatures @xmath305 and @xmath306 , respectively . the inset of figure [ tcos_profile ] shows a contour plot of the order 0 solution . in the two - dimensional example considered in section [ two_dimensional_example ] , for @xmath272 . the inset shows a contour plot of the order 0 solution . ] ( 1,2.5 ) node[left ] ; ( -4.3,6.2 ) node[left ] ; ( 2d ) at ( -1.7,7.7 ) in the two - dimensional example considered in section [ two_dimensional_example ] , for @xmath272 . the inset shows a contour plot of the order 0 solution.,title="fig : " ] ; ( -3,7.7 ) node[left ] ; ( -1.3,6.3 ) node[left ] ; ( -1.64,6.64 ) ( -1.64,8.84 ) ; [ tcos_profile ] in what follows , we construct the asymptotic solution of this problem up to @xmath307 , both using the `` order - by - order '' approach and the implicit approach discussed in the previous section . we will then compare these solutions with mc simulation results , both visually along the line @xmath89 but also very precisely at location ( @xmath308 ) to compare the order of convergence of the asymptotic solution with the theoretically expected one . the order 0 solution for the temperature field is given by @xmath309 the order 1 bulk temperature field can be obtained by solving the laplace equation with the boundary conditions : @xmath310 resulting in @xmath311 the order 2 bulk temperature field is then obtained by solving the laplace equation with the boundary conditions : @xmath312 leading to @xmath313 the solution is complete to second order once the boundary layer contributions are added . the order 1 and order 2 boundary layer correction terms in the vicinity of boundaries @xmath314 are respectively given by : @xmath315 as explained in the previous section , a solution of a similar order can be achieved by directly looking for the solution of the laplace equation @xmath316 with boundary conditions : @xmath317 this is the case here because , as shown in section [ order2bl_prescribed ] , second - order derivatives do not appear in the jump conditions or the temperature boundary layer . applying these `` implicit '' boundary conditions , we obtain @xmath318 the kinetic boundary layer corrections in the vicinity of the boundaries at @xmath314 are given by @xmath319 . evaluating the combined ( bulk and boundary layer correction ) solution at @xmath320 , we obtain @xmath321 this solution is compared to a highly - resolved mc simulation result in fig . [ tcos_profile ] for the case @xmath322 . the material model used is the single - relaxation - time model defined in section [ bl_results ] . the mc solution was obtained using the adjoint monte carlo method described in @xcite and will be denoted @xmath323 below . figure [ cos_sinh_o2 ] plots @xmath324 for 3 asymptotic solutions , namely , the first - order solution @xmath325 , the second - order solution @xmath326 , and the implicit solution . the figure shows that the implicit formulation leads to an order 2 solution overall which additionally features slightly improved accuracy compared to the `` regular '' order 2 solution . as explained in section [ implicit ] , the solution would be `` infinite '' order if no higher order derivative appeared in the jump boundary conditions . the third - order convergence observed for the implicit solution seems to suggest that a non - zero jump coefficient appears in front of the third - order derivative ( @xmath327 ) . in the two - dimensional example considered in section [ two_dimensional_example ] . ] ( 1,2 ) node[left ] ; ( -4.6,6 ) node[left ] ; [ cos_sinh_o2 ] although the analysis presented here has so far been limited to steady problems , extension to unsteady problems is relatively straightforward . in the field of rarefied gas dynamics the hilbert expansion has been extended to time - dependent problems by sone @xcite and takata @xcite , who showed that , other than the additional time - derivative in the governing equation , time dependence does not introduce any new physics up to order 1 in @xmath328 . in this section we show that this is also true for phonon transport for the case of prescribed temperature boundaries by introducing the dimensionless time - dependent boltzmann equation @xmath329 where @xmath330 is a dimensionless time , defined by @xmath331 , where @xmath332 is a characteristic time of variation and the strouhal number is given by @xmath333 we analyze cases where @xmath334 , where the average strouhal number , @xmath335 , follows from an analogous definition to that of @xmath328 in . the condition @xmath334 can be rewritten as @xmath336 , which implies an assumption of diffusive scaling in time . expanding the time dependent function @xmath48 as in eq . results in the same forms for orders 0 and 1 ( equations to ) . differences appear at order 2 . specifically , the form of the order 2 solution reads : @xmath337 applying the solvability condition results in @xmath338 which , after integration , yields the heat equation for the order 0 temperature field : @xmath339 applying the solvability condition to the order 3 solution similarly yields the heat equation for the order 1 temperature field . although not strictly needed for our purpose here , we may solve for @xmath340 in eq . by writing : @xmath341 which , combined with , yields @xmath342 which in the general case differs from @xmath274 . we note that @xmath343 holds in the case where the relaxation time does not depend on frequency and polarization . the order 0 boundary condition was obtained in section [ boundary_conditions ] by noticing that the order 0 distribution matches the distribution emitted by the boundary with no boundary layer correction . introducing time dependence does not modify this result . _ therefore the dirichlet boundary condition @xmath118 remains unmodified at order 0 in the time - dependent case_. at order 1 , we showed that the jump boundary condition emerges from the analysis of the boundary layer correction required by the mismatch between the order 1 bulk distribution and the boundary emitted distribution . as before , time - dependence does not modify the form of the order 1 bulk distribution . therefore , _ the order 1 jump condition remains unmodified in the presence of time dependence_. similarly , the derivation of the order 0 condition for diffuse reflective walls results from an order 1 analysis . _ the neumann condition is unmodified_. the order 2 boundary layer analysis presented in section [ diffuse_order1 ] that yields condition requires a modification since the relation @xmath344 is replaced by the diffusion equation . in this work , we did not proceed to analyze in detail how the order 1 boundary condition for diffuse reflective walls is modified . this shows that the theory developed in this article may be applied to time - dependent problems ( exhibiting diffusive scaling in time ) up to order 1 in the presence of prescribed temperature boundaries , with the only change being that the laplace equation is replaced by the unsteady heat equation . to illustrate and briefly validate some of the conclusions of the previous section , we consider here a square particle heated to a uniform temperature of 301 k and placed in a thermal bath at 300 k , such that its boundary is well described by a prescribed temperature of @xmath345 k. we also assume that the knudsen number is small such that we can calculate the temperature field inside the particle by solving the heat equation @xmath346 with the first - order boundary conditions derived in this work . for convenience we use the `` implicit form '' described in section [ implicit ] @xmath347 in figure [ dirichlet_and_slip_01 ] , we show a measure of this temperature relaxation process , namely @xmath348 , for @xmath349 , where @xmath350 denotes the temperature at the center of the particle . the heat equation solution was obtained using a finite difference scheme . here we note that the particle center is sufficiently far from the boundary that no kinetic boundary layer correction is required . the material model adopted here is that of silicon with a single relaxation time ( @xmath351 ) . this solution is compared with results obtained using the adjoint monte carlo method presented in ref . we also show the solution obtained from the ( traditional ) heat equation with the dirichlet boundary conditions @xmath352k . the figure shows that the asymptotic solution is in excellent agreement with the mc solution , while , as expected , the traditional approach ( with dirichlet boundary conditions)which corresponds to the zeroth - order solution significantly overpredicts the particle cooling rate . ( .9,1.3 ) node[left ] ; ( -3,4.1 ) node[left ] ; [ dirichlet_and_slip_01 ] the theoretical and numerical considerations presented in this paper are quite general and can be extended to a variety of problems where boundaries introduce `` size effects '' by injecting inhomogeneity into the problem . a classic example of such a problem is the interface between two materials : the presence of the interface results in a temperature jump , already shown in this work to be the signature of the kinetic correction required due to the inhomogeneity associated with the presence of a boundary . in this section we show how the asymptotic theory enables us to rigorously relate the kapitza conductance to the kinetic properties of the interface ( e.g. reflection / transmission coefficients ) . our aim here is not to conduct an exhaustive study but rather to demonstrate the applicability of the ideas presented earlier . as a result , we will focus on one specific transmission model and the single relaxation time model . we assume the following : * the interface separating the two media , denoted @xmath353 and @xmath354 , is sharp ( infinitely thin ) and planar . * when a phonon encounters the interface , it is either reflected or transmitted . in either case , its traveling direction is randomized while it keeps the same frequency and polarization . we denote the transmission probability from material @xmath353 to material @xmath354 by @xmath355 , while @xmath356 denotes the probability of reflection at the interface while traveling from @xmath353 to @xmath354 . similarly , @xmath357 and @xmath358 denote the transmission and reflection probabilities for travel from @xmath354 to @xmath353 , respectively . in what follows , we will use @xmath359 , @xmath360 , @xmath361 and @xmath362 , @xmath363 , @xmath364 to denote the relaxation time , frequency - dependent specific heat and magnitude of the ( frequency and polarization dependent ) group velocity in materials @xmath353 and @xmath354 , respectively . as before and without loss of generality , let us align the interface with the @xmath365 plane ( at @xmath89 ) and let the positive @xmath90 direction point from material @xmath353 to material @xmath354 . in this notation , the kinetic boundary condition associated with the interface is given by @xmath366 where superscript `` @xmath367 '' ( resp . `` @xmath368 '' ) refers to particles moving in the positive ( resp . negative ) @xmath90 direction . the order 0 solution in each material phase is solution to the laplace equation @xmath369 with the condition @xmath370 at the interface . replacing @xmath371 and @xmath372 by @xmath373 in and performing the integrations , we obtain : @xmath374 which implies @xmath375 the principle of detailed balance guarantees that the above is true for all @xmath145 . note that the condition @xmath370 does not determine the value of @xmath376 . the additional required condition is given by heat flux continuity : @xmath377 following the procedure of section [ boundary_conditions ] , we find that the order 1 solutions @xmath378 can not satisfy condition without the introduction of boundary layers . here @xmath379 denotes @xmath380 , while @xmath328 is a `` reference '' knudsen number calculated from the properties of one of the two materials ( results are independent of the chosen reference ) . we introduce two boundary layer functions @xmath381 and @xmath382 , and two constants @xmath383 and @xmath384 , anticipating temperature jumps from the order 0 at the interface of the form @xmath385 limiting our analysis to variations only in the @xmath90 direction , we insert the order 1 solution ( boundary layer included ) in condition , to obtain @xmath386 we then solve this boundary layer problem numerically to obtain the condition @xmath387 with @xmath388 , describing the first - order temperature jump across the interface . the numerical procedure used is described in @xcite . we test the asymptotic solution method outlined here on a simple one - dimensional problem with the following features : * the total length of the system is @xmath301 . the two materials are aluminum ( @xmath389 , hence , material @xmath353 ) and silicon ( @xmath390 , hence , material @xmath354 ) . here , we emphasize that we perform this calculation to _ validate _ asymptotic theory describing _ phonon transport _ across the interface . as a result , the aluminum model used here does not include electronic transport , which leads to @xmath391 w / mk . the choice of aluminum was motivated by the fact that this metal is frequently used as a transducer in transient thermoreflectance experiments @xcite and thus a reliable and well understood monte carlo simulation model a priority for validation studies exists @xcite for this material . we use @xmath392 and @xmath393 to denote the range of frequencies of the two material dispersion relations , respectively . we also use a constant relaxation time model in each material ; specifically , we take @xmath394s in al and @xmath395 s in si . * a temperature difference of 1 k is applied across the system by imposing a prescribed temperature of 301 k at @xmath396 , while the boundary at @xmath397 is maintained at 300 k. we note that the prescribed temperatures are used here to impose a temperature gradient onto the system . they are in no way linked to the interface model . * we define @xmath1 as the ratio between the mean free path in the silicon phase and @xmath261 . we choose @xmath261 such that @xmath272 . * the phonon transmissivities at the interface @xmath89 are adapted from the model described in @xcite , which given a `` target '' interface conductance @xmath398 ( as input ) , predicts @xmath399 for frequencies in @xmath400 ( 0 otherwise ) . coefficients @xmath357 are deduced from the principle of detailed balance . due to the one - dimensional nature of the problem studied here and the absence of higher than first - order derivatives of temperature in either material , an `` infinite '' order solution is possible : it can be obtained by solving the following system of four equations in four unknowns ( @xmath401 , @xmath402 , @xmath403 and @xmath404 ) @xmath405 we emphasize here that the temperature jump relations at @xmath406 ( first and last lines in ) appear only because of the particular formulation used here for imposing the temperature gradient , namely using prescribed temperature boundaries far from the interface . here , but also in general , the dynamics of the interface are solely described by the second and third lines of , namely heat flux continuity and the temperature jump across the interface . our numerical results are shown in figure [ fig : interface_asymptotic ] . the figure compares the temperature profile obtained with the deviational monte carlo method @xcite to the order 0 , order 1 and `` infinite '' order asymptotic solution . the order 1 solution provides significant improvement with respect to order 0 . after adding the corresponding boundary layer functions we find that the infinite order solution agrees very well with the monte carlo result . using this model , we obtain the actual conductance value @xmath407 mwm@xmath408k@xmath409 , which is very close to the `` target '' value 110 mwm@xmath408k@xmath409 used as input to the _ model _ described in @xcite . perhaps more importantly , we note that the mc simulation also predicts a conductance value ( obtained by extrapolating the bulk temperature profiles in order to calculate the temperature difference at the interface ) of 108 mwm@xmath408k@xmath409 , which is in perfect agreement with the ( infinite order ) asymptotic result . by comparison , the diffuse mismatch model predicts an interface conductance of @xmath410 mwm@xmath408k@xmath409 . this is consistent with the fact that the diffuse mismatch model results in an upper bound for the interface conductance @xcite . we note that the `` infinite '' order solution may not be available in the general , higher - dimensional case . related treatments of `` connection '' problems associated with different carriers have appeared in @xcite . ( 1,1.2 ) node[left ] ; ( -3,3.8 ) node[left ] ; [ fig : interface_asymptotic ] we have presented an asymptotic solution of the boltzmann equation in the small knudsen - number limit . the resulting solution provides governing equations and boundary conditions that determine the continuum temperature and heat flux fields in arbitrary three - dimensional geometries . our results show that , for steady problems , the equation governing the _ bulk _ temperature field up to second order in the knudsen number is the steady heat conduction equation . we also show that , up to first order in the knudsen number , the equation governing the _ bulk _ temperature field in transient problems is the transient heat conduction equation . although this result is expected ( at least to first order in the knudsen number ) courtesy of traditional kinetic theory analysis @xcite ( expanding the distribution function about the _ local _ equilibrium and giving no consideration to boundaries ) , the present work additionally derives the boundary conditions that complement this equation so that the resulting solutions of this system are _ rigorously _ consistent with solutions of the boltzmann equation . in particular , the present work shows that the constitutive relation is only valid in the bulk , while a few mean free paths from the boundaries kinetic effects are always present . these effects not only modify the local constitutive relation ( which is no longer of the fourier - type ) , they also have a significant effect on the bulk solution by modifying the effective boundary condition subject to which the heat conduction equation is to be solved . these effective boundary conditions are derived for a variety of kinetic boundary conditions and shown to generally be of the jump type thus explaining the temperature jumps at the boundaries previously observed and remarked upon @xcite . we note here that the jump conditions are universal ( non - adjustable ) , while the jump coefficients and kinetic boundary layers are universal for a given material and material - interface interaction model ; in other words , they are independent of system dimensionality and once calculated they can be used in any geometry of interest . tabulated data for the various boundary layers derived not available in analytical form are available upon request . these results provide no evidence or justification for modifying the material constitutive relation ( thermal conductivity ) as a means of extending the applicability of the traditional continuum description to the transition regime the underlying physics is considerably more complex . according to the asymptotic theory presented here , in the regime @xmath411 ( strictly speaking @xmath412 ) solutions consistent with the boltzmann equation are obtained using a thermal conductivity that is equal to the bulk value ; the modified ( typically reduced ) transport rate associated with size effects _ due to boundary presence _ is captured by the additional resistance introduced by the jump boundary conditions as well as kinetic corrections that are to be linearly superposed to the final heat conduction result . on the other hand , by virtue of the expansion considered here , this work pertains to breakdown and extension of the classical fourier description due to the inhomogeneity introduced by boundaries . as a result , it does not treat kinetic effects appearing in a spatially homogeneous material such as those arising from temporal variations that are fast compared to the relaxation time , or spatial variations that have characteristic lengthscales that are on the order of , or smaller than , the phonon mean free path , that are also of interest to the scientific community @xcite . ultimately , a theory that captures both classes of kinetic effects under a unified framework needs to be developed ; we hope that this work is a step towards that goal . our results are extensively validated using deviational monte carlo simulations of multidimensional problems . studies in rarefied gas dynamics @xcite show that second - order asymptotic formulations are reliable to engineering accuracy up to @xmath413 and in some cases , depending on the problem simplicity , beyond . our numerical validations support this finding . we note that the theory presented here assumes boundaries to be flat ( no curvature ) . curvature effects are expected to introduce additional terms in the effective boundary condition expressions and associated boundary layer corrections @xcite . this will be the subject of future work . due to its ability to capture the inhomogeneity in the distribution function associated with presence of boundaries , the present theory lends itself naturally to the description of the kapitza resistance _ and temperature jump _ associated with the interface between two materials . we have shown that the asymptotic description produces results that are in excellent agreement with deviational monte carlo simulations . in other words , given transmission and reflection coefficients at the interface , the asymptotic theory may be used to _ predict _ the kapitza resistance without any assumption on the form of the distribution in the interface vicinity . in this section , we show that @xmath177 and @xmath274 are solution to the laplace equation . we start with the case of @xmath177 . we apply the solvability condition to @xmath75 to obtain : @xmath414 integration over @xmath35 removes terms containing odd powers of @xmath415 , yielding @xmath416 from which we conclude that @xmath417 to obtain the laplace equation for @xmath274 , we apply to @xmath418 . after carrying out the angular integration and cancelling terms containing odd powers of @xmath415 we are left with @xmath419 thus , in order to show that the laplace equation holds for @xmath274 , we need to show that @xmath420 performing the angular integration , we obtain @xmath421 as desired . it appears that this procedure can be applied to all higher order terms ( @xmath422 , @xmath423 , etc . ) . coefficients @xmath424 and @xmath425 ( in ) and functions @xmath426 and @xmath427 ( in ) are determined by boundary value problems of the same form as the ones discussed in section [ boundary_conditions ] satisfying equation . the problems that determine the coefficients @xmath428 include the source terms from that appear on the rhs of . in the interest of brevity , we only discuss the ones associated with the source term @xmath429 . the remaining two ( associated with the term @xmath430 ) may be deduced by analogy . for @xmath137 coefficient @xmath431 is solution to : @xmath432 where @xmath433 denotes the heaviside function . two results can be obtained immediately : * coefficients @xmath434 , @xmath435 and @xmath436 are solutions to the same problems as @xmath157 , @xmath437 and @xmath438 . consequently , they are equal and their associated boundary layers are the same , provided @xmath439 is replaced by @xmath177 in eqs . , and . * coefficients @xmath425 and @xmath428 for @xmath440 are zero . for instance , it can be verified that @xmath441 is a solution of . solutions for all @xmath425 and @xmath442 can be systematically obtained by solving the associated problem without the @xmath443 term and then verifying that @xmath444 we are left with five undetermined coefficients , namely @xmath445 , @xmath446 , @xmath447 , @xmath448 and @xmath449 . these can be determined using the numerical approach described in @xcite ( suitably modified in order to accommodate the volumetric source terms which appear in the mathematical formulation ) . instead of following this approach , here we prove that @xmath450 and that , therefore , the temperature jump associated with the second order derivative is zero , while the boundary layer , although not zero , integrates into a zero temperature . the remaining five coefficients in relation , can be determined by finding the function @xmath451 that satifies : @xmath452 let @xmath453 , where @xmath454 for @xmath455 correspond , respectively , to the five boundary layer functions that are the counterparts of the five temperature jump terms in relation . we proceed with a strategy similar to the one used above , namely , solve for each @xmath454 individually , ignoring the contribution of @xmath456 , and then evaluating @xmath456 . in the present case @xmath457 but @xmath458 ; more details can be found in @xcite . this proves that @xmath459 is the solution of with the specified source terms and boundary conditions , and that the resulting boundary layer satisfies the boundary conditions without requiring a temperature jump correction , that is , relation is proved . the authors would like to thank professor t.r . akylas for many helpful comments and discussions . n.g.h would also like to thank k. aoki and s. takata for many useful discussions . the preparation of this manuscript as well as the work on extension to time - dependent problems ( section [ time - dep ] ) and the conductance of the interface between two materials ( section [ kapitza ] ) was supported by the solid - state solar - thermal energy conversion center ( s3tec ) , an energy frontier research center funded by the u.s . department of energy , office of science , basic energy sciences under award # de - sc0001299 and de - fg02 - 09er46577 . the remainder of the work was supported by the singapore - mit alliance .

we report on-band snapshot observations of 157 sloan digital sky survey ( sdss ) quasars at using the advanced camera for surveys on the _ hubble space telescope ( hst ) _ to search for evidence of gravitational lensing of these sources . none of the quasars appear to be strongly lensed and multiply imaged at the angular resolution ( ) and sensitivity of _ hst_. the non - detection of strong lensing in these systems constrains the luminosity function to an intrinsic slope of , assuming a break in the quasar luminosity function at . this constraint is considerably stronger than the limit of obtained from the absence of lensing in four quasars . such constraints are important for our understanding of the true space density of high - redshift quasars and the ionization state of the early universe .

astro-ph0509135

investigations of multiply imaged high - redshift quasars are important for our basic understanding of the formation and growth of supermassive black holes in galactic centers ( turner 1991 ; haiman & loeb 2001 ) and the ionization state of the universe as a function of time ( madau , haardt , & rees 1999 ; wyithe & loeb 2003a ) . gravitational lensing changes the apparent flux coming from a quasar and thus changes our interpretation of flux - dependent properties ; the discovery that a given quasar is gravitationally lensed means that the naively estimated luminosity is too high . thus , gravitational lensing would modify the expected size of ionized ( ) regions around individual quasars ( cen & haiman 2000 ) and would weaken the lower limits on the neutral fraction in the igm as inferred from the size of regions ( wyithe & loeb 2004 ; mesinger & haiman 2004 ) . this paper concentrates on the effect that gravitational lensing has on the apparent shape of the quasar luminosity function ( comerford , haiman , & schaye 2002 ; wyithe & loeb 2002 ; fan et al . 2003 ) . in a previous paper ( richards et al . 2004 , hereafter paper i ) we investigated whether the known @xmath9 quasars are gravitationally lensed and discussed how the lack of lenses in this sample affects our understanding of the growth of black holes in the early universe ( e.g. , haiman & loeb 2001 ) and the ionization history of the universe at the end of the reionization period ( e.g. , fan et al . 2002 ) . in this paper , we examine the constraints that can be placed on the intrinsic slope of the quasar luminosity function ( qlf ) at @xmath1045 from a search for gravitational lenses in a sample of @xmath1045 quasars ( fan et al . 2001a ; anderson et al . 2001 ; schneider et al . 2001 ; schneider et al . 2005 ) in the sloan digital sky survey ( sdss ; 2000 ) . much theoretical effort has been devoted to placing constraints on the slope of the quasar luminosity function from the fraction of lenses found amongst high-@xmath11 quasars ( e.g. , comerford et al . 2002 ; wyithe & loeb 2002 ) . ( 2002 ) showed that modest constraints could be obtained from a sample of @xmath12 @xmath9 quasars from ground - based imaging ( i.e. , sensitive to @xmath13 splittings ) , and that similar limits could be derived from _ hubble space telescope ( hst ) _ resolution imaging of the ( then ) four known @xmath9 quasars . paper i presented _ hst _ imaging showing that none of those four quasars are strongly gravitationally lensed and derived a limit on the slope of the bright end quasar luminosity function of @xmath14 . ( 2004 ) found even stronger constraints for this sample by including the observation that two quasars appear to be lensed by foreground galaxies but are not multiply imaged . these limits come from magnification bias ( turner , ostriker , & gott 1984 ; narayan 1989 ) , which would be strong if the slope of the qlf were steep and there were many quasars with intrinsic luminosities below the detection threshold . as discussed in paper i , the expected fraction of multiply - imaged quasars at a given redshift depends both on the cosmological model and on the luminosity function of quasars . in the wmap cosmology ( 2003 , using a halo distribution taken from the large n - body simulations of 2001 ) , the fraction of random lines of sight out to @xmath3 that produce multiple images at all splitting angles is of order 0.2% ; this fraction rises to @xmath15% at @xmath16 . if the intrinsic ( unlensed ) qlf has a break in slope , then , in general , the fraction of lensed quasars would increase for apparent fluxes above the break . in a flux - limited survey , there is a strong correlation between luminosity and redshift . if the true qlf possesses a break which moves to fainter luminosities at higher redshift , then the most distant quasars ( which also look through the longest path length ) are expected to be the most likely to be lensed . more explicitly , in the event that no lensing is observed , paper i showed that the tightest constraints on the qlf slope come from @xmath9 quasars , and also that roughly seven @xmath17 quasars have the same statistical power as a single @xmath9 quasar . herein we present the results for a sample of 157 @xmath17 quasars from the sdss that were imaged with _ hst _ ; we find that none of these sources is lensed . as predicted , these quasars provide a constraint on the slope of the qlf that is roughly equivalent to _ hst _ imaging of 22 @xmath9 quasars , limiting the bright end slope of the @xmath35 qlf to @xmath18 . section 2 describes the sample and the data . in 3 we discuss the constraints on the qlf that can be derived from the data . we summarize in 4 . throughout this paper , we adopt the wmap cosmology with @xmath19 , @xmath20 , @xmath21 , an rms mass fluctuation within a sphere of radius @xmath22 mpc of @xmath23 , and power law index @xmath24 for the power spectrum of density fluctuations ( spergel et al . 2003 ) . we also adopt the cosmological transfer function from & hu ( 1999 ) . conversions between @xmath25 and @xmath26 assume @xmath27 ( schmidt , schneider , & gunn 1995 ) with spectral index @xmath28 . there were 281 sdss quasars with @xmath29 as of january 2002 ( fan et al . 1999 , 2000 ; zheng et al . 2000 ; fan et al . 2001a ; schneider et al . 2001 ; anderson et al . 2001 ) , when the sample was defined . most of these were included in sdss spectroscopy , but several were discovered as part of early follow - up spectroscopy on the arc 3.5 m telescope and other telescopes ( e.g. , fan et al . 1999 ) , including some that have not yet been published . we were granted the opportunity to observe 250 of these in hst snapshot mode ; we did so by including in the sample all quasars with @xmath30 , and all quasars with @xmath31 with @xmath32 . snapshot observations are carried out with the understanding that not all objects will be observed . in this context , we were able to give objects priorities for being observed . we put the 48 objects with absolute 1450 magnitude less than @xmath33 , and/or with @xmath34 at highest priority , and those with @xmath35 at lowest priority . all remaining objects , and those with @xmath36 , were placed at medium priority . at the end , 161 objects were observed ( of which four were presented in paper i ) : 48/48 of the high - priority objects , 97/154 of the medium priority objects , and 16/48 of the low - priority objects . the four @xmath8 quasars included in our _ hst _ snapshot program were presented in paper i. in this paper , we describe the results of imaging of 157 sdss quasars with @xmath1 ; see table [ tab : tab1 ] . for the sake of completeness , we also tabulate those 89 sources that were included in our sample , but were never observed by _ hst _ , see table [ tab : tab2 ] . seven objects in table 1 and five objects in table 2 are previously unpublished . images of the 157 quasars were acquired with the high resolution camera ( hrc ) on the advanced camera for surveys ( acs ) . observations were taken in the sdss @xmath0-band ( f775w ) . the wide - field camera on acs has higher sensitivity in @xmath0 than does the hrc , but has substantially higher overhead and is mildly undersampled . the exposure times were 640 seconds for each object , 320 seconds in each of two exposures to help in cosmic ray rejection . the data processing was discussed in detail in paper i , but we briefly review the process here . the raw images were calibrated by the calacs package in iraf as part of on - the - fly - reprocessing ( otfr ) at the time of download . the images that we present are the `` cosmic ray rejected '' ( crj ) images that are output by the otfr algorithms at stsci . the crj files have all been reduced in the standard manner , including having been overscan- , bias- and dark - corrected , flat - fielded and photometrically calibrated , in addition to having bad pixels masked and cosmic rays removed ( see the acs manual for more details ) . _ hst _ resolution data are needed for this project as the median expected splitting of gravitationally lensed quasars is predicted ( and observed ) to be somewhat less than @xmath37 ( turner et al . 1984 ; hinshaw & krauss 1987 ; browne et al . 2003 ) and the sdss images themselves have point - spread - function ( psf ) widths of order @xmath38 ( abazajian et al . 2003 ) with an image scale of @xmath39 . our acs images , on the other hand , have an image scale of @xmath40 and the point spread function is narrow enough that any lens with a separation greater than @xmath41 will be obvious by visual inspection , as we show explicitly in paper i. these observations should therefore be sensitive to essentially all expected lenses . we search for faint secondary images at separations smaller than a few tenths of an arcsecond by fitting and removing a model for the point spread function of each image . we use version v6.1 ( which includes on - orbit updates for acs ) of the tiny tim software ( krist 1995 ) , which produces a model psf for the instruments on _ hst _ given the object s ( observed ) spectral energy distribution and position in the focal plane , the filter curve , and knowledge of the optics of the instrument . we fit the psf model to each crj image , allowing the location on the ccd and normalization to vary , and using sinc interpolation when shifting the model psf by fractional pixels . example psf - subtracted images are shown in figure [ fig : fig1 ] . each of the images shows the familiar first airy ring ; on a very hard stretch , the second airy ring is faintly visible . as discussed in paper i , for a secondary image offset by @xmath42 with a flux ratio of 10:1 , the secondary object is visible as an enhancement in the first airy ring of the primary object ; it becomes clear upon psf subtraction ( fig . [ fig : fig1 ] ) . with a flux ratio of 100:1 , we can not discern a pair with @xmath42 separation even after subtraction of the psf ; however , even if we were able to do so , it would not significantly improve the constraints that we derive below since most split images will have larger image separations . none of our 157 targets appears to be gravitationally lensed . figure [ fig : fig1 ] shows only the ten objects ( both with and without psf subtraction ) which displayed any hint of a second object within the 5 field of view . based on their brightness and morphology , these secondary objects are most likely to be galaxies along the line of sight to the target objects or ( in some cases ) uncorrected cosmetic defects . since a gravitationally lensed pair of quasars with separation roughly comparable to the seeing size may be classified as extended ( rather than point ) sources , we must account for the effect that a morphological selection criterion has on our analysis . for the very highest redshift ( @xmath43 ) quasars found from sdss imaging ( e.g. , fan et al . 2001a , and references therein ) , there is no selection bias against lensed quasars since no morphology restriction was imposed . however , for somewhat lower redshift sources , which were selected with the automated pipeline ( richards et al . 2002 ) , there is a strong bias against high-@xmath11 ( @xmath44 ) quasars that appear extended ( whether due to lensing or errors in morphological classification as a result of low s / n ) . this bias is intentional ; the survey can not afford the fibers that would be needed to explore all of the extended sources in the high-@xmath11 quasar portion of the sdss color - color diagram . the vast majority of these objects are faint moderate - redshift ( @xmath45 ) galaxies . as discussed in paper i and ( 2003 ) , in order to test the ability of the sdss photometric pipeline to identify quasar pairs , we created simulated sdss images of pairs of point sources in these observing conditions and at the appropriate signal - to - noise ratio . the sdss star - galaxy separator would have classified as a galaxy , and hence excluded from the spectroscopic sample , any pair of point sources having both an image separation @xmath46 , and a flux ratio less than @xmath47 5:1 . thus the sdss data are suitable for exploring only a fraction of the parameter space of lensing ( in terms of separations and flux ratios ) that is of interest . the morphological selection likelihood depends on the luminosity and on the assumed lensing flux ratio and splitting angle , for any given individual quasar . we therefore fold this bias into our probability calculations below on an object by object basis . for a given quasar in our sample that has an apparent magnitude of @xmath48 at redshift @xmath49 , and for a given hypothesized magnification factor @xmath50 ( which , for a singular isothermal sphere [ sis ] , implies a flux ratio of @xmath51 ) for a lensing halo of mass @xmath52 ( which determines the splitting angle @xmath53 ) , we compute the probability @xmath54 that the object would have been classified as extended and discarded by the photometric pipeline . to compute the impact of the selection on our final constraints , we compute the product of the selection likelihood with the expected lensing probability , which gives the probability that this quasar would have been selected by sdss _ and _ strongly lensed : @xmath55 \times p_{\rm lens,0}(\mu_s , m)$ ] , where @xmath56 is the probability of multiply - imaged lensing in the absence of any other selection effects . finally , we integrate this product over all combinations of magnifications and lensing halo masses that yield two detectable images with _ hst_. we find that typically the inclusion of the morphological selection reduces the total lensing probability by a factor of three ( i.e. , the value of the above defined integral would be three times larger if the factor @xmath54 was excluded ) . more specifically , we find that only @xmath47 1/3 of true lenses would have made it into our sample , with @xmath47 2/3 being classified as extended sources . in table 1 , we list the overall lensing probability for each source . sixteen of the 157 sources were selected without regard to morphology ( as indicated in table 1 ) ; we therefore did not apply the selection function to these objects . we note that there is a further possible selection effect due to the fact that psf magnitudes do not contain 100% of the flux from a gravitationally lensed pair of images , hence effectively reducing the magnification bias . this effect turns out to be less significant , partly because those lensed objects for which it would be most important are already removed from the sample by the star - galaxy morphological bias . overall , we find the effect is likely to shift the luminosity scale by @xmath570.1 magnitudes or less , and we ignore it in what follows . as discussed in paper i , existing constraints on the high - redshift qlf are limited by sample size and to the most luminous objects . ( 1995 ) , using a set of 90 quasars with @xmath58 , found a power - law luminosity function slope ( @xmath59 ) of roughly @xmath60 . ( 2001b ) measured @xmath61 from 39 quasars in the redshift interval 3.6 to 5.0 . however , this relatively shallow apparent slope does not necessarily represent the intrinsic slope of the qlf , which is expected to be much steeper if a substantial fraction of these quasars are magnified by lensing ( e.g. , schneider , ehlers , & falco 1992 ) . in line with paper i and previous work , we describe the intrinsic ( not necessarily observed ) quasar luminosity function as a broken power law ( e.g. , boyle , shanks , & peterson 1988 ; pei 1995 ) : @xmath62 the qlf is described by four parameters : the normalization @xmath63 , the faint - end slope @xmath64 , the bright - end slope @xmath65 , and the characteristic luminosity @xmath66 at which the qlf steepens . the lensing probability is most sensitive to the last two parameters , @xmath65 and @xmath66 . the faint end slope has negligible impact on our analsysis and we set it to @xmath67 ( e.g. , pei 1995 ) . we apply the lensing model from ( 2002 ) , in which lenses are associated with dark matter halos , to compute the total lensing probability , including the effect of magnification bias . in this model , the abundance of halos as a function of potential well depth is adopted from the simulations of ( 2001 ) . as discussed in paper i , we assume that all halos below @xmath68 have sis profiles ( adopting a standard conversion between circular velocity and halo mass ) , while all halos above this mass follow the dark matter density profile suggested by , frenk , & white ( 1997 , hereafter nfw ) . nfw profiles are much less efficient lenses than are sis profiles . this prescription is essentially equivalent to ignoring all lenses above a halo mass of @xmath69 , as the massive halos do not contribute to lensing at small separations . we do not include here more complex lens models , such as those including external shear or ellipticity . as discussed by , kuhlen , & haiman ( 2005 ) , such models can , in general , boost the lensing probabilities ( although that paper considered in detail only the boost that can occur for singly imaged lensing probabilities ) . in the case of @xmath70 and @xmath71 , the mean probability for lensing is about 4% for each source . the morphology selection effect causes this mean probability to be reduced by about a factor of 3 ( as discussed above ) . in table [ tab : tab1 ] we list each of the quasars that were observed by _ hst _ and list their lensing probabilities , which vary relatively little from source to source . the median redshift of the sample is @xmath72 . the lack of lenses among these 157 quasars allows us to place constraints on the qlf at high redshift , shown in figure [ fig : fig2 ] . the lensing probability is a function of both the break and the faint - end slope . assuming a break of @xmath73 ( @xmath74 ) , the above lensing probabilities yield a 3@xmath75 constraint on the bright end slope of @xmath4 . note that we have used a somewhat brighter break luminosity than we did in paper i for the @xmath9 quasars as the characteristic luminosity is thought to evolve with redshift . since this choice of break luminosity is somewhat arbitrary ( as it is an extrapolation from low - redshift and the qlf may not even exhibit a strong break at high-@xmath11 ) , we also show the constraints on @xmath59 as a function of break luminosity in figure [ fig : fig2 ] . finally , it is useful to consider constraints that we can place on the underlying lensing halo population . without any magnification bias , the total optical depth to multiply - imaged lensing at @xmath76 is only 0.25 percent , which is reduced by the sdss morphological selection to 0.08 percent . this makes the lack of lenses among our 157 sources unsurprising . we find that to reduce the probability of not finding even one lens in the whole sample , the single - object lensing probability must be boosted by a factor of about 40 ( this is essentially the typical magnification bias produced by the model qlf we constrain at 3@xmath75 ) . as a result , we rule out ( at 3@xmath75 confidence ) lensing models in which there are @xmath77 times more galaxies than implied by the halo mass function we adopted from jenkins et al . while this is a weak constraint , it is still the best direct limit on the number density of @xmath78 halos at @xmath79 ( the redshift and mass range dominating the expected lensing probability ; see figure 1 in 2002 ) . constraining the high-@xmath11 qlf slope is particularly important to understand the roles of accretion and feedback in the growth of galaxies ( e.g. , silk & rees 1998 ; fabian 1999 ; wyithe & loeb 2003b ; hopkins et al . for the most luminous quasars with @xmath80 ( i.e. , the redshift at which the quasar comoving density peaks ) , the bright end slope has been shown to be @xmath81 ( croom et al . 2004 ; richards et al . 2005 ) . for the most luminous quasars at @xmath82 , the measured slope is @xmath83 ( fan et al . while the slope of the qlf has been well measured , understanding the physical processes behind the qlf and its evolution are still open questions and various explanations have been proposed . & loeb ( 2003a ) suggest that feedback mechanisms may prevent the gas in the most massive dark matter halos from collapsing at low-@xmath11 , suppressing the number of luminous low-@xmath11 quasars relative to high-@xmath11 . alternatively , in the model of ( 2005a , 2005b ) , the bright end of the qlf is defined by near - eddington accretion at the peak luminosity in the history of a quasar , whereas the faint end slope may be due to sub - eddington accretion . tighter limits on the bright end slope of the high - redshift qlf provide important constraints that any such models must account for . in this work , we have obtained high resolution _ hst _ images of 157 @xmath1 redshift quasars ( known prior to september 2001 ) to look for the signature of gravitational lensing . we have found no evidence of multiple images , significantly limiting the amount by which these quasars can be magnified by foreground mass concentrations ( in the absence of microlensing ) . the lack of any strong lenses puts a @xmath84 constraint on the intrinsic bright end slope of the @xmath8545 luminosity function of @xmath86 . our sample has a strong bias against pairs with @xmath87 due to the exclusion of objects which appear extended in sdss images . we are currently exploring methods for cleanly selecting such wide separation pairs from the sdss imaging data . 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 sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , the korean scientist group , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . support for program # 9472 was provided by nasa through a grant ( hst - go-09472.01-a ; m. a. s and g. t. r. ) from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . this work was supported in part by nasa atp grant nng04gi88 g ( z. h. ) , national science foundation grants ast-0307582 ( d. p. s. ) , ast-0307409 ( m. a. s. ) , ast-0307384 ( x. f. ) , ast-0307291 ( z. h. ) and ast-0307200 ( z. h. ) . x. f. and d. j. e. acknowledge alfred p. sloan research fellowships . x. f. further acknowledges support from a david and lucile packard fellow in science and engineering . lllllllll 001115.23 + 144601.8 & 4.924 & @xmath88 & 18.41 & 18.26 & 0.3137 & 0.0906 & aa & 7 + 001134.52 + 155137.3 & 4.394 & @xmath89 & 20.20 & 19.91 & 0.0553 & 0.0164 & ab & 7 + [email protected] & 4.976 & @xmath91 & 19.61 & 19.46 & 0.1103 & 0.0277 & ac & 7 + 001813.88 + 142455.6 & 4.221 & @xmath92 & 19.41 & 19.30 & 0.0945 & 0.0269 & ad & 7 + 001918.43 + 150611.3 & 4.134 & @xmath93 & 19.91 & 19.94 & 0.0563 & 0.0160 & ae & 7 + [email protected] & 4.352 & @xmath94 & 19.50 & 19.36 & 0.0842 & 0.0345 & af & 3 + 002618.67 + 140946.7 & 4.578 & @xmath95 & 20.11 & 19.68 & 0.0868 & 0.0221 & ag & 7 + 003525.28 + 004002.7@xmath96 & 4.750 & @xmath97 & 19.42 & 19.90 & 0.0999 & 0.0999 & ai & 1 + [email protected] & 4.040 & @xmath98 & 19.89 & 19.57 & 0.0662 & 0.0175 & aj & 0 + [email protected]@xmath96 & 4.350 & @xmath99 & 20.27 & 20.55 & 0.0727 & 0.0727 & ak & 6 + 003749.19 + 155208.4 & 4.076 & @xmath100 & 19.72 & 19.72 & 0.0443 & 0.0182 & al & 6 + [email protected] & 4.972 & @xmath101 & 19.35 & 19.05 & 0.2124 & 0.0531 & am & 7 + [email protected] & 4.416 & @xmath102 & 19.53 & 19.39 & 0.1104 & 0.0303 & an & 6 + 005922.65 + 000301.4 & 4.161 & @xmath103 & 19.43 & 19.24 & 0.1069 & 0.0394 & ao & 3 + 010619.25 + 004823.4 & 4.430 & @xmath101 & 18.70 & 18.59 & 0.1980 & 0.0546 & aq & 6 + 012004.82 + 141108.2 & 4.720 & @xmath104 & 20.22 & 20.45 & 0.0375 & 0.0120 & ar & 6 + 012019.99 + 000735.5 & 4.085 & @xmath105 & 19.82 & 19.87 & 0.0579 & 0.0230 & as & 3 + 012211.11 + 150914.3 & 4.510 & @xmath106 & 18.67 & 18.65 & 0.1835 & 0.0574 & at & 7 + [email protected] & 4.008 & @xmath107 & 19.56 & 19.42 & 0.0728 & 0.0200 & au & 7 + [email protected] & 4.103 & @xmath108 & 18.30 & 18.11 & 0.2105 & 0.0668 & av & 3 + [email protected] & 4.268 & @xmath109 & 19.71 & 19.53 & 0.0807 & 0.0211 & aw & 7 + [email protected] & 4.496 & @xmath110 & 20.10 & 20.16 & 0.0660 & 0.0190 & ax & 7 + [email protected] & 4.153 & @xmath111 & 19.92 & 19.73 & 0.0593 & 0.0168 & ay & 7 + [email protected] & 4.205 & @xmath112 & 18.74 & 18.81 & 0.1329 & 0.0399 & a0 & 6 + 015704.10 + 122858.2 & 4.191 & @xmath113 & 19.90 & 19.90 & 0.0460 & 0.0151 & a1 & 6 + [email protected] & 4.169 & @xmath114 & 19.45 & 19.15 & 0.1116 & 0.0315 & a3 & 0 + 020651.37 + 121624.3@xmath96 & 4.810 & @xmath115 & 19.89 & 19.84 & 0.1025 & 0.1025 & a4 & 6 + [email protected] & 4.697 & @xmath116 & 19.54 & 19.14 & 0.1313 & 0.0406 & a5 & 3 + [email protected]@xmath96 & 4.900 & @xmath117 & 19.93 & 20.04 & 0.1286 & 0.1286 & a6 & 1 + [email protected] & 5.410 & @xmath118 & 19.53 & 19.55 & 0.0073 & 0.0029 & a8 & 6 + [email protected] & 4.019 & @xmath98 & 19.27 & 19.18 & 0.0658 & 0.0204 & a9 & 6 + [email protected] & 4.030 & @xmath119 & 18.03 & 18.19 & 0.2240 & 0.0785 & ba & 6 + [email protected] & 4.010 & @xmath98 & 18.33 & 18.26 & 0.0657 & 0.0266 & bb & 3 + 025019.78 + 004650.3@xmath96 & 4.760 & @xmath120 & 19.65 & 19.72 & 0.1243 & 0.1243 & bc & 3 + [email protected] & 4.228 & @xmath102 & 19.62 & 19.65 & 0.1066 & 0.0308 & bf & 6 + 030025.22 + 003224.3 & 4.188 & @xmath121 & 20.18 & 19.90 & 0.0489 & 0.0192 & bg & 3 + [email protected] & 4.030 & @xmath122 & 19.08 & 19.08 & 0.1174 & 0.0341 & bj & 6 + [email protected] & 4.175 & @xmath123 & 19.26 & 19.51 & 0.1003 & 0.0279 & bl & 1 + [email protected] & 4.700 & @xmath124 & 19.10 & 19.35 & 0.1420 & 0.0393 & bm & 6 + [email protected] & 4.270 & @xmath125 & 19.79 & 19.83 & 0.0473 & 0.0172 & bn & 6 + 033829.31 + 002156.2 & 5.001 & @xmath126 & 20.15 & 19.96 & 0.0668 & 0.0199 & bo & 1 + [email protected]@xmath96 & 4.180 & @xmath98 & 19.76 & 19.78 & 0.0683 & 0.0683 & bs & 3 + 040550.26 + 005931.1@xmath96 & 4.050 & @xmath127 & 20.07 & 20.07 & 0.0356 & 0.0356 & bt & 3 + [email protected]@xmath96 & 4.020 & @xmath128 & 19.77 & 19.89 & 0.0230 & 0.0230 & bu & 0 + 073147.00 + 364346.5 & 4.030 & @xmath122 & 19.29 & 19.22 & 0.1174 & 0.0369 & bv & 6 + 074640.16 + 344624.7 & 4.032 & @xmath129 & 19.34 & 19.31 & 0.0846 & 0.0259 & bx & 7 + 075103.95 + 424211.5 & 4.160 & @xmath130 & 18.82 & 18.87 & 0.1443 & 0.0450 & bz & 6 + 075618.14 + 410408.5 & 5.090 & @xmath131 & 20.21 & 20.07 & 0.0915 & 0.0234 & b0 & 6 + 075652.07 + 450258.9 & 4.812 & @xmath132 & 20.24 & 20.23 & 0.0700 & 0.0199 & b1 & 6 + 075732.90 + 441424.6 & 4.179 & @xmath133 & 19.38 & 19.42 & 0.1047 & 0.0300 & b2 & 6 + 081054.88 + 460357.8 & 4.087 & @xmath134 & 18.38 & 18.53 & 0.1542 & 0.0478 & b5 & 6 + 083100.68 + 434426.9 & 4.452 & @xmath135 & 20.08 & 19.95 & 0.0694 & 0.0188 & b7 & 7 + 083103.00 + 523533.5 & 4.512 & @xmath136 & 19.37 & 19.28 & 0.1169 & 0.0329 & b8 & 6 + 083212.38 + 530327.4 & 4.081 & @xmath137 & 19.46 & 19.28 & 0.0863 & 0.0258 & b9 & 6 + 083941.45 + 031817.0 & 4.244 & @xmath138 & 18.79 & 18.55 & 0.1769 & 0.0543 & cc & 7 + 083946.21 + 511202.8 & 4.415 & @xmath139 & 18.98 & 18.83 & 0.1949 & 0.0597 & cd & 6 + [email protected] & 4.156 & @xmath140 & 18.99 & 18.93 & 0.1474 & 0.0456 & ce & 6 + 085151.26 + 020756.0 & 4.294 & @xmath141 & 19.41 & 19.27 & 0.0910 & 0.0250 & cf & 6 + 085210.89 + 535948.9 & 4.275 & @xmath142 & 19.82 & 19.77 & 0.0690 & 0.0189 & cg & 6 + 085227.28 + 504510.8 & 4.219 & @xmath143 & 18.82 & 18.63 & 0.1873 & 0.0546 & ch & 7 + 090100.61 + 472536.1 & 4.610 & @xmath144 & 19.76 & 19.56 & 0.0920 & 0.0275 & ck & 7 + 090440.63 + 535038.8 & 4.298 & @xmath145 & 19.25 & 19.48 & 0.1394 & 0.0393 & cm & 6 + 090634.84 + 023433.8 & 4.544 & @xmath139 & 18.77 & 18.66 & 0.1987 & 0.0712 & co & 7 + 091316.55 + 591921.4 & 5.110 & @xmath146 & 20.50 & 20.47 & 0.0751 & 0.0179 & cq & 6 + 092038.49 + 564235.9 & 4.183 & @xmath109 & 19.95 & 19.06 & 0.0793 & 0.0239 & cr & 6 + 092256.19 + 561849.1 & 4.198 & @xmath147 & 18.99 & 19.67 & 0.1286 & 0.0403 & cs & 6 + 093931.90 + 003955.0@xmath96 & 4.500 & @xmath137 & 20.62 & 20.29 & 0.0936 & 0.0936 & cv & 5 + 094056.02 + 584830.1 & 4.660 & @xmath148 & 19.27 & 19.41 & 0.1255 & 0.0361 & cw & 6 + 094108.35 + 594725.8 & 4.820 & @xmath149 & 19.36 & 19.25 & 0.1325 & 0.0369 & cx & 6 + 094917.18 + 602104.4 & 4.290 & @xmath150 & 20.26 & 20.20 & 0.0446 & 0.0136 & cy & 6 + 095151.17 + 594556.2 & 4.860 & @xmath151 & 19.81 & 19.73 & 0.0732 & 0.0218 & c0 & 6 + 095511.33 + 594030.7 & 4.356 & @xmath152 & 18.30 & 18.56 & 0.1892 & 0.0552 & c1 & 7 + 101053.52 + 644832.1 & 4.745 & @xmath153 & 19.87 & 19.98 & 0.0680 & 0.0182 & c3 & 6 + 101549.00 + 002019.9 & 4.399 & @xmath114 & 19.27 & 19.23 & 0.1166 & 0.0336 & c4 & 6 + 102043.82 + 000105.7 & 4.160 & @xmath154 & 19.83 & 19.88 & 0.0466 & 0.0182 & c5 & 6 + [email protected]@xmath96 & 4.696 & @xmath155 & 20.08 & 20.29 & 0.0417 & 0.0417 & c6 & 4 + 102332.08 + 633508.1 & 4.880 & @xmath136 & 19.69 & 19.44 & 0.1232 & 0.0342 & c7 & 6 + [email protected] & 4.380 & @xmath156 & 19.90 & 20.05 & 0.0535 & 0.0150 & c9 & 7 + 104008.10 + 651429.2 & 4.583 & @xmath157 & 19.83 & 19.83 & 0.0584 & 0.0183 & da & 6 + [email protected] & 4.320 & @xmath158 & 18.80 & 18.94 & 0.1543 & 0.0532 & db & 6 + 104351.19 + 650647.6 & 4.542 & @xmath159 & 19.09 & 18.89 & 0.1324 & 0.0371 & dc & 6 + [email protected] & 4.291 & @xmath160 & 19.33 & 19.30 & 0.1134 & 0.0343 & dg & 2 + 105902.73 + 010404.1 & 4.060 & @xmath123 & 19.21 & 19.32 & 0.0978 & 0.0292 & di & 6 + 110247.29 + 663519.5 & 4.810 & @xmath121 & 20.47 & 20.57 & 0.0546 & 0.0142 & dj & 6 + [email protected] & 4.029 & @xmath129 & 19.25 & 19.41 & 0.0845 & 0.0243 & dk & 5 + 110826.31 + 003706.7 & 4.502 & @xmath115 & 19.84 & 19.55 & 0.0978 & 0.0348 & dm & 6 + 111224.18 + 004630.3 & 4.032 & @xmath161 & 19.66 & 19.62 & 0.0588 & 0.0172 & dn & 6 + [email protected] & 4.590 & @xmath144 & 19.57 & 19.70 & 0.0917 & 0.0271 & do & 2 + [email protected]@xmath96 & 4.795 & @xmath162 & 20.38 & 20.43 & 0.0719 & 0.0719 & dp & 4 + 112253.51 + 005329.7 & 4.570 & @xmath163 & 19.12 & 19.28 & 0.0998 & 0.0409 & dq & 2 + [email protected]@xmath96 & 5.000 & @xmath164 & 20.46 & 20.20 & 0.0661 & 0.0661 & dr & 0 + 113354.89 + 022420.9 & 4.066 & @xmath165 & 18.84 & 18.88 & 0.1480 & 0.0444 & dt & 7 + 113559.93 + 002422.7 & 4.040 & @xmath105 & 19.87 & 19.64 & 0.0572 & 0.0163 & du & 5 + 115547.83 + 022716.1 & 4.355 & @xmath92 & 19.51 & 19.29 & 0.0970 & 0.0273 & dx & 7 + 120439.42 + 663549.7 & 4.052 & @xmath166 & 19.66 & 19.62 & 0.0842 & 0.0252 & dy & 7 + [email protected] & 5.030 & @xmath147 & 19.28 & 19.28 & 0.1471 & 0.0646 & dz & 2 + 120823.81 + 001027.6@xmath96 & 5.280 & @xmath167 & 20.79 & 20.59 & 0.1164 & 0.1164 & d1 & 1 + 121422.02 + 665707.5 & 4.680 & @xmath168 & 18.91 & 18.96 & 0.0755 & 0.0238 & d2 & 7 + 122600.68 + 005923.5 & 4.264 & @xmath169 & 18.90 & 18.83 & 0.1354 & 0.0524 & d3 & 2 + 122622.03 + 662017.9 & 4.017 & @xmath170 & 20.07 & 20.05 & 0.0531 & 0.0192 & d4 & 7 + 122657.97 + 000938.4 & 4.140 & @xmath120 & 19.24 & 19.31 & 0.1118 & 0.0356 & d5 & 7 + [email protected] & 4.700 & @xmath171 & 20.07 & 20.04 & 0.0465 & 0.0134 & d8 & 2 + [email protected] & 4.150 & @xmath172 & 18.55 & 18.40 & 0.1964 & 0.0613 & ec & 6 + [email protected] & 4.031 & @xmath173 & 19.57 & 19.45 & 0.0000 & 0.0000 & ee & 7 + 130002.16 + 011823.0 & 4.614 & @xmath174 & 18.82 & 18.73 & 0.1446 & 0.1446 & ef & 7 + 130619.38 + 023658.9 & 4.874 & @xmath175 & 19.66 & 19.61 & 0.1153 & 0.0298 & ej & 7 + [email protected] & 4.150 & @xmath176 & 18.83 & 18.85 & 0.1408 & 0.0451 & ek & 2 + 131831.83 + 653929.4 & 4.286 & @xmath177 & 20.07 & 20.17 & 0.0484 & 0.0180 & el & 7 + 132110.82 + 003821.6 & 4.700 & @xmath98 & 20.04 & 20.22 & 0.0750 & 0.0255 & em & 2 + 134134.19 + 014157.7 & 4.725 & @xmath178 & 18.93 & 18.98 & 0.1776 & 0.0513 & eq & 7 + 134723.08 + 002158.8 & 4.270 & @xmath179 & 18.96 & 18.71 & 0.1692 & 0.0499 & er & 6 + 140146.52 + 024434.6 & 4.455 & @xmath180 & 18.58 & 18.53 & 0.1974 & 0.0574 & ev & 7 + 140248.07 + 014634.1 & 4.206 & @xmath181 & 18.26 & 18.12 & 0.2215 & 0.0612 & ew & 7 + 141306.09 + 644149.0 & 4.255 & @xmath164 & 19.49 & 19.34 & 0.0583 & 0.0170 & ey & 7 + 141315.36 + 000032.4 & 4.071 & @xmath161 & 19.73 & 19.54 & 0.0594 & 0.0191 & ez & 2 + 142911.84 + 632344.9 & 4.431 & @xmath107 & 20.33 & 20.33 & 0.0795 & 0.0182 & e4 & 7 + 143352.21 + 022714.0 & 4.749 & @xmath119 & 18.33 & 18.32 & 0.2502 & 0.0763 & e5 & 7 + 144255.56 + 590949.9 & 4.333 & @xmath182 & 20.19 & 20.04 & 0.0564 & 0.0198 & e8 & 7 + 144340.71 + 585653.3 & 4.242 & @xmath183 & 18.02 & 18.09 & 0.2966 & 0.0864 & e9 & 7 + 144413.26 + 004836.7@xmath96 & 4.780 & @xmath154 & 21.00 & 21.08 & 0.0522 & 0.0522 & fb & 0 + [email protected] & 4.185 & @xmath124 & 19.09 & 19.06 & 0.1303 & 0.0376 & fd & 6 + 144717.97 + 040112.4 & 4.508 & @xmath184 & 19.33 & 19.21 & 0.1686 & 0.0449 & fe & 7 + 145107.94 + 025615.6 & 4.486 & @xmath117 & 19.11 & 19.08 & 0.1212 & 0.0354 & ff & 7 + 145229.37 + 595156.3 & 4.029 & @xmath91 & 19.54 & 19.32 & 0.0938 & 0.0317 & fi & 7 + 145350.38 + 610109.0 & 4.134 & @xmath185 & 19.50 & 19.49 & 0.0763 & 0.0235 & fj & 7 + 145747.66 + 575332.1 & 4.355 & @xmath168 & 20.04 & 19.93 & 0.0714 & 0.0230 & fk & 7 + 150527.34 + 573632.0 & 4.385 & @xmath110 & 19.98 & 19.75 & 0.0647 & 0.0197 & fl & 7 + 150847.60 + 571501.3 & 4.876 & @xmath186 & 20.06 & 19.99 & 0.0821 & 0.0240 & fm & 7 + 151002.92 + 570243.3 & 4.310 & @xmath187 & 20.06 & 19.95 & 0.0682 & 0.0201 & fn & 7 + 151155.98 + 040802.9 & 4.621 & @xmath188 & 19.85 & 19.71 & 0.0866 & 0.0321 & fp & 7 + 161616.25 + 513336.9 & 4.528 & @xmath189 & 19.58 & 19.54 & 0.1153 & 0.0329 & fz & 7 + 163257.07 + 441110.2 & 4.105 & @xmath130 & 18.93 & 18.54 & 0.1427 & 0.0431 & f1 & 7 + 163950.51 + 434003.7 & 4.007 & @xmath190 & 17.91 & 17.79 & 0.2738 & 0.0860 & f2 & 7 + 165354.62 + 405402.2 & 4.966 & @xmath112 & 18.79 & 18.54 & 0.1491 & 0.0481 & f3 & 7 + 170804.89 + 602201.9 & 4.350 & @xmath191 & 19.78 & 19.99 & 0.0762 & 0.0225 & f4 & 6 + 171014.51 + 592326.4 & 4.536 & @xmath141 & 19.67 & 19.73 & 0.0950 & 0.0288 & f5 & 6 + 171224.92 + 560624.9 & 4.229 & @xmath182 & 20.02 & 20.07 & 0.0552 & 0.0195 & f6 & 6 + 171808.67 + 551511.2 & 4.620 & @xmath153 & 19.99 & 20.06 & 0.0667 & 0.0201 & f7 & 6 + 172007.20 + 602823.8 & 4.400 & @xmath89 & 20.20 & 20.08 & 0.0554 & 0.0188 & f8 & 6 + 173744.87 + 582829.5 & 4.940 & @xmath130 & 19.33 & 19.29 & 0.1626 & 0.0476 & f9 & 6 + [email protected] & 4.224 & @xmath192 & 19.71 & 19.52 & 0.1004 & 0.0232 & ga & 7 + [email protected] & 4.341 & @xmath177 & 20.20 & 19.95 & 0.0490 & 0.0130 & gb & 7 + 210216.52 + 104906.5 & 4.175 & @xmath193 & 19.22 & 18.96 & 0.1436 & 0.0470 & gc & 7 + 220008.66 + 001744.8 & 4.791 & @xmath194 & 19.30 & 19.19 & 0.1150 & 0.0324 & gg & 6 + [email protected] & 4.160 & @xmath116 & 19.41 & 18.99 & 0.1198 & 0.0506 & gh & 7 + 221644.01 + 001348.2@xmath96 & 4.990 & @xmath195 & 20.30 & 20.31 & 0.1726 & 0.1726 & gj & 6 + 221855.11 + 134708.6 & 4.277 & @xmath131 & 19.83 & 19.53 & 0.0793 & 0.0219 & gl & 7 + [email protected] & 4.861 & @xmath176 & 19.31 & 19.07 & 0.1576 & 0.0602 & gn & 7 + [email protected] & 5.151 & @xmath196 & 20.16 & 19.83 & 0.1684 & 0.0576 & gp & 7 + [email protected] & 4.223 & @xmath93 & 19.95 & 19.88 & 0.0574 & 0.0156 & gr & 7 + 224630.86 + 131706.7 & 4.116 & @xmath166 & 19.70 & 19.65 & 0.0854 & 0.0223 & gs & 7 + [email protected] & 4.041 & @xmath192 & 19.35 & 18.95 & 0.0965 & 0.0277 & gw & 7 + 234003.50 + 140257.2 & 4.535 & @xmath197 & 19.12 & 18.91 & 0.1552 & 0.0467 & g1 & 7 + 234025.97 + 135009.1 & 4.173 & @xmath198 & 20.18 & 19.93 & 0.0399 & 0.0111 & g2 & 7 + 234750.31 + 134102.7 & 4.265 & @xmath100 & 19.75 & 19.63 & 0.0462 & 0.0152 & g3 & 7 + 235152.80 + 160048.9 & 4.668 & @xmath105 & 19.79 & 19.83 & 0.0647 & 0.0189 & g4 & 7 + 235344.26 + 143525.2 & 4.257 & @xmath199 & 19.87 & 19.66 & 0.0776 & 0.0206 & g5 & 7 + llll|llll 003126.80 + 150739.6 & 4.291 & 19.97 & 7 & [email protected] & 4.620 & 19.91 & 7 + 010326.89 + 005538.6 & 4.159 & 20.07 & 0 & 133211.90 + 031556.3 & 4.727 & 19.29 & 7 + 015032.87 + 143425.5 & 4.284 & 20.09 & 6 & [email protected] & 4.427 & 19.91 & 6 + [email protected] & 4.026 & 20.29 & 7 & [email protected] & 4.034 & 19.88 & 6 + [email protected] & 4.592 & 20.53 & 0 & [email protected] & 4.420 & 20.15 & 6 + [email protected] & 4.505 & 19.84 & 6 & 140404.63 + 031403.9 & 4.969 & 19.53 & 7 + 025204.28 + 003136.9 & 4.119 & 19.91 & 6 & [email protected] & 4.213 & 19.30 & 2 + 030437.21 + 004653.6 & 4.281 & 20.12 & 0 & 141534.91 + 033132.1 & 4.451 & 19.89 & 7 + [email protected]@xmath96 & 4.630 & 19.89 & 1 & 142004.11 + 022708.7 & 4.187 & 20.09 & 7 + [email protected]@xmath96 & 4.800 & 20.69 & 3 & 142408.34 + 024219.9 & 4.342 & 19.79 & 7 + [email protected] & 4.142 & 20.20 & 6 & 144117.46 + 035910.5 & 4.312 & 19.60 & 7 + [email protected] & 4.062 & 19.52 & 6 & 144231.72 + 011055.3 & 4.560 & 19.93 & 6 + [email protected] & 4.041 & 20.23 & 6 & [email protected] & 4.552 & 19.29 & 6 + 073354.93 + 321241.5 & 4.446 & 20.38 & 7 & [email protected] & 4.160 & 19.47 & 2 + 074907.57 + 355543.8 & 4.257 & 20.09 & 7 & [email protected]@xmath96 & 4.660 & 20.70 & 1 + 080159.24 + 433624.9 & 4.166 & 20.26 & 6 & 145212.86 + 023526.4 & 4.916 & 19.92 & 7 + 080549.94 + 482345.8 & 4.208 & 19.86 & 6 & 151041.79 + 031810.5 & 4.230 & 19.63 & 7 + 081241.12 + 442129.0 & 4.334 & 20.21 & 6 & 151909.09 + 030633.7 & 4.391 & 19.47 & 7 + 083824.32 + 460443.7 & 4.011 & 19.60 & 7 & 152245.19 + 024543.8 & 4.087 & 18.91 & 7 + 085430.19 + 004213.6 & 4.079 & 20.02 & 6 & 152443.19 + 011358.9@xmath96 & 4.114 & 19.98 & 6 + 085634.93 + 525206.3 & 4.790 & 20.31 & 6 & [email protected] & 4.410 & 19.95 & 2 + [email protected] & 4.450 & 19.77 & 6 & 152743.86 + 035301.3 & 4.226 & 19.86 & 7 + [email protected] & 4.254 & 19.89 & 6 & [email protected]@xmath96 & 4.620 & 19.73 & 2 + 091016.79 + 575331.0 & 4.005 & 19.50 & 6 & 160207.95 + 523717.9 & 4.898 & 19.85 & 7 + 092303.53 + 024739.4 & 4.669 & 20.39 & 7 & [email protected] & 4.920 & 19.78 & 2 + 092819.28 + 534024.2 & 4.413 & 19.67 & 7 & 161544.13 + 010401.8@xmath96 & 4.013 & 20.20 & 5 + 095000.17 + 620318.5 & 4.062 & 20.22 & 6 & 162048.74 + 002005.7 & 4.199 & 19.36 & 6 + 100413.14 + 630437.3 & 4.130 & 20.24 & 6 & [email protected] & 4.268 & 19.38 & 7 + [email protected] & 4.031 & 19.05 & 6 & [email protected] & 4.192 & 19.43 & 7 + [email protected] & 4.173 & 19.54 & 6 & [email protected] & 4.132 & 19.73 & 7 + 105602.36 + 003222.0 & 4.064 & 19.66 & 6 & 221320.00 + 134832.5 & 4.129 & 19.78 & 7 + [email protected] & 4.604 & 19.87 & 6 & 221705.72 + 135352.7 & 4.348 & 20.09 & 7 + [email protected] & 4.850 & 19.64 & 7 & 222050.81 + 001959.0@xmath96 & 4.700 & 20.21 & 6 + 113745.66 + 012715.1 & 4.070 & 19.72 & 7 & 222807.58 + 003526.2 & 4.631 & 19.80 & 0 + 115158.23 + 030341.7 & 4.701 & 20.47 & 7 & [email protected] & 4.372 & 20.07 & 7 + 120640.73 + 033414.9 & 4.386 & 19.86 & 7 & [email protected] & 4.174 & 20.28 & 7 + [email protected] & 4.144 & 19.75 & 7 & [email protected] & 4.003 & 19.71 & 0 + [email protected] & 4.259 & 19.04 & 7 & 225246.44 + 142525.7 & 4.920 & 19.91 & 7 + 123937.17 + 674020.7 & 4.425 & 20.18 & 6 & [email protected] & 4.337 & 20.16 & 7 + [email protected] & 4.187 & 19.83 & 7 & [email protected] & 4.507 & 20.35 & 7 + [email protected] & 4.291 & 20.10 & 6 & 232112.39 + 143312.0 & 4.026 & 20.24 & 7 + 125802.61 + 022721.2 & 4.269 & 19.84 & 7 & 233255.72 + 141916.4 & 4.751 & 20.09 & 7 + 130039.13 + 032203.8 & 4.155 & 20.09 & 7 & 235403.85 + 155630.3 & 4.573 & 20.34 & 7 + 130216.13 + 003032.0 & 4.607 & 19.86 & 6 & 235718.36 + 004350.3 & 4.363 & 20.13 & 1 + [email protected] & 4.052 & 19.79 & 6 & & & & +

the problem of error correction in both coherent and noncoherent network coding is considered under an adversarial model . for coherent network coding , where knowledge of the network topology and network code is assumed at the source and destination nodes , the error correction capability of an ( outer ) code is succinctly described by the rank metric ; as a consequence , it is shown that universal network error correcting codes achieving the singleton bound can be easily constructed and efficiently decoded . for noncoherent network coding , where knowledge of the network topology and network code is not assumed , the error correction capability of a ( subspace ) code is given exactly by a new metric , called the _ injection metric _ , which is closely related to , but different than , the subspace metric of ktter and kschischang . in particular , in the case of a non - constant - dimension code , the decoder associated with the injection metric is shown to correct more errors then a minimum - subspace - distance decoder . all of these results are based on a general approach to adversarial error correction , which could be useful for other adversarial channels beyond network coding . adversarial channels , error correction , injection distance , network coding , rank distance , subspace codes .

0805.3824

the problem of error correction for a network implementing linear network coding has been an active research area since 2002 @xcite . the crucial motivation for the problem is the phenomenon of error propagation , which arises due to the recombination characteristic at the heart of network coding . a single corrupt packet occurring in the application layer ( e.g. , introduced by a malicious user ) may proceed undetected and contaminate other packets , causing potentially drastic consequences and essentially ruling out classical error correction approaches . in the basic multicast model for linear network coding , a source node transmits @xmath0 packets , each consisting of @xmath1 symbols from a finite field @xmath2 . each link in the network transports a packet free of errors , and each node creates outgoing packets as @xmath2-linear combinations of incoming packets . there are one or more destination nodes that wish to obtain the original source packets . at a specific destination node , the received packets may be represented as the rows of an @xmath3 matrix @xmath4 , where @xmath5 is the matrix whose rows are the source packets and @xmath6 is the transfer matrix of the network . errors are incorporated in the model by allowing up to @xmath7 error packets to be added ( in the vector space @xmath8 ) to the packets sent over one or more links . the received matrix @xmath9 at a specific destination node may then be written as @xmath10 where @xmath11 is a @xmath12 matrix whose rows are the error packets , and @xmath13 is the transfer matrix from these packets to the destination . under this model , a coding - theoretic problem is how to design an outer code and the underlying network code such that reliable communication ( to all destinations ) is possible . this coding problem can be posed in a number of ways depending on the set of assumptions made . for example , we may assume that the network topology and the network code are known at the source and at the destination nodes , in which case we call the system _ coherent network coding_. alternatively , we may assume that such information is unavailable , in which case we call the system _ noncoherent network coding_. the error matrix @xmath11 may be random or chosen by an adversary , and there may be further assumptions on the knowledge or other capabilities of the adversary . the essential assumption , in order to pose a meaningful coding problem , is that the number of injected error packets , @xmath7 , is bounded . error correction for coherent network coding was originally studied by cai and yeung @xcite . aiming to establish fundamental limits , they focused on the fundamental case @xmath14 . in @xcite ( see also @xcite ) , the authors derive a singleton bound in this context and construct codes that achieve this bound . a drawback of their approach is that the field size required can be very large ( on the order of @xmath15 , where @xmath16 is the number of edges in the network ) , and no efficient decoding method is given . similar constructions , analyses and bounds appear also in @xcite . in section [ sec : coherent - network - coding ] , we approach this problem ( for general @xmath1 ) under a different framework . we assume the pessimistic situation in which the adversary can not only inject up to @xmath7 packets but can also freely choose the matrix @xmath13 . in this scenario , it is essential to exploit the structure of the problem when @xmath17 . the proposed approach allows us to find a metric the rank metric that succinctly describes the error correction capability of a code . we quite easily obtain bounds and constructions analogous to those of @xcite , and show that many of the results in @xcite can be reinterpreted and simplified in this framework . moreover , we find that our pessimistic assumption actually incurs no penalty since the codes we propose achieve the singleton bound of @xcite . an advantage of this approach is that it is _ universal _ , in the sense that the outer code and the network code may be designed independently of each other . more precisely , the outer code may be chosen as any rank - metric code with a good error - correction capability , while the network code can be designed as if the network were error - free ( and , in particular , the field size can be chosen as the minimum required for multicast ) . an additional advantage is that encoding and decoding of properly chosen rank - metric codes can be performed very efficiently @xcite . for noncoherent network coding , a combinatorial framework for error control was introduced by ktter and kschischang in @xcite . there , the problem is formulated as the transmission of subspaces through an operator channel , where the transmitted and received subspaces are the row spaces of the matrices @xmath5 and @xmath9 in ( [ eq : matrix - model ] ) , respectively . they proposed a metric that is suitable for this channel , the so - called subspace distance @xcite . they also presented a singleton - like bound for their metric and subspace codes achieving this bound . the main justification for their metric is the fact that a minimum subspace distance decoder seems to be the necessary and sufficient tool for optimally decoding the disturbances imposed by the operator channel . however , when these disturbances are translated to more concrete terms such as the number of error packets injected , only decoding guarantees can be obtained for the minimum distance decoder of @xcite , but no converse . more precisely , assume that @xmath7 error packets are injected and a general ( not necessarily constant - dimension ) subspace code with minimum subspace distance @xmath18 is used . in this case , while it is possible to guarantee successful decoding if @xmath19 , and we know of specific examples where decoding fails if this condition is not met , a general converse is not known . in section [ sec : noncoherent - network - coding ] , we prove such a converse for a new metric called the _ injection distance_under a slightly different transmission model . we assume that the adversary is allowed to arbitrarily select the matrices @xmath6 and @xmath13 , provided that a lower bound on the rank of @xmath6 is respected . under this pessimistic scenario , we show that the injection distance is the fundamental parameter behind the error correction capability of a code ; that is , we can guarantee correction of @xmath7 packet errors _ if and only if _ @xmath7 is less than half the minimum injection distance of the code . while this approach may seem too pessimistic , we provide a class of examples where a minimum - injection - distance decoder is able to correct more errors than a minimum - subspace - distance decoder . moreover , the two approaches coincide when a constant - dimension code is used . in order to give a unified treatment of both coherent and noncoherent network coding , we first develop a general approach to error correction over ( certain ) adversarial channels . our treatment generalizes the more abstract portions of classical coding theory and has the main feature of mathematical simplicity . the essence of our approach is to use a single function called a _ discrepancy function_to fully describe an adversarial channel . we then propose a distance - like function that is easy to handle analytically and ( in many cases , including all the channels considered in this paper ) precisely describes the error correction capability of a code . the motivation for this approach is that , once such a distance function is found , one can virtually forget about the channel model and fully concentrate on the combinatorial problem of finding the largest code with a specified minimum distance ( just like in classical coding theory ) . interestingly , our approach is also useful to characterize the error detection capability of a code . the remainder of the paper is organized as follows . section [ sec : preliminaries ] establishes our notation and review some basic facts about matrices and rank - metric codes . section [ sec : adversarial - model ] presents our general approach to adversarial error correction , which is subsequently specialized to coherent and noncoherent network coding models . section [ sec : coherent - network - coding ] describes our main results for coherent network coding and discusses their relationship with the work of yeung et al . @xcite . section [ sec : noncoherent - network - coding ] describes our main results for noncoherent network coding and discusses their relationship with the work of ktter and kschischang @xcite . section [ sec : conclusion ] presents our conclusions . define @xmath20 and @xmath21^+ = \max\{x,0\}$ ] . the following notation is used many times throughout the paper . let @xmath22 be a set , and let @xmath23 . whenever a function @xmath24 is defined , denote @xmath25 if @xmath26 is called a `` distance '' between @xmath27 and @xmath28 , then @xmath29 is called the _ minimum _ `` distance '' of @xmath30 . let @xmath2 denote the finite field with @xmath31 elements . we use @xmath32 to denote the set of all @xmath33 matrices over @xmath2 and use @xmath34 to denote the set of all subspaces of the vector space @xmath8 . let @xmath35 denote the dimension of a vector space @xmath36 , let @xmath37 denote the row space of a matrix @xmath5 , and let @xmath38 denote the number of nonzero rows of @xmath5 . recall that @xmath39 . let @xmath40 and @xmath36 be subspaces of some fixed vector space . recall that the sum @xmath41 is the smallest vector space that contains both @xmath40 and @xmath36 , while the intersection @xmath42 is the largest vector space that is contained in both @xmath40 and @xmath36 . recall also that @xmath43 the rank of a matrix @xmath44 is the smallest @xmath45 for which there exist matrices @xmath46 and @xmath47 such that @xmath48 . note that both matrices obtained in the decomposition are full - rank ; accordingly , such a decomposition is called a full - rank decomposition @xcite . in this case , note that , by partitioning @xmath49 and @xmath50 , the matrix @xmath5 can be further expanded as @xmath51 where @xmath52 . another useful property of the rank function is that , for @xmath44 and @xmath53 , we have @xcite @xmath54 let @xmath55 be matrices . the _ rank distance _ between @xmath5 and @xmath9 is defined as @xmath56 it is well known that the rank distance is indeed a metric ; in particular , it satisfies the triangle inequality @xcite . a _ rank - metric code _ is a matrix code @xmath57 used in the context of the rank metric . the singleton bound for the rank metric @xcite ( see also @xcite ) states that every rank - metric code @xmath58 with minimum rank distance @xmath59 must satisfy latexmath:[\[\label{eq : singleton - bound } achieve this bound are called _ maximum - rank - distance _ ( mrd ) codes and they are known to exist for all choices of parameters @xmath31 , @xmath0 , @xmath1 and @xmath61 @xcite . this section presents a general approach to error correction over adversarial channels . this approach is specialized to coherent and noncoherent network coding in sections [ sec : coherent - network - coding ] and [ sec : noncoherent - network - coding ] , respectively . an _ adversarial channel _ is specified by a finite input alphabet @xmath22 , a finite output alphabet @xmath62 and a collection of _ fan - out sets _ @xmath63 for all @xmath64 . for each input @xmath27 , the output @xmath65 is constrained to be in @xmath66 but is otherwise arbitrarily chosen by an adversary . the constraint on the output is important : otherwise , the adversary could prevent communication simply by mapping all inputs to the same output . no further restrictions are imposed on the adversary ; in particular , the adversary is potentially omniscient and has unlimited computational power . a code for an adversarial channel is a subset @xmath23 . we say that a code is _ unambiguous _ for a channel if the input codeword can always be uniquely determined from the channel output . more precisely , a code @xmath30 is unambiguous if the sets @xmath66 , @xmath67 , are pairwise disjoint . the importance of this concept lies in the fact that , if the code is _ not _ unambiguous , then there exist codewords @xmath68 that are _ indistinguishable _ at the decoder : if @xmath69 , then the adversary can ( and will ) exploit this ambiguity by mapping both @xmath27 and @xmath28 to the same output . a decoder for a code @xmath30 is any function @xmath70 , where @xmath71 denotes a decoding failure ( detected error ) . when @xmath67 is transmitted and @xmath72 is received , a decoder is said to be _ successful _ if @xmath73 . we say that a decoder is _ infallible _ if it is successful for all @xmath72 and all @xmath67 . note that the existence of an infallible decoder for @xmath30 implies that @xmath30 is unambiguous . conversely , given any unambiguous code @xmath30 , one can always find ( by definition ) a decoder that is infallible . one example is the exhaustive decoder @xmath74 in other words , an exhaustive decoder returns @xmath27 if @xmath27 is the unique codeword that could possibly have been transmitted when @xmath65 is received , and returns a failure otherwise . ideally , one would like to find a large ( or largest ) code that is unambiguous for a given adversarial channel , together with a decoder that is infallible ( and computationally - efficient to implement ) . it is useful to consider adversarial channels parameterized by an _ adversarial effort _ @xmath75 . assume that the fan - out sets are of the form @xmath76 for some @xmath77 . the value @xmath78 , which we call the _ discrepancy _ between @xmath27 and @xmath65 , represents the minimum effort needed for an adversary to transform an input @xmath27 into an output @xmath65 . the value of @xmath7 represents the maximum adversarial effort ( maximum discrepancy ) allowed in the channel . in principle , there is no loss of generality in assuming ( [ eq : output - sets ] ) since , by properly defining @xmath78 , one can always express any @xmath66 in this form . for instance , one could set @xmath79 if @xmath72 , and @xmath80 otherwise . however , such a definition would be of no practical value since @xmath78 would be merely an indicator function . thus , an effective limitation of our model is that it requires channels that are _ naturally _ characterized by some discrepancy function . in particular , one should be able to interpret the maximum discrepancy @xmath7 as the level of `` degradedness '' of the channel . on the other hand , the assumption @xmath81 imposes effectively no constraint . since @xmath82 is finite , given any `` naturally defined '' @xmath83 , one can always shift , scale and round the image of @xmath84 in order to produce some @xmath85 that induces the same fan - out sets as @xmath84 for all @xmath7 . [ ex : t - error - channel ] let us use the above notation to define a @xmath7-error channel , i.e. , a vector channel that introduces at most @xmath7 symbol errors ( arbitrarily chosen by an adversary ) . assume that the channel input and output alphabets are given by @xmath86 . it is easy to see that the channel can be characterized by a discrepancy function that counts the number of components in which an input vector @xmath27 and an output vector @xmath65 differ . more precisely , we have @xmath87 , where @xmath88 denotes the _ hamming distance _ function . a main feature of our proposed discrepancy characterization is to allow us to study a whole family of channels ( with various levels of degradedness ) under the same framework . for instance , we can use a single decoder for all channels in the same family . define the _ minimum - discrepancy decoder _ given by @xmath89 where any ties in ( [ eq : minimum - discrepancy - decoder ] ) are assumed to be broken arbitrarily . it is easy to see that a minimum - discrepancy decoder is infallible provided that the code is unambiguous . thus , we can safely restrict attention to a minimum - discrepancy decoder , regardless of the maximum discrepancy @xmath7 in the channel . given a fixed family of channels specified by @xmath22 , @xmath62 and @xmath90 , and parameterized by a maximum discrepancy @xmath7we wish to identify the largest ( worst ) channel parameter for which we can guarantee successful decoding . we say that a code is _ @xmath7-discrepancy - correcting _ if it is unambiguous for a channel with maximum discrepancy @xmath7 . the _ discrepancy - correction capability _ of a code @xmath30 is the largest @xmath7 for which @xmath30 is @xmath7-discrepancy - correcting . we start by giving a general characterization of the discrepancy - correction capability . let the function @xmath91 be given by @xmath92 we have the following result . the discrepancy - correction capability of a code @xmath30 is given exactly by @xmath93 . in other words , @xmath30 is @xmath7-discrepancy - correcting if and only if @xmath94 . suppose that the code is not @xmath7-discrepancy - correcting , i.e. , that there exist some distinct @xmath95 and some @xmath96 such that @xmath97 and @xmath98 . then @xmath99 . in other words , @xmath100 implies that the code is @xmath7-discrepancy - correcting . conversely , suppose that @xmath101 , i.e. , @xmath102 . then there exist some distinct @xmath95 such that @xmath103 . this in turn implies that there exists some @xmath96 such that @xmath104 . since this implies that both @xmath97 and @xmath98 , it follows that the code is not @xmath7-discrepancy - correcting . at this point , it is tempting to define a `` distance - like '' function given by @xmath105 , since this would enable us to immediately obtain results analogous to those of classical coding theory ( such as the error correction capability being half the minimum distance of the code ) . this approach has indeed been taken in previous works , such as @xcite . note , however , that the terminology `` distance '' suggests a geometrical interpretation , which is not immediately clear from ( [ eq : function - exact - capability ] ) . moreover , the function ( [ eq : function - exact - capability ] ) is not necessarily mathematically tractable . it is the objective of this section to propose a `` distance '' function @xmath106 that is motivated by geometrical considerations and is easier to handle analytically , yet is useful to characterize the correction capability of a code . in particular , we shall be able to obtain the same results as @xcite with much greater mathematical simplicity which will later turn out to be instrumental for code design . for @xmath107 , define the _ @xmath108-distance _ between @xmath27 and @xmath28 as @xmath109 the following interpretation holds . consider the complete bipartite graph with vertex sets @xmath22 and @xmath62 , and assume that each edge @xmath110 is labeled by a `` length '' @xmath78 . then @xmath111 is the length of the shortest path between vertices @xmath107 . roughly speaking , @xmath111 gives the minimum total effort that an adversary would have to spend ( in independent channel realizations ) in order to make @xmath27 and @xmath28 both plausible explanations for some received output . let us compute the @xmath108-distance for the channel of example [ ex : t - error - channel ] . we have @xmath112 , since the hamming distance satisfies the triangle inequality . this bound is achievable by taking , for instance , @xmath113 . thus , @xmath114 , i.e. , the @xmath108-distance for this channel is given precisely by the hamming distance . the following result justifies our definition of the @xmath108-distance . [ prop : correction - guarantee ] for any code @xmath30 , @xmath115 . this follows from the fact that @xmath116 for all @xmath117 . proposition [ prop : correction - guarantee ] shows that @xmath118 gives a lower bound on the correction capability of a code therefore providing a correction guarantee . the converse result , however , is not necessarily true in general . thus , up to this point , the proposed function is only partially useful : it is conceivable that the @xmath108-distance might be too conservative and give a guaranteed correction capability that is lower than the actual one . nevertheless , it is easier to deal with addition , as in ( [ eq : delta - distance - definition ] ) , rather than maximization , as in ( [ eq : function - exact - capability ] ) . a special case where the converse is true is for a family of channels whose discrepancy function satisfies the following condition : [ def : magic - property ] a discrepancy function @xmath119 is said to be _ normal _ if , for all @xmath107 and all @xmath120 , there exists some @xmath96 such that @xmath121 and @xmath122 . [ thm : magic - property - converse ] suppose that @xmath90 is normal . for every code @xmath23 , we have @xmath123 . we just need to show that @xmath124 . take any @xmath107 . since @xmath90 is normal , there exists some @xmath96 such that @xmath125 and either @xmath126 or @xmath127 . thus , @xmath128 and therefore @xmath129 . theorem [ thm : magic - property - converse ] shows that , for certain families of channels , our proposed @xmath108-distance achieves the goal of this section : it is a ( seemingly ) tractable function that precisely describes the correction capability of a code . in particular , the basic result of classical coding theory that the hamming distance precisely describes the error correction capability of a code follows from the fact that the hamming distance ( as a discrepancy function ) is normal . as we shall see , much of our effort in the next sections reduces to showing that a specified discrepancy function is normal . note that , for normal discrepancy functions , we actually have @xmath130 , so theorem [ thm : magic - property - converse ] may also be regarded as providing an alternative ( and more tractable ) expression for @xmath131 . to give a nontrivial example , let us consider a binary vector channel that introduces at most @xmath132 erasures ( arbitrarily chosen by an adversary ) . the input alphabet is given by @xmath133 , while the output alphabet is given by @xmath134 , where @xmath135 denotes an erasure . we may define @xmath136 , where @xmath137 the fan - out sets are then given by @xmath138 . in order to compute @xmath111 , observe the minimization in ( [ eq : delta - distance - definition ] ) . it is easy to see that we should choose @xmath139 when @xmath140 , and @xmath141 when @xmath142 . it follows that @xmath143 . note that @xmath78 is normal . it follows from theorem [ thm : magic - property - converse ] that a code @xmath30 can correct all the @xmath132 erasures introduced by the channel if and only if @xmath144 . this result precisely matches the well - known result of classical coding theory . it is worth clarifying that , while we call @xmath145 a `` distance , '' this function may not necessarily be a metric . while symmetry and non - negativity follow from the definition , a @xmath108-distance may not always satisfy `` @xmath146 '' or the triangle inequality . nevertheless , we keep the terminology for convenience . although this is not our main interest in this paper , it is worth pointing out that the framework of this section is also useful for obtaining results on error _ detection_. namely , the @xmath108-distance gives , in general , a lower bound on the discrepancy detection capability of a code under a bounded discrepancy - correcting decoder ; when the discrepancy function is normal , then the @xmath108-distance precisely characterizes this detection capability ( similarly as in classical coding theory ) . for more details on this topic , see appendix [ sec : detection - capability ] . the basic channel model for coherent network coding with adversarial errors is a matrix channel with input @xmath44 , output @xmath147 , and channel law given by ( [ eq : matrix - model ] ) , where @xmath53 is fixed and known to the receiver , and @xmath148 is arbitrarily chosen by an adversary . here , we make the following additional assumptions : * the adversary has unlimited computational power and is omniscient ; in particular , the adversary knows both @xmath6 and @xmath5 ; * the matrix @xmath149 is arbitrarily chosen by the adversary . we also assume that @xmath150 ( more precisely , we should assume @xmath151 ) ; otherwise , the adversary may always choose @xmath152 , leading to a trivial communications scenario . the first assumption above allows us to use the approach of section [ sec : general - approach ] . the second assumption may seem somewhat `` pessimistic , '' but it has the analytical advantage of eliminating from the problem any further dependence on the network code . ( recall that , in principle , @xmath13 would be determined by the network code and the choice of links in error . ) the power of the approach of section [ sec : general - approach ] lies in the fact that the channel model defined above can be _ completely _ described by the following discrepancy function @xmath153 the discrepancy @xmath154 represents the minimum number of error packets that the adversary needs to inject in order to transform an input @xmath5 into an output @xmath9 , given that the transfer matrix is @xmath6 . the subscript in @xmath155 is to emphasize the dependence on @xmath6 . for this discrepancy function , the minimum - discrepancy decoder becomes @xmath156 similarly , the @xmath108-distance induced by @xmath155 is given by @xmath157 for @xmath158 . we now wish to find a simpler expression for @xmath155 and @xmath159 , and show that @xmath155 is normal . [ lem : coherent - discrepancy - rank ] @xmath160 consider @xmath154 as given by ( [ eq : coherent - discrepancy ] ) . for any feasible triple @xmath161 , we have @xmath162 . this bound is achievable by setting @xmath163 and letting @xmath164 be a full - rank decomposition of @xmath165 . @xmath166 from ( [ eq : coherent - delta - between - codewords ] ) and lemma [ lem : coherent - discrepancy - rank ] , we have @xmath167 . since the rank metric satisfies the triangle inequality , we have @xmath168 . this lower bound can be achieved by choosing , e.g. , @xmath4 . note that @xmath169 is a metric if and only if @xmath6 has full column rank in which case it is precisely the rank metric . ( if @xmath170 , then there exist @xmath171 such that @xmath172 . ) [ thm : coherent - magic ] the discrepancy function @xmath173 is normal . let @xmath158 and let @xmath174 . then @xmath175 . by performing a full - rank decomposition of @xmath176 , we can always find two matrices @xmath177 and @xmath178 such that @xmath179 , @xmath180 and @xmath181 . taking @xmath182 , we have that @xmath183 and @xmath184 . note that , under the discrepancy @xmath155 , a @xmath7-discrepancy - correcting code is a code that can correct _ any _ @xmath7 packet errors injected by the adversary . using theorem [ thm : coherent - magic ] and theorem [ thm : magic - property - converse ] , we have the following result . [ thm : coherent - correction - capability ] a code @xmath185 is guaranteed to correct any @xmath7 packet errors if and only if @xmath186 . theorem [ thm : coherent - correction - capability ] shows that @xmath187 is indeed a fundamental parameter characterizing the error correction capability of a code in our model . note that , if the condition of theorem [ thm : coherent - correction - capability ] is violated , then there exists at least one codeword for which the adversary can certainly induce a decoding failure . note that the error correction capability of a code @xmath185 is dependent on the network code through the matrix @xmath6 . let @xmath188 be the column - rank deficiency of @xmath6 . since @xmath189 , it follows from ( [ eq : bound - rank - product ] ) that @xmath190 and @xmath191 thus , the error correction capability of a code is strongly tied to its minimum rank distance ; in particular , @xmath192 if @xmath193 . while the lower bound @xmath194 may not be tight in general , we should expect it to be tight when @xmath30 is sufficiently large . this is indeed the case for mrd codes , as discussed in section [ ssec : coherent - optimality ] . thus , a rank deficiency of @xmath6 will typically reduce the error correction capability of a code . taking into account the worst case , we can use theorem [ thm : coherent - correction - capability ] to give a correction guarantee in terms of the minimum rank distance of the code . [ prop : coherent - guarantee - rank - metric ] a code @xmath185 is guaranteed to correct @xmath7 packet errors , under rank deficiency @xmath132 , if @xmath195 . note that the guarantee of proposition [ prop : coherent - guarantee - rank - metric ] depends only on @xmath132 and @xmath7 ; in particular , it is independent of the network code or the specific transfer matrix @xmath6 . in this subsection , we investigate the model for coherent network coding studied by yeung et al . in @xcite , which is similar to the one considered in the previous subsection . the model is that of a matrix channel with input @xmath44 , output @xmath147 , and channel law given by @xmath196 where @xmath53 and @xmath197 are fixed and known to the receiver , and @xmath198 is arbitrarily chosen by an adversary provided @xmath199 . ( recall that @xmath200 is the number of edges in the network . ) in addition , the adversary has unlimited computational power and is omniscient , knowing , in particular , @xmath6 , @xmath201 and @xmath5 . we now show that some of the concepts defined in @xcite , such as `` network hamming distance , '' can be reinterpreted in the framework of section [ sec : general - approach ] . as a consequence , we can easily recover the results of @xcite on error correction and detection guarantees . first , note that the current model can be completely described by the following discrepancy function @xmath202 the @xmath108-distance induced by this discrepancy function is given by @xmath203 where the last equality follows from the fact that @xmath204 , achievable if @xmath205 . let us now examine some of the concepts defined in . for a specific sink node , the decoder proposed in ( * ? ? ? * eq . ( 2 ) ) has the form @xmath206 the definition of the objective function @xmath207 requires several other definitions presented in @xcite . specifically , @xmath208 , where @xmath209 , @xmath210 , and @xmath211 . substituting all these values into @xmath207 , we obtain @xmath212 thus , the decoder in @xcite is precisely a minimum - discrepancy decoder . in @xcite , the `` network hamming distance '' between two messages @xmath213 and @xmath214 is defined as @xmath215 , where @xmath216 . again , simply substituting the corresponding definitions yields @xmath217 thus , the `` network hamming distance '' is precisely the @xmath108-distance induced by the discrepancy function @xmath218 . finally , the `` unicast minimum distance '' of a network code with message set @xmath185 @xcite is precisely @xmath219 . let us return to the problem of characterizing the correction capability of a code . the discrepancy function @xmath220 is normal . let @xmath221 and let @xmath222 . let @xmath223 be a solution to the minimization in ( [ eq : yeung - discrepancy ] ) . then @xmath224 and @xmath225 . by partitioning @xmath226 , we can always find two matrices @xmath227 and @xmath228 such that @xmath229 , @xmath230 and @xmath231 . taking @xmath232 , we have that @xmath233 and @xmath234 . since @xmath235 , it follows that @xmath236 and @xmath237 . it follows that a code @xmath30 is guaranteed to correct any @xmath7 packet errors if and only if @xmath238 . thus , we recover theorems 2 and 3 in @xcite ( for error detection , see appendix [ sec : detection - capability ] ) . the analogous results for the multicast case can be obtained in a straightforward manner . we now wish to compare the parameters devised in this subsection with those of section [ ssec : coherent - main ] . from the descriptions of ( [ eq : matrix - model ] ) and ( [ eq : channel - model - full ] ) , it is intuitive that the model of this subsection should be equivalent to that of the previous subsection if the matrix @xmath201 , rather than fixed and known to the receiver , is arbitrarily and secretly chosen by the adversary . a formal proof of this fact is given in the following proposition . [ prop : coherent - yeung - comparison ] @xmath239 consider the minimization @xmath240 for any feasible @xmath241 , we have @xmath242 . this lower bound can be achieved by taking @xmath243 where @xmath244 is a full - rank decomposition of @xmath165 . this proves the first statement . the second statement follows from the first by noticing that @xmath245 and @xmath246 . the third statement is immediate . proposition [ prop : coherent - yeung - comparison ] shows that the model of section [ ssec : coherent - main ] is indeed more pessimistic , as the adversary has additional power to choose the worst possible @xmath201 . it follows that any code that is @xmath7-error - correcting for that model must also be @xmath7-error - correcting for the model of yeung et al . let us now evaluate the performance of an mrd code under the models of the two previous subsections . the singleton bound of @xcite ( see also @xcite ) states that @xmath247 where @xmath50 is the size of the alphabet . ] from which packets are drawn . note that @xmath248 in our setting , since each packet consists of @xmath1 symbols from @xmath2 . using proposition [ prop : coherent - yeung - comparison ] , we can also obtain @xmath249 on the other hand , the size of an mrd code , for @xmath250 , is given by @xmath251 where ( [ eq : coherent - mrd - proof-1 ] ) follows from ( [ eq : coherent - delta - rank - distance ] ) . since @xmath248 , both ( [ eq : singleton - yeung ] ) and ( [ eq : singleton - coherent ] ) are achieved in this case . thus , we have the following result . [ thm : coherent - optimality - mrd ] when @xmath250 , an mrd code @xmath57 achieves maximum cardinality with respect to both @xmath252 and @xmath253 . theorem [ thm : coherent - optimality - mrd ] shows that , if an alphabet of size @xmath254 is allowed ( i.e. , a packet size of at least @xmath255 bits ) , then mrd codes turn out to be optimal under both models of sections [ ssec : coherent - main ] and [ ssec : coherent - comparison ] . it is straightforward to extend the results of section [ ssec : coherent - main ] for the case of multiple heterogeneous receivers , where each receiver @xmath256 experiences a rank deficiency @xmath257 . in this case , it can be shown that an mrd code with @xmath250 achieves the refined singleton bound of @xcite . note that , due to ( [ eq : singleton - coherent ] ) , ( [ eq : coherent - mrd - singleton ] ) and ( [ eq : coherent - mrd - proof-1 ] ) , it follows that @xmath258 for an mrd code with @xmath250 . thus , in this case , we can restate theorem [ thm : coherent - correction - capability ] in terms of the minimum rank distance of the code . [ thm : coherent - complete - capability ] an mrd code @xmath58 with @xmath250 is guaranteed to correct @xmath7 packet errors , under rank deficiency @xmath132 , if and only if @xmath195 . observe that theorem [ thm : coherent - complete - capability ] holds regardless of the specific transfer matrix @xmath6 , depending only on its column - rank deficiency @xmath132 . the results of this section imply that , when designing a linear network code , we may focus solely on the objective of making the network code feasible , i.e. , maximizing @xmath259 . if an error correction guarantee is desired , then an outer code can be applied end - to - end without requiring any modifications on ( or even knowledge of ) the underlying network code . the design of the outer code is essentially trivial , as any mrd code can be used , with the only requirement that the number of @xmath2-symbols per packet , @xmath1 , is at least @xmath0 . consider the decoding rule ( [ eq : coherent - decoding - rule ] ) . the fact that ( [ eq : coherent - decoding - rule ] ) together with ( [ eq : coherent - discrepancy - rank ] ) is equivalent to ( * ? ? ? * eq . ( 20 ) ) implies that the decoding problem can be solved by exactly the same rank - metric techniques proposed in @xcite . in particular , for certain mrd codes with @xmath250 and minimum rank distance @xmath18 , there exist efficient encoding and decoding algorithms both requiring @xmath260 operations in @xmath2 per codeword . for more details , see @xcite . our model for noncoherent network coding with adversarial errors differs from its coherent counterpart of section [ ssec : coherent - main ] only with respect to the transfer matrix @xmath6 . namely , the matrix @xmath6 is unknown to the receiver and is freely chosen by the adversary while respecting the constraint @xmath261 . the parameter @xmath132 , the maximum column rank deficiency of @xmath6 , is a parameter of the system that is known to all . note that , as discussed above for the matrix @xmath13 , the assumption that @xmath6 is chosen by the adversary is what provides the conservative ( worst - case ) nature of the model . the constraint on the rank of @xmath6 is required for a meaningful coding problem ; otherwise , the adversary could prevent communication by simply choosing @xmath262 . as before , we assume a minimum - discrepancy decoder @xmath263 with discrepancy function given by @xmath264 again , @xmath265 represents the minimum number of error packets needed to produce an output @xmath9 given an input @xmath5 under the current adversarial model . the subscript is to emphasize that @xmath265 is still a function of @xmath132 . the @xmath108-distance induced by @xmath266 is defined below . for @xmath158 , let @xmath267 we now prove that @xmath266 is normal and therefore @xmath268 characterizes the correction capability of a code . first , observe that , using lemma [ lem : coherent - discrepancy - rank ] , we may rewrite @xmath265 as @xmath269 also , note that @xmath270 & \hspace{8.5em } { } + \operatorname{\sf rank\hspace{0.1em}}(y - a'x ' ) \ } \nonumber \\[2ex ] & = \min_{\substack{a , a ' \in { \mathbb{f}_q}^{n \times n}:\\ \operatorname{\sf rank\hspace{0.1em}}a \geq n - \rho\\ \operatorname{\sf rank\hspace{0.1em}}a ' \geq n - \rho } } \operatorname{\sf rank\hspace{0.1em}}(a'x ' - ax ) \label{eq : noncoherent - delta - expression - rank } \end{aligned}\ ] ] where the last equality follows from the fact that @xmath271 , achievable by choosing , e.g. , @xmath4 . [ thm : noncoherent - magic ] the discrepancy function @xmath272 is normal . let @xmath158 and let @xmath273 . let @xmath274 be a solution to the minimization in ( [ eq : noncoherent - delta - expression - rank ] ) . then @xmath275 . by performing a full - rank decomposition of @xmath276 , we can always find two matrices @xmath177 and @xmath178 such that @xmath277 , @xmath180 and @xmath181 . taking @xmath182 , we have that @xmath278 and @xmath279 . since @xmath280 , it follows that @xmath281 and @xmath282 . as a consequence of theorem [ thm : noncoherent - magic ] , we have the following result . [ thm : noncoherent - capability ] a code @xmath185 is guaranteed to correct any @xmath7 packet errors if and only if @xmath283 . similarly as in section [ ssec : coherent - main ] , theorem [ thm : noncoherent - capability ] shows that @xmath284 is a fundamental parameter characterizing the error correction capability of a code in the current model . in contrast to section [ ssec : coherent - main ] , however , the expression for @xmath266 ( and , consequently , @xmath285 ) does not seem mathematically appealing since it involves a minimization . we now proceed to finding simpler expressions for @xmath266 and @xmath285 . the minimization in ( [ eq : noncoherent - discrepancy - rank ] ) is a special case of a more general expression , which we give as follows . for @xmath44 , @xmath147 and @xmath286 , let @xmath287 the quantity defined above is computed in the following lemma . [ lem : min - rank - quantity ] @xmath288^+.\ ] ] see appendix [ sec : proof - lemma ] . note that @xmath289 is independent of @xmath290 , for all valid @xmath290 . thus , we may drop the subscript and write simply @xmath291 . we can now provide a simpler expression for @xmath265 . [ thm : noncoherent - discrepancy - expression ] @xmath292 this follows immediately from lemma [ lem : min - rank - quantity ] by noticing that @xmath293 . from theorem [ thm : noncoherent - discrepancy - expression ] , we observe that @xmath265 depends on the matrices @xmath5 and @xmath9 only through their row spaces , i.e. , only the transmitted and received row spaces have a role in the decoding . put another way , we may say that the channel really accepts an input subspace @xmath37 and delivers an output subspace @xmath294 . thus , all the communication is made via subspace selection . this observation provides a fundamental justification for the approach of @xcite . at this point , it is useful to introduce the following definition . [ def : modif - subspace - distance ] the _ injection distance _ between subspaces @xmath295 and @xmath296 in @xmath34 is defined as @xmath297 the injection distance can be interpreted as measuring the number of error packets that an adversary needs to inject in order to transform an input subspace @xmath37 into an output subspace @xmath294 . this can be clearly seen from the fact that @xmath298 . thus , the injection distance is essentially equal to the discrepancy @xmath265 when the channel is influenced only by the adversary , i.e. , when the non - adversarial aspect of the channel ( the column - rank deficiency of @xmath6 ) is removed from the problem . note that , in this case , the decoder ( [ eq : noncoherent - decoding - rule ] ) becomes precisely a minimum - injection - distance decoder . [ prop : injection - metric ] the injection distance is a metric . we delay the proof of proposition [ prop : injection - metric ] until section [ ssec : noncoherent - comparison ] . we can now use the definition of the injection distance to simplify the expression for the @xmath108-distance . [ prop : noncoherent - delta - expression ] @xmath299^+.\ ] ] this follows immediately after realizing that @xmath300 . from proposition [ prop : noncoherent - delta - expression ] , it is clear that @xmath301 is a metric if and only if @xmath193 ( in which case it is precisely the injection metric ) . if @xmath302 , then @xmath301 does not satisfy the triangle inequality . it is worth noticing that @xmath303 for any two matrices @xmath5 and @xmath304 that share the same row space . thus , any reasonable code @xmath185 should avoid this situation . for @xmath58 , let @xmath305 be the subspace code ( i.e. , a collection of subspaces ) consisting of the row spaces of all matrices in @xmath30 . the following corollary of proposition [ prop : noncoherent - delta - expression ] is immediate . [ cor : noncoherent - delta - min - expression ] suppose @xmath185 is such that @xmath306 , i.e. , no two codewords of @xmath185 have the same row space . then @xmath307^+.\ ] ] using corollary [ cor : noncoherent - delta - min - expression ] , we can restate theorem [ thm : noncoherent - capability ] more simply in terms of the injection distance . [ thm : noncoherent - complete - capability ] a code @xmath185 is guaranteed to correct @xmath7 packet errors , under rank deficiency @xmath132 , if and only if @xmath308 . note that , due to equality in corollary [ cor : noncoherent - delta - min - expression ] , a converse is indeed possible in theorem [ thm : noncoherent - complete - capability ] ( contrast with proposition [ prop : coherent - guarantee - rank - metric ] for the coherent case ) . theorem [ thm : noncoherent - complete - capability ] shows that @xmath309 is a fundamental parameter characterizing the _ complete _ correction capability ( i.e. , error correction capability and `` rank - deficiency correction '' capability ) of a code in our noncoherent model . put another way , we may say that a code @xmath185 is good for the model of this subsection if and only if its subspace version @xmath310 is a good code in the injection metric . let @xmath311 be a subspace code whose elements have maximum dimension @xmath0 . in @xcite , the network is modeled as an operator channel that takes in a subspace @xmath312 and puts out a possibly different subspace @xmath313 . the kind of disturbance that the channel applies to @xmath296 is captured by the notions of `` insertions '' and `` deletions '' of dimensions ( represented mathematically using operators ) , and the degree of such a dissimilarity is captured by the subspace distance @xmath314 the transmitter selects some @xmath315 and transmits @xmath296 over the channel . the receiver receives some subspace @xmath295 and , using a _ minimum subspace distance decoder _ , decides that the subspace @xmath316 was sent , where @xmath317 this decoder is guaranteed to correct all disturbances applied by the channel if @xmath318 , where @xmath319 is the minimum subspace distance between all pairs of distinct codewords of @xmath320 . first , let us point out that this setup is indeed the same as that of section [ ssec : noncoherent - main ] if we set @xmath321 , @xmath322 and @xmath323 , where @xmath185 is such that @xmath306 . also , any disturbance applied by an operator channel can be realized by a matrix model , and vice - versa . thus , the difference between the approach of this section and that of @xcite lies in the choice of the decoder . indeed , by using theorem [ thm : noncoherent - discrepancy - expression ] and the definition of subspace distance , we get the following relationship : [ prop : delta - and - subspace - distance ] @xmath324 thus , we can see that when the matrices in @xmath185 do not all have the same rank ( i.e. , @xmath320 is a _ non - constant - dimension code _ ) , then the decoding rules ( [ eq : noncoherent - decoding - rule ] ) and ( [ eq : subspace - decoding - rule ] ) may produce different decisions . using @xmath325 in the above proposition ( or simply using ( [ eq : injection - distance - definition ] ) and ( [ eq : subspace - distance - definition ] ) ) gives us another formula for the injection distance : @xmath326 we can now prove a result that was postponed in the previous section . the injection distance is a metric . since @xmath327 is a metric on @xmath34 and @xmath328 is a norm on @xmath329 , it follows from ( [ eq : injection - and - subspace - distance ] ) that @xmath330 is also a metric on @xmath34 . we now examine in more detail an example situation where the minimum - subspace - distance decoder and the minimum - discrepancy decoder produce different decisions . [ ex : decoding - comparison ] for simplicity , assume @xmath193 . consider a subspace code that contains two codewords @xmath331 and @xmath332 such that @xmath333 satisfies @xmath334 , where @xmath335 . suppose the received subspace @xmath322 is such that @xmath336 and @xmath337 , as illustrated in fig . [ fig : subspaces ] . then @xmath338 and @xmath339 , while proposition ( [ prop : delta - and - subspace - distance ] ) gives @xmath340 and @xmath341 . since , by assumption , @xmath342 and @xmath343 , it follows that @xmath344 but @xmath345 , i.e. , the decoders ( [ eq : subspace - decoding - rule ] ) and ( [ eq : noncoherent - decoding - rule ] ) will produce different decisions . this situation can be intuitively explained as follows . the decoder ( [ eq : subspace - decoding - rule ] ) favors the subspace @xmath346 , which is closer in subspace distance to @xmath295 than @xmath347 . however , since @xmath346 is low - dimensional , @xmath295 can only be produced from @xmath346 by the _ insertion _ of @xmath348 dimensions . the decoder ( [ eq : noncoherent - decoding - rule ] ) , on the other hand , favors @xmath347 , which , although farther in subspace distance , can produce @xmath295 after the _ replacement _ of @xmath343 dimensions . since one packet error must occur for each inserted or replaced dimension , we conclude that the decoder ( [ eq : noncoherent - decoding - rule ] ) finds the solution that minimizes the number of packet errors observed . . two spaces are joined with a dashed line if one is a subspace of the other.,title="fig : " ] + the subspace metric of @xcite treats insertions and deletions of dimensions ( called in @xcite `` errors '' and `` erasures '' , respectively ) symmetrically . however , depending upon the position of the adversary in the network ( namely , if there is a source - destination min - cut between the adversary and the destination ) then a single error packet may cause the replacement of a dimension ( i.e. , a simultaneous `` error '' and `` erasure '' in the terminology of @xcite ) . the injection distance , which is designed to `` explain '' a received subspace with as few error - packet injections as possible , properly accounts for this phenomenon , and hence the corresponding decoder produces a different result than a minimum subspace distance decoder . if it were possible to restrict the adversary so that each error - packet injection would only cause either an _ insertion _ or a _ deletion _ of a dimension ( but not both ) , then the subspace distance of @xcite would indeed be appropriate however , this is not the model considered here . let us now discuss an important fact about the subspace distance for general subspace codes ( assuming for simplicity that @xmath325 ) . the packet error correction capability of a minimum - subspace - distance decoder , @xmath349 , is not necessarily equal to @xmath350 or @xmath351 , but lies somewhere in between . for instance , in the case of a constant - dimension code @xmath320 , we have @xmath352 thus , theorem [ thm : noncoherent - complete - capability ] implies that @xmath353 exactly . in other words , in this special case , the approach in @xcite coincides with that of this paper , and theorem [ thm : noncoherent - complete - capability ] provides a converse that was missing in @xcite . on the other hand , suppose @xmath320 is a subspace code consisting of just two codewords , one of which is a subspace of the other . then we have precisely @xmath354 , since @xmath355 packet - injections are needed to get past halfway between the codewords . since no single quantity is known that perfectly describes the packet error correction capability of the minimum - subspace - distance decoder ( [ eq : subspace - decoding - rule ] ) for general subspace codes , we can not provide a definitive comparison between decoders ( [ eq : subspace - decoding - rule ] ) and ( [ eq : noncoherent - decoding - rule ] ) . however , we can still compute bounds for codes that fit into example [ ex : decoding - comparison ] . let us continue with example [ ex : decoding - comparison ] . now , we adjoin another codeword @xmath356 such that @xmath357 and where @xmath358 satisfies @xmath359 . also we assume that @xmath360 is sufficiently large so as not to interfere with the problem ( e.g. , @xmath361 ) . let @xmath362 and @xmath363 denote the packet error correction capabilities of the decoders ( [ eq : subspace - decoding - rule ] ) and ( [ eq : noncoherent - decoding - rule ] ) , respectively . from the argument of example [ ex : decoding - comparison ] , we get @xmath364 , while @xmath365 , where @xmath366 . by choosing @xmath367 and @xmath368 , we get @xmath369 and @xmath370 . thus , @xmath371 , i.e. , we obtain a 1/3 increase in error correction capability by using the decoder ( [ eq : noncoherent - decoding - rule ] ) . we have addressed the problem of error correction in network coding under a worst - case adversarial model . we show that certain metrics naturally arise as the fundamental parameter describing the error correction capability of a code ; namely , the rank metric for coherent network coding , and the injection metric for noncoherent network coding . for coherent network coding , the framework based on the rank metric essentially subsumes previous analyses and constructions , with the advantage of providing a clear separation between the problems of designing a feasible network code and an error - correcting outer code . for noncoherent network coding , the injection metric provides a measure of code performance that is more precise , when a non - constant - dimension code is used , than the so - called subspace metric . the design of general subspace codes for the injection metric , as well as the derivation of bounds , is left as an open problem for future research . when dealing with communication over an adversarial channel , there is little justification to consider the possibility of error detection . in principle , a code should be designed to be unambiguous ( in which case error detection is not needed ) ; otherwise , if there is any possibility for ambiguity at the receiver , then the adversary will certainly exploit this possibility , leading to a high probability of decoding failure ( detected error ) . still , if a system is such that ( a ) sequential transmissions are made over the same channel , ( b ) there exists a feedback link from the receiver to the transmitter , and ( c ) the adversary is not able to fully exploit the channel at all times , then it might be worth using a code with a lower correction capability ( but higher rate ) that has some ability to detect errors . following classical coding theory , we consider error detection in the presence of a bounded error - correcting decoder . more precisely , bounded - discrepancy decoder with correction radius @xmath7 _ , or simply a _ @xmath7-discrepancy - correcting decoder _ , by @xmath372 of course , when using a @xmath7-discrepancy - correcting decoder , we implicitly assume that the code is @xmath7-discrepancy - correcting . the _ discrepancy detection capability _ of a code ( under a @xmath7-discrepancy - correcting decoder ) is the maximum value of discrepancy for which the decoder is guaranteed not to make an undetected error , i.e. , it must return either the correct codeword or the failure symbol @xmath373 . for @xmath75 , let the function @xmath374 be given by @xmath375 the discrepancy - detection capability of a code @xmath30 is given exactly by @xmath376 . that is , under a @xmath7-discrepancy - correction decoder , any discrepancy of magnitude @xmath377 can be detected if and only if @xmath378 . let @xmath379 . suppose that @xmath64 is transmitted and @xmath96 is received , where @xmath380 . we will show that @xmath381 , for all @xmath382 . suppose , by way of contradiction , that @xmath98 , for some @xmath382 , @xmath383 . then @xmath384 , which is a contradiction . conversely , assume that @xmath385 , i.e. , @xmath386 . we will show that an undetected error may occur . since @xmath386 , there exist @xmath95 such that @xmath387 . this implies that there exists some @xmath96 such that @xmath98 and @xmath388 . by assumption , @xmath30 is @xmath7-discrepancy - correcting , so @xmath389 . thus , if @xmath27 is transmitted and @xmath65 is received , an undetected error will occur , even though @xmath390 . the result above has also been obtained in @xcite , although with a different notation ( in particular , treating @xmath391 as a `` distance '' function ) . below , we characterize the detection capability of a code in terms of the @xmath108-distance . for any code @xmath30 , we have @xmath392 . for any @xmath107 , let @xmath96 be a solution to the minimization in ( [ eq : function - exact - detection ] ) , i.e. , @xmath65 is such that @xmath98 and @xmath393 . then @xmath394 , which implies that @xmath395 . suppose that @xmath90 is normal . for every code @xmath23 , we have @xmath396 . we just need to show that @xmath397 . take any @xmath107 . since @xmath90 is normal , there exists some @xmath96 such that @xmath398 and @xmath399 . thus , @xmath400 . first , we recall the following useful result shown in ( * ? ? ? * proposition 2 ) . let @xmath401 . then @xmath402 using ( [ eq : rank - difference - bound ] ) and ( [ eq : bound - rank - product ] ) , we have @xmath403 we will now show that this lower bound is achievable . our approach will be to construct @xmath6 as @xmath404 , where @xmath405 and @xmath406 are both full - rank matrices . then ( [ eq : bound - rank - product ] ) guarantees that @xmath407 . the matrix @xmath408 will be constructed similarly : @xmath409 , where @xmath410 and @xmath411 are both full - rank . let @xmath412 , @xmath413 , and @xmath414 . let @xmath415 be such that @xmath416 , let @xmath417 be such that @xmath418 and let @xmath419 be such that @xmath420 . then , let @xmath421 and @xmath422 be such that @xmath423 now , choose any @xmath424 and @xmath425 that have full row rank , where @xmath426^+$ ] and @xmath427^+$ ] . for instance , we may pick @xmath428 and @xmath429 . finally , let @xmath430 where , in both cases , the upper identity matrix is @xmath431 . we have @xmath432^+ . \nonumber\end{aligned}\ ] ] the authors would like to thank the anonymous reviewers for their helpful comments . s. jaggi , m. langberg , s. katti , t. ho , d. katabi , m. mdard , and m. effros , `` resilient network coding in the presence of byzantine adversaries , '' _ ieee trans . inf . theory _ , 54 , no . 6 , pp . 25962603 , jun . 2008 . s. yang , c. k. ngai , and r. w. yeung , `` construction of linear network codes that achieve a refined singleton bound , '' in _ proc . information theory _ , nice , france , jun . 2429 , 2007 , pp . 15761580 . danilo silva ( s06 ) received the b.sc . degree from the federal university of pernambuco , recife , brazil , in 2002 , the m.sc . degree from the pontifical catholic university of rio de janeiro ( puc - rio ) , rio de janeiro , brazil , in 2005 , and the ph.d . degree from the university of toronto , toronto , canada , in 2009 , all in electrical engineering . frank r. kschischang ( s83m91sm00f06 ) received the b.a.sc . degree ( with honors ) from the university of british columbia , vancouver , bc , canada , in 1985 and the m.a.sc . and ph.d . degrees from the university of toronto , toronto , on , canada , in 1988 and 1991 , respectively , all in electrical engineering . he is a professor of electrical and computer engineering and canada research chair in communication algorithms at the university of toronto , where he has been a faculty member since 1991 . during 19971998 , he was a visiting scientist at the massachusetts institute of technology , cambridge , and in 2005 he was a visiting professor at the eth , zrich , switzerland . his research interests are focused on the area of channel coding techniques . kschischang was the recipient of the ontario premier s research excellence award . from 1997 to 2000 , he served as an associate editor for coding theory for the ieee transactions on information theory . he also served as technical program co - chair for the 2004 ieee international symposium on information theory ( isit ) , chicago , il , and as general co - chair for isit 2008 , toronto .

we adapt the smooth tests of goodness of fit developed by to the study of the non - gaussianity of interferometric observations of the cosmic microwave background ( cmb ) . the interferometric measurements ( visibilities ) are transformed into signal - to - noise eigenmodes , and then the method is applied directly in fourier space . this transformation allows us to perform the analysis in different subsets of eigenmodes according to their signal - to - noise level . the method can also deal with non - uniform or incomplete coverage of the plane . we explore here two possibilities : we analyse either the real and imaginary parts of the complex visibilities ( gaussianly distributed under the gaussianity hypothesis ) or their phases ( uniformly distributed under the gaussianity hypothesis ) . the power of the method in discriminating between gaussian and non - gaussian distributions is studied by using several kinds of non - gaussian simulations . on the one hand , we introduce a certain degree of non - gaussianity directly into the fourier space using the edgeworth expansion , and afterwards the desired correlation is introduced . on the other hand , we consider interferometric observations of a map with topological defects ( cosmic strings ) . to these previous non - gaussian simulations we add different noise levels and quantify the required signal - to - noise ratio necessary to achieve a detection of these non - gaussian features . finally , we have also studied the ability of the method to constrain the so - called nonlinear coupling constant using simulations . the whole method is illustrated here by application to simulated data from the very small array interferometer . [ firstpage ] methods : data analysis methods : statistical cosmic microwave radiation

astro-ph0411154

the study of the probability distribution of the primordial density fluctuations in the universe is one of the most fundamental challenges in the present day cosmology . the determination of this probability distribution could constrain the ensemble of cosmological theories of the formation of primordial density fluctuations . in particular , standard inflationary theories establish that these density fluctuations are gaussianly distributed @xcite . so the detection of some non - gaussian features would question these kinds of theories . a powerful tool in the determination of cosmological parameters is the cosmic microwave background ( cmb ) . the primordial density fluctuations left their imprint in the cmb because of the thermal equilibrium existing before matter radiation decoupling . in this manner , the initial density fluctuations led to anisotropies in the cmb , and these inherited the probability distribution of the former . in short , testing cmb gaussianity is equivalent to testing the gaussianity of the primordial fluctuations and hence the validity of standard inflation . for these reasons there is great interest in the implementation of statistical methods to analyse the gaussianity of current and future cmb data . a representative sample of these methods ( containing real - space statistics , wavelets and fourier - based analyses ) have been recently applied to probe the gaussianity of the data from the wmap mission @xcite . in this paper , we shall concentrate on the smooth tests of goodness of fit developed by @xcite . these methods have been already adapted to cmb data analysis , and applied @xcite to the study of the gaussianity of maxima data @xcite . we further adapt this method here to deal with data from interferometers . with their intrinsic stability ( only the correlated signal is detected ) and their ability to reject atmospheric signals ( e.g. @xcite ) interferometers have proved their worth in ground - based observations of the cmb . over the last few years several interferometers have provided high - sensitivity measurements of the power spectrum at intermediate angular scales . in particular , these experiments include the very small array ( vsa , @xcite ) , dasi @xcite and cbi @xcite ; for a more detailed compilation of interferometric experiments measuring the cmb , e.g. @xcite . these instruments sample the fourier modes of the measured intensity , so they provide a direct measurement of the power spectrum on the sky and allow us to study the gaussianity directly in harmonic space . non - gaussianity analyses have been already performed on interferometer datasets @xcite . in the first case these analyses are mainly based on real - space statistics ( they are applied to the maximum - entropy reconstructed maps in real space ) ; while in the second case , the analysis is performed in fourier space by computing the bispectrum in a similar way to the computation of the power spectrum . we follow the second approach and present here the smooth tests of goodness - of - fit adapted to the study of the gaussianity of interferometric data directly in the visibility space , so we test directly the gaussian nature of the fourier modes on the sky . since an interferometer measures complex quantities ( visibilities ) , we present two different analyses in this paper , one on the real and imaginary parts , and another on the phases . the experimental data are a combination of the signal plus noise . in principle , it would be desirable to analyse data in which signal dominates over the noise ; that is , data with a high signal - to - noise ratio . one suitable approach to select only data with high signal - to - noise ratio is the transformation of the data into signal - to - noise eigenmodes @xcite . this formalism allows us to eliminate the data dominated by noise . moreover , as we explain below , the signal - to - noise eigenmodes are not correlated , allowing the direct implementation of the method . as it is shown below , our method looks for deviations with respect to the null hypothesis ( gaussianity ) in statistics related to the measured moments ( mean , variance , skewness , kurtosis , ... ) of the visibility signal - to - noise eigenmodes . thus , it is clear that the test is specially tailored for detecting non - gaussian signals directly in the fourier domain . this gaussianity analysis of the fourier components completes the analysis of gaussianity in the real space , given that some non - gaussian detection in the fourier space could indicate some degree of non - gaussianity in the real space and vice versa ( a linear transformation of gaussian variables is also gaussian ) . in fact , for example , the analysis of gaussianity directly in the fourier space by means of the bispectrum has demonstrated to be a powerful tool in the detection of the @xmath1 parameter @xcite . then , the analysis in fourier space could detect non - gaussianity that , in principle , should be more evident in real space . as an illustration of this fact , we will show how our method detects the non - gaussianity of a cosmic strings map directly in fourier space , even though it seems more natural to search the non - gaussian features of these objects in real space . the organization of the paper is as follows . smooth tests of goodness - of - fit are introduced in section [ section_2 ] . we focus our interest in the statistics developed by @xcite and their application to the gaussian and uniform distributions that we are going to test . in section [ section_3 ] signal - to - noise eigenmodes are reviewed and applied to the case of interferometer observations . the simulations we are using are described in section [ sec : vsa ] , where the very small array ( vsa ) is taken as a reference experiment . the application to gaussian simulations is described in section [ sec : gauss ] . the power of the test in discriminating between gaussian and non - gaussian signals affected by gaussian noise is studied in section [ section_4 ] . we consider three kinds of non - gaussian simulations : those performed using the edgeworth expansion introduced by @xcite , a string simulation created by @xcite and @xmath2 simulations with a @xmath1 coupling parameter . finally , section [ section_5 ] is dedicated to discussion and conclusions . in this section we present the smooth tests of goodness - of - fit and the work of @xcite applied to a gaussian and a uniform variable . let us suppose @xmath3 independent realizations , @xmath4 , of a statistical variable @xmath5 ( @xmath6 . our aim is to test whether the probability density function of @xmath5 is equal to a prefixed function @xmath7 . the smooth tests of goodness - of - fit are constructed to discriminate between the predetermined function @xmath7 and another that deviates smoothly from the former and is called the _ alternative density function_. we consider an alternative density function given by @xmath8 , where @xmath9 is a parameter vector whose @xmath10th component is @xmath11 and there exists a value @xmath12 such that @xmath13 . the alternative function deviates smoothly from @xmath7 when @xmath9 is displaced from @xmath12 . the probability density function of the @xmath3 independent realizations of @xmath5 is given by @xmath14 . given these realizations , we calculate the estimated value of @xmath9 by means of the maximum likelihood method and denote this value by @xmath15 . the quantity @xmath16 is defined such that @xmath17 where @xmath16 is a measurement of the difference between @xmath15 and @xmath12 and is therefore a test of the hypothesis that @xmath18 . assuming @xmath15 to be close to @xmath12 and @xmath3 to be very large ( @xmath19 ) , the quantity @xmath16 is equal to the so - called _ score statistic _ @xcite , @xmath20 where @xmath21 is the vector whose components are @xmath22 , with the log - likelihood function defined as @xmath23 . @xmath24 is the transposed vector of @xmath25 and the components of the matrix @xmath26 are equal to @xmath27 . among all the possible choices of alternative density functions , we select that presented in the work of @xcite . in the work of @xcite an alternative density function is defined as @xmath28 where @xmath29 is a normalization constant and the @xmath30 functions are orthonormal on @xmath31 with @xmath32 ( note that @xmath33 ) . it can be demonstrated that the score statistic associated with the @xmath34 alternative is given by @xmath35 note that the previous @xmath36 quantity is not the @xmath10th component of the vector @xmath25 in expression ( [ eq:001 ] ) ( although they are closely related ) . we denote the two quantities with the same letter to follow the notation of @xcite . as equation ( [ eq:001 ] ) shows , the expression ( [ eq:004 ] ) gives the score statistic only when the distribution of the data is @xmath7 ( because the condition @xmath37 is imposed ) . in this case ( that is , when the distribution of the data is equal to the prefixed one ) , the @xmath38 statistic is distributed as @xmath39 when @xmath19 . this holds because @xmath36 is gaussianly distributed when @xmath19 ( the central limit theorem ) . moreover , it is easy to prove that the mean value of @xmath40 is equal to unity , independently of the value of @xmath3 . we see that we can work with the @xmath38 quantities or alternately with @xmath40 ( if the distribution of the data is equal to @xmath7 , these are distributed following a @xmath41 function ) . in this paper we apply these statistics to two cases : on one hand , to the real and imaginary parts of the visibilities and on the other hand to the phases of the visibilities . in the first case , the function @xmath7 is a gaussian distribution , and in the second case the distribution is a uniform . if @xmath7 is a gaussian distribution with zero mean and unit variance , the @xmath30 functions are the normalized hermite chebyshev polynomials @xmath42 with @xmath43 , @xmath44 , @xmath45 , and , for @xmath46 , @xmath47 . the @xmath40 statistics are given by : @xmath48 ^ 2 /24 \nonumber\end{aligned}\ ] ] where @xmath49 is the estimated moment of order @xmath50 . bearing in mind the relation between the @xmath40 and the @xmath38 quantities , we see that the statistic @xmath38 is related to moments of order @xmath51 , so that this test is _ directional _ ; that is , it indicates how the actual distribution deviates from gaussianity . for example , if @xmath52 and @xmath53 are small and @xmath54 is large . the data then have a large @xmath55 value because of the relation between @xmath55 and @xmath54 . then , the data have a large skewness value . this holds for all distributions and not only for the gaussian case . these expressions ( or the equivalent @xmath38 ) are used in @xcite to test the gaussianity of the maxima data . the skewness and the kurtosis of these data are constrained in @xcite using the same method . the smooth test of goodness - of - fit applied to the uniform distribution was developed by the first time by @xcite , and his work is the starting point of the work of @xcite . if we want to test whether @xmath7 is a uniform distribution in the interval @xmath56 $ ] , we define @xmath57 and then we test if @xmath58 is a uniform variable in @xmath59 $ ] . in this case we take the legendre polynomials : @xmath60 , @xmath61 and @xmath62 for @xmath46 . thus the normalized functions are @xmath63 with @xmath64 . in this case : @xmath65 ^ 2 /64 \nonumber\end{aligned}\ ] ] where @xmath66 . as in the gaussian variable case , we see that the test is directional : deviations in the estimated moments @xmath67 from the expected values for a uniform distribution give deviations in the statistics . if we assume a small field size ( where the flat - sky approximation of the observed region is valid ) , an interferometer measures complex visibilities at a frequency @xmath68 ( the van cittert zernicke theorem ) : @xmath69 where @xmath70 is the angular position of the observed point on the sky and @xmath71 is a baseline vector in units of the wavelength of the observed radiation ( monochromatic radiation is assumed , and hereinafter we omit the dependence on the frequency , @xmath68 ) . @xmath72 is the primary beam of the antennas ( normalized to @xmath73 ) , and the function @xmath74 is the brightness distribution on the sky , which can be easily translated into temperature contrast values @xmath75 for the case of the cmb . if we denote by @xmath76 the fourier transform of @xmath77 and by @xmath78 the cmb power spectrum ( in units of brightness , @xmath79 ) then the correlations between the visibilities observed at @xmath80 and @xmath81 are given by @xmath82 where @xmath83 denotes the complex conjugate of @xmath76 . these quantities can be computed semi - analytically as described in @xcite , who show how to perform a maximum - likelihood analysis to extract the cmb power spectrum from interferometer observations . note that for the common case in which the primary beam can be approximated by a gaussian function @xmath84 , it can be demonstrated that the real and the imaginary parts of the observed complex visibilities are uncorrelated and thus independent , if the gaussianity of the cmb holds . our aim is to test if the visibilities are gaussianly distributed . this can be done by analysing the real and imaginary parts of the visibilities or studying their phases . the following subsections explain how we work with the data . we work with signal - to - noise eigenmodes as they are defined in the work of @xcite . let us suppose an observed variable @xmath85 , where the subscript @xmath34 denotes a pixel or a position in the space where that variable is defined ( for example the real or the imaginary parts of the visibility @xmath86 are measured in the position @xmath87 of the called @xmath0 space ) . this variable is the sum of a signal and a noise component ( @xmath88 ) whose respective correlation matrices have components given by @xmath89 and @xmath90 . the brackets indicate average to many realizations . the signal - to - noise eigenmodes are defined by @xmath91 where @xmath92 is the called _ square root matrix _ of the noise correlation matrix ( i.e. @xmath93 ) and @xmath94 is the rotation matrix which diagonalizes the matrix @xmath95 . the eigenvalues of this diagonalization are denoted by @xmath96 . in the case we are studying , the noise has zero mean and is not correlated , i.e. @xmath97 , so the components of @xmath92 are @xmath98 . ( [ eq:003 ] ) gives us _ transformed _ signal and noise @xmath99 and @xmath100 such that : @xmath101 with @xmath102 and @xmath103 . so , the correlation matrix of @xmath104 is given by @xmath105 with @xmath106 . we now have a clear characterization of the signal - to - noise relation of our ( transformed ) data . the noise dominates over the signal if @xmath107 . so signal - to - noise eigenmodes @xmath104 with a very low value of the associated eigenvalue @xmath96 are very much dominated by the noise and we do not want to include them in our analysis . this is the main interest of this approach , but there is another interesting point : the signal - to - noise eigenmodes are uncorrelated data . this latter fact will be very useful in the application of the tests of gaussianity described in this paper as explained in the next subsection . we write the visibilities in terms of their real and imaginary parts : @xmath108 testing the gaussianity of the visibilities is equivalent to testing the joint gaussianity of their real and imaginary parts . as indicated above , when the primary beam is gaussian , the real and imaginary parts are independent if the gaussianity of the cmb holds . our data to be analysed are the set of real parts of the visibilities which are correlated among them ( that is : @xmath109 ) and imaginary parts which also are correlated ( @xmath110 ) . moreover , we only have one realization . the test here presented works with a large amount of independent data ( see the hypothesis in the deduction of the score statistic , section 2 ) . to convert our correlated ( and then dependent ) data to a sample of independent data we proceed in the following way . given the real part of the visibilities , we perform the transformation in signal - to - noise eigenmodes ( expression [ eq:003 ] ) and define the variable @xmath111 these new data are uncorrelated and normalized ( zero mean , unit dispersion ) . we operate in the same way with the imaginary parts and add the resulting quantities to those obtained with the real parts . finally , we have the @xmath112 data ( @xmath113number of visibilities ) which are uncorrelated and normalized . moreover , if the visibility distribution is multinormal then the @xmath114 data are independent and gaussian distributed with zero mean and unit dispersion . then , if gaussianity holds , the @xmath112 data fulfill the hypothesis of the smooth tests of goodness - of - fit and their @xmath40 statistics must have a @xmath115 distribution . the test is then applied to the @xmath114 quantities . the analysis can be performed on the visibility phases . it is well - known that if the visibility , @xmath116 , is gaussianly distributed , its phase has a uniform distribution . the phases of the visibilities are not independent because there are correlations of the real parts among themselves , and among the imaginary parts ( the real and imaginary parts are not correlated if the primary beam is a gaussian function ) . thus , and for the case of a gaussian primary beam , given two points @xmath117 and @xmath118 in the @xmath0 plane , the joint probability of finding the phase value @xmath119 in @xmath117 and @xmath120 in @xmath118 is given by @xmath121 where : @xmath122 thus @xmath123 . as an example , we have calculated the form of the previous probability for the particular case @xmath124 and @xmath125 . the plot is shown in figure [ fig1 ] . and @xmath120 in the positions @xmath117 and @xmath118.,width=321 ] the statistics of subsection 2.3 can not be applied directly to the phases because these are not independent . we therefore construct a related independent quantities in the following way . we decorrelate the different visibilities so that , under the assumption of gaussianity , they will be independent . to do this we proceed as in the previous section : we decorrelate the real parts of the visibilities . that transformation gives us the quantities @xmath126 . we proceed in the same way with the imaginary parts and obtain the associated quantities @xmath127 . in this way , we have the complex number @xmath128 . these quantities are independent for different @xmath10 . we calculate the phase of each one and apply to them the statistics explained in subsection 2.3 for a uniform variable . to illustrate the method described above , in what follows we apply it to simulated observations with the very small array . the vsa is a 14-element heterodyne interferometer array , operating at frequencies between 28 and 36 ghz with a 1.5 ghz bandwidth and a system temperature of approximately 30 k , sited at 2400 m altitude at the teide observatory in tenerife ( see @xcite for a detailed description ) . it can be operated in two complementary configurations , referred as the compact array ( which probes the angular range @xmath129 150900 ; see @xcite for observations in this configuration ) and the extended array ( @xmath130 = 3001500 ; see @xcite and @xcite for observations in this configuration ) . for the simulations in this paper , we will use as a template experiment the extended array configuration , which corresponds to the antenna arrangement shown in figure [ fig_ant ] . the observing frequency is 33 ghz , and the primary beam at this frequency has a fwhm of 2.1@xmath131 ( corresponding to a @xmath132 diameter of the aperture function of @xmath133 ) . for the vsa , the shape of the antenna can be approximated by a gaussian function , so we can calculate the correlation matrix using the expressions ( 11 ) and ( 12 ) presented in @xcite . to a tilting table hinged along its northern edge . ] we use a template observation corresponding to @xmath1344 hr integration time , and we explore several values of @xmath135 . the corresponding noise level will be simulated by adding to each visibility a random gaussian number with an amplitude of @xmath1368 jy per visibility ( one channel , one polarization ) in an integration time of 1 min , to reproduce the observed sensitivity of the vsa . thus , in a single day s 4 hr observation we get an rms of @xmath13654 mjy / beam in the real - space map . a simulated 1 day vsa file typically contains 25000 64 s visibilities . in order to perform the analyses here , we proceed to bin these data into cells of a certain size , as described in @xcite . for the analyses carried out in this paper , we have used a bin size of 9@xmath137 , where @xmath137 is the wavelength of the measured signal , which is similar to that used for the power spectrum evaluation in @xcite . we have explored different values of the bin size , and the number has been chosen according to two criteria . if the bin size is too small , the signal - to - noise ratio per pixel is very low . on the other hand , for large bin sizes , the number of binned visibilities is too small to apply the test . the value of 9@xmath137 provides both a reasonable number of visibilities and good power in the detection . nevertheless , we have checked that small variations in this number ( using values from 9 to 14@xmath137 ) do not produce significant changes . our template observation of the vsa contains a total of 895 visibility points after binning in 9@xmath137 cells . this template will be used throughout the paper . in this section we calibrate the method by using simulated gaussian observations . it can be shown that the correlation matrix of the real parts of the visibilities can be written as @xmath138 with @xmath139 , where @xmath140 are the signal - to - noise eigenvalues and @xmath141 is the rotation matrix defined in subsection [ sec:002 ] . to generate gaussian simulations with the desired correlation we start with a set @xmath142 ( @xmath143number of visibilities ) . the real parts of the visibilities are given by @xmath144 . in an analogous way we construct @xmath145 . after that , a gaussian realization of the noise is added to each visibility . given the simulated visibilities we decorrelate their real and imaginary parts as explained in subsection [ sec:002 ] . after that , we calculate the distributions of the @xmath40 quantities ( subsection [ sec:001 ] ) . as we have gaussian simulations , if the amount of data is relatively large , these distributions must be very close to @xmath41 functions as it is explained above . the form of the distributions is shown in fig . [ fig2 ] and they are compared with @xmath41 functions normalized to the number of simulations used ( dashed line ) . we have used 10000 simulations of a vsa field , with a noise level corresponding to @xmath146 hr for each one . this is the typical noise level achieved in a single - field vsa observation @xcite . we bin these data using a cell size of 9@xmath137 , to obtain 895 binned visibilities . in table [ tab1 ] the mean value and the standard deviation ( @xmath147 ) of the distributions is compared with the same quantities of the @xmath41 distribution . as mentioned above , the mean value of the @xmath40 must be equal to 1 independently of the number of data . from table [ tab1 ] we see that the convergence to this value is very accurate . the value of the standard deviation must be @xmath148 only when the number of data tends to infinity ( asymptotic case ) . in the case we are simulating ( 895 visibilities ) we see that the values of the standard deviation are very close to the @xmath41 values ( only @xmath149 seems to separate slightly from the asymptotic value ) . the vsa fields we are analysing have a number of visibilities between @xmath1361000 and @xmath1364000 and then the approximation to the @xmath41 functions will be better than the case shown in table [ tab1 ] . because of that , we can approximate the @xmath40 to @xmath41 distributions in a very accurate way . we then suppose the distributions to be @xmath41 for the @xmath40 statistics or @xmath39 for the @xmath38 . we have applied to these simulations the method described for the phases and we also find @xmath41 distributions for the @xmath40 statistics . the advantage of this assumption is that we do not have to make simulations to calculate the statistical distributions for every vsa field for the gaussian or uniform null hypothesis . we have checked that the distribution of the @xmath40 statistics is also very close to @xmath41 functions even when the number of data is @xmath136200 . in the following sections we cut data with low signal - to - noise eigenvalues and the statistics are calculated with @xmath136200 data ; also in this case , the distributions of the statistics will then be approximated by @xmath2 functions . c c c c c c & @xmath150 & @xmath151 & @xmath152 & @xmath149 & @xmath115 @xmath153 & 0.9863 & 0.9983 & 1.0370 & 1.0131 & 1.0000 @xmath147 & 1.4077 & 1.4426 & 1.4848 & 1.5265 & 1.4142 [ tab1 ] in this section , given our template of a vsa observation , we generate non - gaussian simulations of the visibilities and add the ( gaussian ) vsa noise to them . we analyse three kinds of non - gaussian simulations . first , we consider simulations created by means of the edgeworth expansion in the visibility space ; in second place , we analyse a cosmic strings simulation ; finally , @xmath2 simulations are considered as an exercise to estimate the power of the test to constrain the @xmath1 parameter . in order to characterize the power of the method in each of these cases , we use the `` power of the test '' . roughly speaking , the power , @xmath154 , of a given test at a certain significance level @xmath50 ( @xmath155 ) is parameterized as the area of the alternative distribution function outside the region which contains the @xmath156 area of distribution function of the null hypothesis ( the gaussian case ) . thus , a large value of @xmath154 for a prefixed small value of @xmath50 indicates that the two distributions have a small overlap , so we can distinguish between them ( e.g. @xcite ) . for definiteness , in what follows we adopt for the significance levels the values @xmath157 ( i.e. significance levels of 5% and 1% respectively ) , so we shall quote these pair of values for each one of the cases considered . we first construct simulations with a certain degree of non - gaussianity by using the edgeworth expansion @xcite . we then analyse these simulations and calculate the power of the test to discriminate between a gaussian distribution ( the null hypothesis ) and a distribution with skewness and kurtosis injected via the edgeworth expansion ( the alternative hypothesis ) . the aim is to quantify which signal - to - noise level is required to detect a certain degree of non - gaussianity in the data with our method . we have two options in preparing these simulations : we can inject the skewness and kurtosis either in real space or in visibility space . if we include the non - gaussianity through the edgeworth expansion in real space , this non - gaussianity is diluted when we transform to the fourier ( or visibility ) space ( this is illustrated in appendix a ) . thus , we decide to inject the skewness / kurtosis directly in fourier space , and calibrate our method in terms of the ability to detect deviations of the moments of the signal - to - noise eigenmodes of the visibilities with respect to the gaussian case . a similar approach has been adopted by other authors ( e.g. in @xcite they explore the non - gaussianity of the visibilities simultaneously to the estimation of the power spectrum , by using a modified non - gaussian likelihood function ) . to use non - gaussian simulations with given skewness / kurtosis in fourier space generated from the edgeworth expansion is not motivated by any physical model , but it is a reasonable approach to calibrate the power of a non - gaussianity test because these non - gaussian maps are easy to prepare with specified statistical properties . moreover , systematic effects could introduce a non - gaussian signal in the visibilities which could be mimic with the edgeworth expansion . the probability distribution of these systematic effects could be , for example , asymmetric and , in a first approximation , this fact can be modeled by a distribution with a non - zero skewness . in addition , the distribution could be more peaked or flat that the gaussian one , and that can be modeled by a distribution with a non - zero kurtosis . in this manner the edgeworth simulations are taken as a benchmark to quantify the power of our method to detect some kind of non - gaussian signals . these non - gaussian simulations are generated as follows . we generate a realization of independent values @xmath158 with zero mean , unit variance , and skewness and kurtosis @xmath159 and @xmath160 , respectively . for definiteness , we adopt here the values @xmath161 . as in subsection [ sec : gauss ] , the real parts of the visibilities are @xmath144 . the imaginary parts are generated in an analogous way . after that , we add different noise levels according to a vsa observation . we then decorrelate the data of the simulations and calculate the power of the tests . again , we use the same vsa template observation , binned into cells of side 9@xmath137 to study the power of our tests . figure [ fig3 ] shows the logarithm of the eigenvalues associated with the real visibilities ( those resulting from the diagonalization of @xmath162 ) and those associated with the imaginary parts ( the diagonalization of @xmath163 ) . only the @xmath164 of the data have an eigenvalue signal - to - noise higher than 0.01 ; that is , a signal - to - noise ratio higher than 0.1 . obtained from the template observation used in figure [ fig2 ] . we plot the values associated with both the real and imaginary parts of the visibilities ( see the text for details ) . we have @xmath165 eigenvalues , only 420 of which have a value above 0.01 , which corresponds to a signal - to - noise of @xmath166.,width=302 ] first , we consider the case of a low signal - to - noise ratio , and we adopt the noise levels corresponding to a single - field vsa observation ( @xmath146 hr , @xcite ) . we analyse the non - gaussian simulations with different cuts in the eigenvalues ; that is , we include in the analysis only those eigenmodes whose eigenvalue is higher than a value @xmath167 . in table [ tab2 ] is shown the power of the @xmath150 , @xmath151 , @xmath152 and @xmath149 statistics to discriminate between a gaussian and a non - gaussian distribution . it can be demonstrated that the simulations are constructed such that @xmath112 has zero mean and unit variance , so only the @xmath152 and @xmath149 distributions are notably different from a @xmath41 distribution ( see section [ sec:001 ] ) . in fact , @xmath151 also deviates slightly from a @xmath41 distribution when the analysed data have some degree of kurtosis ( @xmath151 is related to @xmath168 , i.e. to the fourth power of the data ) , but , as we can see in table [ tab2 ] , this deviation is very low . the corresponding probability function is shown in fig . [ ng_vs_g ] ( dotted line ) and is compared with a gaussian function ( dashed line ) . we have used 5000 simulations in this computation . note that although the input skewness and kurtosis is 1.0 , the output simulations have a mean skewness equal to @xmath169 and a mean kurtosis equal to @xmath170 ; i.e. we are analysing some simulations with kurtosis @xmath171 . so we expect less power in the detection of the kurtosis that in the detection of the skewness . the first column in table [ tab2 ] indicates the minimum value of the signal - to - noise eigenvalues that we use to perform the analysis . we indicate the square root of the latter quantity ( signal - to - noise ratio ) in parentheses . the second column is the number of eigenmodes that we use to calculate the statistics ; that is , the number of eigenvalues such that @xmath172 . finally , the last columns indicate the power of the different statistics ( in percentages ) when the significance level is @xmath173 . we see that our test is not able to detect non - gaussianity when every eigenmode is included in the analysis . when eigenmodes with low signal - to - noise ratio are excluded the power is improved but the values are very poor . this tells us that the noise level of a single vsa pointing ( @xmath146 hr ) is too high to detect the non - gaussian simulations we have constructed via the edgeworth expansion . and @xmath174 . dashed line : gaussian distribution . histogram : distribution obtained from a single simulation ( signal plus noise ) involving all the eigenvalues , for a noise level corresponding to a @xmath175 hr integration . the three distributions have zero mean and unit variance . the non - gaussianity in the single simulation is hardly seen in its histogram ; however , it is easily detected when we used our method , as shown in table [ tab3].,width=302 ] c r c c c c @xmath167(@xmath176 ) & num.&@xmath150&@xmath151&@xmath152&@xmath149 0.0 ( 0.0 ) & 1790 & 5.10/1.12 & 4.92/1.06 & 9.06/2.46 & 6.44/ 1.90 0.1 ( 0.32 ) & 188 & 5.12/0.82 & 5.96/1.54&26.20/13.26&12.30/ 7.48 0.2 ( 0.45 ) & 133 & 5.08/0.92 & 6.68/1.70&31.26/17.68&13.76/ 8.76 0.3 ( 0.55 ) & 107 & 4.84/1.04 & 6.38/1.74&32.06/18.46&13.70/ 9.70 0.4 ( 0.63 ) & 89 & 5.06/0.94 & 6.56/2.14&33.64/20.22&15.02/10.90 0.5 ( 0.71 ) & 76 & 5.82/1.10 & 7.14/2.30&34.22/20.56&14.82/10.94 [ tab2 ] we therefore now explore the case of an integration time of @xmath175 hr , which is comparable to the whole signal - to - noise level achieved in the dataset presented in @xcite . in this case , the noise levels are reduced by a factor @xmath177 with respect to the previous case . the results are shown in table [ tab3 ] . as is to be expected , the detection of non - gaussianity is better . for example , we could detect our non - gaussian simulations with a power equal to @xmath13699@xmath178 ( for the @xmath152 statistic ) with the signal - to - noise achieved during @xmath179 hours and analysing eigenmodes with eigenvalues higher than 0.4 or 0.5 . in this last case the distribution of all the data ( every eigenmode ) for one simulation is shown by the histogram in figure [ ng_vs_g ] . we see that this distribution is closer to a gaussian ( dashed line ) than the initial one ( dotted line ) for two reasons : first , the loss of skewness and kurtosis by the construction of the edgeworth simulations ( the actual skewness and kurtosis of the simulations is lower than the input values , see above ) and , second , by the addition of gaussian noise . when we only plot the autovalues higher than 0.4 or 0.5 , the histogram is closer to the dotted line ; that is , we see the non - gaussianity . c r c c c c @xmath167(@xmath176 ) & num.&@xmath150&@xmath151 & @xmath152 & @xmath149 0.0 ( 0.0 ) & 1790 & 5.38/0.86&5.52/1.22 & 57.96/36.08&18.20/10.04 0.1 ( 0.32 ) & 513 & 5.62/1.22&6.48/1.62 & 95.92/88.62&35.58/25.64 0.2 ( 0.45 ) & 441 & 5.94/1.52&7.36/1.82 & 97.66/91.94&36.24/26.82 0.3 ( 0.55 ) & 402 & 5.60/1.08&7.60/2.22 & 98.26/93.28&37.42/28.52 0.4 ( 0.63 ) & 371 & 5.90/1.54&7.52/2.02 & 98.72/94.28&38.56/29.68 0.5 ( 0.71 ) & 245 & 5.52/1.40&7.80/2.38 & 98.82/94.44&39.24/30.80 [ tab3 ] in the tables shown in this section we have presented only the power of the @xmath40 statistics . in the case we are studying , @xmath54 has less power than @xmath152 because @xmath180 also has information about @xmath150 and @xmath151 , which do not have power and compensate for the power of @xmath152 . however , @xmath54 is better than the @xmath152 statistic if we also have detection of non - gaussianity in @xmath151 and/or @xmath150 . for the same reason , @xmath181 is slightly better than @xmath149 because it has combined information about @xmath149 and @xmath152 . in our case , however , @xmath181 is not better than @xmath152 , so we show only the @xmath40 statistics . finally , we have analysed the edgeworth simulations with the test applied to the phases ( subsection [ sec:3_3 ] ) . we have found that the non - gaussianity of this kind of simulation is not detected with the phases method ; that is , the phases are compatible with a uniform distribution . a deeper analysis of the phases shows that the moments @xmath67 ( section [ sec : unif ] ) are sligthly different from the values corresponding to a uniform distribution . however , the quantity , @xmath3 of data with high signal - to - noise eigenvalues is not large enough to give values of the @xmath40 statistics that differ sufficiently for the values obtained from a uniform distribution , so the power of the test has very low values . ( the distribution for the phases can be calculated analytically by assuming , for example , a distribution for the real and imaginary parts given by equation ( 25 ) in @xcite . we have found this ditribution to oscillate about the uniform one , but the moments are very close to those of the uniform distribution . ) in this section , we probe if our method is able to detect the non - gaussian signature introduced by cosmic strings . to this end , we apply our method to the analysis of the string simulation shown in fig . [ strings_map ] @xcite . we first analyse the simulation without noise . in this way we can learn what happens when we have only the string map . after that we add different noise levels to see how the method is able to detect the strings . although in theory the correlation matrix depends only of the power spectrum and the beam ( see expression [ eq:002 ] ) , the ( finite ) pixel size of the real string map we are analysing and the irregular coverage of the visibility plane could introduce noticeable imprecisions . the correlation matrix of the strings is then calculated using gaussian maps simulated with the same pixel size : we calculate the power spectrum of the string real map and construct gaussian real maps with this power spectrum ( we use 40000 gaussian simulations ) . from these gaussian maps we calculate the corresponding visibility values ( expression [ eq:005 ] ) . the correlation of these visibility simulations is that used for the analysis of the strings . note that in the expression ( [ eq:002 ] ) there is no hypothesis concerning the statistic of the temperature field . the only hypothesis is the homogeneity of the field . the template observation of the vsa is the same as in previous sections ( 895 binned visibilities in 9@xmath137 cells ) . in this case we diagonalize the correlation matrix of the signal because we do not have noise . with this diagonalization we decorrelate the data in a similar way as is done in sections [ sec:002 ] and [ sec:004 ] . we analyse the real and imaginary parts of the visibilities obtained from the string map without noise . the values found for the statistics are : @xmath182 , @xmath183 , @xmath184 and @xmath185 . the estimated moments , @xmath186 , are @xmath187 , @xmath188 , @xmath189 and @xmath190 . these values are clearly incompatible with a gaussian realization . the @xmath112 quantities ( see expression [ eq : y_k ] ) associated with the real parts of the string visibilities are shown in the fig . [ fig_y_s ] ( those associated with the imaginary parts have similar features ) . every @xmath112 has an associated signal eigenvalue @xmath191 whose value decreases with @xmath10 . we see , then , that for the strings the absolute value ( or _ amplitude _ ) of @xmath112 grows when @xmath191 decreases ( this explains why the moments @xmath186 are so large ) . this feature shows that the @xmath112 are not well decorrelated or normalized . we analyse a gaussian map with the same procedure . figure [ y_s_gauss ] shows the resulting @xmath112 quantities for this gaussian case . we see that they have a width of order of unity . thus , we conclude that the behaviour of fig . [ fig_y_s ] is a feature of the non - gaussianity of the string signal . the fact that the @xmath112 quantities are not well decorrelated or normalized indicates that we are not able to estimate their correlation matrix accurately . this seems to be because we have a model with a very large cosmic variance , and the correlation matrix ( or equivalently the power spectrum ) can not be well estimated from only one realization . moreover , we are estimating the power spectrum with the expression that maximizes the likelihood under the hypothesis of gaussianity for the temperature field @xcite . but the temperature field is not gaussian for the strings , and so the estimated power spectrum for the strings ( that which maximizes the likelihood in this case ) can differ from the one we are using . for a study of the dependence of the power spectrum on non - gaussianity see @xcite . in the gaussian case , however , we are able to calculate the power spectrum and then we decorrelate and normalize the visibilities properly . in this way , the @xmath112 quantities have a width equal to unity , i.e. @xmath192 , as shown in figure [ y_s_gauss ] . summarizing , when there is no noise , in principle we should detect the strings by means of all the @xmath40 statistics ( even @xmath150 and @xmath151 ) because their values are not compatible with a @xmath41 distribution ( see table [ tab1 ] ) . associated with the real part of the visibilities for an observation of the string map presented in fig . [ strings_map ] using our vsa template without adding noise.,width=321 ] , but for a simulated gaussian cmb map.,width=321 ] finally , we have applied the smooth tests of goodness - of - fit developed by neyman to the phases of the visibilities of the noiseless map of strings ( section [ sec:3_3 ] ) . the @xmath40 values are : @xmath193 ( @xmath194 ) , @xmath195 ( @xmath196 ) , @xmath197 ( @xmath198 ) and @xmath199 ( @xmath200 ) . and for the @xmath201 statistics : @xmath202 ( @xmath194 ) , @xmath203 ( @xmath204 ) , @xmath205 ( @xmath206 ) and @xmath207 ( @xmath208 ) . the probability of obtaining a lower or equal statistic value under the null hypothesis ( the visibilities are gaussian ) is showed in parentheses . in the following subsection we discuss the effect of gaussian noise on the detectability of strings with our method . the features in fig . [ fig_y_s ] change when ( gaussian ) noise is added . the behaviour for low values of @xmath10 remains unchanged because these values are associated with high signal - to - noise values so that these data are dominated by the signal . however , the behaviour for high @xmath10 is dominated by the noise ( low signal - to - noise eigenvalues ) so that the behaviour will resemble that of a gaussian signal . the analysis with different noise levels is given in table [ tab4 ] ( noise corresponding to an integration time equal to @xmath209 hr ) and table [ tab5 ] ( noise corresponding to an integration time equal to @xmath210 hr ) . we have analysed 5000 simulations involving the string simulation plus a noise realization . the analysis is done for different cuts of the eigenvalues ( @xmath167 ) . the power to distinguish between a gaussian distribution and the strings plus noise when the significance level is @xmath211 is shown in the tables . l c c c c @xmath167 & @xmath150 & @xmath151 & @xmath152 & @xmath149 0 . & 5.16/0.82 & 5.72/1.20 & 4.92/1.00 & 4.50/0.98 0.1 & 3.72/0.58 & 6.00/0.86 & 2.28/0.36 & 1.70/0.28 0.2 & 3.10/0.30 & 4.16/0.44 & 1.70/0.24 & 1.20/0.34 0.3 & 5.10/0.66 & 3.36/0.22 & 1.48/0.26 & 1.24/0.30 0.4 & 1.46/0.14 & 1.78/0.12 & 1.08/0.36 & 1.28/0.30 0.5 & 1.00/0.10 & 2.04/0.08 & 0.80/0.18 & 0.90/0.22 [ tab4 ] l c c c c @xmath167 & @xmath150 & @xmath151 & @xmath152 & @xmath149 0 . & 5.30/0.98 & 24.26/8.44 & 4.20/0.86 & 4.28/1.04 0.1 & 6.56/0.84 & 79.56/49.32 & 1.58/0.12 & 1.30/0.22 0.2 & 7.18/0.92 & 86.72/59.60 & 1.08/0.08 & 1.24/0.34 0.3 & 2.94/0.26 & 87.62/60.44 & 0.80/0.04 & 0.70/0.10 0.4 & 3.74/0.10 & 90.74/64.16 & 0.48/0.02 & 0.66/0.08 0.5 & 3.20/0.12 & 92.90/69.58 & 0.48/0.08 & 0.38/0.08 [ tab5 ] the results in table [ tab4 ] indicate that the vsa noise of a single field observed during @xmath209 hrs is too high for detecting the strings with our method . when the integration time is multiplied by 25 ( table [ tab5 ] ) we start to detect the strings by means of the @xmath212 statistic . it is important to note that , even if the data were well decorrelated , the @xmath212 statistic would also be a indicator of non - gaussianity . for example , given realizations of normalized and independent data with kurtosis @xmath213 , the mean value of @xmath212 is equal to @xmath214 . another example of the use of @xmath212 to detect non - gaussianity can be found in @xcite in the formalism of the multinormal analysis . as a test , we repeat the same procedure using a gaussian simulation instead of the string map and , as expected , we obtain a negligible power . nowadays , non - gaussian simulations constructed by the addition of a gaussian field and its square have acquired a notable relevance because they could represent perturbations produced in several inflationary scenarios ( for a review on the subject , see @xcite and references therein ) . the parameter which measures the coupling with the non - linear part is denoted by @xmath1 and , then , an analysis to study the power of our method to detect this parameter seems to be very appropriate . realistic simulations in the flat - sky approximation would require to develop appropriate software ( see e.g. @xcite for simulations on the full sphere ) , and this is beyond the scope of the present work . however , we can estimate the power of our method in detecting a non - zero @xmath1 component using a simple approximation . we follow @xcite , and we generate what they call `` @xmath2 maps '' by assuming that the observed temperature contrast field on sky ( @xmath215 ) can be written as @xmath216 where @xmath217 is the gaussian ( linear ) component . in reality , this approximation is only valid in the sachs - wolfe regime ( see e.g. @xcite ) , but we use it as a toy model in our exercise of the study of the detection of the @xmath1 parameter , so the definition of the non - linear coupling parameter in our @xmath2 model gives @xmath218 . therefore , the coupling parameter used by @xcite and @xcite is related with our definition as @xmath219 , and the one used by @xcite is @xmath220 . ] . we first consider the ideal case where there is no instrumental noise , and we have analyzed different sets of simulations with @xmath1 values @xmath221 ( gaussian case ) , @xmath222 , @xmath223 and @xmath224 . we have run @xmath224 simulations for every case . only the case with @xmath225 has been detected via the @xmath151 statistic . [ fig_chi2 ] shows the distribution function of the @xmath151 statistic for the gaussian case ( dashed boxes ) , compared with the corresponding one for the non - gaussian case with @xmath225 ( solid boxes ) . from here , we infer that the power of this statistic to discriminate between the non - gaussian simulations and the gaussian ones is about the @xmath226 ( @xmath227 ) with a significance level of @xmath228 ( @xmath229 ) . analysing the same previous simulations but adding the noise level corresponding to an observation time of @xmath230 hr , we find that the power of our statistic is reduced roughly by a factor 2 . @xcite constrained the @xmath1 parameter by using the bispectrum of the vsa extended array data ( note that this is exactly the same configuration that we have adopted to build our visibility template ) . with their definition of @xmath1 , they found an upper limit of 7000 with a confidence level of @xmath231 . in the sachs - wolfe regime , this would correspond to a value of 31500 according to our @xmath1 definition . however , this value can not be direclty compared with the one obtained in our work , because our model does not correspond to the realistic case which was considered in @xcite . for the gaussian case ( dashed boxes ) and for the non - gaussian case with @xmath225 ( solid boxes ) . the solid line is a @xmath115 function normalized to the number of simulations ( 10000 simulations ) . ( for the sake of clarity , the histogram for the gaussian case has been cut at the top.),width=302 ] in this paper we have presented a method of searching for non - gaussianity in data from interferometric cmb observations , directly in the visibility space . this method can be adapted to other interferometric experiments ( e.g. cbi and dasi ) if we know the correlation matrices of the signal and of instrumental noise . note that in this paper , we have dealt with decorrelated noise ( which is usually the case for these experiments ) , but the case of correlated noise can be studied in an analogous way . we have applied the method to work with the real and imaginary parts of the visibilities . the method tests whether they are gaussian distributed . however , we have applied the method to the phases of the visibilities to test whether they are uniformly distributed , but we found that it is not very sensitive for detecting the kind of non - gaussianity we have analysed here . we have integrated the signal - to - noise formulation into the smooth goodness - of - fit tests . in this way we can deal only with the data dominated by the signal we want to analyse . in the text it is noted that the correlation matrices of the signal - to - noise eigenmodes and of the signal are descomposed as the product of a matrix and its transpose ( subsection [ sec:002 ] and section [ sec : gauss ] ) . this decomposition is analogous to that of cholesky . this latter decomposition is computationally faster than the one used here ; however , the decomposition we use allows us to deal with better quality data , that is , with a higher signal - to - noise ratio . the analysis with the cholesky decomposition takes every eigenvalue ; that is , @xmath232 . it is important to stress that smooth goodness - of - fit tests do not require the data to be on a regular grid , but can be applied to any data set . in that sense the test is perfectly adapted to interferometric data because the coverage of the @xmath0 plane is neither regular nor complete . the method could therefore also be applied by selecting those visibilities in a certain @xmath0 range , thus allowing study of the non - gaussianity as a function of angular scale . in addition , although the method has been presented here as a tool to study the non - gaussian properties of the sky signal , it could also be used as a powerful diagnostic for detecting systematics in the data , as we pointed out in section 6.1 . for short integration periods , a stack of visibilities will be dominated by instrumental noise , so this method could be used to trace the presence of spurious signals in the data ( e.g. those comming from cross - talk between antennas ) , or to study the correlation properties of the noise . summarizing , to study the power of our method in detecting non - gaussian signals on the sky , we have analysed three kinds of simulations . first we have analysed non - gaussian visibilities created by inserting some degree of skewness and kurtosis ( @xmath233 ) with the edgeworth expansion directly in fourier space . using the vsa as a reference experiment , we have shown the performance of the method in detecting those levels of non - gaussianity in the data with realistic values of the integration time for this experiment . all these results can be easily adapted to other instruments , just by rescaling the integration times according to the square of the ratio of the different sensitivities . in addition , we have also shown the performance of the test in the detection of the non - gaussian signal introduced by cosmic strings . even though those kind of signals are usually detected using real - space statistics or wavelets @xcite , we have demonstrated that the signal - to - noise eigenmodes approach allows us to detect them because the method is very sensitive to the characterization of the covariance matrix . in this particular case , due to the non - gaussian nature of the strings , a complete decorrelation can not be achieved and the statistics show huge deviations with respect to the gaussian case . for the case of the vsa , this translates into the fact that cosmic strings can hardly be detected in single - field observations ( integration times of @xmath209 hr ) , but they could be detected ( if present ) using this method , given the current sensitivity achieved by the whole data set published by the vsa team ( with a sensitivity of @xmath13625 @xmath234 hr integration ) . finally , we have studied the power of the method to detect a non - zero @xmath1 component in a toy model based on @xmath2 maps . we would like to thank f. r. bouchet for kindly providing the string map , and to r. rebolo and the anonymous referee for useful comments . we also acknowledge terry mahoney for revising the english of the manuscript . jar - m acknowledges the hospitality of the ifca during two visits . ama acknowledges the iac for its hospitality during a visit . rbb acknowledges the mcyt and the uc for a ramn y cajal contract . we acknowledge partial financial support from the spanish mcyt project esp2002 - 04141-c03 - 01 . aghanim , n. , kunz , m. , castro , p. g. , & forni , o. 2003 , a&a , 406 , 797 aliaga a. m. , martnez - gonzlez e. , cayn l. , argeso f. , sanz j. l. , barreiro r. b. , 2003a , new astronomy reviews , 47 , 821 aliaga a. m. , martnez - gonzlez e. , cayn l. , argeso f. , sanz j. l. , barreiro r. b. , gallegos j. e. , 2003b , new astronomy reviews , 47 , 907 amendola , l. 2002 , apj , 569 , 595 balbi , a. et al . 2000 , apj , 545 , l1 barreiro , r. b. & hobson , m. p. 2001 , mnras , 327 , 813 bartolo , n. , komatsu , e. , matarrese , s. , & riotto , a. 2004 , phys . , accepted ( astro - ph/0406398 ) bond , j. r. 1995 , phys . rev . lett . , 74 , 4369 bond , j. r. , jaffe , a. h. & knox , l. 2000 , apj , 533 , 19 bouchet , f. r. , bennett , d. p. & stebins , a. 1988 , nat , 335 , 410 cayn , l. , martnez - gonzlez , e. , argeso , f. , banday , a.j . & grski , k.m . , 2003a , mnras , 339 , 1189 cayn , l. , argeso , f. , martnez - gonzlez , e. & sanz j. l. 2003b , mnras , 344 , 917 chiang , l. , naselsky , p. d. , verkhodanov , o. v. & way , m. j. 2003 , apj , 590 , l65 cox d. r. & hinkley , d. v. 1974 , theoretical statistics ( london : chapman & hall ) cruz , m. , martnez - gonzlez , e. , vielva , p. , & cayn , l. 2004 , mnras , in press ( astro - ph/0405341 ) dickinson , c. et al . 2004 , mnras , 353 , 732 eriksen , h. k. , hansen , f. k. , banday , a. j. , grski , k. m. , & lilje , p. b. 2004a , apj , 604 , 14 eriksen , h. k. , novikov , d. i. , lilje , p. b. , banday , a. j. , & grski , k. m. 2004b , apj , 612 , 64 grainge , k. et al . 2003 , mnras , 341 , l23 guth , a. h. , 1981 , phys . d. , 23 , 347 hanany , s. et al . 2000 , apj , 545 , l5 hobson , m. p. , jones , a. w. & lasenby , a. n. 1999 , mnras , 309 , 125 hobson , m. p. & maisinger , k. 2002 , mnras , 334 , 569 komatsu , e. & spergel , d. n. 2001 , phys.rev.d , 63 , 063002 komatsu , e. et al . 2003 , apjs , 148 , 119 lay , o. p. & halverson , n. w. 2000 , apj , 543 , 787 leitch , e. m. et al . 2002 , apj , 568 , 28 liguori , m. , matarrese , m. & moscardini , l. 2003 , apj , 597 , 57 martnez - gonzlez , e. , gallegos , j. e. , argeso , f. , cayn , l. & sanz , j. l. 2002 , mnras , 336 , 22 melhuish , s. j. , dicker , s. , davies , r. d. , gutierrez , c. m. , watson , r. a. , davis , r. j. , hoyland , r. & rebolo , r. 1999 , mnras , 305 , 399 neyman , j. 1937 , skandinavisk aktuarietidskrift , 20 , 150 padin , s. et al . 2001 , apj , 549 , l1 park , c. 2004 , mnras , 349 , 313 rayner , j. c. w. & best d. j. 1989 , smooth tests of goodness of fit ( new york : oxford university press ) rayner , j. c. w. & best , d. j. 1990 , international statistical rev . , 58 , 9 rocha , g. , magueijo , j. , hobson , m. & lasenby , a. , 2001 , phys . d. , 64 , 063512 savage , r. et al . 2004 , mnras , 349 , 973 smith , s. et al . 2004 , mnras , 352 , 887 taylor , a. c. et al . 2003 , mnras , 341 , 1066 vielva , p. , martnez - gonzlez , e. , barreiro , r. b. , sanz , j. l. , & cayn , l. 2004 , apj , 609 , 22 watson , r. a. et al . 2003 , mnras , 341 , 1057 white , m. , carlstrom , j. e. , dragovan , m. , & holzapfel w. l. 1999 , apj , 514 , 12 suppose a square real map of pixel coordinates @xmath235 and pixel data @xmath236 . these values are generated via the edgeworth expansion such that they are independent , normalized and have skewness @xmath159 . the map has @xmath237 pixels on a side and the area of the pixel is @xmath238 . we now perform its fourier transform : @xmath239 whose bispectrum is @xmath240 taking into account that @xmath236 are independent and @xmath241 we have @xmath242 where @xmath243 is the kronecker delta . the fourier modes could be normalized to the desired spectrum : @xmath244 such that @xmath245 . finally , the ( normalized ) bispectrum is given by @xmath246 note that the gaussianity associated to the third moment in the fourier space is reduced by a factor @xmath237 compared to that in real space . when we introduce some degree of non - gaussianity ( skewness or kurtosis ) by means of the edgeworth expansion in the visibilities , it is clear that each eigenmode will show a different degree of non - gaussianity which will depend on the associated signal - to - noise eigenvalue . this relation can be obtained as follows . as it is explained in the main text , we generate a realization of independent values @xmath158 with zero mean , unit variance , and skewness and kurtosis @xmath159 and @xmath160 , respectively . as in subsection [ sec : gauss ] , the real parts of the visibilities are @xmath144 . the imaginary parts are generated in an analogous way . after that , we add the corresponding different noise levels and decorrelate the data by using the transformation given by equation ( [ eq:003 ] ) . then , bearing in mind the definition of the @xmath96 quantities , it can be easily demonstrated that for the final decorrelated and normalized data we have as one would expect , the skewness and the kurtosis are lost if the ( gaussian ) noise dominates the data ( @xmath248 ) . in contrast , the skewness and the kurtosis are preserved for the case @xmath249 . in the previous expressions , the angled brackets ( @xmath250 ) denote an average over realizations .

we study sub - riemannian and sub - lorentzian geometry on the lie group and on its universal cover . in the sub - riemannian case we find the distance function and completely describe sub - riemannian geodesics on both and , connecting two fixed points . in particular , we prove that there is a strong connection between the conjugate loci and the number of geodesics . in the sub - lorentzian case , we describe the geodesics connecting two points on , and compare them with lorentzian ones . it turns out that the reachable sets for lorentzian and sub - lorentzian normal geodesics intersect but are not included one to the other . a description of the timelike future is obtained and compared in the lorentzian and sub - lorentzain cases . contents \1 . introduction . \2 . sub - riemannian and sub - lorentzian geometry . \3 . structure of the lie groups and . \4 . sub - riemannian geometry on and . \5 . sub - lorentzian geometry on . \6 . proofs of main results .

0910.0945

sub - riemannian geometry is proved to play an important role in many applications , e.g. , in mathematical physics and control theory . sub - riemannian geometry enjoys major differences from the riemannian being a generalization of the latter at the same time , e.g. , geodesics may be singular , the hausdorff dimension is larger than the manifold topological dimension , the exponential map is never a local diffeomorphism . there exists a large amount of literature developing sub - riemannian geometry . typical general references are @xcite . the sub - lorentzian case is less studied and the first works in this directions appeared rather recenlty , see @xcite . in the development of sub - riemannian geometry , one observes several examples , which are mainly nilpotent lie groups , with either a left or right invariant distribution and metric . a sample representative is the heisenberg group ( see , e.g. , @xcite ) . analysis of these groups in the sub - riemannian setting , is already well studied . while these groups enjoy the advantage that explicit results are easier to obtain , their properties are sometimes too nice to be good examples to reveal all specific features of sub - riemannian geometry in its generality . for instance , the cut locus and the conjugate loci for the heisenberg group globally coincide . a natural next step after considering nilpotent groups is to consider semisimple lie groups . let @xmath2 be such a group with the lie algebra @xmath3 . let @xmath4 be the cartan involution sending an element @xmath5 from the lie algebra to minus its conjugate transpose . then there is a splitting @xmath6 , where @xmath7 and @xmath8 are the @xmath9 and @xmath10 eigenspaces of @xmath11 . remember that the killing form @xmath12 is non - degenerate when @xmath3 is semisimple . if @xmath2 is compact ( or more generally , if @xmath13 is compact , where @xmath14 denotes the center of @xmath2 ) , then @xmath15 , and @xmath16 , is positive definite on @xmath3 . so , we can use this to define a bi - invariant riemannian metric on @xmath2 . the restriction of this metric to a distribution on @xmath2 gives a sub - riemannian manifold . an example of such a manifold , namely @xmath17 ( or @xmath18 considered as unit quaternions ) , was considered in @xcite . the problem of geodesic connectivity on @xmath18 was addressed in @xcite . if @xmath2 is non - compact and @xmath19 ( i.e. , @xmath13 is non - compact ) , then the killing form restricted to @xmath8 is positive definite . we consider a left translation of @xmath8 as our horizontal distribution , and using the metric induced by the killing form , we obtain a sub - riemannian manifold equipped with a bi - invariant metric . let @xmath20 be a subgroup of @xmath2 with the lie algebra @xmath7 . then @xmath2 is diffeomorphic to @xmath21 , by @xmath22 , and if @xmath2 has a finite center , then @xmath20 is a maximal compact subgroup . in addition , the above mentioned distribution is horizontal with respect to the quotient map @xmath23 it follows that all normal geodesics are liftings of geodesics from @xmath24 , hence they are of the form @xmath25 where @xmath26 is the initial point . here we consider an example with @xmath27 and @xmath28 . although non - holonomic geometry on @xmath0 ( or the isometric case of @xmath29 ) was first considered earlier in , e.g. , @xcite , we will obtain new results and a more complete description of geodesics both in sub - riemannian and sub - lorentzian settings . much of all meaningful results come from the analysis of the universal cover of @xmath0 , which we denote by @xmath1 , and which is of its own interest as a new representative of a non - nilpotent lie group over the topological space @xmath30 . we remark also that the khler manifold @xmath31 , is of particular importance in quantum field theories describing a black hole in two - dimensional spacetime by means of an @xmath31-gauged wess - zumino - novikov - witten model , see e.g. , @xcite . let @xmath32 be the lie algebra of @xmath0 . when considered as a bilinear form on the entire @xmath32 , the killing form is an index 1 pseudo - metric . furthermore , the induced lorentzian metric on the lie group makes @xmath0 isometric to what is called the anti - de sitter space @xmath33 in general relativity . this makes it tempting to study sub - lorentzian geometry on @xmath0 . apart from the fact that the hamiltonian approach was proposed in @xcite to study sub - lorentzian structures on @xmath0 , the authors are not aware of other concrete examples so far , where sub - lorentzian geometry is studied on any other manifold different from the heisenberg group @xcite or an extension of it to a @xmath34-type carnot group @xcite . the notion of distance in sub - lorentzian geometry , as well as in lorentzian geometry , is given by the supremum of length over timelike curves . since timelike loops may appear in @xmath0 , the distance function behaves badly ( more specifically the distance from a point to itself is @xmath35 ) . therefore , sub - lorentzian geometry on @xmath1 is more interesting and meaningful than on @xmath0 . we are also interested in comparison of sub - lorentzian and lorentzian geometries on @xmath1 , to try to understand somewhat more how the geometry changes when the lorentzian metric is restricted to a distribution . the strong interplay between @xmath1 and @xmath0 will , however , be practical for all explicit calculations . the structure of the paper is as follows . section 2 is devoted to general definitions and relations between sub - riemannian and sub - lorentzian geometry , and optimal control . section 3 describes the lie groups @xmath0 and @xmath1 . section 4 contains results concerning sub - riemannnian geometry . we describe the number of geodesics connecting two points and give explicit formulas for the distance functions . the cut and conjugate loci on both lie groups are given . we discuss the connection between the conjugate locus and the behavior of the geodesics . in section 5 , we completely describe the two - point connectivity problem by sub - lorentzian geodesics , and compare it with the lorentzian case . the lorentzian and sub - lorentzian future for @xmath1 are compared . it turns out that the reachable sets for lorentzian and sub - lorentzian normal geodesics intersect but are not included one to the other . section 6 contains the proofs of main results . the authors express special thanks to mauricio godoy and irina markina for many interesting discussions during seminars at the university of bergen . a sub - riemannian manifold is an @xmath36-dimensional manifold @xmath37 , with a fiber metric @xmath38 on an @xmath39-dimensional smooth distribution @xmath40 ( @xmath41 ) . by distribution , we mean a sub - bundle of the tangent bundle . absolutely continuous curves that are almost everywhere tangent to @xmath42 are called horizontal . the length of a horizontal curve @xmath43 \rightarrow m$ ] , is defined by @xmath44 the _ carnot - carathodory distance _ between points @xmath45 is defined as @xmath46 where the infimum is taken over all horizontal curves @xmath47 satisfying @xmath48 and @xmath49 . if there are no such curves connecting @xmath50 and @xmath51 , then the distance is @xmath35 . if the minimum in the above relation is attained , then the curve is called a length minimizing curve . define @xmath52 and iteratively @xmath53 $ ] , @xmath54 , where @xmath55 $ ] denotes the lie brackets , @xmath56 = xy - yx$ ] for @xmath57 , @xmath58 . if there exists a positive integer @xmath59 , such that @xmath60 , then the distribution is called bracket generating . it is called regular if @xmath61 is independent of the choice of @xmath62 for all @xmath63 . we say that @xmath42 is step @xmath64 regular , if it is regular and @xmath64 is the smallest number for which @xmath60 . chow - rashevski theorem @xcite states that a bracket generating distribution @xmath42 guarantees that any two points may be connected by a horizontal curve . in addition , we have the following generalizations of the corresponding properties from the riemannian case : suppose that @xmath42 satisfies the bracket generating condition . then * any @xmath65 has a neighborhood @xmath66 , such that there exists a minimizing curve joining the points @xmath51 and @xmath50 for every @xmath67 ; * if @xmath37 is complete regarding to @xmath68 , then any two points can be joined by a minimizing curve . remark that the length minimizing curve may be singular for arbitrarily close points ( see @xcite ) . a curve is called geodesic if it is locally a length minimizneighborhooder . by a normal geodesic in the sub - riemannian case we mean an integral curve of the hamiltonian system generated by a hamiltonian function @xmath69 in some neighborhood of a point @xmath62 with respect to any local orthonormal basis @xmath70 in this neighborhood . here , and in rest of the paper , if @xmath5 is a vector field , then @xmath71 denotes the hamiltonian function @xmath72 with respect to the vector field @xmath5 . if we have a basis @xmath73 , then we simplify notations by writing @xmath74 instead of @xmath75 . one of the principal differences between sub - riemannian and riemannian geometries is that the function @xmath76 is not differentiable in any neighborhood of @xmath50 when @xmath77 . a sub - lorentzian manifold is defined similarly to a sub - riemannian manifold , but with @xmath38 now being an index 1 pseudo - metric on @xmath42 . we will say that a vector @xmath78 is * timelike if @xmath79 , * lightlike or null if @xmath80 , * spacelike if @xmath81 , * causal or nonspacelike if @xmath82 . a chosen vector field @xmath83 in @xmath42 , is said to be the time - orientation on @xmath37 , if @xmath84 for any @xmath85 . a causal vector is called future directed if @xmath86 , and past directed if @xmath87 . a horizontal curve @xmath88\ \rightarrow \mathbb{r}$ ] is called timelike , null , spacelike , causal , future directed or past directed , respectively , if @xmath89 is such a vector for almost every @xmath90 $ ] . we define the timelike future @xmath91 of @xmath50 with respect to @xmath38 as the set of all points @xmath92 , such that there is a horizontal , timelike future directed curve @xmath47 , with @xmath93 and @xmath49 . the causal future , @xmath94 , is defined similarly , with timelikeness interchanged with causality . analogous definitions are valid for the timelike or causal past , which we denote by @xmath95 and @xmath96 respectively . we define the length of a horizontal causal curve by @xmath97 . the sub - lorentzian distance is defined by @xmath98 the supremum is taken over all horizontal future directed causal curves from @xmath50 to @xmath51 . similarly to the lorentzian distance , the sub - lorentzian distance satisfies the reverse triangle inequality , and may not be very well behaving . for instance , if there is a timelike loop trough a point @xmath85 , then @xmath99 . a curve @xmath88 \rightarrow \mathbb{r}$ ] is called a length maximizer , if @xmath100 . similarly , a curve @xmath47 is called a relative maximizer with respect to an open set @xmath101 , if @xmath102 ) \subseteq w$ ] and @xmath103 , where the supremum is taken over all horizontal future directed causal curves contained in @xmath101 , connecting @xmath104 and @xmath105 . by using the maximum principle , for @xmath42 bracket generating , we know that all relative maximizers are either normal geodesics or strictly abnormal maximizers @xcite , and that the relative maximizers always exist locally . by normal sub - lorentzian geodesics @xmath47 we mean curves , such that for any local orthonormal basis @xmath70 of @xmath42 , with @xmath106 as the time - orientation , @xmath47 is an integral curve of the hamiltonian system generated by a hamiltonian function @xmath107 . the question of whether length maximizers exist between two points , is a much more complicated than the question of the existence of length minimizers in riemannian or sub - riemannian geometries . the most common sufficient condition for the global existence of maximizing curves on a lorentzian or sub - lorentzian manifold @xmath37 is a relatively strict requirement that @xmath37 should be globally hyperbolic , i.e. , strongly causal ( every point has an arbitrarily small neighborhood , such that causal curves that leave the neighborhood never return back ) , and @xmath108 is compact for any @xmath109 . determination of curves whose length is equal to the distance , either in the sub - riemannian or sub - lorentzian setting , can be formulated as a solution to an optimal control problem . let @xmath37 be an @xmath36-dimensional manifold , let a submersion @xmath110 be a fiber bundle with the fiber @xmath111 , @xmath58 , and let @xmath112 be a fiber preserving map . an _ admissible pair _ @xmath113 is a bounded measurable mapping @xmath114 \rightarrow v , 0 < \tau < \infty$ ] such that @xmath115 is a lipschitzian curve on @xmath37 and @xmath116 . for some fixed value of @xmath117 , we denote the space of all admissible pairs by @xmath118 . the space @xmath118 is a smooth banach submanifold of @xmath119,v)$ ] , which is modeled on @xmath120,\mathbb{r}^{\dim v})$ ] . an admissible pair is said to connect @xmath45 , if @xmath121 , @xmath122 . such a system is called _ controllable _ , if any two points in @xmath37 are connected by at least one admissible pair . let @xmath123 be a functional , given by @xmath124 where @xmath20 is some differentiable function . suppose we are given two points @xmath125 , such that there is at least one admissible pair connecting them . an _ optimal control problem _ with respect to the functional @xmath126 , is a problem of finding an element @xmath127 connecting @xmath128 , such that for any @xmath129 connecting the same two points , we have @xmath130 . analogously , we may look for the maximum of @xmath126 . we call @xmath131 the optimal control , and @xmath132 the optimal trajectory . we may also consider _ free - time _ optimal control problems where @xmath117 is allowed to vary . the main tool to solve optimal control problems is the following first order condition known as the _ pontryagin maximum principle _ ( pmp ) . for a given value of @xmath117 , let @xmath133 be an optimal pair for the above problem , i.e. , @xmath134 . let us define a pseudo - hamiltonian function as @xmath135 then there exists a curve @xmath136 \rightarrow t^*m$ ] , and a number @xmath137 , such that @xmath138,\ ] ] and @xmath139 moreover , if @xmath140 , then @xmath141 does not vanish almost everywhere in @xmath142 $ ] . we will write @xmath143 to denote the hamiltonian vector field associated to a hamiltonian function @xmath144 . for a pseudo - hamiltonian function , @xmath145 is defined so that for a fixed @xmath146 , @xmath147 is the hamiltonian vector field associated with the hamiltonian @xmath148 . for the problem of the maximum of @xmath126 , the above theorem has the same formulation , changing only the requirement @xmath137 to @xmath149 . if we consider a free - time problem , then we also require @xmath150 . if @xmath151 , then the solution is called normal ( in this case we may just choose @xmath152 ) . if @xmath140 , the solution is called abnormal . if @xmath153 is defined and is @xmath154 on @xmath155 , where @xmath156 is the zero - section , then @xmath157 . let us turn to specific sub - riemannian and sub - lorentzian settings . we only need the formulation for the case when @xmath158 is a lie group . this case is somewhat simpler since the tangent bundle of lie groups is trivial , so we can always find a global basis to span a distribution . let @xmath159 be a distribution , and @xmath38 be either a sub - riemannian or sub - lorentzian metric on @xmath42 . we may assume that the collection @xmath160 forms an orthonormal basis . let @xmath161 , where @xmath162 . we choose @xmath163 , @xmath164 , to ensure that @xmath165 is a horizontal curve . the determination of the sub - riemannian distance , comes down to finding an optimal pair @xmath166 \rightarrow u \times g$ ] that minimizes the functional @xmath167 the corresponding pseudo - hamiltonian is given by @xmath168 as a consequence of fillipov s theorem @xcite , there always exists a length minimizing horizontal curve between two points in this setting . similarly , finding the sub - lorentzian distance with @xmath106 as a time - orientation , means finding a pair @xmath169 that maximizes @xmath170 and the pseudo - hamiltonian related to this problem becomes @xmath171 the latter case is more complicated because we optimize a non - convex functional . in our example of @xmath0 , we only consider left - invariant distributions , that lead to left - invariant hamiltonians . let us , therefore , include some sketch of the theory of hamiltonian systems in this case . generally , assume that @xmath144 is a left - invariant hamiltonian function on a lie group ( i.e. , @xmath172 ) . from the isomorphisms of bundles , we have @xmath173 @xmath174 the traditional hamiltonian equation @xmath175 , which locally has the form @xmath176 may be written as @xmath177 @xmath178 if @xmath144 is left - invariant , and hence is independent of @xmath179 , then the first term in the equation for @xmath180 vanishes . in particular , if @xmath5 is a left - invariant vector field , then @xmath181 , so @xmath182 . furthermore , @xmath183 \right ) \\ & = & p\left(\left [ \frac{dh}{dp}(p ) , \frac{dh_j}{dp}(p ) \right ] \right ) = \{h_j , h \}(p).\end{aligned}\ ] ] here @xmath184 denotes the poisson brackets . for more details , see @xcite ( observe that the difference in sign in our formulation and in @xcite happens because of different definitions of @xmath185 $ ] . ) notice that similar considerations can be done by interchanging the left and the right actions . we will start by defining the following functions , which will simplify our notations later . first , let @xmath186 be defined by @xmath187 furthermore , let @xmath188 where @xmath189 . we remark that @xmath190 @xmath191 also @xmath192 observe that @xmath193 does not exist , because @xmath194 when @xmath195 , we will write @xmath196 and @xmath197 . [ chi ] define @xmath198 , as a unique number satisfying @xmath199 . then , @xmath200 , implies that @xmath201 , while @xmath202 implies @xmath203 . these numbers will become important later . finally , we define @xmath204 in the following way . if @xmath205 , then @xmath206 , that is @xmath207 when @xmath208 , and @xmath209 when @xmath210 . furthermore , if @xmath211 , then @xmath212 here , @xmath213 denotes the ceiling of @xmath214 ( i.e. , @xmath215 ) and @xmath216 we consider the lie group @xmath0 of unitary complex @xmath217 matrices of the form @xmath218 with the operation of usual matrix multiplication . simplifying notation , we will sometimes view @xmath219 as an element from @xmath220 , writing @xmath221 . in this notation , the group operation is @xmath222 the identity is @xmath223 , and the inverse element becomes @xmath224 . @xmath0 is isomorphic to the lie group @xmath29 of real @xmath225 matrices of determinant 1 , by the mapping @xmath226 we will often use the isomorphism from the tangent bundle @xmath227 to a subspace of the complex tangent bundle @xmath228 , given by @xmath229 where @xmath230 , @xmath231 and @xmath232 . let @xmath233 denote the left actions by @xmath179 . the tangent map of the left action is given by @xmath234 @xmath235 the tangent space at the identity is spanned by the vectors @xmath236 viewed as elements of the lie algebra @xmath32 of @xmath0 , they have the following form @xmath237 analogously to the pauli matrices in relation to the lie algebra @xmath238 of the lie group @xmath17 . so we can obtain the left - invariant basis for the tangent bundle as @xmath239 the dual basis for the cotangent bundle is @xmath240 @xmath241 @xmath242 the bracket relations yield @xmath243 = - 2 z , \qquad [ x , z ] = - 2 y , \qquad [ y , z ] = 2 x.\ ] ] it follows that any distribution spanned by two of three vector fields is bracket generating . the exponential map for this lie group is @xmath244 @xmath245 we define the metric @xmath38 as a left - invariant metric , whose restriction to the @xmath32 has the following form @xmath246 remark that in @xmath32 , the killing form is equal to @xmath247 in the basis of @xmath248 , the metric tensor of @xmath38 has the form @xmath249 the metric @xmath38 is then lorentzian , and in fact , bi - invariant . hence , as a lorentzian manifold , @xmath0 may be considered as a subset of @xmath250 , which is @xmath251 with an index 2 pseudo - metric . this subset @xmath252 is called the 3-dimensional anti - de sitter space . the restriction of @xmath38 to the distribution @xmath253 , makes it positive definite , and makes @xmath254 a sub - riemannian manifold . similarly , the restriction @xmath38 to @xmath255 makes @xmath256 a sub - lorentzian manifold , if we define the time - orientation by @xmath257 . the latter case however , contains timelike loops . an example is @xmath258 which is a loop through 1 . to avoid this problem we will also study the universal cover @xmath1 of @xmath0 this will also be helpful in order to study sub - riemannian geometry on @xmath0 . in addition , it is an interesting example on its own . in fact , sometimes the attribution anti - de sitter space is used for the universal cover instead of @xmath33 itself . since @xmath0 is diffeomorphic to @xmath259 , the universal cover must be diffeomorphic to @xmath260 . we represent the covering space @xmath1 as @xmath261 , with the covering map @xmath262 @xmath263 where @xmath264 and @xmath265 . we define the product on @xmath1 , to be the unique product for which @xmath266 is the identity , and which makes @xmath267 a group homomorphism . it is obvious that the lie algebra of @xmath1 is also @xmath32 . [ groupop ] let @xmath268 . then @xmath269 , where @xmath270 @xmath271 the fact that @xmath272 , trivially implies the expression for @xmath273 . the value of @xmath274 must satisfy the relation @xmath275 @xmath276 now since @xmath277 we know that @xmath278 and the formula for @xmath274 follows . it is clear that @xmath279 is the identity under this product . observe that @xmath280 . the associativity of the product remains to be proved . due to the associativity of the product on @xmath0 , we only need to show that if @xmath281 then @xmath282 . let @xmath283 , and denote @xmath284 @xmath285 then @xmath286 similarly to the reasoning above , @xmath287 so since @xmath288 , we know that @xmath289 , and it follows that @xmath290 . the left action is given by ( @xmath291 ) @xmath292 @xmath293 @xmath294 the lifted vector fields are @xmath295 @xmath296 @xmath297 the exponential map in @xmath1 is @xmath298 which can be found by lifting the exponential map from @xmath0 . we lift the metric @xmath38 to a metric on @xmath1 . denote this metric by @xmath299 . analogously to @xmath0 , let us define the distributions @xmath300 and @xmath301 . the restriction of @xmath299 to @xmath302 and @xmath303 , defines sub - riemannian and sub - lorentzian structures on @xmath1 respectively . in the sub - lorentzian case , we let @xmath304 define the time orientation . we can construct a sub - lorentzian manifold , by considering the distribution @xmath305 , but the geodesics are very similar to the sub - lorentzian manifold @xmath306 so we omit this choice of distribution . we will now take advantage of the fact that the pseudo - metric induced by the killing form is bi - invariant . [ thmon ] let @xmath2 be a lie group with the lie algebra @xmath3 , and with a bi - invariant pseudo - metric @xmath38 . let @xmath20 be a subgroup of @xmath2 , with the lie algebra @xmath7 , and let us denote @xmath307 . define a left - invariant distribution @xmath42 by @xmath308 . then all normal geodesics on the non - holonomic manifold @xmath309 are lifting of the normal geodesics on @xmath24 with the induced metric . this means that all normal geodesics starting at @xmath26 are of the form @xmath310 where @xmath311 is the projection . this theorem is a special case of the corresponding result from @xcite . we may use theorem [ thmon ] with @xmath312 . the normal geodesics starting at the identity admit the form @xmath313 in @xmath0 and @xmath314 in @xmath315 any other normal geodesic starting at some point , is a left translation of a normal one starting at the identity . we define _ the vertical space _ in a sub - riemannian manifold @xmath316 with respect to a point @xmath62 , as the set of all points in @xmath37 that can be reached from @xmath62 by curves tangent to the orthogonal complement of @xmath42 . we define _ the horizontal space _ with respect to @xmath62 , to be the set of all point that can be reached by horizontal curves of constant speed ( that is , @xmath47 is a horizontal curve , such that if any horizontal vector field @xmath5 is of unit length , then @xmath317 is constant ) . the vertical space ( or vertical line , since it is one dimensional ) in @xmath0 with respect to 1 is @xmath318 and the horizontal space is @xmath319 similarly , in @xmath1 the vertical line is defined by @xmath320 ( and from proposition [ groupop ] it follows that @xmath20 is isomorphic to @xmath321 ) , and the horizontal space is defined by @xmath322 . recall that @xmath323 is the unique number satisfying the equality @xmath324 in the interval @xmath325 . [ srnumbergeo ] up to reparameterization , we have the following results regarding the number of geodesics connecting @xmath326 and @xmath327 . * if @xmath328 is in the vertical line , that is @xmath329 for some @xmath274 , then the number of geodesics is uncountable ( countably many geometrically different ) . * if @xmath328 is in the horizontal space , then there exists a unique geodesic connecting @xmath328 and @xmath326 . this geodesic is contained in the horizontal space . * if @xmath328 is any other point , then we can obtain the number of geodesics in the following way . let @xmath64 be the largest positive integer , such that @xmath330 if there exists @xmath64 , such that the above inequality is strict , then there are @xmath331 geodesics connecting these two points . if @xmath64 gives the equality , then the number of geodesics is @xmath332 . if no such @xmath64 exists , then there is a unique geodesic . the number of geodesics in the case ( c ) above may be difficult to determine , but due to the fact that the value of @xmath333 , belongs to the interval @xmath334 , this number remains between @xmath335 an @xmath336 , where @xmath337 the proof is a long case by case analysis , and therefore , we leave it to section [ proofs ] . regarding to @xmath338 , we say that two geodesics are _ geometrically similar _ , if one is the image of another under an isometry . the isometry considered in @xmath338 is @xmath339 , @xmath340 . since @xmath341 is a step 2 regular sub - riemannian manifold , there are no abnormal length minimizers . every geodesic may be extended indefinitely , therefore the carnot - carathodory metric is complete ( see @xcite and @xcite ) . hence , every point is connected to @xmath326 by a length minimizing geodesic . the same holds for @xmath0 . the above information along with the proof of proposition [ srnumbergeo ] , leads to the following result . [ corlengthsr ] * if @xmath342 , then @xmath343 . * if @xmath344 , then @xmath345 . * if @xmath346 is neither in the vertical , nor in the horizontal space , and @xmath347 , then @xmath348 where @xmath349 is a unique number satisfying @xmath350 * if @xmath351 , then @xmath352 * if @xmath346 satisfies the inequality @xmath353 , then @xmath354 where @xmath355 is a unique number satisfying @xmath356 * if @xmath346 satisfies the inequality @xmath357 , then @xmath358 where @xmath355 is a unique number satisfying @xmath359 the details are again left to section [ proofs ] . notice that in all cases , the distance is independent of the sign of @xmath274 , and of the argument of @xmath273 . from these results for the universal cover , we easily obtain some conclusions for @xmath0 . the number of geodesics connecting 1 and @xmath219 is * uncountable ( there are countably many geometrically different geodesics ) , if @xmath360 . * countable otherwise . the next result we prove here . [ ngeosu ] if @xmath361 , then @xmath362 . since both @xmath0 and its cover are step 2 regular , it follows that @xmath363 let @xmath364 and let @xmath365 , @xmath366 . we need to show that @xmath367 , for @xmath368 . when @xmath369 , this is a trivial consequence of corollary [ corlengthsr ] , case ( a ) . if @xmath370 , we start with the case ( f ) @xmath371 . then @xmath372 let us consider the case ( e ) with @xmath373 . the value @xmath374 is less than @xmath375 or @xmath376 ( the lower bounds for @xmath377 are defined as follows : the first for @xmath378 , the second for @xmath379 ) ) . if @xmath380 , then @xmath381 if @xmath382 , then we have the inequality @xmath383 which follows from the fact that @xmath384 is increasing . so we only need to look at the case when @xmath385 . we define @xmath386 to be a number such that @xmath387 we argue from the contrary . assume that @xmath388 . then from the fact that @xmath389 is increasing , it turns out that @xmath390 . but then @xmath391 which is a contradiction . for sub - riemannian lie groups satisfying theorem [ thmon ] , we define a sub - riemannian analogue of the exponential map about the identity @xmath392 by @xmath393 for a more general definition of the exponential map in the sub - riemannian setting , see @xcite . we define _ the conjugate locus _ of @xmath2 from the identity 1 , as the set of critical values of @xmath394 . we often split the conjugate locus in several sets , defining the @xmath36-th conjugate locus by the set @xmath395 , where @xmath396 , and t is so that there exist exactly @xmath36 values @xmath397 , so that @xmath398 are all critical points . we define _ the cut locus _ from 1 as the set reachable by more than one minimizing geodesic . the cut locus for @xmath1 is the vertical line . the cut locus of @xmath0 consists of the points where either @xmath399 , or @xmath400 , and the points are different from the identity . the cut locus for @xmath1 follows from the proof of proposition [ srnumbergeo ] and corollary [ corlengthsr ] . for @xmath0 , if a point @xmath401 is in the cut locus , then either @xmath402 is in the cut locus for @xmath1 ( which is the set of points @xmath403 for arbitrary @xmath404 ) or there exist more than one @xmath327 , such that @xmath405 and @xmath406 . this only happens when @xmath407 . in this case there are exactly two points @xmath328 . the following proposition was proved for @xmath0 in @xcite ( for the isometric case of @xmath29 ) , but here we generalize it , including the universal cover @xmath1 . the @xmath36-th conjugate locus of @xmath408 consists of the vertical line , if @xmath36 is odd . if @xmath409 , then it consists of the points given by the equation @xmath410 first , observe that @xmath411 exists only for @xmath412 . put @xmath413 . then we have @xmath414 @xmath415 the values of the elements of the matrix @xmath416 become @xmath417 here , we have simplified by writing just @xmath418 and @xmath419 for @xmath420 and @xmath421 . the determinant of the above matrix is @xmath422 . the value of @xmath418 vanishes at the points , for which @xmath423 , and the image of such points is the vertical line ( see the proof for proposition [ srnumbergeo ] for more details ) . moreover , @xmath424 vanishes only for @xmath425 . hence , for a generic @xmath396 , the point @xmath426 can be singular for @xmath394 , only if @xmath427 . let us use the normalization @xmath428 . let the @xmath36-th value @xmath429 be such that @xmath430 , is a singular point . then it is clear that @xmath431 let @xmath432 be a point that does not belong to the vertical line ( i.e. @xmath433 ) . notice that if the value of @xmath434 from proposition [ srnumbergeo ] increases ( or the value of @xmath435 decreases ) , then ( and only then ) the number of geodesics increases when we pass through the even - indexed conjugate loci . the @xmath36-th conjugate locus of @xmath436 for odd @xmath36 is the vertical line , and if @xmath409 , then it consist of the points @xmath437 in contrast with the sub - riemannian case , we only know that the relative maximizers exist locally . we have no guarantee of global existence of maximizers . let us consider the distribution @xmath438 , with the metric @xmath299 restricted to @xmath303 . we formulate an optimal control problem of maximizing @xmath439 where @xmath440 . we only do the case when @xmath441 . if @xmath140 one gets the same results , so it follows that there are no strictly abnormal geodesics . in order to use pmp , let us define the pseudo - hamiltonian @xmath442 @xmath443 where @xmath444 . for the existence of @xmath445 , we need that @xmath446 and that @xmath447 . in this case @xmath448 this is from the fact that the optimal control is @xmath449 since @xmath144 can not be extended to @xmath450 ( the tangent bundle with the zero - section removed ) , we solve the equation using the pseudo - hamiltonian function @xmath451 instead , @xmath452 @xmath453 @xmath454 @xmath455 is a first integral in this case , and from the condition @xmath456 , @xmath455 is equal to 1 . we find the solutions ( denote @xmath457 ) @xmath458 @xmath459 @xmath460 observe that @xmath461 , and @xmath462 . note also that @xmath463 has to be equal to @xmath464 in order to simplify our calculation , we first solve it for @xmath0 , and then lift it to @xmath1 . we have to solve the following differential equation . @xmath465 make the following observation that @xmath466 from this we notice that @xmath467 @xmath468 @xmath469 if we expand equation , we can write @xmath470 @xmath471 and multiplying from the right by @xmath472 and adding @xmath473 on both sides , we obtain @xmath474 it follows that @xmath475 i.e. , @xmath476 if we lift this curve and consider theorem [ thmon ] , then we easily get to the following proposition . assume that there exists a global length maximizing curve between @xmath326 and @xmath328 . then , this curve is a timelike , future - directed normal geodesic on the form @xmath477 where @xmath478 and @xmath479 . explicitly , these geodesics have the form @xmath480 @xmath481 the projection of them to @xmath0 is given by @xmath482 @xmath483 we will discuss the situation when these curves correspond to length maximizers in subsection [ future ] . the result in the sub - lorentzian case is more complicated , than in the sub - riemannian case . so we have to give some definitions in order to describe geodesics in a reasonable way . first , let us define a function @xmath484 , by @xmath485 also , we define the numbers @xmath486 as the numbers satisfying the equations @xmath487 finally , we define the function @xmath488 let us construct the following subsets of @xmath1 : @xmath489 consist of all points @xmath490 where @xmath491 , @xmath492 , and @xmath493\bigcup [ \sqrt{1-s^2 } , \ , 1 - \frac{s^2}{2})$ ] when @xmath494 , and @xmath495 otherwise . further , for @xmath366 , define @xmath496 as a set of all points @xmath497 where @xmath491 , @xmath492 , and @xmath498 for @xmath494 , and @xmath499 $ ] otherwise . we split them into the following sets * @xmath500 consist of all points @xmath497 where @xmath491 , @xmath492 , and * * @xmath501 , when @xmath502 , * * @xmath503 \cup ( f(s,\omega_k ) , \infty)$ ] , when @xmath504 , * * @xmath505 \cup \{f(1,\omega_k)\}$ ] , when @xmath506 , * * @xmath507 , when @xmath508 . * @xmath509 consist of all points @xmath497 where @xmath491 , @xmath492 , and * * @xmath510 , when @xmath511 , * * @xmath512 , when @xmath506 , * * @xmath513 , when @xmath514 . * @xmath515 consist of all points @xmath497 where @xmath491 , @xmath504 , and @xmath516 . let @xmath517 . finally , define @xmath518 @xmath519 [ maxgeo ] there are timelike , future directed geodesics connecting @xmath326 and @xmath328 , only if @xmath520 . more precisely , * the geodesic is unique if @xmath521 ; * there are two geodesics if @xmath522 ; * there are three geodesic if @xmath523 ; * there are countable many geodesics if @xmath524 . this case by case analysis is left to section [ proofs ] . let @xmath323 be the constants defined in subsection [ notationsec ] . then @xmath525 , so these numbers play a role here as well . however , unlike what was the case when we considered the sub - riemannian geodesics , the mapping @xmath526 does not have any critical points satisfying @xmath527 . we also remark the following interesting comparison with lorentzian geometry . [ prop6 ] the set reachable by lorentzian geodesics starting at @xmath326 , is the planes @xmath528 , along with all points satisfying @xmath529 , @xmath530 , and @xmath531 . more precisely , * there are uncountably many geodesics if @xmath320 ( all geometrically similar , except for the case @xmath532 , when they are uncountably many geometrically different ) . * there is a unique geodesic if @xmath433 . a time - like geodesic @xmath533 , starting at @xmath326 satisfies @xmath534 , and we know that @xmath478 and @xmath535 ( and hence @xmath427 ) . we rewrite the geodesics in terms of the parameters @xmath536 and @xmath537 , where @xmath538 $ ] and @xmath539 . the geodesics in these coordinates look like @xmath540 now @xmath537 can be determined by @xmath541 , and @xmath542 . rewriting the equation for @xmath435 as @xmath543 and inserting it in the equation for @xmath274 , we get that @xmath544 for @xmath545 . for the remaining cases , @xmath546 for @xmath547 , and @xmath548 for @xmath549 . it is obvious that there are solutions only if @xmath550 . the equality can never be attained , because @xmath551 , which follows from @xmath552 the equation @xmath553 determines @xmath554 for @xmath555 . when @xmath556 , there are no restrictions on @xmath554 . also , if @xmath369 , there are no restrictions on @xmath537 . proposition [ prop6 ] yields that the set of points reachable by the lorentzian geodesics starting at @xmath326 neither is contained , nor contains the set reachable from @xmath326 by the sub - lorentzian geodesics . this contrasts the fact that the lorentzian timelike future always contains the sub - lorentzian one . using the information we have collected , concerning sets reachable by geodesics , we obtain some results for the timelike future in general . * the timelike future of @xmath1 with respect to the lorentzian metric @xmath299 , is given by latexmath:[\[\mathcal{i}^+(\tilde 1 , \tilde \rho ) = \left\{(c , w ) \in \operatorname{csu}(1,1)| \ , c < - \tan^{-1 } * the timelike future of @xmath1 with respect to the sub - lorentzian metric @xmath558 , is given by @xmath559 let us denote by @xmath560 , and @xmath561 in both sub - lorentzian and lorentzian settings , we show by using the information about the geodesics , that @xmath562 and @xmath563 are included in their respective timelike futures . to show the opposite inclusion , remark that in both lorentzian and sub - lorentzian settings , we use the fact that there exist relative length maximizers locally . so there exists a neighborhood @xmath564 of @xmath326 , such that @xmath565 define @xmath566 . now from left invariance , we know that @xmath567 , and it follows that @xmath568 hence it is sufficient to show that @xmath569 , for every @xmath570 . since every timelike curve from @xmath326 to a point outside @xmath562 has to pass through @xmath571 , it follows by continuity that it is sufficient to show that @xmath572 , for every @xmath573 . here , @xmath574 means the closure of @xmath66 . finally , since @xmath575 is an isomorphism , we have that @xmath576 , so all remaining arguments turn down to show that @xmath577 for every @xmath578 . the same holds for @xmath579 . in order to prove ( a ) we show that there are timelike geodesics connecting every point satisfying @xmath580 with the origin let @xmath581 be an arbitrary point satisfying @xmath582 . then we can construct a timelike curve from @xmath326 , by taking a geodesic from @xmath326 to @xmath583 , and then continue by a left translation of a geodesic from @xmath326 to @xmath584 . let @xmath578 , @xmath585 . first , if @xmath586 and @xmath587 , then @xmath588 @xmath589 @xmath590 so in this case @xmath591 . here @xmath592 . we now turn to the case when either @xmath593 or @xmath554 is less than 1 . if we denote @xmath594 then we have @xmath595 @xmath596 @xmath597 @xmath598 since @xmath599 or @xmath600 , we conclude that @xmath601 . it follows that @xmath602 , because @xmath603 . as a consequence , @xmath604 . turning to ( b ) , let us observe that @xmath489 consist of all points , satisfying @xmath605 clearly all the points in @xmath489 are in the timelike future . pick up any point @xmath581 satisfying @xmath606 . let @xmath47 be a left translation by @xmath607 of any geodesic connecting @xmath326 and @xmath608 . the endpoint of @xmath47 is @xmath581 . since @xmath609 , the curve @xmath47 must at some point intersect @xmath489 . pick up any point @xmath610 , travel from @xmath326 to @xmath611 along a geodesic , and continue along @xmath47 to @xmath581 . let @xmath612 be of the form @xmath613 assume first that @xmath614 . if we again denote @xmath615 , we then obtain @xmath616 @xmath617 we only need to show that @xmath618 . denote @xmath619 . then @xmath620 @xmath621 since @xmath622 . if @xmath623 , the equality @xmath624 @xmath625 where @xmath626 , and @xmath627 tells us that @xmath328 may be written as a finite product of boundary elements , all with real part of the second coordinate of absolute value less than or equal to @xmath628 . since we know that locally , the lorentzian or sub - lorentzian causal future is the closure of the timelike future ( see @xcite ) , it follows from left invariance that @xmath629 and @xmath630 are the closure of @xmath631 and @xmath632 , respectively . both with respect to the lorentzian @xmath299 and the sub - lorentzian metric @xmath633 , @xmath1 is strongly causal . the proof is the same for both the lorentzian and sub - lorentzian cases . by left invariance , it is sufficient to find a strongly causal neighborhood . let @xmath634 be the set of all points @xmath581 in the timelike future of @xmath326 satisfying @xmath635 . observe that inequality @xmath636 must hold at the same time . since , the causal future of all the elements in @xmath637 is contained in the causal future of @xmath326 , any causal curve must exit through the surface @xmath638 . then we need to show that for any such element @xmath639 . observe that if @xmath640 and @xmath641 , then @xmath642 @xmath643 @xmath644 hence , @xmath637 is strongly causal neighborhood of its elements . the distance between @xmath326 and @xmath581 with respect to the lorentzian metric @xmath299 is equal to @xmath645 when @xmath646 , we know that @xmath647 . the identity for @xmath648 is trivial . for @xmath649 , the formula follows from the fact that this space is globally hyperbolic , and from the proof of proposition [ prop6 ] . the lower bound for the distance when @xmath646 follows from the reverse triangle inequality , and from the fact that every such element has points in its timelike past of distance arbitrarily close to @xmath267 . to prove that @xmath650 , we need to show that for every pair @xmath651 in this subset , @xmath652 is compact . it is trivial that @xmath653 , and by the same reasoning @xmath654 . for some @xmath655 . clearly , @xmath656 it follows that @xmath657 , is compact for @xmath658 , but not when @xmath659 . one would expect an analogous statement for the sub - lorentzian metric presenting a description of the distance function in @xmath489 , given the proof of proposition [ maxgeo ] , but it is more difficult to prove whether it is globally hyperbolic or not . the technique of this proof is to consider a general geodesic satisfying @xmath660 and @xmath661 . each geodesic is determined by their initial conditions @xmath662 , where @xmath663 . our task will be , given the final point of the geodesic , @xmath581 , to find how many choices of initial conditions do we have . we start by proving a ) . if @xmath320 , then @xmath667 ( and only then ) , so @xmath668 , where @xmath366 . furthermore , @xmath669 , @xmath670 and @xmath671 , so there are countably many choices . also , there are no restrictions to @xmath537 . to obtain the remaining results , let us define @xmath672 . it is easy to see that determines @xmath673 , so we only need to look at the number of solutions to . if @xmath674 ( i.e. @xmath675 ) , then the geodesics are contained in the horizontal space , and there is a unique one for every point ( @xmath676 ) . in order to find a solution , we must solve the equations @xmath680 @xmath681 observe that @xmath682 . we rewrite as @xmath683 and inserting it in , we obtain @xmath684 substituting @xmath685 back into , we have that @xmath686 is a solution to the equation @xmath687 which has a solution only for @xmath688 , and it is unique . to show this , take the derivative of @xmath689 as @xmath690 the limits @xmath691 imply that the necessary condition of the existence of the solution to is vanishing of the derivative @xmath692 at some point . assume first that @xmath693 . then has no zeros for @xmath694 . the derivative vanishes at most once on the interval @xmath695 . in fact , it vanishes exactly once if @xmath696 . therefore , has one solution in @xmath695 . similarly , it can be shown that has exactly one solution , when @xmath697 , and this solution @xmath686 is between -1 and 0 . if @xmath322 , never vanishes in @xmath698 . equations are now written in a more complicated form @xmath716 @xmath717 an immediate observation from , is that @xmath718 . let us introduce some notations . we write @xmath719 \setminus \ { 0 \}$ ] , and also @xmath720 , @xmath721 . if @xmath722 , then @xmath723 . @xmath724 so @xmath725 @xmath726 then @xmath686 is a solutions to the equation @xmath727 where @xmath728 $ ] , and @xmath729 . also @xmath686 can only have the value @xmath730 , when @xmath723 . from now we will just assume that @xmath731 , since the considerations for @xmath732 are totally analogous . from lemma [ boundlemma ] , the left - hand side of decreases or increases from @xmath733 to @xmath734 . the derivative @xmath735 @xmath736 vanishes once if @xmath737 , and otherwise it is negative . if @xmath722 , then yields that a solution can occur only for @xmath738 . observe that @xmath739 if @xmath740 and @xmath741 , then there is a unique root if and only if , @xmath742 . if @xmath741 and @xmath723 , then there is clearly no solutions for @xmath743 , and from , it follows that there is only one solution for @xmath744 . otherwise , let @xmath745 be a root of the expression @xmath746 then , there is a solution only if @xmath747 if the equality is attained in the above inequality , then @xmath748 , i.e. , @xmath749 . hence , there is only one solution when @xmath750 and there are two solutions when @xmath434 is between this value and @xmath751 . all the cases has now been examined , so this ends the proof . let us use the same notation as in the proof of proposition [ srnumbergeo ] . if @xmath533 is a geodesic that goes from @xmath326 to @xmath752 in a unit time interval , then @xmath753 . this implies ( a)(e ) rather easily from the previous proof . to prove ( f ) , let @xmath754 be a geodesic corresponding to a choice of @xmath64 and @xmath755 , with @xmath756 . the length becomes @xmath757 then , from the equation @xmath758 , we know that @xmath759 so @xmath760 @xmath761 it follows that @xmath762 is minimal . since @xmath763 and @xmath479 , let us denote @xmath764 , and @xmath765 . then @xmath766 . similarly to the sub - riemannian geodesics , we consider @xmath767 , and look at how the final point defines the initial conditions . we are going to use the projection of @xmath581 to @xmath0 . along with the coordinates @xmath768 and @xmath769 , we shall use new coordinates given by @xmath770 observe that @xmath771 . then the projection of the geodesics to these coordinates becomes @xmath772 the relationship between the projected coordinates and the original ones on the coving space is @xmath773 @xmath774 furthermore , @xmath775 @xmath776 @xmath777 @xmath778 first , let us consider the case when @xmath779 . the only points that can be reached by this type of geodesics are @xmath780 . then @xmath781 and there are no restrictions on @xmath537 . otherwise , if @xmath782 , then @xmath783 . it follow that @xmath784 when @xmath785 , we find @xmath554 as @xmath786 , where @xmath787 . so we only need to find the number of solutions to the equation with respect to @xmath788 , and then , the values of @xmath554 and @xmath537 will be determined by . let us analyze a special case first . there is exactly one geodesic , satisfying @xmath785 , when @xmath581 is of the form @xmath789 where @xmath790 \setminus \ { 0 \}$ ] or @xmath791 $ ] if @xmath792 . we continue the case by case analysis of the remaining possibilities . from , we know that @xmath793 a quick look at and yields that @xmath795 and @xmath796 . also from , we know that at least one of @xmath797 and @xmath798 is positive . we define @xmath799 . then , @xmath800 here , we denote @xmath801 notice , that the first part of the above equation is not valid if @xmath802 , and the second part is not valid if @xmath803 . the relation implies @xmath804 . if @xmath805 , then we know that @xmath806 . furthermore , from , we obtain that @xmath807 is a solution to the equation @xmath808 and @xmath809 increases from @xmath810 to 1 . so , a solution exists only if @xmath811 . then the points , which are connected by these types of geodesics are @xmath812 where @xmath813 . to get some idea of the meaning the above equations , notice that @xmath814 , so @xmath815 is an increasing function of @xmath798 when @xmath514 . also , @xmath816 is a decreasing function of @xmath798 when @xmath494 . using this information , we know that there are only geodesic of these types connecting the points of the form @xmath817 with the identity of the group , where @xmath818 and @xmath819 satisfy one of the following conditions when @xmath828 , we know that @xmath829 , and @xmath807 is a solution to the equation @xmath830 . so the points reachable by these types of geodesics are @xmath831 where @xmath832 , and @xmath819 satisfy one of the following conditions we introduce the notation @xmath843 \setminus \{0 \}$ ] . let @xmath721 . notice that @xmath844 , and hence , @xmath845 . now define @xmath846 so that @xmath847 . from the fact that @xmath848 , we know that @xmath849 , which yields that @xmath850 now write @xmath851 . then , @xmath852 , so @xmath853 given a value of @xmath64 , the value of @xmath274 is completely defined by @xmath854 . we also know that if two geodesics has the same endpoint , then the maximal difference in their value of @xmath64 is 1 . we need to find @xmath788 as a solution to @xmath855 let us denote @xmath856 . it follows that @xmath857 as in the previous cases , there are some obvious restrictions on the values of @xmath858 or @xmath859 . here we used that @xmath860 because of . from , we know that @xmath861 . if @xmath805 , then @xmath862 , and so @xmath807 is a solution to @xmath863 where @xmath864 . again we use the notation @xmath865 . the upper bound for @xmath807 is strict , unless @xmath866 and @xmath723 , because @xmath867 can be @xmath707 , but not @xmath868 . moreover , from , @xmath869 implies that either @xmath785 or @xmath870 , the case which already excluded . similarly , when @xmath828 , so @xmath807 is a solution to @xmath871 . * we start with the special case @xmath722 . then @xmath872 decreases from @xmath873 to @xmath874 for @xmath875 , and from @xmath876 to @xmath877 for @xmath878 . * if @xmath879 , then @xmath880 decreases from @xmath35 to @xmath881 for @xmath878 , or to @xmath882 otherwise . * if @xmath723 , then @xmath880 increases from @xmath883 to @xmath881 for @xmath884 . continuing with @xmath885 , if @xmath886 , then @xmath880 increases from @xmath883 to @xmath887 , and decreases to @xmath881 for @xmath878 , or to @xmath888 otherwise . further , if @xmath805 , then @xmath862 , and @xmath889 , and the same holds when @xmath828 , interchanging @xmath797 and @xmath798 . also , from @xmath890 we know that @xmath891 is an increasing function of @xmath797 and @xmath798 , when @xmath892 . when @xmath893 , @xmath840 increases in @xmath797 and decreases in @xmath798 . as a result , we know which point is reachable by which type of geodesics . we only show the results for @xmath492 , since for @xmath894 , the results are the same , only with different sign for @xmath840 . * if @xmath722 , then * * if @xmath822 , then @xmath821 , @xmath895 ; * * if @xmath896 , @xmath897 , @xmath898 . * if @xmath741 and @xmath879 , then * * if @xmath822 , @xmath899 , and @xmath900 ; * * if @xmath875 , @xmath901 , and @xmath902 . * if @xmath741 and @xmath723 , then there is one geodesic ( list 1 ) * * if @xmath903 , @xmath899 , and @xmath904 ; * * if @xmath504 , @xmath899 , and @xmath905 ; * * if @xmath824 and @xmath906 ; * * if @xmath514 , @xmath901 , and @xmath907 + and two geodesics ( list 2 ) * * if @xmath504 , @xmath899 , and @xmath908 * * if @xmath824 , and @xmath909 ; * * if @xmath514 , @xmath901 , and @xmath910 the above list can be rewritten so the interval is determined for @xmath891 only . note that the term `` one geodesic '' and `` two geodesics '' is a bit misleading , since , if we include both signs of @xmath911 , list 1 and list 2 are not disjoint . * @xmath912 consist of all points @xmath913 where @xmath914 , @xmath915 , @xmath916 , when @xmath917 , and @xmath918 , when @xmath514 . they are the points that are reachable by geodesics with @xmath205 , and with @xmath427 , @xmath722 . * @xmath919 , consist of the points @xmath920 where @xmath921 $ ] , and define @xmath922 , by . @xmath923 where @xmath924 . they are the points reachable by geodesics with @xmath785 . * @xmath925 , consist of all points @xmath926 where @xmath914 , @xmath915 and @xmath927 . they are the points reachable by geodesic with a certain choice of @xmath64 , and with @xmath879 . * @xmath928 , consist of all points @xmath929 where @xmath914 , @xmath930 , and * * @xmath931 , when @xmath514 ; * * @xmath932 if @xmath824 ; * * @xmath933 , when @xmath934 ; * * @xmath935 , when @xmath936 . + they are the points from list 1 , for @xmath723 and for a choice of @xmath64 ( except for the points @xmath937 , for @xmath504 , that appear twice and which are therefore included in @xmath938 ) . * @xmath939 , consist of all points @xmath929 where @xmath914 , @xmath940 and * * @xmath941 $ ] when @xmath504 ; * * @xmath942 when @xmath824 ; * * @xmath943 when @xmath944 . + they are the points from list 2 for a certain choice of @xmath64 and with @xmath723 . notice that @xmath945 is on the boundary of @xmath912 , @xmath946 is on the boundary of @xmath947 and @xmath948 is on the boundary of @xmath949 for @xmath950 . the result follows merely by comparing the sets and counting number of the sets in which a value of @xmath891 appears . points in @xmath938 are counted twice . we use @xmath912 , @xmath945 , @xmath951 , @xmath952 , and @xmath953 to define @xmath489 , and use @xmath954 , @xmath955 , @xmath956 , @xmath957 , and @xmath938 to define @xmath500 , @xmath958 , and @xmath515 .

we investigate a microscopic model for strongly correlated electrons with both on - site and nearest neighbor coulomb repulsion on a 2d square lattice . this exhibits a state in which electrons undergo a `` somersault '' in their internal spin - space ( spin - flux ) as they traverse a closed loop in external coordinate space . when this spin-1/2 antiferromagnetic ( afm ) insulator is doped , the ground state is a liquid of charged , bosonic meron - vortices , which for topological reasons are created in vortex - antivortex pairs . the magnetic exchange energy of the distorted afm background leads to a logarithmic vortex - antivortex attraction which overcomes the direct coulomb repulsion between holes localized on the vortex cores . this leads to the appearance of pre - formed charged pairs . we use the configuration interaction ( ci ) method to study the quantum translational and rotational motion of various charged magnetic solitons and soliton pairs . the ci method systematically describes fluctuation and quantum tunneling corrections to the hartree - fock approximation ( hfa ) . we find that the lowest energy charged meron - antimeron pairs exhibit d - wave rotational symmetry , consistent with the symmetry of the cuprate superconducting order parameter . for a single hole in the 2d afm plane , we find a precursor to spin - charge separation in which a conventional charged spin - polaron dissociates into a singly charged meron - antimeron pair . this model provides a unified microscopic basis for ( i ) non - fermi - liquid transport properties , ( ii ) d - wave preformed charged carrier pairs , ( iii ) mid - infrared optical absorption , ( iv ) destruction of afm long range order with doping and other magnetic properties , and ( v ) certain aspects of angled resolved photo - emission spectroscopy ( arpes ) .

cond-mat0001235

the microscopic understanding of the effect of charge carrier doping on spin@xmath0 antiferromagnetic ( afm ) mott insulators is the central issue of the high - temperature superconducting cuprates . @xcite many puzzling experimental features of these systems @xcite suggest that a fundamental law of nature remains to be recognized . extremely low doping ( @xmath1 charge carriers per site ) leads to a complete destruction of the long - range afm order , and a transition to an unusual non - fermi - liquid metal . this unusual metal becomes superconducting , with the transition temperature @xmath2 strongly dependent on the doping @xmath3 . the maximum @xmath2 is reached for dopings around @xmath4 . for higher dopings the critical temperature decreases to zero , and in the overdoped region a crossover towards a ( non - superconducting ) fermi - liquid takes place . two central questions require resolution . the first one concerns the nature of the charge carriers responsible for this non - fermi - liquid metallic behavior . this is a fundamental issue , since it lies outside the framework of landau s fermi - liquid theory and it necessitates understanding the appearance of non - quasiparticle - like charge carriers in a system of interacting electrons . the second question concerns the nature of strong attractive pairing between these charge carriers , given the purely repulsive interaction between the constituent electrons . in conventional superconductors , the pairing attraction is due to overscreening of the electron - electron coulomb repulsion by the ionic lattice . in the case of the high - temperature superconducting cuprates , it has been suggested @xcite that pairing is an intrinsic property of the electron gas itself mediated by afm spin - fluctuations of the doped system . accordingly , the challenge is to identify a strong attractive force based purely on repulsive coulomb interactions . in this paper , we derive such a force and demonstrate that it leads to d - wave pairing of charge carrying holes introduced by doping a quantum , spin-@xmath5 , mott - hubbard antiferromagnet . the simplest model hamiltonians used to investigate the cuprate physics are the hubbard model and the closely related t - j model . unlike the 1d problem , an exact solution for the 2d hubbard hamiltonian is not known . as a result , various approximations are necessary . although the application of the mean - field theory has been severely criticized in this context , it provides a valuable reference point for incorporating fluctuation effects . moreover , even for the 1d hubbard model , essential features of the exact solution may be recaptured by judiciously incorporating fluctuation and tunneling effects into mean - field theory . @xcite the most straightforward mean - field theory is the hartree - fock approximation ( hfa ) . at half - filling ( @xmath6 ) the hfa predicts an afm mott insulator ground - state . as the system is doped , hfa suggests that charge carrier holes in the afm background assemble in charged stripes , which are quasi - one - dimensional structures . @xcite a large effort has been devoted to studying these charged stripes and relating them to certain features of the cuprates . @xcite recently , a more fundamental investigation of the many - electron problem has suggested the possibility of an alternative model hamiltonian for the cuprate physics . this model hamiltonian , called the spin - flux model , @xcite suggests that the long - range coulomb interaction between spin-@xmath5 electrons leads to qualitative new physics , not apparent in the conventional hubbard model ( see section 2 ) . the results of the hartree - fock study @xcite of this spin - flux model are summarized in section 3 . they suggests that the undoped parent compound is also an afm mott insulator . however , unlike the conventional hubbard model , the one - electron dispersion relations of the afm mean - field in the spin - flux model match those measured experimentally through angle - resolved photo - emission for undoped cuprates . a proper description of the highest occupied electronic states ( as provided by the spin - flux model ) , is crucial to considerations of doping . the spin - flux model and the conventional hubbard model differ dramatically in this regard . at the hf level , the doping holes added to the afm background of the spin - flux model are trapped in the core of antiferromagnetic spin vortices . this composite object ( meron - vortex ) is a bosonic charged collective mode of the many - electron system ( the total spin of the magnetic vortex is zero ) . the reversal of the spin - charge connection provides a microscopic basis for non - fermi - liquid behavior . a magnetic vortex is strongly attracted to an antivortex . this attraction increases logarithmically with the distance between the vortex cores , and is stronger than the unscreened coulomb repulsion between the charge meron - vortex cores . in effect , the increase in coulomb energy between a given pair of holes is more than offset by the lowering in exchange energy between the background electrons as their vortices approach each other from far away . as the inter - vortex distance increases , more and more spins are rotated out of their afm background orientation and the total energy of the system increases . thus , even at the hf level , the spin - flux model provides a fundamental underpinning for the origin of both non - fermi - liquid behavior , and strong pairing between the charge carriers . while providing a good starting point , the hartree - fock approximation also has serious shortcomings . for instance , the ground - state wavefunction in the presence of doping is non - homogeneous ( the static meron - vortices of the spin - flux model , or the charged stripes of the conventional hubbard model , break translational symmetry ) . physically , one expects that these charge carriers can move along the planes , resulting in a wavefunction which preserves the translational symmetry of the original hamiltonian . quantum dynamics of the charge carriers also determines whether the doped ground - state is really a metal . charge carriers in the optimally doped cuprates are quite mobile excitations , although their scattering rates are radically different from electrons in a conventional fermi liquid . a consistent way of treating the quantum dynamics of the charge carriers is provided by the configuration interaction ( ci ) method,@xcite described in section 4 . here , a linear combination of hf wavefunctions is used in order to restore the various broken symmetries . for instance , in a doped system the ci wavefunction is chosen to be a linear combination of hf wavefunctions with the charge carrier localized at different sites . certain types of charge carriers can lower their total energy substantially by quantum mechanically hopping from one site to the next . we tested the accuracy of the ci method against the exact solution @xcite of the one - dimensional hubbard model in reference 3 . in the 1d hubbard model the ci method describes the quantum dynamics of charged domain wall solitons in the afm background . by including these effects as fluctuation corrections to the hartree - fock mean - field theory , the ci method provides excellent agreement with the exact bethe ansatz solution for the ground - state energy of the doped 1d hubbard chain , over the entire @xmath7 range . the ci method also leads to a clear demonstration of the spin - charge separation in 1d . addition of one doping hole to the half - filled antiferromagnetic chain results in the appearance of two different carriers : a charged bosonic domain - wall ( which carries the charge but no spin ) and a neutral spin-1/2 domain wall ( which carries the spin but no charge ) . this study @xcite demonstrates the effectiveness of the ci method . in this paper we use the ci method to investigate dynamics of the charged meron - vortices in the spin - flux model . throughout this paper we exploit and refer to the analogy between the charge excitations of the 1d hubbard model and the 2d spin - flux model , @xcite apparent in the ci approach . @xcite the ci results for the spin - flux model ( presented in section 4 ) confirm that the meron - vortices are very mobile , suggesting that a collection of such mobile bosonic charge carriers is a non - fermi - liquid metal . the ci method also allows us to identify the rotational symmetry of the meron - antimeron pair wavefunction to be d - wave for the most stable pairs . an energetically more expensive metastable s - wave pairing is also possible . the possibility of spin - charge separation in 2d is elucidated . a summary of the results , their interpretation and conclusions is provided in section 5 . the effective 2d hamiltonian that we use to describe the strongly correlated electrons residing in the @xmath8 orbitals of the isolated cuo@xmath9 plane is the tight - binding model : @xmath10 where @xmath11 creates an electron at site @xmath12 with spin @xmath13 , @xmath14 is the hopping amplitude from site @xmath15 to site @xmath12 on the square lattice , @xmath16 is the total number of electrons at site @xmath12 , and @xmath17 is the coulomb interaction between electrons at sites @xmath12 and @xmath15 . the dominant terms are the nearest - neighbor hopping @xmath18 and the on - site coulomb repulsion @xmath19 . if only these two terms are considered , and we shift the chemical potential by @xmath20 , this reduces to the well - known hubbard model . the neglect of all longer range coulomb interaction ( @xmath21 , if @xmath22 ) , in the hubbard model , is based on the fermi - liquid theory notion of screening of the effective electron - electron interaction . however , fermi liquid theory fails to explain many of the crucial features of the high - t@xmath23 cuprates . in our description , we include the nearest - neighbor coulomb repulsion , which we assume is on the energy scale of @xmath24 . this leads to the generation of spin - flux , an entirely new type of broken symmetry in the many - electron system , which we show leads naturally to bosonic charge carriers in the form of meron - vortices , non - fermi - liquid behavior and a strong attractive pairing force between holes in the afm background . in order to extract the relevant physics from our starting hamiltonian , @xmath25 we introduce bilinear combination of electron operators @xmath26 , @xmath27 , for @xmath22 ( summation over multiple indexes is assumed ) . here @xmath28 is the @xmath29 identity matrix and @xmath30 are the usual pauli spin matrices . the notation @xmath31 means that the sites @xmath12 and @xmath15 are nearest neighbors . the quantum expectation value @xmath32 of the @xmath33 operators are associated with charge - currents ( @xmath34 ) and spin - currents ( @xmath35 ) . non - vanishing charge currents lead to appearance of electromagnetic fields , which break the time - reversal symmetry of the hamiltonian . experimentally , this does not occur in the cuprates . in the following , we adopt the ansatz that there is no charge current in the ground state @xmath36 but circulating spin - currents may arise and take the form @xmath37 , where @xmath38 for all @xmath12 and @xmath15 , and @xmath39 is a unit vector . these spin - currents provide a transition state to the spin - flux mean - field that we use in this paper . using the pauli spin - matrix identity , @xmath40 , it is possible to rewrite the nearest - neighbor electron - electron interaction terms as @xmath41 . if we neglect fluctuations in the spin - currents , we can use the mean - field factorization , @xmath42 . thus , the quartic nearest - neighbor coulomb interaction term is reduced to a quadratic term that is added to the hopping term leading to the effective hamiltonian : @xmath43 here , @xmath44 are spin - dependent @xmath45 hopping matrix elements defined by the mean - field theory , and @xmath46 . in deriving ( [ 1.104 ] ) we have dropped constant terms which simply change the zero of energy as well as terms proportional to @xmath47 which simply change the chemical potential . it was shown previously @xcite that the ground state energy of the hamiltonian of eq.([1.104 ] ) depends on the su(2 ) matrices @xmath48 only through the plaquette matrix product @xmath49 . here , @xmath50 is the spin - flux which passes through each plaquette and @xmath51 is the angle through which the internal coordinate system of the electron rotates as it encircles the plaquette . since the electron spinor wavefunction is two - valued , there are only two possible choices for @xmath50 . if @xmath52 we can set @xmath53 and the hamiltonian ( [ 1.104 ] ) describes conventional ordered magnetic states of the hubbard model . the other possibility is that a spin - flux @xmath54 penetrates each plaquette , leading to @xmath55 . this means that the one - electron wavefunctions are antisymmetric around each of the plaquettes , i.e. that as an electron encircles a plaquette , its wavefunction in the internal spin space of euler angles rotates by @xmath56 in response to strong interactions with the other electrons . in effect , the electron performs an internal `` somersault '' as it traverses a closed path in the cuo@xmath9 plane . this spin - flux phase is accompanied by a afm local moment background ( with reduced magnitude relative to the afm phase of the conventional hubbard model ) . in the spin - flux phase , the kinetic energy term in ( [ 1.104 ] ) exhibits broken symmetry as though a spin - orbit interaction has been added . however , it is distinct from the smaller , conventional spin - orbit effects which give rise to anisotropic corrections to superexchange interactions between localized spins in the afm . @xcite in the presence of charge carriers this mean - field is unstable to the proliferation of topological fluctuations ( magnetic solitons ) which eventually destroy afm long range order . in this sense , the analysis which we present below goes beyond simple mean field theory . the quantum dynamics of these magnetic solitons described by the configuration interaction ( ci ) method , corresponds to tunneling effects not contained in the hartree - fock approximation . for simplicity , throughout this paper we assume that the mean - field spin - flux parameters @xmath48 are given by the simplest possible choice @xmath57 ( see fig . [ fig1.55 ] ) . in order to go beyond a mean - field description of the spin - flux , these matrices may also be treated as dynamical variables . in this paper , we go beyond mean - field theory in describing the antiferromagnetic degrees of freedom but restrict ourselves to a mean - field model of the spin - flux . the configuration interaction method utilizes a linear combination of judiciously chosen hartree - fock wavefunctions.@xcite in this section , we provide a short review of the relevant hartree - fock results for the spin - flux model . a full comparison between the hfa for the spin - flux model and the conventional hubbard model has been published elsewhere . @xcite one of the most widely used approximations for the many - electron problem is the static hartree - fock approximation ( hfa ) . in this approximation the many - body problem is reduced to one - electron problems in which each electron moves in a self - consistent manner depending on the mean - field potential of the other electrons in the system . while this method is insufficient , by itself , to capture all of the physics of low dimensional electronic systems with strong correlations , it provides a valuable starting point from which essential fluctuation corrections can be included . in particular , we use the hartree - fock method to establish the electronic structure and the static energies of various magnetic soliton structures . in the more general configuration interaction ( ci ) variational wavefunction , the solitons acquire quantum dynamics and describe large amplitude tunneling and fluctuation effects that go beyond mean field theory . in the hf approximation , the many - body wavefunction @xmath58 is decomposed into a slater determinant of effective one - electron orbitals . the one - electron orbitals are found from the condition that the total energy of the system is minimized @xmath59 in order to approximate the ground state of the spin - flux hamiltonian ( [ 1.104 ] ) , we consider a slater determinant trial - wavefunction of the form @xmath60 is the vacuum state , @xmath61 is the total number of electrons in the system and the one - electron states are given by @xmath62 here , the one - particle wave - functions @xmath63 form a complete and orthonormal system . using the wavefunction ( [ 2.3 ] ) in equation ( [ 2.1 ] ) , and minimizing with respect to the one - particle wavefunctions @xmath63 , we obtain the hartree - fock eigen - equations : where @xmath64 are the pauli spin matrices and the charge density , @xmath65 and the spin density , @xmath66 must be computed self - consistently . the notation @xmath67 appearing in ( [ 2.5 ] ) means that the sum is performed over the sites @xmath15 which are nearest - neighbors of the site @xmath12 . the self - consistent hartree - fock equations ( [ 2.5],[2.6],[2.7 ] ) must be satisfied by the occupied orbitals @xmath68 , but can also be used to compute the empty ( hole ) orbitals . the ground - state energy of the system in the hfa is given by @xmath69 where the single particle energies are obtained from ( [ 2.5 ] ) . the approximation scheme described above is called the unrestricted hartree - fock approximation , because we did not impose constraints on the wavefunction @xmath58 which would require it to be an eigenfunction of various symmetry operations which commute with the hamiltonian ( [ 1.104 ] ) . if these symmetries are enforced , the method is called the restricted hartree - fock approximation . we use the unrestricted hfa since it leads to lower energies . the breaking of symmetries in our case implies that electronic correlations are more effectively taken into account . @xcite the restoration of these symmetries is deferred until the ci wavefunction is introduced . in the undoped ( half - filled ) case , the self - consistent hartree - fock equations can be solved analytically for the infinite system , using plane - wave one - particle wave - functions . in the unrestricted hartree - fock approach , doping the system leads to the appearance of inhomogeneous solutions , which break the translational invariance . in this case , we solve the unrestricted self - consistent hartree - fock equations numerically on a finite lattice . starting with an initial spin and charge distribution @xmath70 and @xmath71 , we numerically solve the eigenproblem ( [ 2.5 ] ) and find the hf eigenenergies @xmath72 and wavefunctions @xmath73 . these are used in eqs . ( [ 2.6 ] ) and ( [ 2.7 ] ) to calculate the new spin and charge distribution , and the procedure is repeated until self - consistency is reached . numerically , we define self - consistency by the condition that the largest variation of any of the charge or spin components on any of the sites of the lattice is less that @xmath74 between successive iterations . for the undoped system , the hartree - fock equations ( [ 2.5 ] ) for an infinite system are easily solved . in the cuprates , long - range afm order is experimentally observed . accordingly , we choose a spin distribution at the site @xmath75 of the form @xmath76 , where @xmath77 is the unit vector of some arbitrary direction , while the charge distribution is @xmath78 . in the spin - flux phase , it is convenient to choose a square unit cell , in order to simplify the description of the @xmath48 phase - factors . we make the simplest gauge choice compatible with the spin - flux condition for the @xmath79-matrices , namely that @xmath80 ( see fig . ( [ fig1.55 ] ) ) . this leads to the reduced square brillouin zone @xmath81 . from the hartree - fock equations we find two electronic bands , characterized by the dispersion relations : @xmath82 where each level is four - fold degenerate and @xmath83 are the noninteracting electron dispersion relations in the presence of spin - flux . the hf ground - state energy of the spin - flux afm background is given by ( see eq . ( [ 2.8 ] ) ) : @xmath84 where the afm local moment amplitude is determined by the self - consistency condition ( [ 2.7 ] ) @xmath85 at half - filling the valence band ( @xmath86 ) is completely filled , the conduction band ( @xmath87 ) is completely empty , and a mott - hubbard gap of magnitude @xmath88 opens between the valence and the conduction bands . the ground - state of the undoped spin - flux model is an afm mott insulator . it is interesting to note that the quasi - particle dispersion relation obtained in the presence of the spin - flux ( eq . ( [ 2.27 ] ) ) closely resembles the dispersion as measured through angle - resolved photo - emission spectroscopy ( arpes ) in a compound such as sr@xmath9cuo@xmath9cl@xmath9 @xcite ( see fig . [ fig2.13 ] ) . there is a a large peak centered at @xmath89 with an isotropic dispersion relation around it , observed on both the @xmath90 to @xmath91 and @xmath92 to @xmath93 lines . the spin - flux model in hfa exhibits another smaller peak at @xmath94 which is not resolvable in existing experimental data . this minor discrepancy may be due to next nearest neighbor hopping or other aspects of the electron - electron interaction which we have not yet been included in our model . @xcite the quasi - particle dispersion relation of the conventional hubbard model ( @xmath95 ) has a large peak at @xmath96 on the @xmath90 to @xmath91 line ( see fig . [ fig2.13 ] ) , but it is perfectly flat on the @xmath92 to @xmath93 line ( which is part of the large nested fermi surface of the conventional 2d hubbard model ) . also , it has a large crossing from the upper to the lower band - edge on the @xmath90 to @xmath92 line . this dispersion relation is very similar to that of the @xmath97 model ( see ref . 17 ) . while both the conventional and spin - flux model predict afm insulators at half - filling ( at least at the hf level ) , the spin - flux also provides a much better agreement with the dispersion relations , as measured by arpes . as in the 1d case , @xcite the effect of doping is the appearance of discrete levels deep inside the mott - hubbard gap . these levels are drawn into the gap from the top ( bottom ) of the undoped valence ( conduction ) bands . accordingly , the type of excitations created by doping depends strongly on topology of the electronic structure near the band edges . if we introduce just one hole in the plane , the self - consistent hfa solution is a conventional spin - polaron or `` spin - bag '' ( see fig . this type of excitation is the 2d analog of the 1d spin - polaron . @xcite the doping hole is localized around a particular site , leading to the appearance of a small ferromagnetic core around that site . the spin and charge distribution at the other sites are only slightly affected . in fact , the localization length of the charge depends on @xmath7 , and becomes very large as @xmath98 , since in this limit the mott - hubbard gap closes . for intermediate and large @xmath7 , the doping hole is almost completely localized on the five sites of the ferromagnetic core . the spin - bag is a charged fermion , as can be seen by direct inspection of its charge and spin distributions . this is also confirmed by its electronic structure ( see fig.3b ) . thus , the 2d spin - bag is indeed the analog of the 1d spin - polaron . @xcite the 2d analog of the 1d charged domain wall is the meron - vortex ( see fig . [ fig2.22 ] ) . like the 1d domain - wall , the meron - vortex is also a topological excitation , characterized by a topological ( winding ) number @xmath99 ( the spins on each sublattice rotate by @xmath100 on any closed contour surrounding the center of the meron ) . as such , a single meron - vortex can not be created in an extended afm background with cyclic boundary conditions by the introduction of a single hole ( just as a single isolated charged domain wall can not be created on an afm ( even ) chain with cyclic boundary conditions , by the introduction of a single hole ) . from a topological point of view , this is so because the afm background has a winding number 0 , and the winding number must be conserved , unless topological excitations migrate over the boundary into the considered region . however , excitations can be created in pairs of total topological number 0 . in the 1d case , this means creation of pairs of domain walls , while in 2d this means the creation of vortex - antivortex pairs . from figs . 4(a ) and 4(b ) we can see that the meron - vortex is a charged boson . the total spin of such a configuration is zero , while it carries the doping charge trapped in the vortex core . moreover , from its electronic spectrum ( fig . 4b ) we can see that only the extended states of the valence band are occupied . they are the only ones contributing to the total spin . since only one state is drawn from the valence band into the gap , to become a discrete bound level , it appears that an odd ( unpaired ) number of states remains in the valence band . however , one must remember that for topological reasons , merons must appear in vortex - antivortex pairs . therefore , the valence band has an even number of ( paired ) levels , and the total spin is zero . this argument for the bosonic character of the meron - vortex is identical to that for the charged domain wall in polyacetylene . @xcite unlike in the 1d case,@xcite we can not directly compare the excitation energy of the spin - bag with the excitation energy of the meron - vortex . the reason is that the excitation energy of the latter increases logarithmically with the size of the sample , and therefore an isolated meron - vortex is always energetically more expensive than a spin - bag . however , topology requires that merons and antimerons are created in pairs . the excitation energy of such a meron - antimeron pair is finite , allowing a meaningful comparison between excitation energies of a pair of spin - bags and a meron - antimeron pair . in figs . 5(a ) and 5(b ) we show the self - consistent spin and charge distributions for the lowest energy self - consistent hf configuration found when we add 2 holes to the afm background , in the spin - flux model , for @xmath101 . this configuration consists of a meron and an antimeron centered on second nearest - neighbor sites . as a result of interactions , the cores of the vortices are somewhat distorted . if the vortices were uncharged , vortex - antivortex pair annihilation would be possible . however , for charged vortices , the fermionic nature of the underlying electrons prevents two holes from being localized at the same site , in spite of the bosonic character of the collective excitation . a very interesting feature of this configuration is the strong topological attraction between the vortex and the antivortex . the closer the two cores are to each other , the smaller is the region in which the spins are rotated out of their background afm orientation by the vortices , and therefore the smaller is the excitation energy of the pair . since the holes are localized in the cores of the vortices , this topological attraction between vortices is an effective attraction between holes in the purely repulsive 2d electron system . this effect is unique to the spin - flux phase . in the conventional hubbard model , vortices are not stable excitations . the vortex - antivortex attraction increases as the logarithm of the distance between the cores . therefore , the pair of vortices should remain bound even if full unscreened @xmath102 coulomb repulsion exists between the charged cores , providing a compelling scenario for the existence of strongly bound pre - formed pairs in the underdoped regime . there is another possible self - consistent state for the system with two holes , consisting of two spin - bags far from each other ( such that their localized wave functions do not overlap ) . the excitation energy of such a pair of spin - bags is simply twice the excitation energy of a single spin - bag . when this excitation energy is compared to the excitation energy of the tightly bound meron - antimeron pair , we find that it is higher by @xmath103 ( for @xmath101 ) . in fact , for @xmath104 the hfa predicts that the meron - antimeron pair is the low - energy charged excitation , while for @xmath105 , the spin - bag is the low - energy charge carrier . this is analogous with the situation in 1d , where the spin - bag was predicted to be the low energy excitation for @xmath106 , in the hfa . @xcite as in 1d , however , we expect that this conclusion will be drastically modified once the charged solitons are allowed to move along the planes and the lowering of kinetic energy through translations is also taken into consideration . we complete this review of the hf results by pointing out that the strong analogy between the 1d hubbard model and the 2d spin - flux model is due to the similarity between the electronic structures at zero doping . as seen from fig.[fig2.13 ] , the 2d spin - flux model has isotropic dispersion relations about the @xmath96 point . this acts as a fermi point for the noninteracting system as it does in the 1d system . the two empty discrete levels drawn deep inside the mott - hubbard gap in the presence of the meron - vortex split from the @xmath96 peaks of the electron dispersion relation . the different topology of the large nested fermi surface of the conventional hubbard model leads to instability of the meron - antimeron pair . in fact , in the conventional hubbard model doping holes assemble in charged stripes , as opposed to the liquid of meron - antimeron pairs which is the low - energy state of the doped spin - flux model . the essence of the configuration interaction ( ci ) method is that the ground - state wavefunction , for a system with @xmath61 electrons , is not represented by just a single @xmath107 slater determinant ( as in the hfa ) , but a judiciously chosen linear combination of such slater determinants . @xcite given the fact that the set of all possible slater determinants ( with all possible occupation numbers ) generated from a complete set of one - electron orbitals constitute a complete basis of the @xmath61-particle hilbert space , our aim is to pick out a subset of slater determinants which captures the essential physics of the exact solution . consider the ci ground - state wavefunction given by @xmath108 is a distinct @xmath107 slater determinant and the coefficients @xmath109 are chosen to satisfy the minimization principle : @xmath110 this leads to the system of ci equations @xmath111 where @xmath112 is the energy of the system in the @xmath113 state , @xmath114 are the matrix elements of the hamiltonian in the basis of slater determinants @xmath115 , and @xmath116 are the overlap matrix elements of the slater determinants ( which are not necessarily orthogonal ) . the ci solution is easily found by solving the linear system of equations ( [ 3.3 ] ) , once the basis of slater determinants @xmath117 is chosen . if we denote by @xmath118 the @xmath119 one - electron occupied orbitals of the slater determinant @xmath120 , these matrix elements are given by : @xmath121{ccc } \beta_{1,1}^{nm } & ... & \beta_{1,n_e}^{nm } \\ \vdots & & \vdots\\ \beta_{n_e,1}^{nm } & ... & \beta_{n_e , n_e}^{nm } \\ \end{array } \right|\ ] ] the matrix elements of the hamiltonian ( [ 1.104 ] ) can be written as : @xmath122 where the expectation values of the hopping and on - site interaction terms are : @xmath123{ccccc } \beta_{1,1}^{nm}& ... &t_{1,p}^{nm}& ... &\beta_{1,n_e}^{nm}\\ \vdots & & \vdots & & \vdots\\ \beta_{n_e,1}^{nm}& ... &t_{n_e , p}^{nm}& ... &\beta_{n_e , n_e}^{nm } \\ \end{array } \right|\ ] ] and @xmath124{ccccccc } \beta_{1,1}^{nm } \!\!& ... \!\!&u_{1,p_1}^{nm}(i)\!\ ! \!\!&d_{1,p_2}^{nm}(i)\!\!& ... \!\ ! & \beta_{1,n_e}^{nm } \\ \vdots & & \vdots & & \vdots & & \vdots\\ \beta_{n_e,1}^{nm}\!\!& ... \!\!&u_{n_e , p_1}^{nm}(i)\!\ ! & ... \!\!&d_{n_e , p_2}^{nm}(i)\!\!& ... \!\ ! & \beta_{n_e , n_e}^{nm}\\ \end{array } \right|_{.}\ ] ] here , @xmath125 @xmath126 @xmath127 and @xmath128 we now consider the specific choice of the slater determinant basis @xmath115 . strictly speaking , one may choose an optimized basis of slater determinants from the general variational principle : @xmath129 however , implementation of this full trial - function minimization scheme ( also known as a multi - reference self - consistent mean - field approach @xcite ) is numerically cumbersome even for medium - sized systems . instead , we select the slater determinant basis @xmath115 from the set of broken symmetry , unrestricted hartree - fock wavefunctions ( [ 2.3 ] ) , their symmetry related partners and their excitations . clearly , ( [ 2.3 ] ) satisfies ( [ 3.4 ] ) by itself , provided that the @xmath130 coefficients corresponding to the other slater determinants in eq.([3.1 ] ) are set to zero ( see eq . ( [ 2.1 ] ) ) . since this unrestricted hf wavefunction is not translationally invariant ( the doping hole is always localized somewhere on the lattice ) , we can restore the translational invariance of the ci ground - state wavefunction by also including in the basis of slater determinants all the possible lattice translations of this unrestricted hf wavefunction . furthermore , if the self - consistent configuration is not rotationally - invariant ( e.g. a meron - antimeron pair ) , all possible rotations must be performed as well . by rotation we mean changing the relative position of the meron and antimeron while keeping their center of mass fixed . clearly , all the translated hf slater determinants lead to the same hf ground - state energy @xmath131 as defined by eq . ( [ 2.8 ] ) . the ci method lifts the degeneracy between states with the hole - induced configuration localized at different sites ( see eq . ( [ 3.3 ] ) ) , thereby restoring translational invariance . we may identify the lowering in the total energy due to the lifting of this degeneracy as quantum mechanical kinetic energy of deconfinement which the doping - induced configuration saves through hopping along the lattice . in addition , quantum fluctuations in the internal structure of a magnetic soliton can be incorporated by including the lowest order excited state configurations of the static hartree - fock energy spectrum . such wavefunctions are given by @xmath132 , where @xmath133 labels an excited particle state and @xmath134 labels the hole which is left behind ( see eq . ( [ 2.3 ] ) ) . once again , all possible translations ( and non - trivial rotations ) of this `` excited '' configurations must be included in the full ci wavefunction . these additions can describe changes in the `` shape '' of the soliton as it undergoes quantum mechanical motion along the plane . the ci method is described in more detail in reference 3 , where it is used to study the 1d hubbard chain in order to gauge its accuracy by comparing its results with the exact bethe ansatz solution . we showed that the ci method recaptures the essential physical features of the exact solution of the 1d hubbard chain , such as spin - charge separation , as well as leading to remarkable agreement of ground state energies of doped chains for all values of @xmath7 . the main difference between the 1d case and the 2d case is the computation time required . the computation time for one matrix element @xmath135 scales roughly like @xmath136 , where @xmath137 is the number of sites . the number of configurations included in the ci set scales as @xmath138 when @xmath139 solitons are present . for both an @xmath137-site chain and an @xmath140 lattice , the hf `` bulk '' limit is reached for @xmath141 . in the 1d case @xcite we used chains with @xmath142 , and numerical calculations can be easily performed . however , in 2d the smallest acceptable system has 100 sites , leading to an enormous increase in the computation time . nevertheless , our sample of results in 2d suggest a simple and clear physical picture which we describe below . the charged spin - bag carries a spin of 1/2 . let @xmath143 be the hf determinants for the spin - bag centered at any two nearest neighbor sites , respectively , and let @xmath144 be the total spin operator in the @xmath145-direction . then , @xmath146 . since the hubbard hamiltonian commutes with @xmath147 , it follows that @xmath148 . from the ci equation ( [ 3.3 ] ) we conclude that there is no mixing between states with different total spin . as a result , it is enough to include in the ci set only those configurations with the spin bag localized on the same magnetic sublattice . let us denote by @xmath149 the initial static hartree - fock configuration , and by @xmath150 the configuration obtained through its translation by @xmath151 sites in the x - direction and @xmath152 sites in the y - direction ( cyclic boundary conditions are imposed ) . the condition that only configurations on the same sublattice are included means that @xmath153 must be an even number , and the cyclic boundary conditions mean that @xmath154 , for a @xmath155 lattice . as explained in detail in the 1d analysis , @xcite mixing configurations with the charged spin - bag localized at different sites and then subtracting out the contribution of the undoped afm background allows us to calculate the dispersion band of the spin - bag itself : @xmath156 here , the total energy of the lattice with the spin - bag @xmath157 and the ci wave - function @xmath158 ) . the finite size of the lattice and cyclic boundary conditions restricts the calculation to @xmath159-points of the form @xmath160 , where @xmath161 is any pair of integer numbers . as usual , only @xmath159-points inside the first brillouin zone need to be considered . an analysis of the dependence of the spin - bag dispersion relation @xmath162 on the size @xmath140 of the lattice is shown in fig . [ fig3.20 ] , for the conventional hubbard model ( upper panel ) and spin - flux model ( lower panel ) , and @xmath101 . we used 6x6 , 8x8 , 10x10 and ( only for the spin - flux model ) 12x12 lattices . the dispersion relation is plotted along lines of high symmetry of the full brillouin zone . for comparison , we also show the excitation energy @xmath163 obtained in the static hfa as a full line . for both models , we see that the spin - bag dispersion band is almost converged , even though we used quite small lattices . the convergence is somewhat slower in the spin - flux case , as seen most clearly at the ( 0,0 ) point . although the values obtained from the four lattices all differ at ( 0,0 ) , the extremum values correspond to the 6x6 and the 8x8 lattices , while the values for the 10x10 and 12x12 lattices are indistinguishable . we conclude that the fit ( [ 3.40 ] ) is legitimate . from fig . [ fig3.20 ] we also see that the dispersion relations for the spin - bag in the two different models are very different . the dispersion relations over the full 2d brillouin zone are shown in fig . [ fig3.21 ] , and they are seen to mimic the electronic dispersion relation of the underlying undoped afm background , shown in fig . [ fig2.13 ] . this is consistent with the quasi - particle nature of this charged spin-1/2 spin - bag . in the conventional hubbard model , the undoped afm background has a large nested fermi surface along the @xmath92 to @xmath93 line , and it is exactly along this line that the spin - bag dispersion band has a minimum . similarly , the lowest energy of the spin - bag of the spin - flux model is at @xmath89 , corresponding to the fermi points of the underlying undoped spin - flux afm background . the extra kinetic energy @xmath164 saved by the spin - bag through quantum hopping is @xmath165 in the conventional model and @xmath166 in the spin - flux model ( for @xmath101 ) . since the spin - bag is confined to one magnetic sublattice , it must tunnel two lattice constants to the next allowed site . consequently , the energy gained through hopping ( of order @xmath167 ) is small . this is displayed , for the spin - bag of the spin - flux model , in fig . [ fig3.24 ] , where we plot the lowering in kinetic energy of the deconfined spin - bag @xmath168 as a function of @xmath169 . a similar dependence for the spin - bag of the conventional hubbard model is presented elsewhere . @xcite as in the 1d case , we conclude that the spin - bag in 2d is a rather immobile quasiparticle - like excitation . in the 1d model it is energetically favorable for the immobile spin - bag to decay into a charged bosonic domain wall and a neutral fermionic domain wall , resulting in spin - charge separation . @xcite the analog of the 1d charged bosonic domain wall is the 2d charged bosonic meron - vortex of the spin - flux model . if the spin - bag decays into a charged meron - vortex , a magnetic antivortex must also be created for topological reasons . unlike the pair of domain walls in the 1d case , the vortex - antivortex pair is tightly bound by a topological binding potential that increases as the logarithm of the vortex - antivortex separation . therefore , we expect that the doping charge is shared between the two magnetic vortices . one technical problem for testing this hypothesis is that such a configuration ( a vortex - antivortex pair sharing one doping hole ) is not self - consistent at the static hartree - fock level . in the static approximation we require two doping holes to stabilize two vortex cores and create a meron - antimeron pair . we can , however , construct a trial wavefunction to describe the singly - charged vortex - antivortex pair , by adding one electron in the first empty state of the self - consistent doubly - charged meron - antimeron configuration . the first empty levels of the meron - antimeron pair are the localized levels bound in the vortex cores , two for each vortex ( see fig . because of degeneracy between the two lower discrete levels of the pair , we have in fact two distinct trial wave - functions , obtained by adding one electron in either of these two lower localized gap electronic states of a self - consistent meron - antimeron pair . these wavefunctions are not invariant to rotations ( see fig . [ fig2.25 ] ) . therefore we must include in the ci set of slater determinants the configurations obtained through @xmath170 rotations of the vortex - antivortex pair about its fixed center of mass in addition to translated configurations . as a result , we have a total of @xmath171 configurations describing the singly charged vortex - antivortex pair localized at all possible sites with all possible orientations about the center of mass . we performed this ci analysis for a @xmath172 lattice and @xmath101 . the hf energy of a simple static spin - bag is @xmath173 ( measured with respect to the hf energy of the undoped afm background , equal to @xmath174 ) . the energy of the static singly - charged vortex - antivortex pair is @xmath175 . thus , we see that because this singly - charged pair trial wavefunction is not self - consistent , in the static case this configuration is energetically much more costly than the self - consistent spin - bag configuration . however , if we allow for quantum motion of these configurations , the situation changes dramatically . performing the ci analysis for the set of all possible translated spin - bag configurations , we find that the energy of the spin - bag is lowered to @xmath176 . performing the ci analysis for the set of all translated and rotated singly - charged vortex - antivortex pairs we find that this configuration s energy is lowered to @xmath177 . this shows that the vortex - antivortex pair has lowered its translational and rotational kinetic energy by almost @xmath178 , thereby becoming the low - energy charge carrier . this large number is not surprising , since unlike the spin - bag , the vortex - antivortex pair is not constrained to motion on one magnetic sublattice . as a result , such configurations lower their kinetic energy by an amount on the scale of @xmath24 , as opposed to @xmath167 for the spin - bag configuration . for larger @xmath7 values this effect is even more pronounced . we conclude that these results strongly support the hypothesis of spin - bag dissociation into a much more mobile singly - charged vortex - antivortex pair , analogous to the 1d spin - bag dissociation into a pair of a charged bosonic domain - wall and a neutral fermionic domain - wall . @xcite unlike in the 1d case , however , we do not have distinct charge and spin carriers for the composite excitation . instead the spin and the charge are shared equally between the vortex and the antivortex . if , on the other hand , there was a mechanism whereby the vortices became unbound , complete spin - charge separation could occur , in which one vortex traps the hole ( and is therefore a charged meron ) and the other vortex carries the spin , in a lotus - flower@xcite ( or undoped magnetic meron ) configurations . in the absence of the corresponding self - consistent static hf configuration we are not able to settle this question . at very low doping , the strong vortex - antivortex topological attraction binds the spin and charge together . this is different from the 1d case , where the absence of long - range interactions between the domain - walls allow for a complete spin - charge separation at any doping and even at zero temperature . this scenario opens a new avenue for research into how the system evolves with doping . if each hole is dressed into a singly - charged vortex - antivortex pair , when two such pairs overlap it is possible that both doping charges move to the same pair , creating a meron - antimeron pair of charged bosons . such pre - formed charge pairs may condense into a superconducting state at low temperatures . the other uncharged vortex - antivortex pair may either collapse and disappear ( this is likely to happen at low - temperatures ) or remain as a magnetic excitation of the system ( at higher temperatures ) , mediating the destruction of the long - range afm order , the renormalization of the spin - wave spectrum and the opening of the spin pseudogap . from the static hf analysis we found that the most stable static self - consistent configuration with two doping holes added to the afm background of the spin - flux model is the meron - antimeron pair , for @xmath179 . at larger @xmath7 , two charged spin - bags become more stable , in the static hf approximation . this is in close analogy to the prediction that the spin - bag is energetically more favorable than the static charged domain - wall for @xmath180 , in the hfa of the 1d hubbard model . @xcite however , in the 1d case the charged domain - wall is considerably more mobile than the charged spin bag , gaining a kinetic energy of the order of @xmath24 as opposed to @xmath167 energy gained by the spin - bag . as a result , when this kinetic energy of deconfinement is taken into account within the ci method , the charged domain - wall is found to be the relevant charged excitation for all values of @xmath7 . a similar picture emerges in the 2d case , because the meron - vortices are much more mobile than the spin - bags . for the 2d system , we have shown that the charged spin - bag has very similar behavior to the 1d charged spin - bag . the analog of the 1d charged bosonic domain - wall is the 2d charged bosonic meron - vortex . we now consider the properties of the doubly - charged meron - antimeron pair . all the numerical results quoted in the rest of this section refer to a meron - antimeron pair on a 10x10 lattice , in the spin - flux model with @xmath101 . as already discussed , the meron - antimeron pair is not rotationally invariant . we can find the rotational kinetic energy saved by the pair as it rotates about its center of mass . in the present case , only 4 configurations need to be included , corresponding to the four possible self - consistent arrangements of the meron and antimeron about their fixed center of mass ( see fig . [ fig3.28 ] ) . simple rotation by @xmath170 of the one - particle orbitals @xmath181 about the center of mass is not , however , sufficient to generate the rotated configurations . first of all , the @xmath170 rotation also changes the spin - flux parameterization . if the spin - flux of the initial configuration is @xmath57 , a @xmath170 rotation leads to a state corresponding to the rotated configuration @xmath182 . thus , following the @xmath170 rotation , a unitary transformation must be performed in order to restore the initial spin - flux parameterization . for the case cited above , this simply implies the change in the one - particle orbitals @xmath183 for all sites @xmath184 which are a type 2 site of the unit cell , in other words sites with even @xmath185 and odd @xmath186 ( also , see fig . [ fig1.55 ] ) . the second observation is that the rotation by @xmath170 also changes ( flips ) all the spins of the afm background surrounding the pair . thus , an extra @xmath187 rotation about an axis perpendicular to the lattice plane is necessary to restore the alignment of the afm background . following these transformations it is straightforward to generate the slater determinants @xmath188 and @xmath189 from the initial self - consistent hf meron - antimeron pair described by @xmath190 . the ci method can be used to find the rotational energy saved by superposing these rotated meron - antimeron configurations . the lowest ci energy found is @xmath191 below the energy of the static pair , and corresponds to d - wave symmetry . by this we mean that the coefficients @xmath109 multiplying the 4 rotated states in the ci wave - function @xmath192 . translation of a pair over the whole lattice can also be investigated . since the pair does not carry any spin , all possible translations must be included ( there is no restriction to same magnetic sublattice configurations ) . this leads to a total of @xmath193 possible configurations for a @xmath140 lattice . again , when various configurations are generated from the initial self - consistent hf meron - antimeron state @xmath194 , care must be taken to preserve the same spin - flux parameterization and the same afm background orientation . this can be achieved performing transformations similar to the ones described above . as a result of performing the ci method on the set of translated states , we find the dispersion relation of the ( un - rotated ) meron - antimeron pair . this is shown in fig . [ fig3.25 ] . in this plot we show the total energy of the lattice with the moving meron - antimeron pair , as a function of the total momentum of the pair . quantum hopping of the meron - antimeron pair lowers its total energy by an extra @xmath195 . two other interesting features are observed in fig . [ fig3.25 ] . the first one is that the dispersion relation of this rigidly polarized pair is not invariant to rotations by @xmath170 , as expected . more important is that the minima of the dispersion relation occur at the @xmath91 points . since the momentum of the pair is twice the momentum of either the meron or the antimeron , this is consistent with the fact that in their lowest energy state , both the meron and the antimeron have momenta of @xmath89 , in the spin - flux model . the doubling of the size of the brillouin zone is also a direct consequence of this doubling of total momentum ( for comparison with undoped dispersion relation , see fig . [ fig2.13 ] ) . however , to obtain the true energy of the charged pair , we must mix all @xmath196 rotated and translated meron - antimeron configurations . all have the same static hf energy and are equally important in the ci method . let @xmath197 denote the initial self - consistent static hartree - fock meron - antimeron configuration , and @xmath198 denote the configuration obtained through translation by @xmath151 sites in the x - direction and @xmath152 sites in the y - direction , as well as a rotation by an angle of @xmath199 of the pair about its center of mass . here @xmath200 and @xmath201 ( cyclic boundary conditions are imposed ) . the ci wavefunctions are then given by : @xmath202 the dispersion relation @xmath203 obtained from this complete set is shown in fig . [ fig3.26 ] . the reference point is the hf energy of the self - consistent meron - antimeron pair @xmath204 . thus , we see that the total kinetic energy saved by the freely moving and rotating meron - antimeron pair is @xmath205 , when the total momentum of the pair is @xmath91 . this energy equals the sum @xmath206 of rotational and translational kinetic energies found before ( the number of significant figures indicates the estimated accuracy of the computational method ) . the rotational invariance of the dispersion band is also restored . besides the absolute minima about the @xmath91 points , there is a more shallow minimum region about the @xmath90 point . the rotational symmetry of the meron - antimeron pair wavefunction , defined by the coefficients @xmath207 , is a function of the total momentum carried by the pair , as shown in fig . [ fig3.27 ] . the absolute minima points @xmath91 and the area around them correspond to pairs with d - wave symmetry . by this , we mean that the coefficients @xmath208 have the form @xmath209 with @xmath210 , i.e. @xmath211 . the core area , about the local minimum @xmath90 point , corresponds to s - wave symmetry . in this region the coefficients @xmath212 again satisfy eq . ( [ 3.101 ] ) , but for @xmath213 , i.e. @xmath214 . the intermediary area appears to be a mixture of different @xmath215 values . a simple decomposition of the form ( [ 3.101 ] ) is no longer possible . instead , a sum of such terms corresponding to different @xmath215 values is required . since we only have rotations by @xmath170 , a unique identification of the composite symmetry is not possible . moreover , the energy of the states in this intermediary area is at the top of the dispersion band . in order to find the correct ci states for energies well above the static hf value ( i.e. larger than zero , in this case ) we must add to the ci set the first set of excited hf states . for a meron - antimeron pair , excitation of an electron from the valence band onto the empty localized levels inside the mott - hubbard gap costs about 1.5 t of energy , for @xmath101 , so such states should contribute significantly in the ci states with positive energies and modify their dispersion and symmetry ( for this reason , we do not show the upper three high - energy bands in fig . [ fig3.26 ] ) . consequently , both the energy and the symmetry of the states in the intermediary area may be modified from the ones shown in fig . [ fig3.26 ] . however , the minima at @xmath91 and @xmath90 are at energies well below zero . their energies and rotational symmetry are unaffected by additions of higher energy configurations to the ci set . the fact that we obtain two distinct minima is not very surprising . as argued before , we expect that individual merons and antimerons are created with momenta of @xmath216 . as a result , two different couplings are possible . a @xmath217 ) meron can pair with a @xmath217 ) antimeron , creating a pair of total momentum @xmath91 . this is the most stable coupling , leading to the lowest possible energy of @xmath218 below the static hf energy . this pair has d - wave symmetry . the second possible coupling is between a @xmath217 ) meron and a @xmath219 ) antimeron . this pair has a total momentum of @xmath90 , and s - wave symmetry . however , this coupling is less strong . for the @xmath101 case considered , the energy of the s - wave @xmath90 pair is @xmath220 above the energy of the d - wave @xmath91 pair . the existence of both d - wave and s - wave pairing , and the dominance of the d - wave pairing , have been established experimentally for the high - t@xmath23 cuprates . @xcite we are not aware of any other microscopic theory that predicts the two types of pairing to appear in different regions of the brillouin zone . the total kinetic energy saved by the meron - antimeron pair through quantum hopping and rotation is of order @xmath24 , as expected , since the pair is not restricted to one magnetic sublattice and tunneling is not required for motion . consequently , we expect that the energy saved by the meron - antimeron pair for larger values of @xmath7 is comparably large . on the other hand , the energy saved by the spin bag through tunneling motion scales like @xmath167 . in fact , we argued that a spin - bag may dissociate into a singly - charged vortex - antivortex pair in order to enhance its mobility . however , even if dissociation does not occur , the kinetic energy saved by a pair of spin - bags is significantly smaller than the kinetic energy saved by the meron - antimeron pair . this shows that for @xmath101 the meron - antimeron pair is even more favorable energetically than the hfa predicts and suggests that the @xmath7 range where meron - antimeron pair formation occurs may extend well beyond the @xmath221 limit found within the hfa . @xcite in the 1d case , the range of stability of the charged domain wall versus the charged spin - bag is extended ( from the hf prediction of @xmath222 ) to all @xmath7 range.@xcite a numerical analysis is needed to determine if the limit is extended to infinity in the 2d case as well . the configuration interaction approximation incorporates crucial quantum translational and rotational motion of the charge carriers , which are absent in the static hartree - fock approximation . given the accuracy of the ci method in recapturing certain features of the exact bethe ansatz solution of the 1d hubbard model , @xcite we believe that the ci method is likewise a very powerful tool for describing effects beyond mean - field theory in 2d . for 2d systems , numerical calculations are much more time consuming . however , our small sample of results is quite suggestive of a simple physical picture . in direct analogy with the 1d results , we find that the bosonic charged meron - vortices are much more mobile than the fermionic charged spin - bags . the extra kinetic energy gained by the meron - vortices very likely extends their region of stability beyond the @xmath221 limit suggested in the hfa . @xcite there are also strong indications that a charged spin - bag is unstable to decay into a singly - charged vortex - antivortex pair , analogous to the spin - charge separation in the 1d case . nucleation of such pairs of vortices with doping is expected to further influence the magnetic behavior of the cuprates . the bound state and the unbound continuum states of the singly charged meron - antimeron pair may account for the anomalous `` quasiparticle '' spectral widths observed on angle - resolved photo - emission experiments . the symmetry of the meron - antimeron pairs emerges very clearly from the ci treatment . we find two regions of stability of the meron - antimeron pair . pairs with total momentum of @xmath91 have d - wave symmetry , and are the most stable . pairs with total momentum @xmath90 have s - wave symmetry and have a smaller gap . thus , we find that different pairing appears in different regions of the brillouin zone . these results appear to have a direct bearing on numerous experiments , which show a mixture of strong d - wave component and a smaller s - wave component in the superconducting state of the cuprates . @xcite many other features of our model are in agreement with experimentally observed properties of the cuprate superconductors . nucleation of magnetic vortices with doping explains a variety of magnetic properties , starting with complete destruction of the long - range afm order for very low doping concentration . as we can see from fig . 5(a ) a tightly - bound meron - antimeron pairs disturbs the long - range afm ordering of most of the spins on the 10x10 lattice . for very low dopings , these pairs are far from each other , and there are many spins on the plane whose orientations are not affected by any pair . thus , most of the spins maintain the long range afm order . however , as the doping increases to about @xmath223 ( which is roughly equivalent to having two meron - antimeron pairs on the 10x10 lattice ) the areas occupied by each meron - antimeron pairs start to overlap with those occupied by the neighboring pairs . at this doping the orientation of all the spins on the cuo@xmath9 planes is affected by at least one pair of vortices , and therefore the lro is severely disrupted . the local ordering , however , is still afm . this picture explains the extremely low doping necessary for the disappearance of lr afm order , as well as the fact that the spin correlation length is basically equal to the average distance between holes ( vortices ) and does not depend strongly on the temperature.@xcite each hole carries its vortex with it , and the spins in each vortex are correlated with each other . the correlation length is thus roughly equal to the average size of the vortex which is defined by the average inter - vortex ( inter - hole ) distance . the nucleation of magnetic vortices also explains the split of the @xmath91 afm bragg peak into the four incommensurate peaks whose positions shift with doping,@xcite as observed in lacuo and , more recently , in ybacuo.@xcite the form factor of any given vortex already gives rise to an apparent splitting of the neutron scattering peak . as demonstrated in ref . 9 , even at the mean - field level we recapture the neutron scattering data using a random distribution of meron - vortices . optical behavior of the cuprates is also explained naturally using our model . two features develop in the optical absorption spectra with doping : a broad mid - infrared temperature - independent absorption band , and a strongly temperature - dependent low - frequency drude tail . @xcite in our model the broad mid - infrared band is related to excitation of electrons from the valence band onto the empty levels bound in the vortex cores , @xcite which are localized deep inside the mott - hubbard gap ( see fig . the number of localized levels scales with the number of vortices , and inter - vortex interactions lead to their splitting into a broad band . as such , this mechanism is similar to the one leading to a broad mid - infrared absorption band in polyacetylene with doping . @xcite the polyacetylene band is due to electronic excitations inside the cores of the polyacetylene domain - wall solitons,@xcite which are the topological analogues of meron - vortices.@xcite another strong argument in favor of this interpretation is provided by photoinduced absorption experiments . @xcite if the undoped parent compounds are illuminated with intense visible light , they develop absorption bands that resemble the mid - infrared bands of the doped compounds . similar behavior is observed in polyacetylene , and is attributed to the nucleation of solitons by photoexcited electron - hole pairs . @xcite the second component of the optical spectrum is the drude tail . it results from the response of the freely moving charged vortices to the external electric field . the strong temperature dependence of this tail is determined by the scattering mechanism for merons ( presumably due to interactions with other merons and spin - waves ) . this interpretation is also supported by the fact that the superconducting transition leaves the mid - infrared absorption band unchanged . merons with internal electronic structure are still present on the planes but pair condensation leads to a collapse of the drude tail into a @xmath224 response . as already discussed , nucleation of charged meron - vortices with doping provides a microscopic basis for non - fermi - liquid behavior , due to the bosonic nature of the mobile charged excitations . the model also predicts the existence of pre - formed pairs with d - wave symmetry , which are thought to be responsible for the pseudo - gap effects . @xcite as the number of pairs of charged bosons increases with doping and the temperature is lowered , the meron - antimeron pairs bose condense and become coherent , leading to superconductivity . this mechanism naturally explains the puzzling scaling of the superfluid density with doping @xmath225 , in other words with the number of holes , not of electrons . in our model , it is the doping - induced positively charged meron - vortices that are the mobile charge carriers . as a result , the density of preformed meron - antimeron cooper pairs is obviously proportional to doping . finally , for large dopings ( @xmath226 ) the average vortex size become extremely small and the very cores of the merons start to overlap . in this limit the mott - hubbard gap is completely filled in by the discrete levels , and the spin - flux state becomes energetically unstable relative to a normal fermi liquid . @xcite it is noteworthy that all of the independent features described above are in qualitative agreement with our model , which has essentially no free or adjustable parameters . the choice of @xmath7 is fixed by the experimentally measured size of the mott - hubbard charge transfer gap at zero doping . more detailed comparisons with specific experiments may require the incorporation of specific ( smaller energy scale ) interactions which are not included in this simplest version of the spin - flux hamiltonian . a derivation of the explicit consequences of this picture appears to be worthy . a more comprehensive and quantitative e comparison with the experiments may prove quite fruitful . m.b . acknowledges support from the ontario graduate scholarship program and a fellowship from william f. mclean . this work was supported in part by the natural sciences and engineering research council of canada . k. yamada , c.h . lee , k. kurahashi , j. wada , s. wakimoto , s. ueki , h. kimura , y. endoh , s. hosoya , g. shirane , r.j . birgeneau , m. greven , m.a . karstner and y.j . kim , phys . rev . * b 60 * , 6165 ( 1998 ) . d. vaknin _ et al . . lett . * 58 * , 2802 ( 1987 ) . for a review of optical properties , tanner and t. timusk , in _ physical properties of high temperature superconductors iii _ , edited by donald m. ginsberg , ( world scientific press , singapore , 1992 ) ; g.a . thomas , in _ high temperature superconductivity _ , edited by d.p . tunstall and w. barford , proceedings of the 39@xmath227 scottish universities summer school in physics , st . andrews , ( adam hilger , new york , 1991 ) .

in this paper we obtain multifractal generalizations of classical results by lvy and khintchin in metrical diophantine approximations and measure theory of continued fractions . we give a complete multifractal analysis for stern brocot intervals , for continued fractions and for certain diophantine growth rates . in particular , we give detailed discussions of two multifractal spectra closely related to the farey map and to the gauss map .

math0509603

in this paper we give a multifractal analysis for stern brocot intervals , continued fractions and certain diophantine growth rates . we apply and extend the multifractal formalism for average growth rates of @xcite to obtain a complete multifractal description of two dynamical systems originating from the set of real numbers . recall that the process of writing an element @xmath0 of the unit interval in its regular continued fraction expansion @xmath1=\cfrac{1}{a_{1}(x)+\cfrac{1}{a_{2}(x)+\cfrac{1}{a_{3}(x)+\cdots}}}\ ] ] can be represented either by a uniformly hyperbolic dynamical system which is based on an infinite alphabet and hence has infinite topological entropy , or by a non - uniformly hyperbolic dynamical system based on a finite alphabet and having finite topological entropy . obviously , for these two systems the standard theory of multifractals ( see e.g. @xcite ) does not apply , and therefore it is an interesting task to give a multifractal analysis for these two number theoretical dynamical systems . there is a well known result which gives some information in the generic situation , that is for a set of full @xmath2-dimensional lebesgue measure @xmath3 . namely with @xmath4 $ ] referring to the @xmath5-th approximant of @xmath0 , we have for @xmath3-almost every @xmath6 , @xmath7 note that by employing the analogy between regular continued fraction expansions of real numbers and geodesics on the modular surface , the number @xmath8 can be interpreted as the hyperbolic length associated with the approximant @xmath9 . also , the parameter @xmath5 represents the word length associated with @xmath9 with respect to the dynamical system on the infinite alphabet , whereas @xmath10 can be interpreted as the word length associated with @xmath9 with respect to the dynamical system on the finite alphabet . there are two classical results by khintchin and lvy @xcite , @xcite , @xcite , @xcite which allow a closer inspection of the limit @xmath11 . that is , for @xmath3-almost every @xmath6 we have , with @xmath12 , @xmath13 clearly , dividing the sequence in @xmath14 by the sequence in @xmath15 leads to the sequence in @xmath11 . therefore , if we define the level sets @xmath16 then these classical results by lvy and khintchin imply for the hausdorff dimensions ( @xmath17 ) of these level sets @xmath18 a natural question to ask is what happens to this relation between these hausdorff dimensions for prescribed non - generic limit behavior . our first main results in this paper will give an answer to this question . namely , with @xmath19 referring to the golden mean , we show that for each @xmath20 $ ] there exists a number @xmath21 such that , with the convention @xmath22 and @xmath23 , @xmath24 furthermore , for the dimension function @xmath25 given by @xmath26 we show that @xmath25 can be expressed explicitly in terms of the legendre transform @xmath27 of a certain pressure function @xmath28 , referred to as the stern brocot pressure . for the function @xmath28 we obtain the result that it is real - analytic on the interval @xmath29 and vanishes on the complement of this interval . we then show that the dimension function @xmath25 is continuous and strictly decreasing on @xmath30 $ ] , that it vanishes outside the interval @xmath31 , and that for @xmath20 $ ] we have @xmath32 before we state the main theorems , let us recall the following classical construction of stern brocot intervals ( cf . @xcite , @xcite ) . for each @xmath33 , the elements of the @xmath5-th member of the stern brocot sequence @xmath34 are defined recursively as follows . * @xmath35 and @xmath36 ; * @xmath37 for @xmath38 ; * @xmath39 , for @xmath40 . with this ordering of the rationals in @xmath41 $ ] we define the set @xmath42 of stern brocot intervals of order @xmath5 by @xmath43 clearly , for each @xmath33 we have that @xmath42 represents a partition of the interval @xmath44 . the first members in this sequence of sets are the following , and it should be clear how to continue this list using the well known method of mediants . @xmath45 as already mentioned above , our multifractal analysis will make use of the stern brocot pressure function @xmath28 . this function is defined for @xmath46 by @xmath47 in here , @xmath48 refers to the euclidean length of the interval @xmath49 . we will see that @xmath28 is a well defined convex function ( cf . proposition [ pro : analyticpropertiesp ] ) . also , note that we immediately have that@xmath50 the following theorem gives the first main results of this paper . in here , @xmath27 refers to the legendre transform of @xmath28 , given for @xmath51 by @xmath52 . [ thm : main ] _ ( see fig . [ cap : the - stern brocot - pressure ] ) _ 1 . the stern brocot pressure @xmath28 is convex , non - increasing and differentiable throughout @xmath53 . furthermore , @xmath28 is real analytic on the interval @xmath29 and is equal to @xmath54 on @xmath55 . 2 . for every @xmath20 $ ] , there exist @xmath56 and @xmath21 related by @xmath57 such that , with the conventions @xmath58 and @xmath22 , @xmath59 furthermore , the dimension function @xmath25 is continuous and strictly decreasing on @xmath30 $ ] , it vanishes outside the interval @xmath31 , and for @xmath20 $ ] we have@xmath60 where @xmath61 . also , for the left derivative of @xmath25 at @xmath62 we have @xmath63 . theorem [ thm : main ] has some interesting implications for other canonical level sets . in order to state these , recall that the elements of @xmath42 cover the interval @xmath44 without overlap . therefore , for each @xmath6 and @xmath64 there exists a unique stern brocot interval @xmath65 containing @xmath0 . the interval @xmath66 is covered by two neighbouring intervals from @xmath67 , a left and a right subinterval . if @xmath68 is the left of these then we encode this event by the letter @xmath69 , otherwise we encode it by the letter @xmath70 . in this way every @xmath6 can be described by a unique sequence of nested stern brocot intervals of any order that contain @xmath0 , and therefore by a unique infinite word in the alphabet @xmath71 . it is well known that this type of coding is canonically associated with the continued fraction expansion of @xmath0 ( see section 2 for the details ) . in particular , this allows to relate the level sets @xmath72 and @xmath73 to level sets given by means of the stern brocot growth rate @xmath74 of the nested sequences @xmath75 , and to level sets of certain diophantine growth rates @xmath76 and @xmath77 . these growth rates are given by ( assuming that these limits exist ) @xmath78 @xmath79 [ thm:2 ] let @xmath80 be given . if one of the limits in @xmath81 exists then also the other two do exist , and @xmath82 furthermore , @xmath83 exists if and only if @xmath84 exists , and if one of these exists then @xmath85 by theorem [ thm : main ] , it therefore follows that for each @xmath20 $ ] , @xmath86 note that the level sets @xmath87 have already been under consideration in @xcite . there they were introduced in terms of homological growth rates of hyperbolic geodesics ( see remark [ rem : vergleich ] ( 2 ) ) . clearly , theorem [ thm : main ] and proposition [ thm:2 ] consider the dynamical system associated with the finite alphabet , a system which is closely related to the farey map . now , our second main result gives a multifractal analysis for the system based on the infinite alphabet , and this system is closely related to the gauss map . in here , the relevant pressure function is the _ diophantine pressure _ @xmath88 , which is given by @xmath89 } q_{k}\left(\left[a_{1},\ldots , a_{k}\right]\right)^{-2\theta } , \textrm { for } \theta>\frac{1}{2}.\ ] ] we remark that a very detailed analysis of the function @xmath88 can be found in @xcite . our second main result is the following . [ thm : main3 ] _ ( see fig . [ cap : diaophantine - pressure ] ) _ the function @xmath88 has a singularity at @xmath90 , and @xmath88 is decreasing , convex and real - analytic on @xmath91 . furthermore , for @xmath92 we have@xmath93 also , the dimension function @xmath94 is real - analytic on @xmath95 , it is increasing on @xmath96 $ ] and decreasing on @xmath97 . in particular , @xmath94 has a point of inflexion at some point greater than @xmath98 and a unique maximum equal to @xmath2 at @xmath98 . additionally , @xmath99 , @xmath100 and @xmath101 . the paper is organized as follows . in section 2 we first recall two ways of coding elements of the unit interval . one is based on a finite alphabet and the other on an infinite alphabet , and both are defined in terms of the modular group . these codings are canonically related to regular continued fraction expansions , and we end the section by commenting on a 1 - 1 correspondence between stern brocot sequences and finite continued fraction expansions . in section 3 we introduce certain cocycles which are relevant in our multifractal analysis . in particular , we give various estimates relating these cocycles with the geometry of the modular codings and with the sizes of the stern brocot intervals . this will then enable us to prove the first part of proposition [ thm:2 ] . section 4 is devoted to the discussion of several aspects of the stern brocot pressure and its legendre transform . in section 5 we give the proof of theorem [ thm : main ] , which we have split into the parts _ the lower bound _ , _ the upper bound _ , and _ discussion of boundary points of the spectrum_. finally , in section 6 we give the proof of theorem [ thm : main3 ] by showing how to adapt the multifractal formalism developed in section 4 and 5 to the situation here . throughout , we shall use the notation @xmath102 to denote that for two non - negative functions @xmath103 and @xmath104 we have that @xmath105 is uniformly bounded away from infinity . if @xmath102 and @xmath106 , then we write @xmath107 . one immediately verifies that the results of theorem [ thm : main ] and proposition [ thm:2 ] can be expressed in terms of the farey map @xmath108 acting on @xmath41 $ ] , and then @xmath25 represents the multifractal spectrum of the measure of maximal entropy ( see e.g. @xcite ) . likewise , the results of theorem [ thm : main3 ] can be written in terms of the gauss map @xmath109 , and then in this terminology @xmath110 describes the lyapunov spectrum of @xmath109 . for the definitions of @xmath108 and @xmath109 and for a discussion of their relationship we refer to remark [ rem : fg ] . since the theory of multifractals started through essays of mandelbrot @xcite @xcite , frisch and parisi @xcite , and halsey et al . @xcite , there has been a steady increase of the literature on multifractals and calculations of specific multifractal spectra . for a comprehensive account on the mathematical work we refer to @xcite and @xcite . essays which are closely related to the work on multifractal number theory in this paper are for instance @xcite , @xcite , @xcite , @xcite , @xcite , @xcite and @xcite . we remark that brief sketches of some parts of theorem [ thm : main3 ] have already been given in @xcite . the results there do for instance not cover the boundary points of the spectra . furthermore , note that for the @xmath77spectrum partial results have been established in @xcite ( corollary 2 ) . let @xmath111 refer to the modular group acting on the upper half - plane @xmath112 . it is well known that @xmath113 is generated by the two elements @xmath28 and @xmath114 , given by@xmath115 defining relations for @xmath113 are @xmath116 , and a fundamental domain @xmath117 for @xmath113 is the hyperbolic quadrilateral with vertices at @xmath118 and @xmath119 . for @xmath120 such that @xmath121 , one easily verifies that @xmath122 is a subgroup of @xmath113 of index @xmath123 and that @xmath124 is a fundamental domain for @xmath125 , for @xmath126 the ideal triangle with vertices at @xmath127 and @xmath128 ( see fig . [ cap : funddomain ] ) . consider the two elements @xmath129 given by@xmath130 and let @xmath131 denote the free semi - group generated by @xmath69 and @xmath70 . it is easy to see that for @xmath132 we have that the cayley graph of @xmath131 with respect to @xmath133 coincides with the restriction to @xmath134 of the the cayley graph of @xmath125 with respect to @xmath133 ( see fig . [ cap : fig1 ] ) . let @xmath135 denote the full shift space on the finite alphabet @xmath136 , for @xmath137 given as above . also , let @xmath138 be equipped with the usual left - shift @xmath139 . then @xmath138 is clearly isomorphic to the completion of @xmath131 , where the completion is taken with respect to a suitable metric on @xmath131 ( see @xcite ) . one immediately verifies that the canonical map @xmath140,\\ ( x_{1},x_{2},\ldots ) & \mapsto & \lim_{n\rightarrow\infty}x_{1}\cdots x_{n}(z_{0}),\end{aligned}\ ] ] is 11 almost everywhere , in the sense that it is 21 on the rationals in @xmath41 $ ] and 11 on @xmath141 , for @xmath141 referring to the irrational numbers in @xmath142 $ ] . note that the stern brocot sequence @xmath143 coincides with the set of vertices at infinity of @xmath144 of word length @xmath145 , for each @xmath64 . for the infinite alphabet @xmath146 we define the shift space of finite type @xmath147 which we assume to be equipped with the usual left - shift @xmath148 . then there exists a canonical bijection @xmath149 , given by@xmath150 this coding is closely related to the continuous fraction expansion . namely , if @xmath151 then @xmath152 & \,\,\textrm{for } & x = a\\ { } [ 1,n_{1},n_{2},\ldots ] & \,\,\textrm{for } & x = b.\end{array}\right.\ ] ] also , if @xmath153\rightarrow[0,1]$ ] and @xmath154 are given by , for @xmath155 $ ] and @xmath156 , @xmath157 then by symmetry we have that @xmath158 . to overcome the fact that @xmath159 is not topological transitive , we also require the full shift space @xmath160 over @xmath161 . in here , @xmath162 and @xmath163 refers to the left - shift map on @xmath164 . clearly , @xmath160 is _ finitely primitive _ in the sense of @xcite , and we remark that this property is a necessary preliminary for the thermodynamical formalism used throughout this paper . note that the two shift spaces @xmath165 and @xmath160 are related by the 2 - 1 factor map @xmath166 , which is given by @xmath167 [ rem : fg ] note that the finite coding is in 1 - 1 correspondence to the coding of @xmath41 $ ] via the inverse branches @xmath168 and @xmath169 of the _ farey map _ @xmath108 . in here , @xmath168 and @xmath169 are given by @xmath170 and @xmath171 , for @xmath155 $ ] . one easily verifies that @xmath172 and @xmath173 , and hence @xmath138 can be interpreted as arising from a ` twisted farey map ' . similarly , one notices that @xmath174 is closely related to the coding of @xmath41 $ ] via the infinitely many branches of the gauss map @xmath109 , which is given by @xmath175 for @xmath155 $ ] . more precisely , we have that the dynamical system @xmath176 is a topological 21 factor of the dynamical system @xmath177 , where the factor map can be established either on the symbolic level via @xmath166 or on the geometric level via @xmath178 . the situation is summarized in the following commuting diagram . @xmath179_{\mathfrak{g } } & & \ar[ll]_{\pi_{\mathrm{cf}}}\overline{\sigma}\ar[d]_{\overline{\sigma } } & & \ar[ll]_{p}\sigma^ * \ar[rr]^{\pi^*}\ar[d]_{\sigma^ * } & & { \mathbb { i}}\ar[d]^{{\mathfrak{g}_s } } \ar[rr]^{\mathfrak{f } } & & { \mathbb { i}}\ar[d]^{{\mathfrak{g } } } \\ { \mathbb { i } } & & \ar[ll]^{\pi_{\mathrm{cf } } } \overline{\sigma } & & \ar[ll]^{p}\sigma^ * \ar[rr]_{\pi^ * } & & { \mathbb { i}}\ar[rr]_{\mathfrak{f } } & & { \mathbb { i}}\\ } \end{aligned}\ ] ] in here , @xmath180 denotes the twisted gauss map and @xmath181 is given by @xmath182 $ ] for @xmath183 . note that both @xmath181 and @xmath184 are bijections and that @xmath185 . we end this section by showing that there is a 11 correspondence between the elements of the stern brocot sequence and finite continued fraction expansions . this will turn out to be useful in the sequel . for @xmath186 , let @xmath187 refer to the set all @xmath188-tuples of positive integers which add up to @xmath5 and whose @xmath188-th entry exceeds @xmath2 . that is , @xmath189 since @xmath190 , we can identify an element @xmath191 in a unique way with the finite continued fraction expansion @xmath192 $ ] . also , one easily verifies that for @xmath193,@xmath194 [ lem : tnbycf ] for all @xmath186 we have@xmath195=\mathfrak{t}_{n-1}\setminus\mathfrak{t}_{n-2}= \left\ { \frac{s_{n-1,2\ell}}{t_{n-1,2\ell}}:1\leq\ell\leq2^{n-2}\right\ } .\ ] ] furthermore , if @xmath196\in\mathfrak{t}_{n}\setminus\mathfrak{t}_{n-1}$ ] then its two siblings in @xmath197 are , for @xmath198 , @xmath199 \quad \textrm{and } \quad \frac{s_{n+1,v}}{t_{n+1,v}}=\left[a_{1},a_{2 } , \ldots , a_{m-1},a_{m}-1,2\right].\ ] ] for the first part of the lemma note that the second equality follows by definition of @xmath200 . the first equality is obtained by induction as follows . we clearly have @xmath201\right\ } = \mathfrak{t}_{1}\setminus\mathfrak{t}_{0}$ ] . then assume that the assertion holds for @xmath202 . since the sets @xmath203 are @xmath204invariant it follows for @xmath205,@xmath206 for @xmath207 \in\mathfrak{t}_{n-2}\setminus\mathfrak{t}_{n-3}$ ] we have by the inductive assumption that @xmath208 , and hence @xmath209\right ) & = & \frac{1}{1/\left[a_{1},\ldots , a_{k}\right]+1}=\left[a_{1}+1,a_{2},\ldots , a_{k}\right]\in a_{k}^{n},\\ bs\left(\left[a_{1},\ldots , a_{k}\right]\right ) & = & \frac{1}{1+\left[a_{1},\ldots , a_{k}\right]}=\left[1,a_{1},a_{2},\ldots , a_{k}\right]\in a_{k+1}^{n}.\end{aligned}\ ] ] by combining the two latter observation , we obtain @xmath210.\ ] ] since @xmath211 the first part of the lemma follows . for the second part note that by the above @xmath212,\ , \left[a_{1},a_{2},\ldots , a_{m}-1,2\right]\in \mathfrak{t}_{n+1}\setminus\mathfrak{t}_{n}.\ ] ] since @xmath213 $ ] , @xmath214 $ ] , and @xmath215 $ ] are consecutive neighbours in @xmath143 , the lemma follows . note that @xmath28 can be written alternatively also in terms of denominators of approximants as follows . @xmath216\right)^{-2\theta}.\ ] ] in order to see this , note that for @xmath217 , @xmath218 on the other hand , using the recursive definition of @xmath219 , we have for @xmath220,@xmath221 therefore , by taking logarithms , dividing by @xmath5 and letting @xmath5 tend to infinity , we obtain @xmath222 hence , using lemma [ lem : tnbycf ] , the result follows . in this section we introduce the dynamical cocycles which will be crucial in the multifractal analysis to come . we show that these cocycles are closely related to stern brocot intervals and continued fractions . finally , we give the proof of the first part of proposition [ thm:2 ] . we remark that the results in this section could be obtained alternatively by using elementary estimates for countinued fractions only . instead , we have put some emphasis on obtaining these results by making use of the hyperbolic metric @xmath223 on @xmath224 . the intension here is that this should make it easier to follow the later transfer of the results of @xcite , which were derived in terms of kleinian groups , into the language of stern brocot intervals and continued fractions . recall that the poisson kernel @xmath225 for the upper half - plane is given by@xmath226 with @xmath133 defined as in section [ sec : coding ] , the cocycle @xmath227 associated with the finite alphabet is given by@xmath228 we remark that @xmath229 is continuous with respect to the standard metric . also , it is well known that @xmath230 is equal to the hyperbolic distance of @xmath133 to the horocycle through @xmath231 based at @xmath232 . furthermore , note that in terms of the theory of iterations of maps , @xmath229 is equal to the logarithm of the modulus of the derivative of the ` twisted farey map ' , mentioned in remark [ rem : fg ] . similar , we define the cocycle @xmath233 associated with the infinite alphabet as follows . for @xmath234 such that @xmath235 , let @xmath236 be given by@xmath237 one immediately verifies that @xmath238 is equal to the hyperbolic distance of @xmath133 to the horocycle based at @xmath239 containing either the point @xmath240 ( if @xmath188 is odd ) or @xmath241 ( if @xmath188 is even ) . note that in terms of the theory of iterations of maps , the function @xmath236 is clearly an analogue of the logarithm of the modulus of the derivative of the gauss map . throughout , we also require the potential function @xmath242 , which is given by @xmath243 note that @xmath244 . finally , the relevant potentials for the shift space @xmath160 are the functions @xmath245 in here , @xmath246 refers to the inverse branch with respect to @xmath247 of the @xmath248-@xmath2 factor map @xmath166 introduced in section [ 2.2 ] . more precisely , we have for @xmath249 such that @xmath250 , @xmath251 the following lemma relates the euclidean sizes of the stern brocot intervals to the hyperbolic distances of @xmath133 to the elements in the orbit @xmath252 . [ lem : uniformlyexcursion ] for each @xmath64 and @xmath253 such that @xmath254 , we have @xmath255 in here , @xmath256 is defined by @xmath257 . for @xmath258 the statement is trivial . for @xmath186 , we first consider the case @xmath259 . if @xmath260 , then @xmath261 is equal to either @xmath262 ( for @xmath263 ) or @xmath264 ( for @xmath265 ) . also , note that for the modulus of the conformal derivative we have @xmath266 combining these two observations , we obtain @xmath267}\right|\asymp\left|\left(g^{-1}\right)'|_{t_{n}(x)}\right|^{-1}\asymp e^{-d(z_{0},g(z_{0}))}\asymp e^{-d(z_{0},gx_{n}(z_{0}))}.\ ] ] this proves the assertion for @xmath259 . for the general situation we only consider the case @xmath268 . the remaining cases can be dealt with in a similar way . in this case @xmath269 , and the above implies , for @xmath270 , @xmath271 also , by using the well - known elementary fact that @xmath272 for @xmath273 and @xmath274 , one immediately obtains for @xmath275 , @xmath276 finally , one also immediately verifies that for @xmath277 , @xmath278 combining the three latter observations , the statement of the lemma follows . the previous lemma has the following immediate implication . [ cor : vergleichtsni ] for each @xmath64 and @xmath279 such that @xmath280 , we have @xmath281 the following lemma relates the cocycle @xmath236 to the sizes of the stern brocot intervals and to the denominators @xmath282 of the approximants . [ lem : vergleichbar ] for each @xmath283 and @xmath253 we have , with @xmath284 , @xmath285 we only consider the case @xmath188 even and @xmath286 . the remaining cases can be obtained in a similar way . let @xmath287 , and note that then @xmath288 combining this with the fact that for @xmath289 we have @xmath290 ( which follows since on @xmath68 we have that @xmath291 is comparable to @xmath292 ) , we obtain @xmath293 finally , note that by lemma [ lem : uniformlyexcursion ] and since @xmath294 , we have @xmath295 combining these estimates , the lemma follows . we are now in the position to prove the first part of proposition [ thm:2 ] . [ proof of first part of proposition [ thm:2 ] ] the equalities @xmath296 and @xmath297 are immediate consequences of the following well known diophantine inequalities . for all @xmath298 $ ] and @xmath283 , we have ( see e.g. @xcite ) @xmath299 in order to show that @xmath300 , let @xmath301 , for @xmath253 and @xmath283 . obviously , @xmath302 is a subsequence of the sequence @xmath303 . therefore , if @xmath304 exists then so does @xmath305 , and both limits must coincide . for the reverse , suppose that @xmath305 exists such that @xmath306 . let @xmath256 be defined as in the statement of lemma [ lem : uniformlyexcursion ] , and put @xmath307 . by combining ( [ eq : asympparabexcursion ] ) and lemma [ lem : vergleichbar ] , we then have @xmath308 this gives the upper bound , and hence finishes the proof in particular for @xmath309 . for the opposite inequality we can therefore assume without loss of generality that @xmath310 . first , observe that @xmath311 hence , it is now sufficient to show that @xmath312 , or what is equivalent @xmath313 . for this , observe that @xmath314 if we would have that @xmath315 , then there exists a subsequence @xmath316 such that @xmath317 , for some @xmath318 $ ] . it follows that @xmath319 , and hence by combining this with the calculation above , we obtain @xmath320 this shows that @xmath321 , and hence finishes the proof . the main goal in this section is to derive various analytic properties of the stern brocot pressure function @xmath28 . these properties are derived by considering the pressure functions associated with the systems @xmath322 and @xmath323 . in order to introduce these functions , let @xmath324 refer to the set of all @xmath5cylinders @xmath325 likewise , let @xmath326 ( resp . @xmath327 ) refer to the set of @xmath5-cylinders for the system @xmath328 ( resp . @xmath329 ) . the pressure function @xmath330 associated with @xmath138 is then given by @xmath331 also , for the system @xmath174 we define the pressure functions @xmath332 and @xmath333 , for @xmath334 and @xmath335 continuous , by @xmath336 finally , the pressure functions @xmath337 and @xmath338 associated with @xmath339 are given completely analogous , for @xmath334 and @xmath340 continuous , by @xmath341 clearly , by recalling the definitions of @xmath342 and @xmath343 in section 3 , we immediately have that @xmath344 . in this subsection we employ important results of hanus , mauldin and urbaski obtained in @xcite and @xcite . the results here will be crucial cornerstones in our subsequent analysis of the stern brocot pressure . studies of analytic properties of pressure functions are usually based on the existence of certain gibbs measures , here on @xmath174 and @xmath323 . the existence of these measures in our situation here is guaranteed by the following proposition , which essentially follows from a result in @xcite . [ prophmu2 ] for each @xmath345 , @xmath346 , and for @xmath347 , there exists a unique completely ergodic @xmath163invariant gibbs measure @xmath348 associated with the potential @xmath349 . that is , we have for all @xmath350 and @xmath351 , @xmath352 in particular , the borel measure @xmath353 is an ergodic @xmath354invariant gibbs measure on @xmath174 such that for all @xmath355 and @xmath356 , @xmath357 the measure @xmath358 is unique with respect to this property , and @xmath359 . by @xcite ( lemma 3.4 ) , the cocycle @xmath236 is hlder continuous in the sense that there exists @xmath360 such that for each @xmath64 , @xmath361 clearly , we also immediately have that @xmath362 is hlder continuous . furthermore , the following summability condition holds for @xmath345 , @xmath346 , and for @xmath347 , @xmath363 hence , all preliminaries of @xcite ( corollary 2.10 ) are fulfilled , which then gives the existence of a unique invariant gibbs measure @xmath348 with properties as stated in the proposition . immediate consequences of the definition of @xmath364 and the definition @xmath353 are that @xmath358 is @xmath365invariant , that @xmath358 fulfills the gibbs property ( [ eq : gibbs*property2 ] ) , and that the equality @xmath359 is satisfied . to prove ergodicity of @xmath366 , let @xmath367 such that @xmath368 . we then have @xmath369 , for @xmath370 such that @xmath371 . this gives @xmath372 . since @xmath348 is completely ergodic , which by definition means that @xmath348 is ergodic with respect to @xmath373 for all @xmath374 , it follows @xmath375 . the @xmath163-invariance of @xmath348 then implies that @xmath376 . consequently , it follows that @xmath377 . the following proposition employs yet another result of hanus , mauldin and urbaski , obtained in their spectral analysis of the perron frobenius operator . [ prophmu3 ] the pressure function @xmath333 is a convex , decreasing and real - analytic function with respect to both coordinates in the second coordinate @xmath333 is strictly decreasing to @xmath378 . in particular , there hence exists a positive real analytic function @xmath379 on @xmath29 such that @xmath380 . furthermore , for the derivative of @xmath379 at @xmath381 we have @xmath382 in here , @xmath383 refers to the unique @xmath354invariant gibbs measure associated with the potential @xmath384 . also , @xmath385 refers to the @xmath386invariant probability measure on @xmath138 absolutely continuous to @xmath387 , whose existence is guaranteed by _ kac s formula _ ( @xcite ; see the proof ) . first , note that it is sufficient to verify the statements in the proposition for @xmath323 only . then note that @xmath388 and @xmath362 are hlder continuous , that the summability condition ( [ summability ] ) is satisfied for all @xmath389 , and that @xmath390 hence , we can apply @xcite ( proposition 6.5 ) ( see also @xcite ( proposition 2.6.13 ) ) , from which it follows that @xmath391 is real - analytic on @xmath392 . next , note that for the partial derivatives of @xmath338 we have @xmath393 this shows that @xmath338 as a function in the second coordinate is strictly decreasing , which then gives the existence of a real - analytic function @xmath394 for which @xmath395 , for all @xmath345 . now , the first equality in ( [ derive ] ) follows from the implicit function theorem . the second equality is a consequence of _ kac s formula _ ( @xcite ) , which guarantees that there exists a @xmath386invariant measure @xmath396 on @xmath138 , given by @xmath397 in here , @xmath398 refers to the canonical injection which maps an element of @xmath174 to its representation by means of the finite alphabet of @xmath138 . we remark that kac s formula gives in fact a 11 correspondence between @xmath386invariant measures on @xmath138 and @xmath354invariant measures on @xmath174 . now , one immediately verifies that @xmath399 hence , this allows to define @xmath400 . in particular , we then have that @xmath401 , from which the second equality follows . finally , note that since @xmath338 and @xmath402 coincide , in the integrals above we can replace @xmath388 and @xmath362 by @xmath236 and @xmath364 . this finishes proof of the proposition . note that the measure @xmath385 in proposition [ prophmu3 ] is in fact a weak gibbs measure for the potential @xmath403 . therefore , the results of @xcite are applicable , and hence in this way one could immediately obtain some large deviation results for the situation here . in this subsection we employ the results of the previous subsection , in order to derive analytic properties of the stern brocot pressure @xmath28 and its legendre transform @xmath27 . a key preliminary observation is stated in the following proposition , which shows that the stern brocot pressure @xmath28 coincides with the function @xmath379 obtained in proposition [ prophmu3 ] . [ prop : p = beta ] for @xmath345 , we have @xmath404 let @xmath387 be the measure obtained in proposition [ prophmu3 ] , for @xmath381 fixed . first , recall from the proof of proposition [ prophmu3 ] that the measure class of @xmath405 contains a @xmath386invariant probability measure @xmath385 on @xmath138 for which @xmath401 . secondly , note that for the measure @xmath406 we have by _ abramov s formula _ ( @xcite , @xcite ) that the measure theoretical entropies @xmath407 and @xmath408 are related by @xmath409 . thirdly , by applying _ pinsker s result on relative entropies _ ( @xcite ) to our situation here , we obtain that the relative entropy @xmath410 of @xmath405 vanishes . this gives @xmath411 . combining these observations with the usual variational principle ( @xcite ) , it now follows @xmath412 in here , the latter equality is obtained as follows . note that @xmath413 is an equilibrium measure on @xmath160 for the potential @xmath414 . also , @xmath413 fulfills the gibbs property [ eq : gibbs*property ] , and @xmath415 . next , recall _ sarig s variational principle _ ( @xcite ) which states that for @xmath340 hlder continuous , @xmath416 in here , @xmath417 refers to the set of @xmath163invariant borel probability measures on @xmath323 . applying this to the situation here , we obtain @xmath418 an elementary rearrangement then gives the result . for the reverse inequality , first note that we can induce @xmath419 on the set @xmath420 . for the resulting induced system the return time to @xmath421 of a point @xmath422 is given by @xmath423 . define @xmath424 , and let @xmath425 be an ergodic equilibrium measure for the potential @xmath403 , that is @xmath426 . in this situation we necessarily have that @xmath427 , and this can be seen as follows . first , we show that @xmath428 implies that @xmath429 is equal to either @xmath430 or @xmath431 , where @xmath430 ( resp . @xmath431 ) refers to the dirac measure at the periodic point @xmath432 ( resp . @xmath433 ) . namely , if @xmath434 then we immediately have @xmath435 . on the other hand , if @xmath436 then ergodicity of @xmath437 gives @xmath438 . now , since @xmath439 for all @xmath440 , where @xmath441 if and only if @xmath442 , we have @xmath443 for @xmath429-almost every @xmath0 , and this again implies that @xmath444 . thus , if @xmath428 then @xmath444 . this shows that @xmath445 @xmath446 , giving @xmath447 , and hence contradicting the fact @xmath448 . therefore , we can assume without loss of generality that @xmath427 . we can now use kac s formula once more , which guarantees that there exists a @xmath354invariant probability measure @xmath449 in the measure class of @xmath437 , such that @xmath450 and @xmath451 for @xmath452 , we argue similar as above and obtain @xmath453 the following proposition collects the properties of @xmath28 and @xmath27 which will be crucial in the analysis to come . [ pro : analyticpropertiesp ] 1 . the stern brocot pressure function @xmath28 coincides with the pressure function @xmath330 associated with @xmath138 . 2 . @xmath28 is convex and non - increasing on @xmath53 and real - analytic on @xmath29 . @xmath447 , for all @xmath454 . @xmath28 is differentiable throughout @xmath53 . the domain of @xmath27 is equal to @xmath455 $ ] , where @xmath456 6 . we have @xmath457 . we have @xmath458 8 . we have @xmath459 . for the proofs of ( 7 ) and ( 8) the following lemma will turn out to be useful . [ lem : qgammainequalitystrict ] for each @xmath460 \in(0,1)$ ] and @xmath461 we have , with @xmath462 , @xmath463 for @xmath464 , and @xmath465 , @xmath466 we give a proof by complete induction of the slightly stronger inequality @xmath467 in which @xmath468 denotes the kronecker symbol . + first note that @xmath469 , @xmath470)=1\leq\gamma^{1}\rho^{1 - 1}$ ] , and if @xmath471 then one immediately verifies that @xmath472=a_{1}\leq\gamma^{a_{1}}\rho^{a_{1}-1}\rho^{-1}$ ] . also , for @xmath274 we have @xmath473)=f_k\leq \gamma^{k}=\gamma ^{\tau_k}\rho^{\tau_k - k},\label{eq : abschgoldenmean}\ ] ] where @xmath474 refers to the @xmath475-th member of the fibonacci sequence @xmath476 , given by @xmath477 for all @xmath274 . recall that @xmath478 . now suppose that ( [ eq : slightlystronger ] ) holds for some @xmath464 and for all @xmath479 . it is then sufficient to consider the following two cases . 1 . [ enu : case1 ] if @xmath480 such that @xmath481 and @xmath482 , for all @xmath483 and some @xmath484 and @xmath485 , then @xmath486 and @xmath487 . hence , an elementary calculation gives @xmath488 2 . if @xmath489 , then either @xmath490 for @xmath491 , or there exists @xmath484 such that @xmath481 and @xmath492 for all @xmath493 with @xmath494 . in the first case we use ( [ eq : abschgoldenmean ] ) , whereas in the second case we employ ( [ enu : case1 ] ) , and obtain @xmath495\right ) & = & q_{k+2}\left(\left[a_{1},\ldots , a_{k},1,1\right]\right)\\ & \leq & \gamma^{\tau_{k+1}}\rho^{\tau_{k+1}-k-1 } \rho^{-1}.\end{aligned}\ ] ] for @xmath496 , the inequality follows by induction over @xmath497 , using ( [ enu : case1 ] ) and the fact that @xmath498\right)=q_{k+2 } \left(\left[a_{1 } , \ldots , a_{k+1}-1,1\right]\right)$ ] . before giving the proof of proposition [ pro : analyticpropertiesp ] , we remark that the statements ( 7 ) and ( 8) in proposition [ pro : analyticpropertiesp ] are in fact equivalent . nevertheless , we shall prove these two statements separately , where the proof of ( 7 ) primarily uses ergodic theory , whereas the proof of ( 8) is of elementary number theoretical nature . [ proof of proposition [ pro : analyticpropertiesp ] ] * _ ad ( 1)_. * the assertion is an immediate consequence of ( [ eq : asympparabexcursion ] ) and corollary [ cor : vergleichtsni ] . * _ ad ( 2)_. * the assertion follows immediately by combining proposition [ prophmu3 ] and proposition [ prop : p = beta ] . alternatively , the statement can also be derived from proposition 2.1 in @xcite . * _ ad ( 3)_. * by definition of @xmath28 we have @xmath499 . also , by ( 2 ) we know that @xmath28 is non - increasing therefore , it is sufficient to show that @xmath28 is non - negative , and indeed this follows since@xmath500 * _ ad ( 4)_. * in order to determine the left derivative @xmath501 of @xmath28 at @xmath2 , recall from proposition [ prophmu3 ] that @xmath387 refers to the unique gibbs measure on @xmath174 such that @xmath502 , for all @xmath503 . for each @xmath64 , let us fix an element @xmath504 such that @xmath505 , for @xmath506 . we then have by lemma [ lem : vergleichbar ] , @xmath507 on the other hand , we have for all @xmath508 $ ] , @xmath509 this shows that @xmath510 , and hence @xmath28 is differentiable everywhere . * _ ad ( 5)_. * since @xmath511 , the upper bound of the domain of @xmath27 is equal to @xmath54 . for the lower bound @xmath512 of the domain we have by @xcite ( proposition 2.3 ) , @xmath513 we are left with to determine the actual value of @xmath514 . for this , first note that for the linear combination @xmath515 of the dirac measures @xmath516 and @xmath517 at the periodic points @xmath518 and @xmath519 , an elementary calculation shows that @xmath520 . this implies that @xmath521 . for the reverse inequality note that @xmath522 , for all @xmath523 , where @xmath524 refers to the set of @xmath386invariant borel probability measures on @xmath138 . in order to calculate the right hand side of the latter inequality , recall that the smallest interval in @xmath42 has the length @xmath525 . using this observation and corollary [ cor : vergleichtsni ] , we obtain @xmath526 note that in here the supremum is achieved at for instance any noble number in @xmath527 , that is at numbers whose continued fraction expansion eventually consists of @xmath2 s only . * _ ad ( 6)_. * the result in ( 3 ) implies that @xmath528 therefore , it is sufficient to show that @xmath2 is the least zero of @xmath28 . for this assume by way of contradiction that @xmath529 , for some @xmath530 . since @xmath28 is non - increasing , it follows that @xmath28 vanishes on the interval @xmath531 . but this contradicts the fact that @xmath28 is real - analytic on @xmath29 and positive at for instance @xmath54 . * _ ad ( 7)_. * for all @xmath64 and @xmath532 , we have @xmath533 therefore , @xmath534 which implies that @xmath535 , for all @xmath20 $ ] . hence , in order to verify that @xmath536 it is sufficient to show that this limit is non - negative . for this , let @xmath537 and recall that by the variational principle ( cf . @xcite ) we have for each @xmath20 $ ] that there exists @xmath538 such that @xmath539 furthermore , by @xcite ( proposition 2.3 ) we have @xmath540 . therefore , if @xmath523 denotes a weak limit of some sequence @xmath541 for @xmath542 tending to @xmath62 from below , then the lower semi - continuity of the entropy ( cf . @xcite ) gives @xmath543 note that we clearly have @xmath544 . now , the final step is to show that for the discrete measure @xmath545 considered in the proof of ( 5 ) we have @xmath546 this will be sufficient , since @xmath547 . therefore , suppose by way of contradiction that there exists @xmath548 such that @xmath549 let us first show that @xmath550 , for some @xmath551 . if this would not be the case , then the @xmath386invariance of @xmath552 would imply @xmath553 and hence we obtain by induction that @xmath554 . this contradicts our assumption @xmath555 , showing that @xmath556 . we can now continue the above argument as follows . since @xmath557 is convex , we can assume without loss of generality that @xmath552 is ergodic . this then immediately implies that @xmath558 for @xmath552-almost every @xmath559 , and furthermore that for some @xmath551 and @xmath5 sufficiently large , @xmath560 let us fix @xmath561 with this property , and define @xmath562 as in lemma [ lem : qgammainequalitystrict ] . combining lemma [ lem : tnbycf ] and inequality ( [ eq : ergodmean ] ) , it follows @xmath563 . hence , using corollary [ cor : vergleichtsni ] , lemma [ lem : vergleichbar ] and lemma [ lem : qgammainequalitystrict ] , @xmath564 * _ ad ( 8)_. * first note that @xmath565 , for each @xmath186 and @xmath566 . this implies that @xmath567 and @xmath568 are both less than @xmath569 . hence , using lemma [ lem : tnbycf ] and lemma [ lem : qgammainequalitystrict ] , it follows for @xmath570 and @xmath571 , @xmath572\right)^{-2\theta}\\ & \leq & 2\sum_{k=1}^{n}{\binom{n-1 } { k-1}}\left(\gamma^{n+1}\rho^{n+1-k-1}\right)^{-2\theta}\\ & = & 2\gamma^{-2\theta(n+1)}\sum_{k=0}^{n-1}{\binom{n-1 } { k}}\left(\rho^{n-1-k}\right)^{-2\theta}\\ & = & 2\gamma^{-2\theta(n+1)}\sum_{k=0}^{n-1}{\binom{n-1 } { k}}\left(\rho^{-2\theta}\right)^{n-1-k}\\ & \leq & 2\gamma^{-2\theta(n+1)}\left(1+\rho^{-2\theta}\right)^{n-1}.\end{aligned}\ ] ] recalling the definition of @xmath28 , we then obtain @xmath573 for the lower bound , note that @xmath574 since @xmath575 , it therefore follows @xmath576 by combining these two bounds for @xmath577 and then letting @xmath578 tend to @xmath378 , the proposition follows . in this section we give the proof of theorem [ thm : main ] , which we have split up into the three separate parts _ the lower bound _ , _ the upper bound _ and _ discussion of boundary points of the spectrum_. we begin with the following important preliminary remarks . [ rem : vergleich ] @xmath579 + ( 1 ) note that by corollary [ cor : vergleichtsni ] and lemma [ lem : vergleichbar ] , we have for @xmath580 and @xmath581 ( assuming in each case that the limit exists ) , @xmath582 ( 2 ) recall that in @xcite and @xcite we in particular considered oriented geodesics @xmath583 from @xmath128 to [ 0 , 1 ) , and coded these by means of their intersections with the tesselation given by the @xmath131-orbit of the fundamental domain @xmath117 . more precisely , if @xmath584 ends at @xmath585 such that @xmath584 intersects @xmath586 in succession , with @xmath587 for all @xmath588 , then @xmath589 is coded by the infinite word @xmath590 . clearly , this type of coding is analogous to the finite coding represented by @xmath138 . hence , the results of @xcite and @xcite for the hausdorff dimensions of the level sets @xmath591 can immediately be transfered to the situation in this paper , and in this way we obtain that for each @xmath592 , @xmath593 therefore , the following proof of theorem [ thm : main ] will in particular also give an alternative proof of the identity in ( [ eq : oldspec ] ) . let us also emphasize that a straight forward inspection of the arguments in the general multifractal analysis of @xcite shows that there we did not make full use of the group structure of the kleinian group . in fact , the arguments there exclusively consider certain rooted sub - trees of the cayley graph of the kleinian group , and therefore they continue to hold if the underlying algebraic structure is only a semi - group acting on hyperbolic space , rather than a group . therefore , the main results of this general multifractal analysis for growth rates can be applied immediately to the setting in this paper . in this way one also immediately obtains that @xmath28 is differentiable everywhere , real - analytic on @xmath29 and equal to @xmath54 otherwise . [ lem : inclusion ] for each @xmath592 there exists a unique gibbs measure @xmath594 on @xmath174 such that for @xmath595 we have @xmath596 in here , the function @xmath597 is given by @xmath537 . the existence of the unique ergodic gibbs measure @xmath594 has already been obtained in proposition [ prophmu2 ] . as shown in proposition [ prophmu3 ] , the significance of @xmath594 is that it allows to represent the lyapunov exponent @xmath542 in terms of @xmath236 and @xmath364 as follows . @xmath598 using remark [ rem : vergleich ] ( 1 ) , it follows that @xmath599 . for the following lemma recall that the hausdorff dimension @xmath600 of a probability measure @xmath552 on some metric space is given by @xmath601 for each @xmath592 we have , with @xmath602 , @xmath603 the first inequality follows , since by ergodicity of @xmath594 we have@xmath604 the second inequality is an immediate consequence of lemma [ lem : inclusion ] . [ lem : lowerbound ] for each @xmath592 we have@xmath605 the aim is to show that the local dimension of @xmath606 exists and is equal to @xmath607 , for each @xmath608 . for this , let @xmath609\cap{\mathbb { i}}$ ] for @xmath610 and @xmath253 , and define @xmath611 we obviously have that @xmath612 is uniformly bounded from above , and hence @xmath613 . combining the gibbs property of @xmath594 , lemma [ lem : vergleichbar ] , ( [ eq : alpha*-alpha ] ) and ( [ eq : alpha - alphaalpha ] ) , it follows for @xmath614-almost every @xmath0 , + + @xmath615@xmath616 the reverse inequality for the ` @xmath617 ' is obtained by similar means , and we omit its proof . [ lem : upperbound ] for each @xmath592 we have@xmath618 note that @xmath619 , for each @xmath592 . by combining this with the gibbs property of @xmath594 , it follows that for each @xmath620 and @xmath621 such that @xmath622 , we have for @xmath5 sufficiently large , @xmath623 therefore , for the sequence of balls @xmath624 with radii @xmath625 and centre @xmath232 , which tends to @xmath626 for @xmath5 tending to infinity , we have @xmath627 applying the mass distribution principle , the proposition follows . [ cor : dimupperbound ] for each @xmath592 we have @xmath628 the assertion is an immediate consequence of combining lemma [ lem : upperbound ] and the fact @xmath629 @xmath579 + _ the case @xmath309 _ : recall the two classical results of lvy and khintchin mentioned in the introduction . > from these we immediately deduce that @xmath630 . also , recall that by proposition [ pro : analyticpropertiesp ] ( 6 ) we have that @xmath631 . this shows that @xmath632 , and hence gives that the dimension function @xmath25 is continuous from the right at @xmath54 . + in order to show that @xmath633 , we argue as follows . for @xmath309 , we already know that @xmath634 and that @xmath635 , for @xmath636-almost every @xmath637 . hence , lvy s result gives that , if @xmath636 is absolutely continuous with respect to the lebesgue measure @xmath3 on @xmath527 then @xmath633 . hence , it remains to show that @xmath636 has this property . for this , consider some @xmath638 for @xmath64 , and fix @xmath581 and @xmath283 such that @xmath639 . using the gibbs property of @xmath640 and lemma [ lem : vergleichbar ] , we obtain @xmath641 _ the case @xmath642 _ : in order to show that the dimension function @xmath25 is continuous from the left at @xmath62 , we proceed as follows . proposition [ pro : analyticpropertiesp ] ( 7 ) implies that @xmath643 . using monotonicity of the hausdorff dimension together with lemma [ lem : upperbound ] , it then follows @xmath644 hence , we have @xmath645 , which gives that @xmath25 is continuous from the left at @xmath62 . + finally , for the left derivative of @xmath25 at @xmath62 , note that a straight forward computation of the derivative of @xmath25 on the interval @xmath646 shows that @xmath647 . since @xmath648 tends to @xmath378 as @xmath542 approaches @xmath62 , it follows @xmath649 in this section we outline the necessary changes which have to be implemented in the proof of theorem [ thm : main ] in order to derive theorem [ thm : main3 ] . the analytic properties of @xmath88 as stated in theorem [ thm : main3 ] can be obtained as follows . in section 4 replace the function @xmath650 ( resp . @xmath651 ) by the function @xmath652 ( resp . @xmath653 ) constant equal to @xmath2 . in this way we obtain for the pressure function @xmath337 associated with @xmath323 , @xmath654 also , note that by lemma [ lem : vergleichbar ] we have @xmath655 ( below , we shall specify the domain of @xmath656 ) . hence , combining these observations with proposition [ prophmu3 ] adapted to the situation here , the alleged analytic properties of @xmath88 follow . also , using the same strategy in section [ section 5.1 ] and [ section 5.2 ] , that is replacing in there the function @xmath364 by the function @xmath657 , one immediately obtains @xmath658 ( below , we shall specify the domain of @xmath110 ) . for clarifying the range of @xmath656 and of @xmath110 , and for the discussion of the boundary points of @xmath110 , we first remark that @xmath88 has a singularity at @xmath90 . this follows , since for every approximant @xmath207 $ ] we have ( see e.g. @xcite ) @xmath659\right)\leq2^{k}\prod_{i=1}^{k}a_{i},\ ] ] which immediately gives @xmath660 here , @xmath661 refers to the riemann zeta - function @xmath662 . this shows that @xmath663 and @xmath664 both tend to infinity for @xmath578 tending to @xmath90 from above . from this we deduce that @xmath665 is well defined for arbitrary large values of @xmath542 , and also that @xmath666 in order to see that the domain of @xmath667 is the interval @xmath668 and that @xmath669 , it is now sufficient to show that @xmath670 indeed , on the one hand we have @xmath671}q_{k } \left(\left[a_{1},\ldots , a_{k}\right]\right)^{-2\theta } & \leq & \lim_{k\to\infty}-\frac{1}{k}2\theta\log q_{k}\left(\gamma\right)=-2\theta\log\gamma.\end{aligned}\ ] ] on the other hand , using lemma [ lem : qgammainequalitystrict ] and [ lem : tnbycf ] , we observe for @xmath672 and @xmath673 , + @xmath674}q_{k}\left(\left[a_{1},\ldots , a_{k}\right]\right)^{-2\theta}}$]@xmath675 a combination of these two observations gives the statement in ( [ eq : toshow ] ) . in order to prove continuity of @xmath94 at @xmath62 , note that by arguing similar as in the proof of lemma [ lem : upperbound ] , we obtain for @xmath676 , @xmath677 combining this with the monotonicity of hausdorff dimension , it follows @xmath678 finally , the same argument as used in section [ sub : discussion - of - boundary - points ] for determining the limit behaviour of @xmath679 , gives that for the left derivative of @xmath94 at @xmath62 we have @xmath680 this finishes the proof of theorem [ thm : main3 ] . u. frisch and g. parisi . on the singularity structure of fully developed turbulence . in _ turbulence and predictability in geophysical fluid dynamics and climate dynamics _ , pages 8488 , north holland amsterdam , 1985 . m. pollicott and h. weiss . multifractal analysis of lyapunov exponent for continued fraction and manneville - pomeau transformations and applications to diophantine approximation . , 207(1):145171 , 1999 .

the spin-1/2 alternating heisenberg chain system nacusbo features two relevant exchange couplings : within the structural cuo dimers and between the dimers . motivated by the controversially discussed nature of , we perform extensive density - functional - theory ( dft ) calculations , including dft+ and hybrid functionals . fits to the experimental magnetic susceptibility using high - temperature series expansions and quantum monte carlo simulations yield the optimal parametersk andk with the alternation ratio . for the closely related system nacuteo , dft yields substantially enhanced , but weaker . the comparative analysis renders the buckling of the chains as the key parameter altering the magnetic coupling regime . numerical simulation of the dispersion relations of the alternating chain model clarify why both antiferromagnetic and ferrromagnetic can reproduce the experimental magnetic susceptibility data .

1402.1091

the vibrant research on magnetic insulators keeps on delivering new examples of exotic magnetic behaviors and unusual magnetic ground states ( gss).@xcite two prominent examples are the spin - liquid system herbertsmithite cu@xmath0zn(oh)@xmath2cl@xmath1 , featuring a kagome lattice of @xmath9 spins,@xcite or the recently discovered ba@xmath0cusb@xmath1o@xmath10 , where the magnetism is likely entangled with the dynamical jahn - teller distortion.@xcite cuprates are a particularly promising playground to study low - dimensional magnetism , since they often combine the quantum spin @xmath9 ensured by the cu @xmath11 electron configuration and the low - dimensionality of the underlying magnetic model . the latter is ensued by the unique variety of lattice topologies realized in cuprates , which includes geometrically frustrated lattices , where quantum fluctuations are additionally enhanced by the competing magnetic interactions . the simplest example of a quantum gs that lacks a classical analog is the quantum - mechanical singlet . such a gs is found experimentally , e.g. , in csv@xmath1o@xmath12 ( ref . ) , cute@xmath1o@xmath12 ( ref . ) , cacuge@xmath1o@xmath2 ( ref . ) , and cu@xmath1(po@xmath0)@xmath1ch@xmath1 ( ref . ) . all these compounds feature pairs of strongly coupled spins ( magnetic dimers ) . an isolated dimer is an archetypical two - level quantum system , which can be solved analytically . compounds with sizable couplings between the dimers can exhibit diverse behaviors . for instance , the non - frustrated@xcite spin lattice of the han purple bacusi@xmath1o@xmath2 is favorable for propagation of triplet excitations , promoting a bose - einstein condensation of magnons , experimentally observed in the magnetic field range between 23.5 and 49t.@xcite in contrast , srcu@xmath1(bo@xmath0)@xmath1 features strongly frustrated interdimer couplings that give rise to a fascinating variety of magnetization plateaus . the remarkable difference between the behavior of bacusi@xmath1o@xmath2 and srcu@xmath1(bo@xmath0)@xmath1 is governed by the difference in the magnetic couplings that constitute the respective spin model . thus , the precise information on the underlying spin model is crucial for understanding the magnetic properties . an evaluation of the microscopic magnetic model can be performed in different ways . the basic features of the spin lattice can be often conceived by applying empirical rules , such as the goodenough kanamori rules.@xcite then , the resulting qualitative model is parameterized by fitting its respective free parameters to the experiment . the main challenge is the limited amount of the available experimental data that may not suffice for a unique and justified fitting of the model - specific free parameters . thus , such a phenomenological approach is generally insecure against ambiguous solutions . microscopic modeling based on density - functional theory ( dft ) calculations is an alternative solution . such calculations require no experimental information beyond the crystal structure , and in contrast to the phenomenological method , provide a microscopic insight . a straightforward application of the dft is impeded by the fact that cuprates are strongly correlated materials . hence the effective one - electron approach of dft generally fails to reproduce their insulating electronic gs.@xcite this shortcoming can be mended in alternative calculational schemes , such as dft+@xmath5 or hybrid functionals , yet these methods are not parameter - free . often , these parameters sensitively depend on the fine structural details of the system under investigation . the low - dimensional @xmath9 heisenberg compound na@xmath0cu@xmath1sbo@xmath2 is an instructive example that demonstrates the performance and the limitations of the phenomenological as well as the microscopic approach . this compound was initially described as a distorted honeycomb lattice , owing to the hexagonal arrangement of the cu atoms in the crystal structure.@xcite however , this purely geometrical analysis neglects the key ingredients of the magnetic superexchange , such as the orientation and the spatial extent of the magnetically active orbitals . indeed , as pointed out by the authors of ref . , the orientation of the cu @xmath13 orbitals readily accentuates the chains formed by structural dimers and hints at two relevant magnetic couplings : @xmath14 within the structural dimers and @xmath15 between the dimers ( fig . [ f - str ] ) , leading to the quasi-1d heisenberg chain model with alternating nearest - neighbor couplings . thermodynamical measurements confirmed the quasi-1d character of the spin model,@xcite yet no agreement was found for the sign of the intradimer coupling @xmath14 : refs . and vouch for a ferromagnetic ( fm ) and antiferromagnetic ( afm ) exchange , respectively . the sign of @xmath14 basically governs the magnetic gs : the afm - afm solution is a disordered dimer state , while the gs of an fm - afm chain is adiabatically connected to the haldane phase with nontrivial topology and sizable string order parameter.@xcite therefore , for the magnetic gs , the sign of @xmath14 is of crucial importance . notably , even dft studies do not concur with each other : ref . reports afm @xmath14 , while an alternative dft - based method in ref . yields fm coupling . to resolve the controversy on the sign of @xmath14 , the authors of ref . performed inelastic neutron scattering ( ins ) experiments on single crystals of na@xmath0cu@xmath1sbo@xmath2 . the resulting values for the exchange couplings ( @xmath14=@xmath16k and @xmath15=161k ) clearly indicate the fm - afm chain scenario . still , the origin of ambiguous solutions in earlier experimental as well as in dft studies has not been sufficiently clarified . in our combined experimental and theoretical study , we evaluate the magnetic model for na@xmath0cu@xmath1sbo@xmath2 and its te sibling na@xmath1cu@xmath1teo@xmath2 ( ref . ) using extensive dft calculations and investigate how the magnetic gs is affected by the structural distortion within the chains . by comparing our dft results to the earlier studies , we explain the origin of ambiguous parameterizations of dft - based spin models in both compounds . simulations of the momentum - resolved spectrum for our microscopic model reveal excellent agreement with the ins experiments ( ref . ) and enlighten the ambiguity of afm afm and fm - afm solutions inferred from the thermodynamical measurements . this paper is organized as follows . the used experimental as well as computational methods are described in sec . [ s - methods ] . the details of the crystal structures of na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 are discussed in sec . [ s - str ] . in sec . [ s - res ] , we present our magnetic susceptibility measurements and extensive dft calculations . peculiarities of the excitation spectrum of the heisenberg chain model is discussed sec . [ s - spectrum ] . finally , a summary and a short outlook are given in sec . [ s - sum ] . [ [ synthesis - and - sample - characterization ] ] synthesis and sample characterization + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + polycrystalline samples of na@xmath0cu@xmath1sbo@xmath2 were prepared by solid state reaction . a stoichiometric amount of na@xmath1co@xmath0 ( chempur , 99.9+% ) , sb@xmath1o@xmath12 ( 99.999% , alfa aesar ) and cuco@xmath0@xmath17cu(oh)@xmath1 ( chempur ) was thoroughly mixed . the homogeneous powder was pressed into a platinum crucible and annealed at 1273k for two weeks in air . finally the crucible was taken out of the furnace at 1273k and cooled down to room temperature in air . for magnetic measurements , the powder sample was pressed into a pellet and heated again at 973k in a platinum boat for several days . the green powder was identified and characterized by powder x - ray diffraction using a high - resolution guinier camera with cu k@xmath18 radiation . the determined lattice parameters @xmath19 , @xmath20 , @xmath21 and @xmath22 are in good agreement with ref . . to control the oxygen content in the sample at different stages of the thermal treatment , we performed coulometric titration of the samples using a commercial oxylyt device . we found that the maximal oxygen content ( close to the stoichiometric na@xmath0cu@xmath1sbo@xmath2 ) is attained right after the thermal treatment at 973k ( fig . s3 in ref . ) . however , a subsequent storage at room temperature and in air leads to a reduction of the oxygen content . this effect can be seen in the magnetic susceptibility by the increased amount of curie impurity ( fig . s4 in ref . ) . therefore , for thermodynamic measurements , we use `` fresh '' samples ( i.e. , we performed measurement right after the thermal treatment ) that feature smallest impurity contribution . magnetic susceptibility @xmath23 of na@xmath0cu@xmath1sbo@xmath2 was measured using a squid magnetometer ( mpms , quantum design ) in a magnetic field of 0.04 t . [ [ dft - calculations ] ] dft calculations + + + + + + + + + + + + + + + + for the electronic structure calculations , the full - potential local - orbital code fplo ( version fplo8.50 - 32 ) within the local ( spin ) density approximation ( l(s)da ) was used.@xcite in the scalar relativistic calculations the exchange and correlation potential of perdew and wang has been applied.@xcite the accuracy with respect to the @xmath24-mesh has been carefully checked . the lda band structure has been mapped onto an effective one - orbital tight binding ( tb ) model based on cu - site centered wannier functions ( wf ) . the strong coulomb repulsion of the cu @xmath25 orbitals was considered by mapping the tb model onto a hubbard model . in the strongly correlated limit and at half - filling , the lowest lying ( magnetic ) excitations can be described by a heisenberg model with @xmath26 for the antiferromagnetic part of the exchange . spin - polarized lsda+@xmath5 supercell calculations were performed using two limiting cases for the double counting correction ( dcc ) : the around - mean - field ( amf ) and the atomic limit ( al , also called the fully localized limit ) . we varied the on - site coulomb repulsion @xmath27 in the physically relevant range ( 48ev in amf and 59ev in al ) , keeping the on - site exchange @xmath28ev . the partial na@xmath29 occupancy and the sb@xmath30/te@xmath31 substitution were modeled using the virtual crystal approximation ( vca).@xcite hse06 ( ref . ) hybrid functional calculations were performed using the pseudopotential code vasp-5.2,@xcite employing the basis set of projector - augmented waves . the default admixture of the fock exchange ( 25% ) was adopted . we used the primitive unit cell with 2 cu atoms and a 6@xmath326@xmath326 @xmath24-mesh with the nkred=3 flag . [ [ simulations - and - fits - to - the - experiment ] ] simulations and fits to the experiment + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we used the high - temperature series expansion ( htse ) to a heisenberg chain with alternating nearest - neighbor couplings @xmath14 and @xmath15 . for the case of afm couplings , the parameterization is given in @xmath33 in table ii of ref . ; the parameters for the case of fm @xmath14 are provided in ref . . quantum monte carlo simulations were performed using the loop algorithm@xcite from the ` alps ` package.@xcite to evaluate the reduced magnetic susceptibility , we used 50000 loops for thermalization and 500000 loops after thermalization for chains of @xmath34 spins @xmath9 using periodic boundary conditions . exact ( lanczos ) diagonalization of the heisenberg hamiltonians was performed using spinpack.@xcite the lowest - lying @xmath35=0 , @xmath35=1 and @xmath35=2 excitations were computed for @xmath36=32 sites chains of @xmath37=1/2 using periodic boundary conditions . ( color online ) top : crystal structure of na@xmath0cu@xmath1sbo@xmath2 . the basic elements are cuo@xmath38 plaquettes and sbo@xmath2 octahedra . bottom : segments of the structural chains of cu@xmath1o@xmath2 dimers for the experimental distorted plaquette geometry ( upper panel ) and an ideal planar arrangement of the cu@xmath1o@xmath2 units in the fictitious structures ( lower panel).,width=328 ] the monoclinic ( space group @xmath39 ) crystal structure of na@xmath0cu@xmath1sbo@xmath2 ( ref . ) features pairs of slightly distorted , edge - shared cuo@xmath38 plaquettes forming structural dimers with the cu o cu bonding angle of 95@xmath40 . the dimers are connected by the equatorial plane of sbo@xmath2 octahedra and form chains running along the @xmath41 axis ( fig . [ f - str ] , bottom ) . the apical o atoms of the sbo@xmath2 octahedra mediate connections to the next cu@xmath1o@xmath2 dimer chain . in this way , the magnetic layers , separated by na atoms , are formed ( fig . [ f - str ] , top ) . the crystal structure of na@xmath1cu@xmath1teo@xmath2 ( ref . ) features a similar motif , with the reduced number of na atoms between the layers , to keep the charge balance . in addition , the smaller size of te@xmath42 compared to sb@xmath43 gives rise to a stronger distortion of the cu@xmath1o@xmath2 dimer chains in na@xmath1cu@xmath1teo@xmath2 . to investigate the influence of this distortion , we also computed fictitious structures with idealized planar arrangements of the cu@xmath1o@xmath2 units ( fig . [ f - str ] bottom , lower panel ) . above 200k , the magnetic susceptibility of na@xmath0cu@xmath1sbo@xmath2 fits reasonably to the curie - weiss law with @xmath44=0.442emuk(molcu)@xmath45 and the antiferromagnetic weiss temperature @xmath46=60@xmath4710k . the effective magnetic moment amounts to @xmath48@xmath49 , slightly exceeding the spin - only value for @xmath37=1/2 ( 1.73@xmath49 ) . the resulting value of the lande factor @xmath50=2.17 is typical for cu@xmath51 compounds . at lower temperatures , antiferromagnetic correlations give rise to a broad maximum in the magnetic susceptibility around @xmath52=96k . the low - temperature upturn below 17k is likely caused by defects , typical for powder samples of quasi-1d magnets @xmath53e.g . , sr@xmath1cu(po@xmath38)@xmath1 from ref . or ( no)cu(no@xmath0)@xmath0 from ref . @xmath54 $ ] , since already a single defect terminates the spin chain . ( color online ) experimental ( exp ) magnetic susceptibility of na@xmath0cu@xmath1sbo@xmath2 ( circles ) and the quantum monte carlo ( qmc ) fits for the fm - afm and the afm - afm solution of the alternating heisenberg chain ( ahc ) model . inset : difference curves emphasize the excellence of the fm - afm solution.,width=325 ] we briefly compare our susceptibility measurements with the published data . the curie - weiss fit from ref . yields a similar @xmath46=55k , but their @xmath50=2.33 exceeds our estimate . this discrepancy likely originates from the difference in the magnetic field ( 0.1 t versus 0.04 t in our work ) as well as different temperature ranges used for the fitting . unfortunately , the authors of ref . do not provide the values of @xmath46 and @xmath50 , but a curie - weiss fit to their data yields @xmath46@xmath5549k and @[email protected] , in good agreement with our findings . a bare comparison of the absolute values of @xmath56 ( table [ t - cw ] ) reveals sizable deviations of the @xmath23 data from ref . compared to the other two data sets . .[t - cw ] na@xmath0cu@xmath1sbo@xmath2 : the curie - weiss temperature @xmath46 ( in k ) and the @xmath50-factor evaluated using the curie - weiss fit for @xmath57k , as well as the experimental position @xmath52 ( in k ) of the susceptibility maximum and its abosolute value @xmath56 @xmath53 in emu(molcu)@xmath58 $ ] . [ cols=">,^,^,^,^",options="header " , ] for a more elaborate analysis , we adopt the ahc model and search for solutions that agree with the experimental @xmath23 curve . to this end , we perform htse considering the physically different scenarios : both @xmath14 and @xmath15 couplings are afm ( `` afm afm '' ) and @xmath14 is fm ( `` fm afm '' ) . the corresponding htse coefficients for the two cases can be found in refs . and , respectively . in both cases , we obtain a solution ( first row of table [ t - chit ] ) which conforms to the experimental data . l r r r r r r & @xmath14 & @xmath15 & @xmath50 & @xmath59 & @xmath60 & @xmath61 + + & @xmath62 & 171 & 2.01 & 3@xmath63 & 4.7@xmath64 & + & 155 & 66 & 2.20 & @xmath65 & 6.1@xmath64 & 1.1 + + & @xmath66 & 165 & 2.01 & @xmath67 & @xmath67 & @xmath67 + & 143 & 39 & 2.13 & @xmath67 & @xmath67 & @xmath67 + + ref . & 160 & 62 & 1.97 & 2.2@xmath68 & & @xmath69 + + & @xmath67217 & 174 & 2.02 & 9@xmath70 & 6@xmath64 & 1 + & 153 & 61 & 2.19 & 3@xmath70 & 6@xmath64 & 1.2 + our solution for the fm - afm case ( table [ t - chit ] , first row ) nearly coincides with the corresponding solution from ref . ( table [ t - chit ] , second row ) , yielding @xmath71 and a considerably smaller @xmath50-factor of about 2 compared to the value from the curie - weiss fits ( 2.17 ) . for the afm - afm case , we obtain @xmath72 which deviates from the result of ref . , but closely resembles the solution from ref . ( table [ t - chit ] , third row ) . the discrepancy can originate from different parameterizations used for the htse fitting . in particular , the afm - afm solutions in refs . and are obtained using the parametrization from ref . . in contrast , we adopt the coefficients from a more recent and extensive study,@xcite valid in the whole temperature range measured . to account for the full temperature range measured , we turn to qmc simulations . thus , we adopt the ratios @xmath73 and @xmath74 from our htse fitting , and calculate the reduced magnetic susceptibility @xmath75 , which can be fitted to the experimental curve using the expression : @xmath76 where @xmath77 and @xmath78 are the avogadro and boltzmann constants , respectively , @xmath79 the bohr magneton , @xmath60 and @xmath80 account for impurity / defect contributions , @xmath59 is a temperature - independent term , and @xmath81 . using a least - squares fitting , we obtain the solutions listed in table [ t - chit ] ( last row ) and shown in fig . [ f - chit ] . the afm afm solution shows sizable deviations at high temperatures and in the vicinity of the low - temperature upturn ( fig . [ f - chit ] , inset ) , while the fm - afm solution yields an excellent fit to the experimental @xmath23 in the whole temperature range , making the latter solution more favorable . still , the choice is impeded by the following issues . first , the ahc model is a minimal model for na@xmath0cu@xmath1sbo@xmath2 , which completely neglects interchain couplings and anisotropies . second , the @xmath50-factor of the fm - afm solution deviates significantly from the estimate based on the curie - weiss fit , while its counterpart from the afm - afm solution shows a better agreement with the curie - weiss fit . finally , the shape of the @xmath23 curve is affected by oxygen deficiency in the sample,@xcite which is difficult to control during the synthesis process . therefore , the afm - afm solution can not be ruled out using the @xmath23 data , only . to resolve the ambiguity between the fm - afm and afm - afm solutions , we perform microscopic magnetic modeling of na@xmath0cu@xmath1sbo@xmath2 and its te sibling na@xmath1cu@xmath1teo@xmath2 using dft calculations . the valence bands feature similar band width and are similarly structured in the two compounds , as revealed by the lda densities of states ( dos ) in fig . [ dos_atom ] . the dos is dominated by cu and o states down to @xmath82ev and @xmath83ev for na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 , respectively . contributions from na , sb and te are marginal in this energy range . only at the lower edge of the valence band , we find a sizable hybridization of sb states for na@xmath0cu@xmath1sbo@xmath2 centered around @xmath83ev . a similar admixture of te states is observed for na@xmath1cu@xmath1teo@xmath2 , where the additional valence electron of te compared to sb shifts the cu o te density down by about 1ev . ( color online ) total and atom - resolved lda density of states for na@xmath0cu@xmath1sbo@xmath2 ( top ) and na@xmath1cu@xmath1teo@xmath2 ( bottom ) . the contribution of na states is negligible on this scale ( not shown).,width=328 ] the lda band structures for both compounds feature a well - separated density of cu and o states centered around the fermi energy . in the local coordinate system of a cuo@xmath38 plaquette , this density is formed by the anti - bonding @xmath84-combination of cu @xmath13 and o @xmath85 states ( @xmath86 combination ) . the orbital - resolved density of states for na@xmath0cu@xmath1sbo@xmath2 is shown in fig . [ dos_orb ] . two aspects should be pointed out . first , the metallic solution ( nonzero dos at the fermi energy ) observed in na@xmath0cu@xmath1sbo@xmath2 , is in contrast with the green color of the compound , indicative of the insulating behavior . similar , the calculated lda band gap of 0.06ev for na@xmath1cu@xmath1teo@xmath2 ( see fig . [ dos_atom ] ) is far too small to account for the green color of the powder and originates from dimerization effects . this drastic underestimation of the band gap is a well - known shortcoming of the lda , which does not account for the strong coulomb repulsion in the cu 3@xmath87 orbitals . the missing part of correlation energy will be accounted for by resorting to a hubbard model , as well as using dft+@xmath5 and hybrid - functional calculations . second , the orbital resolved density of states ( see fig . [ dos_orb ] ) shows small hybridization with the out - of - plane cu - o states due to the distortion of the dimer chains . since these contributions are small compared to the pure antibonding @xmath86 states , the restriction to an effective tb model is still justified . to verify the structural input , we relaxed the crystal structures within lda . for na@xmath0cu@xmath1sbo@xmath2 , the relaxation results in a rather small energy gain of 33mev per formula unit ( f.u . ) , and the respective changes in the crystal structure are negligible . in contrast , a relaxation of the atomic coordinates in na@xmath1cu@xmath1teo@xmath2 lowers the energy by 130mev per f.u . and alters mainly the chain buckling . since the relaxation of na@xmath1cu@xmath1teo@xmath2 affects the magnetically relevant @xmath86 states , we evaluated the magnetic properties for both , the experimental and the relaxed crystal structure . ( color online ) orbital - resolved lda density of states for na@xmath0cu@xmath1sbo@xmath2 . in the local coordinate system of a cuo@xmath38 plaquette , the @xmath88 axis runs along one of the cu o bonds , while the @xmath89 axis is perpendicular to the plaquette plane ( for the ideal planar coordination ) . the states in the vicinity of the fermi energy are dominated by the in - plane cu @xmath25 and o @xmath90 states.,width=328 ] the transfer integrals @xmath91 ( the hopping matrix elements ) are evaluated by a least - squares fit of an effective one - orbital tb model to the two lda @xmath86 bands . using 10 inequivalent @xmath91 terms ( see the bottom panel of fig . [ tbm ] , table [ tab_ts ] , and ref . ) we obtain excellent agreement between the tb model and the lda band structure . the respective fit for na@xmath0cu@xmath1sbo@xmath2 is shown in fig . [ tbm ] ( top ) . ( color online ) top : lda band structure of na@xmath0cu@xmath1sbo@xmath2 and the fit using the effective one - orbital tb model with ten inequivalent transfer integrals @xmath91 . bottom : the superexchange pathways corresponding to the relevant @xmath91 terms.,width=328 ] l ccccccc + @xmath92/mev & @xmath93 & @xmath94 & @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 + exp & * 60.6 * & * 127 * & 18.2 & @xmath6727.8 & 17.0 & 21.8 & 17.4 + relaxed & * 68.2 * & * 134 * & 18.1 & @xmath6732.3 & 20.6 & 20.9 & 19.2 + planar exp & * 45.3 * & * 119 * & 22.4 & @xmath677.8 & 9.4 & 30.1 & @xmath67 + planar relax . & * 55.6 * & * 125 * & 23.8 & @xmath679.2 & 10.7 & 29.2 & @xmath67 + + @xmath92/mev & @xmath93 & @xmath94 & @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 + exp & * 15.6 * & * 162 * & 16.4 & @xmath6738.5 & 24.7 & 13.7&25.5 + relaxed & * 42.5 * & * 152 * & 17.3 & @xmath6742.4 & 26.3 & 14.5 & 23.1 + planar exp . & * 27.3 * & * 152 * & 29.3 & @xmath6712.6 & 12.4 & 25.6 & 1.3 + planar relax . & * 45.2 * & * 148 * & 30.0 & @xmath6712.8 & 12.7 & 26.0 & @xmath67 + in both systems , the leading coupling is @xmath94 , which connects two neighboring structural dimers : @xmath100mev for na@xmath0cu@xmath1sbo@xmath2 and @xmath101mev for na@xmath1cu@xmath1teo@xmath2 , respectively . the coupling within the structural cu@xmath1o@xmath2 dimers ( @xmath102mev for na@xmath0cu@xmath1sbo@xmath2 and @xmath103mev for na@xmath1cu@xmath1teo@xmath2 ) are significantly smaller . besides , several long - range couplings that connect different chains , are comparable to @xmath93 ( table [ tab_ts ] and ref . ) . subsequent mapping of the tb model onto a hubbard model ( adopting @xmath104ev ) and a heisenberg model , yield the following afm contributions : @xmath105k and @xmath106k for na@xmath0cu@xmath1sbo@xmath2 and @xmath107k and @xmath108k for na@xmath1cu@xmath1teo@xmath2 , respectively . the resulting minimal model is incomplete , since it disregards the fm contribution to the exchange integrals , which are expected to be especially large for the @xmath3 coupling within the structural dimers . to estimate the total exchange integrals , comprising afm and fm contributions , we performed lsda+@xmath5 calculations of magnetic supercells . mapping the total energies of different collinear spin arrangements onto a classical heisenberg model yields @xmath109k for na@xmath0cu@xmath1sbo@xmath2 and @xmath110k for na@xmath1cu@xmath1teo@xmath2 , respectively . for the exchange between the structural dimers , we find @xmath111k for na@xmath0cu@xmath1sbo@xmath2 and @xmath112k for na@xmath1cu@xmath1teo@xmath2 ( @xmath113ev ) . all further exchange integrals between different chains and layers are smaller than 10k , and thus can be neglected in the minimal model . unlike the related compounds featuring edge - shared chains@xcite or cu@xmath1o@xmath2 dimers,@xcite na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 exhibit a sizable influence of the coulomb repulsion @xmath27 on the exchange integrals ( see fig . [ js_compare ] ) . however , the variation of @xmath27 within the physically relevant range ( sec . [ s - methods ] ) does not affect the fm nature of @xmath3 . thus , na@xmath0cu@xmath1sbo@xmath2 features alternating chains with the exchange integrals of nearly the same magnitude but different sign ( fm @xmath3 and afm @xmath4 ) , while for na@xmath1cu@xmath1teo@xmath2 , the afm exchange between the structural dimers is dominant . the evaluated exchange integrals are listed in table [ sbte_js ] . for an independent computational method , we use hybrid functional ( hf ) total energy calculations . the absence of the double counting problem and minimal number of free parameters makes hf calculations an appealing alternative to the dft+@xmath5 methods.@xcite here , we employ the hse06 functional to evaluate the leading couplings @xmath3 and @xmath4 in both compounds . in accord with dft+@xmath5 , we obtain fm @xmath3 and afm @xmath4 . for na@xmath0cu@xmath1sbo@xmath2 , the resulting exchange integrals are in excellent agreement with the htse estimates ( table [ sbte_js ] ) . similar to dft+@xmath5 , na@xmath1cu@xmath1teo@xmath2 features a weaker @xmath3 and stronger @xmath4 , thus the @xmath114 value is substantially reduced . we are now in position to compare our results with the previous dft - based studies . derakhshan _ et al_. ( ref . ) evaluated the relevant transfer integrals using @xmath36th - order muffin - tin - orbital downfolding of the lda band structure . although this computational method ( ref . ) as well as the code@xcite used for the calculations differ from our approach , the difference in the resulting @xmath91 values does not exceed 25% . hence , the estimated afm contributions to the exchanges @xmath3 and @xmath4 generally agree with our values . however , in contrast to the present study , the authors of ref . did not perform dft+@xmath5 calculations and therefore completely disregarded the fm contributions , which are especially relevant for the short - range coupling @xmath3 . thus , their afm - afm solution originates from a severe incompleteness of the computational scheme and the respective mapping onto the spin hamiltonian . in contrast to ref . , koo and whangbo performed dft+@xmath5 calculations using vasp , and recovered fm @xmath3 and afm @xmath4 , in qualitative agreement with the experiment . however , the absolute values of the leading couplings are considerably overestimated . we believe that this overestimation stems from the choice of the on - site coulomb repulsion parameter @xmath115 . it is well - known that the parameters of the dft+@xmath5 calculations are not universal,@xcite in particular basis dependent , and should be carefully chosen based on the nature of the magnetic atom and the code used . the @xmath115 range studied in ref . ( 4 .. 7ev ) is too narrow , and larger @xmath115 is likely required to reproduce the correct magnetic energy scale in na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 . llrrr data source & method & @xmath3 & @xmath4 & @xmath114=@xmath3/@xmath4 + + & lsda+@xmath5 & @xmath116 & 150 & @xmath117 + & hse06 & @xmath118 & 163 & @xmath119 + & htse & @xmath62 & 171 & @xmath120 + & qmc & @xmath121 & 174 & @xmath122 + + ref . & ehtb & @xmath123 & 345 & @xmath124 + ref . & htse & @xmath123 & 209 & @xmath125 + ref . & htse & 22 & 169 & 0.13 + + & lsda+@xmath5 & @xmath126 & 232 & @xmath127 + & hse06 & @xmath123 & 291 & @xmath128 + + ref . & ehtb & @xmath129 & 516 & @xmath130 + ref . & htse & @xmath131 & 215 & @xmath132 + ref . & htse & 13 & 127 & 0.1 + next , we study the influence of the structural parameters onto the alternation ratio @xmath133 for na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 . the two compounds differ not only by the nonmagnetic ions ( sb and te ) located between the structural dimers , but also by details of their chain geometry . these subtle differences can have a substantial impact on the magnetic properties . in particular , the substitution of sb by te and the corresponding change of the na content modulates the crystal field . furthermore , the substitution of sb by te has a sizable impact on the buckling of the dimer chains , which is determined by the deviation of o atoms from an ideal planar arrangement . finally , the interatomic distances in the two compounds are different . to separate these effects out , we introduce fictitious compounds containing ideal planar dimer chains ( see fig . [ f - str ] ) , evaluate their electronic structure , and compare them with real compounds . ( color online ) calculated exchange integrals @xmath4 and @xmath3 , as well as frustration ratios @xmath134 as a function of the coulomb repulsion @xmath27 for different structural models of the two compounds.,width=328 ] top : lda band structures showing the magnetically active antibonding dp@xmath135 states for na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 . bottom : comparison of the lda band structures calculated for different structural models of na@xmath0cu@xmath1sbo@xmath2.,width=328 ] the direct comparison of the antibonding dp@xmath135 bands for the experimentally observed crystal structures of na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 ( fig . [ vgl_n2n3 ] , upper panel ) reveals that these bands differ mainly by their width . in contrast , comparing the antibonding @xmath86 bands of na@xmath0cu@xmath1sbo@xmath2 within the experimental crystal structure ( distorted plaquettes ) with the fictitious crystal structure ( planar plaquettes ) reveals similar band widths , but substantially different dispersions ( compare x-@xmath136 or x - z in fig . [ vgl_n2n3 ] ) . the same trend is also observed for na@xmath1cu@xmath1teo@xmath2 . \(i ) to separate out the effect of the sb@xmath137te substitution , we perform vca calculations for the same structural model . in particular , a certain fraction @xmath88 of sb atoms is replaced by te , with a concomitant change in the na content , in order to keep the charge balance . the band structures calculated for different te concentrations exhibit similar dispersions and similar band width , evidencing the minor relevance of the pure substitutional effect for the magnetic exchange couplings.@xcite to estimate the impact of the chain distortion and interatomic distances onto the magnetism in more detail , we evaluated the magnetic model also for two fictitious crystal structures of na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 ( featuring planar dimer chains).@xcite the obtained hopping terms and exchange integrals are given in tables [ tab_ts ] and [ list_js ] . lsda+@xmath5 calculations ( @xmath138ev ) yield @xmath3=@xmath139k and @xmath140k for the fictitious na@xmath0cu@xmath1sbo@xmath2 and @xmath141k and @xmath142k for the fictitious na@xmath1cu@xmath1teo@xmath2 , respectively . the dependence of the exchange integrals on the coulomb repulsion @xmath27 is depicted in fig . [ js_compare ] . analysis of the resulting exchange couplings suggests that the two structural parameters act differently : the distortion of the dimer - chains mainly influences the coupling strength of @xmath3 and the coupling regime between the dimer - chains ( @xmath143 and @xmath144 ) , whereas the interdimer exchange @xmath4 is rather insensitive to this parameter ( table [ list_js ] ) , since the respective superexchange path does not involve o(2 ) atoms that rule the distortion . lrrr & & & + + exp & @xmath116 & 150 & ( 188 ) + relaxed & @xmath145 & 162 & ( 209 ) + planar ( exp ) & @xmath146 & 126 & ( 165 ) + planar ( relax ) & @xmath147 & 142 & ( 182 ) + + exp & @xmath126 & 232 & ( 305 ) + relaxed & @xmath148 & 197 & ( 269 ) + planar ( exp ) & @xmath149 & 212 & ( 269 ) + planar ( relax ) & @xmath150 & 200 & ( 255 ) + \(ii ) comparing the total exchange integrals for na@xmath0cu@xmath1sbo@xmath2 for the experimental crystal structure with the planar system discloses an increase of the nn coupling @xmath3 by nearly a factor of 2 , whereas @xmath4 is decreased by less than 20% . this observation is in line with the intuitive picture derived from geometrical considerations comparing the experimental distorted crystal structure to the fictitious system containing ideal planar chains ( compare fig . [ f - str ] , lower panel ) . locking the o atoms within the chain plane directly alters the exchange path of @xmath3 along cu - o - cu , by a change of the cu - o - cu bridging angle and the orientation of the magnetically active orbitals . in contrast , the superexchange path of @xmath4 ( cu - o - o - cu ) is altered only indirectly by changes of the crystal - field due to the distortion of the sb / teo@xmath2 octahedra ( compare fig . [ f - str ] , lower panel ) . \(iii ) the modulation of interatomic distances influences @xmath3 and @xmath4 in a similar way . the crucial impact of the interatomic distances on @xmath4 manifests itself in the coupling strength of the planar model structures for na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 ( see tab . [ list_js ] ) with the corresponding nnn cu - cu interdimer distance . the about 0.1 shorter nnn cu - cu distance in the fictitious planar na@xmath1cu@xmath1teo@xmath2 structure compared to the fictitious planar na@xmath0cu@xmath1sbo@xmath2 increases the coupling strength by about 60% . however , comparing the experimental distorted crystal structure with the planar model structure of na@xmath1cu@xmath1teo@xmath2 the difference in the nnn cu - cu distance is only half as large ( about 0.05 ) as between the two planar structures and result in an about 1/4 smaller increase of @xmath4 . thus , @xmath4 follows a simple distance relation and scales according to @xmath151 . the same relation holds for @xmath3 ( compare @xmath3 for the two planar structures with the change of the nn cu - cu distance ) . based on the above considerations , we can conclude that the crucial parameter , determining the alteration ratio @xmath133 for na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 , is the distortion of the chains . thus , a directed modification of the chain buckling by the appropriate substitution of ions should allow to tune the magnetism of these systems . furthermore , the chain distortion also influences the interchain coupling regime . in the experimental structure the long - range exchanges mostly operate within the magnetic layers ( in the @xmath152-plane ) , whereas in the planar system the coupling between the layers is enhanced ( @xmath91 s in table [ tab_ts ] ) . as already mentioned , the fm - afm and afm - afm solutions correspond to different magnetic gss . in the former case , the gs is similar to the haldane chain and features sizable string order parameter @xmath153 , indicative of a topological order , while in the latter case the string order is suppressed ( @xmath154).@xcite it is thus tempting to find an observable quantity that would be substantially different in the two phases . theoretical studies of the @xmath9 ahc model suggest that this requirement is fulfilled for the momentum position of the spin gap . indeed , the @xmath155 gap is characteristic for afm - afm chains , except for the narrow parameter range @xmath156 , where the gap shifts to small finite @xmath157.@xcite in contrast , the spin gap in the fm - afm chains is located at @xmath158.@xcite therefore , by measuring momentum resolved excitation spectra , the sign of @xmath3 can be reliably determined . to resolve the ambiguity between the fm - afm and afm - afm cases ultimately , we calculate the low - energy excitations for @xmath73 as well as @xmath159 using lanczos diagonalization of the respective heisenberg hamiltonian . the resulting @xmath160 dependence is plotted in fig . [ f - ins ] . although the two solutions yield similar estimates for the spin gap , its @xmath161-position is very different : @xmath158 and @xmath155 , for the fm - afm and afm - afm solution , respectively . another distinct feature of the excitation spectra is the well - separated branch of lowest - energy excitations ( fig . [ f - ins ] ) . for the fm - afm solution , this branch resembles the behavior of @xmath162 , while the afm - afm solution yields a @xmath163-like behavior . ( color online ) low - energy excitation spectra for the alternating heisenberg chains of @xmath164 spins , corresponding to the qmc solutions ( fm - afm and afm - afm ) from table [ t - chit ] . note the difference in the behavior of the low - lying branch for the two solutions . experimental data from ref . are shown with filled circles ( the gray line is guide to the eye).,width=325 ] to compare with the experimental dispersion from ref . , we scale the two spectra using the values of the exchange couplings from our qmc fits to the magnetic susceptibility ( table [ t - chit ] , last row ) . this way , we find that the fm - afm solution agrees very well with the experimental data ( fig . [ f - ins ] ) , while the afm - afm solution can be safely ruled out . fig . [ f - ins ] also provides an answer to an intriguing question , why both @xmath73 and @xmath159 provide good fits to the susceptibility data . at finite temperature , magnetic susceptibility reflects the thermal - averaged magnetic spectrum integrated over the whole momentum space . thus , at low temperatures , @xmath23 is largely affected by the value of the spin gap , but is insensitive to its @xmath161 position . since the values of the spin gap for the two solutions are very similar ( around 100k ) , the similarity of the low - temperature @xmath23 is also not surprising . moreover , the shape of the low - energy branch is similar ( but reflected around @xmath165 ) , thus the @xmath161-integrated spectrum is nearly the same in both cases . only at elevated temperatures , the contribution of high - lying states gives rise to the difference in @xmath23 . this is in excellent agreement with the enhanced deviation of the afm - afm solution at high temperatures ( fig . [ f - chit ] , inset ) . since the first report on the magnetism of the low - dimensional @xmath37=1/2 systems na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 , their spin models were controversially debated in the literature . the main conundrum was the sign of the exchange coupling @xmath14 operating within the structural cu@xmath1o@xmath2 dimers . to resolve the conflicting reports , we applied a series of different computational methods , including density functional theory ( dft ) band structure , virtual crystal approximation , dft+@xmath5 , and hybrid functional calculations , as well as high - temperature series expansions , quantum monte carlo simulations , and exact diagonalization . our calculations evidence that the magnetism of both compounds can be described by the alternating heisenberg chain model with two relevant couplings : ferromagnetic @xmath14 within the structural dimers , and antiferromagnetic @xmath15 between the dimers . the alternation parameter @xmath166 amounts to about @xmath122 and @xmath167 in na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 , respectively . this parameter regime corresponds to the haldane phase , characterized by the gapped excitation spectrum and a topological string order . using extensive calculations for different structural models , we find that the physically relevant ratio @xmath133 is primarily ruled by the distortion of the structural chains , while the sb@xmath137te substitution and the change in the cu cu distance play a minor role . a comparison of the simulated dispersion @xmath160 with the experimental inelastic neutron scattering data ( ref . ) yields an unequivocal evidence for the fm nature of @xmath14 in na@xmath0cu@xmath1sbo@xmath2 . these spectra facilitate the understanding of the similarity between the magnetic susceptibility curves for mutually exclusive solutions that involve ferromagnetic and antiferromagnetic @xmath14 . it is important to note that the problem of ambiguous solutions appears in the empirical modeling , only . in contrast , the microscopic modeling based on dft calculations readily yields a quasi - one - dimensional model with the ferromagnetic @xmath14 . this clearly indicates that present - day dft calculations are a reliable tool to disclose even rather complicated cases and deliver a reliable microscopic magnetic model . since the correctness of the magnetic model is of crucial importance for its refinement and extension , dft calculations should be an indispensable ingredient of real - material studies . we are grateful to professor m. sato for explaining the details of his previous experimental work on na@xmath0cu@xmath1sbo@xmath2 and na@xmath1cu@xmath1teo@xmath2 . fruitful discussions with a. a. tsirlin are kindly acknowledged . oj was partially supported by the european union through the european social fund ( mobilitas grant no . jr was supported by the dfg through the project ri 615/16 - 3 . 53ifxundefined [ 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 10.1088/0034 - 4885/71/1/012501 [ * * , ( ) ] , link:\doibase 10.1038/nature08917 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.107204 [ * * , ( ) ] , link:\doibase 10.1126/science.1212154 [ * * , ( ) ] link:\doibase 10.1143/jpsj.65.3142 [ * * , ( ) ] link:\doibase 10.1103/physrevb.65.144445 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.106.177203 [ * * , ( ) ] link:\doibase 10.1103/physrevb.74.174421 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.77.224437 [ * * , ( ) ] , link:\doibase 10.1088/0953 - 8984/21/30/305501 [ * * , ( ) ] link:\doibase 10.1103/physrevb.52.3533 [ * * , ( ) ] link:\doibase 10.1103/physrevb.66.054426 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.81.104416 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.107202 [ ] ( ) , link:\doibase 10.1103/physrevlett.93.087203 [ * * , ( ) ] , link:\doibase 10.1038/nature04732 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.76.100406 [ * * , ( ) ] , link:\doibase 10.1143/jpsj.79.011005 [ * * , ( ) ] link:\doibase 10.1103/physrev.100.564 [ * * , ( ) ] link:\doibase 10.1016/0022 - 3697(59)90061 - 7 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.61.433 [ * * , ( ) ] link:\doibase 10.1143/jpsj.75.084707 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.104403 [ * * , ( ) ] link:\doibase 10.1103/physrevb.48.9555 [ * * , ( ) ] link:\doibase 10.1021/ic701153z [ * * , ( ) ] link:\doibase 10.1143/jpsj.77.104709 [ * * , ( ) ] link:\doibase 10.1021/ic0502832 [ * * , ( ) ] link:\doibase 10.1016/j.jssc.2005.02.002 [ * * , ( ) ] @noop @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.85.035207 [ * * , ( ) ] link:\doibase 10.1063/1.1564060 [ * * , ( ) ] link:\doibase 10.1063/1.2204597 [ * * , ( ) ] link:\doibase 10.1103/physrevb.54.11169 [ * * , ( ) ] link:\doibase 10.1016/0927 - 0256(96)00008 - 0 [ * * , ( ) ] link:\doibase 10.1103/physrevb.61.9558 [ * * , ( ) ] , link:\doibase 10.1021/ic00101a006 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.87.047203 [ * * , ( ) ] , link:\doibase 10.1016/j.jmmm.2006.10.304 [ * * , ( ) ] , @noop link:\doibase 10.1016/j.jssc.2003.09.024 [ * * , ( ) ] link:\doibase 10.1103/physrevb.82.054413 [ * * , ( ) ] , link:\doibase 10.1021/ic50218a017 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.205111 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.85.014407 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.76.115109 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.62.r16219 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.53.2571 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.127101 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.035103 [ * * , ( ) ] , link:\doibase 10.1143/jpsj.68.2073 [ * * , ( ) ] , link:\doibase 10.1143/jpsj.63.2514 [ * * , ( ) ] ccccccccc + & & & & + & @xmath169 & @xmath170 & @xmath171 & & @xmath169 & @xmath170 & @xmath171 & + cu & 0 & 0.6667 & 0 & & 0 & 0.6667 & 0 & + sb & 0 & 0 & 0 & & 0 & 0 & 0 & + o(1 ) & 0.2931 & 0.3340 & 0.7750 & & -0.1987 & -0.1667 & -0.2234 & + o(2 ) & 0.2404 & 0.5 & 0.1774 & & -0.2619 & 0 & 0.1734 & + na(1 ) & 0 & 0.5 & 0.5 & & 0 & -0.5 & -0.5 & + na(2 ) & 0.5 & 0.3280 & 0.5 & & 0 & -0.1732 & -0.5 & + + & & & & + o(2 ) & 0.2931 & 0.5 & 0.7750 & & -0.1987 & 0.5 & -0.2234 & + ccccccccc + & & & & + & @xmath169 & @xmath170 & @xmath171 & & @xmath169 & @xmath170 & @xmath171 & + cu & 0 & 0.66475 & 0 & & 0 & -0.3353 & 0 & + te & 0 & 0 & 0 & & 0 & 0 & 0 & + o(1 ) & 0.1936 & 0.1632 & 0.2121 & & 0.1906 & 0.1682 & 0.2156 & + o(2 ) & 0.7574 & 0 & 0.1640 & & -0.2519 & 0 & 0.1648 & + na & 0 & 0.1839 & 0.5 & & 0 & 0.1849 & -0.5 & + + & & & & + o(2 ) & 0.1936 & 0.5 & 0.2121 & & 0.1906 & 0.5 & 0.2156 & + l cccccccccc + @xmath92/mev & @xmath93 & @xmath94 & @xmath95 & @xmath96 & @xmath97 & @xmath172 & @xmath173 & @xmath98 & @xmath174 & @xmath99 + exp & * 60.6 * & * 127 * & 18.2 & @xmath6727.8 & 17.0 & 5.8 & @xmath676.6 & 21.8 & @xmath674.6 & 17.4 + relaxed & * 68.2 * & * 134 * & 18.1 & @xmath6732.3 & 20.6 & 6.4 & @xmath677.2 & 20.9 & @xmath673.8 & 19.2 + planar exp & * 45.3 * & * 119 * & 22.4 & @xmath677.8 & 9.4 & 14.1 & @xmath6718.8 & 30.1 & @xmath6713.2 & @xmath67 + planar relaxed & * 55.6 * & * 125 * & 23.8 & @xmath679.2 & 10.7 & 17.3&@xmath6721.7 & 29.2 & @xmath6719.9 & @xmath67 + + @xmath92/mev & @xmath93 & @xmath94 & @xmath95 & @xmath96 & @xmath97 & @xmath172 & @xmath173 & @xmath98 & @xmath174 & @xmath99 + exp & * 15.6 * & * 162 * & 16.4 & @xmath6738.5 & 24.7 & 2.8 & @xmath67 2.5 & 13.7 & @xmath67 & 25.5 + relaxed & * 42.5 * & * 152 * & 17.3 & @xmath6742.4 & 26.3 & 3.4 & @xmath673.9 & 14.5 & @xmath67 & 23.1 + planar exp . & * 27.3 * & * 152 * & 29.3 & @xmath6712.6 & 12.4 & 14.1 & @xmath6716.0 & 25.6 & @xmath6711.8 & 1.3 + planar relaxed & * 45.2 * & * 148 * & 30.0 & @xmath6712.8 & 12.7 & 16.0 & @xmath6718.6 & 26.0 & @xmath6710.7 & @xmath67 + l ccccccc + @xmath175/mev & @xmath3 & ( @xmath176 ) & @xmath4 & ( @xmath177 ) & @xmath178+@xmath179+@xmath180 & @xmath181+@xmath182 & @xmath183+@xmath184 + exp & -11.6 & & 12.9 & & -0.01 & 0.8 & 0.4 + & & ( 3.7 ) & & ( 16.1 ) & & & + relaxed & -10.7 & & 13.9 & & 0.3 & 0.7 & @xmath67 + & & ( 4.7 ) & & ( 18.0 ) & & & + planar exp & -19.7 & & 10.8 & & 0.9 & 0.2 & 1.2 + & & ( 2.1 ) & & ( 14.2 ) & & & + planar relaxed & -19.5 & & 12.2 & & 1.0 & 0.2 & @xmath67 + & & ( 3.1 ) & & ( 15.7 ) & & & + + @xmath175/mev & @xmath3 & ( @xmath176 ) & @xmath4 & ( @xmath177 ) & @xmath178+@xmath179+@xmath180 & @xmath181+@xmath182 & @xmath183+@xmath184 + exp & -10.3 & & 19.9 & & 0.1 & 1.0 & 0.1 + & & ( 0.2 ) & & ( 26.2 ) & & & + relaxed & -10.5 & & 16.9 & & 0.2 & 1.0 & @xmath67 + & & ( 1.8 ) & & ( 23.1 ) & & & + planar exp . & -26.8 & & 18.2 & & 1.2 & 0.5 & 0.8 + & & ( 0.8 ) & & ( 23.1 ) & & & + planar relaxed & -26.2 & & 17.2 & & 1.5 & 0.6 & @xmath67 + & & ( 2.0 ) & & ( 21.9 ) & & & + lcccccc structure & @xmath185 & @xmath186 & cu o cu & o o & cu o & @xmath187o + + exp & 2.96 & 5.91 & 95.27 & 2.94 & 2.021/2.000 & 0.39 + planar exp & 2.96 & 5.91 & 94.22 & 2.94 & 2.021/2.017 & 0 + relaxed & 2.96 & 5.91 & 96.03 & 2.96 & 1.998/1.988 & 0.43 + planar relaxed & 2.96 & 5.91 & 95.39 & 2.96 & 1.998/1.998 & 0 + + exp & 2.86 & 5.82 & 91.27 & 2.83 & 1.978/1.999 & 0.55 + planar exp & 2.86 & 5.82 & 95.48 & 2.83 & 1.978/1.931 & 0 + relaxed & 2.86 & 5.82 & 92.91 & 2.92 & 1.950/1.972 & 0.53 + planar relaxed & 2.86 & 5.82 & 95.24 & 2.92 & 1.950/1.935 & 0 + dependence of exchange integrals on the double - counting correction : the around - mean - field ( amf ) and the atomic limit ( al ) . for the lsda+@xmath5 calculations we adopted the experimental crystal structure ( exp ) and the structure with planar chains ( planar exp ) . ] oxygen content determined by coulometric titration of a freshly annealed sample ( green curve ) and a sample stored at room temperature and in air ( violet curve ) . note the difference between the initial oxygen content in the two samples . ] dependence of the magnetic susceptibility on the annealing temperature and sample history ( quality ) due to changes in the o occupation . external magnetic field is 0.04 t . note the difference between the 700@xmath188c samples stored in air ( `` old '' ) and measured right after annealing ( `` fresh '' ) . in the paper , we used @xmath23 data for fresh samples annealed at 640@xmath188c , i.e. , the samples featuring the smallest curie - law - type impurities and/or defects ( inset ) . ] comparison of the antibonding @xmath189 band using lda for the experimental crystal structure of na@xmath0cu@xmath1sbo@xmath2 and a model compound where sb is substituted by te ( with an concomitant change in the na content ) within the same crystal structure using vca . the obtained bands show nearly the same band width and shape pointing to minor relevance of pure substitutional effects . ] 2ifxundefined [ 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 10.1016/j.jssc.2005.02.002 [ * * , ( ) ] link:\doibase 10.1021/ic0502832 [ * * , ( ) ]

we present results on the evolution and clustering of lines at low redshift as part of our series `` a uniform analysis of the forest at . '' the sample analyzed in this paper contains 1298 absorption lines from 165 quasar spectra mined from the archives of the faint object spectrograph on the hubble space telescope ( hst ) . our sample extends to , slightly higher than the sample analyzed by the hst quasar absorption line key project . we confirm the result from the key project that the number density evolution of lines at low redshifts can be described by a power law that is significantly flatter than that found at high . we find that the evolution is somewhat steeper than obtained previously . specifically , we find for lines with equivalent widths greater than 0.24 and using a variable equivalent width threshold . we find that the difference between our and key project results is likely attributable to different redshift coverage of the two samples . the results concerning the number density evolution are not significantly affected if one includes lines which are members of metal systems . object to object fluctuations in the number of lines detected are small , indicating a high degree of uniformity in the intergalactic medium on large scales . we find marginal evidence that weak and strong lines undergo different evolution . we find weak clustering for lines at velocity separations , weaker than the level predicted from an earlier analysis by ulmer of a small subsample of the key project data . we see no correlations for metal system forest or extensive metal system forest combinations .

astro-ph0111487

this paper is a continuation of the project of the analysis of absorption lines covering the range of redshifts , @xmath6 . the first two papers in this series ( scott , bechtold , & dobrzycki 2000 ; scott et al . 2000 ; hereafter papers i and ii ) presented the analysis of intermediate resolution ground based data . this and accompanying papers ( bechtold et al . 2001 , paper iii ; scott et al . 2001 , paper v ) report the analysis of hubble space telescope ( hst ) faint object spectrograph ( fos ; ford & hartig 1990 ) archival spectra . paper iii describes the fos data reductions , absorption line analysis , and line identification procedure . in this paper we present the distribution and clustering properties of the absorption lines . all data analyzed in this and other papers from this series can be accessed from our web pages : + ` http://lithops.as.arizona.edu/~jill/quasarspectra/ ` + or + ` http://hea-www.harvard.edu/qedt/quasarspectra/ ` . from data presented in paper iii , we selected all spectra suitable for the analysis of the forest . we excluded spectra observed with a-1 aperture before installation of costar , broad - absorption line quasars , and quasars observed in spectropolarimetry mode ( see paper iii ) . in cases where two images of a gravitationally lensed quasar were observed , or a close pair of quasars or grouping of quasars were observed , we included only one line - of - sight , choosing the spectra with the highest signal - to - noise . for detailed discussion on the detection significance and completeness in our sample , and on the identification of metal lines , see sec . 2 of paper iii . in total , the sample analyzed in this paper contains spectra of 165 quasars ( table [ tab : quasars ] ) . the spectral resolution of all the data is 230280 . of those 165 quasars , 63 were analyzed by weymann et al . ( 1988 ) as part of the hst quasar absorption line key project ( bahcall et al . 1993 , 1996 ; schneider et al . 1993 , jannuzi et al . 1998 , weymann et al . 1998 ) . the sixty fourth object in weymann et al . sample , pg 2302 + 029 , is a broad absorption line quasar ( jannuzi et al . 1996 ) and was excluded from our sample ( as well as from samples 610 in weymann et al . ) . we selected the sample of the absorption lines for further analysis using the following criteria : \1 . the detection significance for the line has to be greater than 5@xmath8 . a line can not be identified as a heavy element absorption feature . regions within 300 of the centers of metal absorption lines were excluded from the analysis . \3 . whenever a `` forest '' sample was called for , we excluded lines identified as members of heavy element systems and regions within 300 from such lines . we excluded lines lying bluewards from the location of the emission line , to avoid confusion with the lines from higher redshift systems . please note that these criteria are more conservative than those of the hst key project . the sample contains 1298 lines if all lines ( forest and metal systems ) are selected , and 1157 forest lines . figure [ fig : eqw ] shows the rest equivalent width , , defined as @xmath9 , of all lines versus redshift , @xmath2 . equivalent widths were measured from gaussian fits to significant absorption features ( see paper iii ) . this plot has not been corrected for the relative sensitivity for detection at each redshift and is only intended to show the redshift distribution of the sample . figure [ fig : eqwhist ] shows the histogram of rest equivalent widths of lines in the full sample . the shaded area shows the lines with detected metal absorption . figure [ fig : eqwhist ] clearly shows that lines associated with metal systems begin to compose a significant part of the sample for rest equivalent widths greater than @xmath101 . we can not rule out the possibility that some of the lines in our sample are in fact unidentified metal system lines . however , we found that the exclusion from the analysis of all lines with @xmath11 which , in view of fig . [ fig : eqwhist ] , are the prime suspects for being unrecognized metal lines produced negligible differences in the number density evolution results . figure [ fig : zcoverage ] shows the comparison of redshift coverage in our sample with the weymann et al . ( 1998 ) sample . the upper panel shows the combined path length available for study in both samples as a function of redshift , while the lower panel shows the number of available lines . the notable differences are that ( 1 ) our sample extends to @xmath12 , while the key project sample effectively ends at @xmath13 , ( 2 ) our sample contains considerably more lines in the @xmath14 range , and ( 3 ) our sample by definition excludes lines below emission line , while the key project sample does not . we note that did not attempt to remove the flat field features in the fos spectra , which the key project went to great lengths to do ( schneider et al . 1993 , jannuzi et al . we identify 38 of their 141 confirmed flat field features as lines with significance greater than 5@xmath8 ( paper iii ) . however , only nine of those features lie in the forest . simple extrapolation of those numbers to the sample analyzed here would indicate that roughly 24 lines out of our total sample of 1298 , or less than 2% , are spurious . however , even this low number is likely overestimated , since majority of redshift path in our sample that is added to the key project sample ( see fig . [ fig : zcoverage ] ) is in regions covered by g130h and g270h gratings , which have less ( none in the case of g130h , actually ) identified flat field features than grating g190h ( jannuzi et al . those should have a negligible effect on our results . we parameterized the distribution of individual clouds using : @xmath16 where , @xmath17 , and are the distribution parameters . for a non - evolving population of clouds @xmath18 , depending on the value of @xmath19 ( peterson 1978 ) . it is well - established that @xmath17 is significantly greater than 1 at @xmath20 , indicating strong evolution ( e.g. bajtlik , duncan , & ostriker 1988 ; lu , wolfe , & turnshek 1991 ; bechtold 1994 ; kim et al.1997 ; paper i ) . however , the hst quasar absorption line key project and other studies showed that @xmath17 is less then 1 at @xmath21 , consistent with the no evolution case ( bahcall et al . 1993 , 1996 ; weymann et al . 1998 ; vanden berk et al . 1999 ; impey , petry , & flint 1999 ) . the proximity effect ( bajtlik , duncan & ostriker 1988 ; bechtold 1994 ; paper ii ) at high redshift causes the derived value of @xmath17 to depend strongly on the inclusion or exclusion of lines with @xmath22 . however , the proximity effect is not expected to influence the results as significantly at low redshift for a few reasons . first , the overall number density of absorbers per unit redshift is smaller than at high redshift , and so the expected number of lines affected will also be small . second , the quasars used at low redshift are intrinsically fainter on average than the quasars observed at high @xmath2 , and thus it is expected that their influence at their environment will also be less prominent ( see paper v ) . we determined the distribution parameters both including and excluding lines with `` ejection velocity '' , velocity displacement from the emission redshift of the quasar , greater than 3000 . a detailed analysis of the proximity effect is presented in paper v. we estimated the parameters , @xmath17 , and from eq . ( [ eq : dndz ] ) for the whole fos sample and several subsamples . we developed and utilized a computer code , based on the algorithm from murdoch et al . ( 1986 ; their equations [ a2a ] and [ a8a ] corrected ) , modified so that the integral in eq . ( a2 ) of murdoch et al . , which can not be calculated analytically in the case of variable threshold , is derived numerically . the murdoch et al . method consists of solving for @xmath17 and by finding the roots of the derivatives of the log of the likelihood function describing equation ( [ eq : dndz ] ) with respect to @xmath17 and . uncertainties in each parameter are found from a parabolic fit to @xmath23 such that 1@xmath8 corresponds to @xmath24 . the normalization , , is calculated by dividing the total number of lines by the sum of the integral of equation ( [ eq : dndz ] ) over the redshift paths in all quasars : @xmath25^{-1}. \label{eq : anaughtintegral}\ ] ] for the considered sample . for each run of our program , we tested the goodness - of - fit of the outcome using kolmogorov - smirnoff test . this treatment of the normalization differs slightly from that of the hst key project . we and the key project both solve for @xmath17 and using all lines above some rest equivalent width that is either fixed to some constant value or varied according to the signal - to - noise ratio in each spectral region . weymann et al . ( 1998 ) determine the normalization , denoted @xmath26 in their table 2 , explicitly in the maximum likelihood solution rather than by using eq . ( [ eq : anaughtintegral ] ) . each @xmath26 value listed in their table is the local line density relative to the local density of lines above 0.24 , regardless of the limiting equivalent width of the sample used . in our software , this relative scaling is not done . in other words , we always normalize to the stated constant rest equivalent width limit and do not quote a value of for the case of a variable rest equivalent width limit . table [ tab : dndz - hst ] presents a summary of our estimates of the parameters from eq . ( [ eq : dndz ] ) in several subsamples of the data set . the main result , seen clearly in table [ tab : dndz - hst ] , is that in all cases the value of the evolution index , @xmath17 , is within few tenths of 0.5 , significantly lower than the value estimated from ground - based samples of forest lines at redshifts greater than @xmath101.7 . our values are consistent with the `` no evolution '' expectations . we show the results for the @xmath27 , @xmath28 sample in figure [ fig : dndzuskp ] . the values of the k - s probability show that the power law is an excellent approximation of the distribution of lines at the redshift range covered in our sample . note that this disagrees with the result of kim , cristiani , & dodorico ( 2001 ) , who found that the break in @xmath29 slope occurred around @xmath30 . the results appear to depend only weakly on whether the lines from metal systems are included in the sample . also , it can be seen in table [ tab : dndz - hst ] that taking the proximity effect into account by excluding the lines with @xmath31 results in an insignificant change in the estimates of the number density evolution index ; the difference is of the order of 1@xmath8 or less . the results agree qualitatively with the findings of the hst qsoal key project . weymann et al . ( 1998 ) have analyzed the complete key project absorption line sample and also found that the evolution index , @xmath17 , is smaller than the one found for high - redshift samples . however , their values of @xmath17 for various subsamples range from @xmath32 to @xmath33 with uncertainties of @xmath34 . the values of @xmath17 that we find from our total fos sample are larger than those found by the key project by @xmath101.9@xmath8 in the case of the constant , 0.24 equivalent width threshold and by @xmath104.5@xmath8 in the case of a variable equivalent width threshold a comparison of these two results is shown in figure [ fig : dndzuskp ] , where we show our data and fit for @xmath35 ( solid line ) , and the @xmath29 vs. @xmath2 curve ( dotted line ) for a comparable sample from weymann et al.(1998 ; their sample # 5 , which had @xmath36 and @xmath37 ) . we see a comparable number of lines per unit redshift at low @xmath2 , but slightly ( ca . 15% ) more lines at @xmath38 . there are a few effects that may account for the discrepancy . as noted earlier ( see fig . [ fig : zcoverage ] ) , our sample has larger redshift coverage and contains markedly more lines at @xmath14 . also , our sample excludes lines below emission line . we determined the parameters of eq . ( [ eq : dndz ] ) in a subsample of our data , containing only the objects we had in common with the key project . the results were in very good agreement with weymann et al.(1998 ) ; we show the data points and the fit on fig . [ fig : dndzuskp ] . specifically , we obtained @xmath39 for the @xmath27 sample , and @xmath40 for the variable threshold case , markedly lowering any discrepancy . also , the results show the same qualitative behavior of lowering the value of @xmath17 when limiting the analysis to the key project objects . the other effect that can influence the results is the exclusion of lines below emission lines . in order to verify whether this can account for at least part of the discrepancy , we analyzed the line lists of the hst key project using our equivalent width threshold information and our metal system identifications for excluding spectral regions . including identified lines blueward of emission as the key project did we find @xmath41 and @xmath42 , for variable and 0.24 thresholds , respectively , in good agreement with the key project results . excluding lines blueward of emission , we find @xmath43 and @xmath44 for variable and 0.24 thresholds , respectively , i.e. no significant difference in both cases . we thus conclude that exclusion or inclusion of lines below does not affect the results , and that the primary source of the discrepancy between our and weyman et al.s ( 1998 ) results lies in the difference in the sample coverage . it is known that @xmath29 steepens around @xmath45 . it is obvious from fig . 5 of weymann et al . ( 1998 ) that if they included the @xmath46 quasar um18 in their sample ( it was excluded because of incomplete line identifications ; note that this object is included in the sample presented here ) , then their derived distribution would have been considerably steeper , in agreement with our result . it is possible that this , combined with the fact that we have better coverage at high-@xmath2 end of the sample , contributes to our deriving larger @xmath17 than the key project . we note , however , that the fact that k - s test shows that for our sample the power law is an excellent approximation of the number density evolution over the entire redshift range of our sample is in apparent contradiction to what inclusion of um18 would do to the key project sample , in which case their fit to a single power law would likely become worse . if the intergalactic medium evolves , then we expect that the observed @xmath17 could depend on equivalent width , or equivalently , the distribution of equivalent widths could be a function of redshift . weymann et al . ( 1998 ) found a trend of increasing @xmath17 with increasing line strength in the key project data . penton , shull , & stocke ( 2000 ) saw no difference in the evolution of strong and weak lines in the hst / ghrs sample . although the statistical significance is low , our results suggest that there appears to be a difference in the rate of evolution for weak and strong lines . in figure [ fig : hstdndz ] the solid line shows the evolution of the number density of absorbers for lines from one of the samples , with @xmath27 . on the same plot , we show the data and fits for two of its subsamples : the dashed line shows the sample with @xmath47 , the dotted line @xmath48 . as in other plots , the data have been arbitrarily binned solely for presentation ; the lines show the fits to unbinned data . while number densities of strong and weak lines seem to be comparable in the present epoch , there are more ( by roughly 25% ) strong lines at @xmath1 . in principle , the dividing line between those two subsamples lies near the point where lines leave the linear part of the curve of growth . therefore , one could try to interpret this result as an indication that the two subsamples are in fact two different populations of absorbers . however , there are two reasons why we regard this result with caution and avoid drawing far - reaching conclusions . first , the result is marginal . second , the same behavior would be expected to occur from line blending . two blending effects associated with increased density of lines at larger @xmath2 may affect the relative number of strong and weak lines : ( 1 ) two lines , both above the threshold , may blend into features indistinguishable from strong lines ( `` blending - out '' ) , or ( 2 ) two weak lines , both with below the threshold , may blend , resulting in a feature that has combined width greater than the threshold ( `` blending - in '' ) . these two effects work in opposite directions , and they are hard to quantify , especially in spectra with limited resolution ( see , e.g. , parnell & carswell 1988 , liu & jones 1988 ) . if the first of those two effects dominates over the second , the result may appear as if there were more strong lines at higher redshifts . impey , petry , & flint ( 1999 ) reported marginal evidence for quasar - to - quasar variations in the number density of lines in their sample of 10 spectra taken with hst / ghrs . our sample , which contains 165 lines of sight , is particularly well suited to test whether there is structure in the distribution of absorbers on very large scales . we find remarkably little scatter in the number of lines detected . for each object , we calculated the expected number of absorption lines by integrating eq . ( [ eq : dndz ] ) over the usable parts of the spectrum , and compared it with the number of lines observed in the spectrum . figure [ fig : dndzfluct ] shows the histogram of the deviations in units of the individual uncertainties . since the number of lines in each object can be small , the individual uncertainties were calculated using the gehrels ( 1986 ) approximation , @xmath49 , where @xmath50 is the predicted number of lines . if the distribution were random , then the histogram would be a gaussian with unit sigma . a dotted line in figure [ fig : dndzfluct ] shows the predicted gaussian distribution . the observed distribution is in fact much _ narrower _ than random : the farthest outliers deviate by only @xmath51 , and there are only a handful of objects outside @xmath52 . the explanation is that the deviations in the number of lines detected in the fos spectra is not given by counting statistics . rather , the deviations reflect the uniformity of the intergalactic medium averaged over large scales . the size scale probed is given roughly by the @xmath53 between and emission lines of each quasar , or @xmath10750 @xmath54 mpc at @xmath55 and @xmath10850 @xmath54 mpc at @xmath56 ( @xmath57 ) . we will explore the uniformity of the line distribution in more detail in a future paper . equation ( [ eq : dndz ] ) describes very well the distribution of rest equivalent widths of lines in the sample . we note , however , that the derived value of the equivalent width distribution parameter , , depends quite strongly on whether lines associated with metal systems are included in the analyzed sample . figure [ fig : wstarfit ] shows the distribution of equivalent widths in the @xmath58 , @xmath59 samples . clearly , inclusion of lines from metal systems , which ( as noted above ) tend to be strong , affects the slope of the observed distribution . it has been found that a simple power law , @xmath60 , describes the distribution of forest column densities . analyses of high redshift data ( petitjean et al.1993 ; press & rybicki 1993 ; hu et al . 1995 ; dobrzycki & bechtold 1996 ; kim , cristiani , & dodorico 2001 ) showed that the value of @xmath61 is in the @xmath62 range . analysis of low redshift data from hst / ghrs ( penton , schull , & stocke 2000 ) indicated @xmath61 near the high end of this range , at @xmath63 . however , recent analysis by dav & tripp ( 2001 ) of high resolution hst / stis spectra of two quasars suggested that at low @xmath2 the distribution is even steeper , with @xmath64 . the fos data lack the spectral resolution to perform direct voigt profile fitting , which is necessary to derive the column density directly . however , because of the large size of the sample we can obtain precise measurements of the equivalent width distribution . since column density and equivalent width are related via the curve of growth ( see , e.g. , barcons & webb 1991 ; chernomordik & ozernoy 1993 ; penton , schull , & stocke 2000 ) , the distribution of line equivalent widths can give insight into the distribution of column densities . our data do not support the @xmath65 value of the column density distribution slope . for all our forest subsamples , we get the equivalent width distribution parameter , , in the @xmath66 range with very low uncertainties , of the order of @xmath67 or less . this value of , when converted to the value of @xmath61 using the curve of growth , yields @xmath68 , depending on the doppler parameter distribution . in order to obtain the value of the slope in the vicinity of 2 , we would have to have from the @xmath69 range , which is inconsistent with our results . however , as barcons & webb ( 1991 ) pointed out , the relation between @xmath61 and @xmath70 may be affected by a subtle problem with line blending , or the presence of clustering of weak lines . since those effects are hard to quantify , it is possible that the discrepancy is in fact smaller . also , as noted above , the derivation of is rather sensitive to the selection criteria in the sample . including / excluding lines associated with metal systems in the sample leads to distinctly different value of the distribution slope . this may help explain some of the discrepancy with the dav and tripp ( 2001 ) result . an important clue to the nature of the absorption in quasar spectra is the clustering of lines in redshift space . in general , it is agreed that a weak clustering at scales of few hundred km / s is present in high redshift forest ( rauch 1998 and references therein ; more recently liske et al . 2000 ; penton , schull , & stocke 2000 ) . at low redshifts , ulmer ( 1996 ) analyzed the line clustering in the spectra of 12 objects from the hst / fos key project ( 100 lines ) and reported a significant excess of line pairs at velocity separations of 250500 . more recently , vanden berk et al . ( 1999 ) reported positive correlation for similar velocity separations , though not as strong as ulmer , in a sample consisting of the combination of ulmer s sample and fos spectra of quasars in the galactic poles ( 217 absorption lines , 22 quasars ) ; they noted that the signal seems to be coming from a small set of line `` clumps '' in a few of the objects . impey , petry , & flint ( 1999 ) found excess of nearest - neighbor pairs at separations of 250500 in their sample of 139 lines from 10 hst / ghrs spectra , but they did not find any correlation function signal on any scales . however , penton , schull , & stocke ( 2000 ) reported a clustering signal for @xmath71 in their sample of 15 hst / ghrs spectra containing 111 lines . the critical factor in the analysis of line clustering is the number of the available objects ( i.e. lines of sight ) , since correlating the line positions is possible only in one dimension . our sample is a superset of the ulmer sample and with the exception of two quasars , ton155 and 1306 + 3021 , which did not meet our sample s criteria of the vanden berk et al . sample . in our data set that has comparable selection criteria ( @xmath27 , @xmath72 ) , we have 84 objects with 622 absorption lines , i.e. seven times more lines of sight and six times more lines than ulmer and ca.four times more lines of sight and three times more lines than vanden berk et al . we calculated the correlation function by comparing the observed number of pairs of lines with the number counted in monte carlo simulations . we performed the simulations by drawing redshifts at random from the power - law distribution of @xmath29 ( eq . [ [ eq : dndz ] ] ) inside ranges identical to those in which our sample was defined . each simulation was run until we had a total number of lines equal to the total number of lines in the real sample . we ran the simulations 10,000 times . to account for blending , we removed pairs with separations smaller than 250 . however , we do expect blending to play at least some role , so that the number of expected pairs in the simulations may be overestimated , primarily in the @xmath73 bin . the clustering results in our sample are shown in figure [ fig : corr ] . solid crosses show the observed number of pairs in each velocity separation bin , while the dashed histogram shows the expected number of pairs , as calculated through the simulations . a key issue is the uncertainties of the observed data points . if the lines are clustered , simply taking the square root of the number of line pairs in each bin will cause the error of the bin to be underestimated . we calculated the error bars for the observed data points using a modified bootstrap technique , similar to the method used by fernndez - soto ( 1996 ) . in this method , one replaces parts of the spectra with randomly selected substitute parts of the spectra and then calculates the distribution of pairs , repeating this procedure a large number of times . it can be shown that the dispersion of pair counts in each bin is an unbiased estimate of the actual variance in this bin . in fig . [ fig : corr ] , a weak correlation can be seen for @xmath74 . the signal is weak , though as mentioned above it is quite possible that line blending was not all accounted for in the simulations , leading to overestimating the expected number of pairs , thus mitigating the clustering signal . the correlation found in our data has lower amplitude than the one reported by ulmer ( 1996 ) or vanden berk et al . at least part of the difference can be explained if one looks into the sample used by ulmer ( and , consequently , vanden berk et al . ) in more detail . we first note that our selection criteria excluded regions containing the cluster of four pairs in ton 153 , since this region is bluewards of the emission line . in regions that we have in common , we note the following differences : \1 . in pg 1352 + 011 , ulmer has four close pairs , coming from lines near 1843 , 1846 , 1849 , 1852 , and 1855 . we present this region in figure [ fig : kp2us1 ] . the line at 1843 was deblended by key project from a wing of a strong si line near 1841 . in our analysis ( paper iii ) it was not identified as a separate feature . we do see a line near 1849 ( paper iii ) , but it has a detection significance of @xmath75 , which is too low to be included in the sample analyzed here . all these factors result in our sample only containing one close pair in this region , the 18521855 pair . \2 . in two pairs in ulmer s sample @xmath76 in 3c 351 and @xmath77 in pg 1634 + 706 we identify one of the pair components as a heavy element line rather than as a line . this means that they are removed from consideration in our analysis of correlations . both these systems were considered by ulmer ( and vanden berk et al . ) to be pairs . in total , in the regions that we had in common , we do see five fewer close pairs . removal of those pairs lowers the number reported by ulmer ( 14 pairs ) by 36% , and the correlation signal ( i.e. the ratio of @xmath78 ) by about half . indeed , as it has been found in previous studies ( heisler , hogan , & white 1989 ; bahcall et al . 1996 ; vanden berk et al . 1999 ) , the correlation signal is dependent on a small handfull of line pairs . in fig . [ fig : corrfun ] we show the results converted to the two point correlation function , @xmath79 . on the same plot , the dashed line shows ulmer s result . we show two data points for ulmer s results in the @xmath73 bin ; the bottom one is what his result would be if the five close pairs mentioned above were removed . please note that we examined in detail only the @xmath73 bin in ulmer s data and so we do not show how other velocity bins would be affected . obviously , the differences in line identifications and deblending accounts for some but not all of the difference between the correlation strengths . clustering of lines at small velocity separations is predicted by cold dark matter ( cdm ) model simulations of the galaxy formation , the growth of large scale structure and subsequent evolution of the intergalactic medium ( cen et al . 1998 ; dav et al . these models predict that the absorbing gas traces the dark matter distribution more closely than galaxies , and so the 2-point correlation function of clouds provides a different and complementary constraint on the formation of large scale structure than the distribution of galaxies and galaxy clusters . the solid histogram on fig . [ fig : corrfun ] shows the galaxy - galaxy correlation function , taken as @xmath80 ( e.g. davis & peebles 1983 ) . we see that the galaxy - galaxy correlation function is stronger than the correlation function , as is predicted by the cdm simulations . analyses of fos data suggested that forest lines may have a tendency to be found preferably near heavy element systems . this tendency appeared to be even stronger if one considered the `` extensive '' metal systems , i.e. metal systems with four or more identified species ; those systems were often seen near `` clumps '' of lines . see bahcall et al . ( 1996 ) , jannuzi ( 1997 ) , and vanden berk et al . ( 1999 ) . given scarcity of available systems , those analyses were , for the most part , qualitative , though vanden berk et al . reported some positive signal for metal system forest correlation at @xmath81 for strong ( @xmath82 ) lines . in the @xmath83 , @xmath59 sample , we have 78 identified metal systems , 41 of which qualify as `` extensive '' . we analyzed the correlation of forest vs. those systems , using approach similar to the one used in the analysis of forest correlations . in the simulations , we fixed the positions of the metal systems at their observed locations and selected locations of forest lines at random . we compared the simulated and observed pairs of absorbers , requiring that one of the pair components be a metal system . we do not see any correlation on any scale in both cases ( fig . [ fig : corrmetals ] ) . it is , however , clear that even the entire fos sample is somewhat too small for this test . we analyzed the redshift distribution of the clouds at @xmath84 in the spectra of 165 quasars . the major results are the following . we confirm that the clouds at redshifts less than 1.7 undergo slow , if any , evolution . a single power law is an excellent approximation of the behavior of the number density evolution of lines at low redshift . we find the evolution of the cloud number density to be somewhat steeper that the one found by the hst key project , probably the result of a different proportion of @xmath85 lines in the two samples . \2 . we see marginal evidence for a difference between the evolution index for weak and strong lines . the weak lines undergo evolution in the opposite direction from the strong lines , i.e. their number per comoving volume increases with decreasing redshift . the strong lines are consistent with the no - evolution case . we see weak clustering in the forest for @xmath86 . however , the correlation is weaker than reported by ulmer ( 1996 ) or vanden berk et al . at least part of the difference can be accounted for by differences in details of the deblending and identification of a small number of absorption features . in all cases , the clustering signal is smaller than that seen for galaxies at the same redshift . the fos data set is useful for many other quasar absorption line studies . paper v reports the analysis of the proximity effect , and the evolution of the uv radiation field which may drive the evolution of @xmath29 for the forest clouds derived here ( see paper v and references therein for further discussion of this point ) . the results for clustering described in this paper can be confirmed or refuted with high resolution spectroscopy , currently planned or underway with stis on board hst . ad would like to thank a. fernndez - soto for helpful comments on the bootstrapping technique . we would like to thank the anonymous referee for several helpful comments . 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 nasa . this project was supported by stsci grants no.ar-05785.02-94a and go-06606.01 - 95 . ad acknowledges support from nasa contract no . nas8 - 39073 ( cxc ) . jb , js and mm received financial support from nsf grant ast-9617060 . js acknowledges support of the national science foundation graduate research fellowship and the zonta foundation amelia earhart fellowship . lccclccc q0002@xmath870507 & 1.900 & 00 02 46.4 & @xmath8705 07 29 & um18 & 1.447 & 1.693 & w + q0003@xmath871553 & 0.450 & 00 03 25.1 & @xmath8715 53 07 & 0003@xmath8715 & 0.223 & 0.450 & w + q0003@xmath871955 & 0.025 & 00 03 45.3 & @xmath8719 55 28 & mrk335 & 0.006 & 0.025 & + q0007@xmath871041 & 0.089 & 00 07 56.7 & @xmath8710 41 47 & iiizw2 & 0.007 & 0.089 & + q0015@xmath871612 & 0.553 & 00 15 56.7 & @xmath8716 12 46 & qso0015@xmath87162 & 0.310 & 0.553 & + q0017@xmath870209 & 0.401 & 00 17 51.1 & @xmath8702 09 46 & q0017@xmath870209 & 0.330 & 0.401 & + q0024@xmath872225 & 1.108 & 00 24 38.6 & @xmath8722 25 23 & nab0024@xmath8722 & 0.779 & 1.108 & w + q0026@xmath871259 & 0.142 & 00 26 38.1 & @xmath8712 59 29 & pg0026@xmath8712 & 0.001 & 0.142 & + q0042@xmath871010 & 0.583 & 00 42 22.8 & @xmath8710 10 28 & mc0042@xmath87101 & 0.368 & 0.583 & + q0043@xmath870354 & 0.384 & 00 43 12.6 & @xmath8703 54 00 & pg0043@xmath87039 & 0.354 & 0.384 & w + q0044@xmath870303 & 0.624 & 00 44 31.5 & @xmath8703 03 32 & pks0044@xmath87030 & 0.370 & 0.624 & w + q0050@xmath871225 & 0.061 & 00 50 58.0 & @xmath8712 25 20 & izw1 & 0.007 & 0.061 & + q0100@xmath870205 & 0.394 & 01 00 38.6 & @xmath8702 05 04 & 0100@xmath870205 & 0.295 & 0.394 & + q0102@xmath882713 & 0.780 & 01 02 16.6 & @xmath8827 13 12 & ct336 & 0.502 & 0.780 & v + q0107@xmath880235 & 0.948 & 01 07 40.3 & @xmath8802 35 51 & q0107@xmath88025a & 0.644 & 0.899 & + q0107@xmath881537 & 0.861 & 01 07 03.2 & @xmath8815 37 50 & qso0107@xmath88156 & 0.831 & 0.861 & + q0117@xmath872118 & 1.493 & 01 17 34.7 & @xmath8721 18 02 & pg0117@xmath87213 & 1.103 & 1.493 & w + q0121@xmath885903 & 0.047 & 01 21 51.2 & @xmath8859 03 59 & fairall9 & 0.013 & 0.047 & + q0122@xmath880021 & 1.070 & 01 22 55.3 & @xmath8800 21 31 & pks0122@xmath88003 & 0.747 & 1.070 & uw + q0137@xmath870116 & 0.260 & 01 37 22.9 & @xmath8701 16 35 & phl1093 & 0.234 & 0.260 & + q0159@xmath881147 & 0.669 & 01 59 30.4 & @xmath8811 47 00 & 3c57 & 0.408 & 0.669 & w + q0214@xmath871050 & 0.408 & 02 14 26.8 & @xmath8710 50 18 & pks0214@xmath8710 & 0.295 & 0.408 & + q0219@xmath874248 & 0.444 & 02 19 30.1 & @xmath8742 48 30 & 3c66a & 0.218 & 0.444 & + q0232@xmath880415 & 1.450 & 02 32 36.6 & @xmath8804 15 10 & pks0232@xmath8804 & 1.067 & 1.450 & w + q0254@xmath883327b & 1.915 & 02 54 39.4 & @xmath8833 27 24 & pks0254@xmath88334 & 1.460 & 1.693 & + q0302@xmath882223 & 1.400 & 03 02 35.7 & @xmath8822 23 29 & 1e0302@xmath88223 & 1.025 & 1.397 & + q0333@xmath873208 & 1.258 & 03 33 22.5 & @xmath8732 08 36 & nrao140 & 0.905 & 1.251 & + q0349@xmath881438 & 0.616 & 03 49 09.6 & @xmath8814 38 06 & 3c95 & 0.364 & 0.616 & uw + q0355@xmath884820 & 1.005 & 03 55 52.5 & @xmath8848 20 48 & 0355@xmath884820 & 0.828 & 1.005 & + q0405@xmath881219 & 0.574 & 04 05 27.5 & @xmath8812 19 31 & pks0405@xmath8812 & 0.328 & 0.574 & w + q0414@xmath880601 & 0.781 & 04 14 49.3 & @xmath8806 01 05 & pks0414@xmath8806 & 0.503 & 0.781 & w + q0420@xmath880127 & 0.915 & 04 20 43.5 & @xmath8801 27 29 & pks0420@xmath8801 & 0.830 & 0.915 & + q0421@xmath870157 & 2.044 & 04 21 32.7 & @xmath8701 57 32 & pks0421@xmath8701 & 1.568 & 1.685 & + q0424@xmath881309 & 2.159 & 04 24 47.7 & @xmath8813 09 33 & pks0424@xmath8813 & 1.678 & 1.693 & + q0439@xmath884319 & 0.593 & 04 39 42.8 & @xmath8843 19 24 & pks0439@xmath88433 & 0.344 & 0.593 & w + q0454@xmath882203 & 0.534 & 04 54 01.2 & @xmath8822 03 49 & pks0454@xmath8822 & 0.294 & 0.534 & + q0454@xmath870356 & 1.345 & 04 54 09.0 & @xmath8703 56 14 & pks0454@xmath87039 & 0.979 & 1.345 & + q0518@xmath884549 & 0.035 & 05 18 23.6 & @xmath8845 49 43 & pks0518@xmath8845 & 0.002 & 0.035 & + q0624@xmath876907 & 0.374 & 06 24 35.3 & @xmath8769 07 03 & hs0624@xmath876907 & 0.159 & 0.374 & w + q0637@xmath887513 & 0.656 & 06 37 23.5 & @xmath8875 13 37 & pks0637@xmath8875 & 0.397 & 0.656 & w + q0742@xmath873150 & 0.462 & 07 42 30.8 & @xmath8731 50 15 & b20742@xmath87318 & 0.329 & 0.462 & w + q0743@xmath886719 & 1.511 & 07 43 22.3 & @xmath8867 19 07 & pks0743@xmath8867 & 1.119 & 1.511 & w + q0823@xmath882220 & 0.910 & 08 23 50.0 & @xmath8822 20 34 & pks0823@xmath8822 & 0.828 & 0.910 & + q0827@xmath872421 & 0.935 & 08 27 54.4 & @xmath8724 21 07 & b20827@xmath8724 & 0.831 & 0.935 & + q0844@xmath873456 & 0.064 & 08 44 34.0 & @xmath8734 56 07 & ton951 & 0.007 & 0.064 & + q0848@xmath871623 & 1.936 & 08 48 53.7 & @xmath8716 23 40 & q0848@xmath87163 & 1.479 & 1.694 & + q0850@xmath874400 & 0.513 & 08 50 13.5 & @xmath8744 00 24 & us1867 & 0.327 & 0.513 & w + q0916@xmath875118 & 0.553 & 09 16 30.1 & @xmath8751 18 52 & ngc2841ub3 & 0.310 & 0.553 & w + q0935@xmath874141 & 1.937 & 09 35 48.8 & @xmath8741 41 55 & pg0935@xmath87416 & 1.478 & 1.693 & + q0945@xmath874053 & 1.252 & 09 45 50.1 & @xmath8740 53 43 & 4c40@xmath8824 & 0.900 & 1.252 & + q0947@xmath873940 & 0.206 & 09 47 44.9 & @xmath8739 40 54 & pg0947@xmath87396 & 0.018 & 0.206 & + q0953@xmath874129 & 0.239 & 09 53 48.3 & @xmath8741 29 40 & pg0953@xmath87414 & 0.045 & 0.239 & w + q0955@xmath873238 & 0.533 & 09 55 25.5 & @xmath8732 38 23 & 3c232 & 0.293 & 0.533 & w + q0957@xmath875608a & 1.414 & 09 57 57.4 & @xmath8756 08 22 & 0957@xmath87561a & 1.041 & 1.414 & + q0958@xmath875509 & 1.750 & 09 58 08.2 & @xmath8755 09 06 & mark132 & 1.327 & 1.694 & + q0959@xmath876827 & 0.773 & 09 59 09.7 & @xmath8768 27 47 & 0959@xmath8768w1 & 0.496 & 0.773 & w + q1001@xmath870527 & 0.161 & 10 01 43.3 & @xmath8705 27 34 & pg1001@xmath87054 & 0.002 & 0.161 & + q1001@xmath872239 & 0.974 & 10 01 58.5 & @xmath8722 39 54 & pks1001@xmath8722 & 0.830 & 0.974 & + q1001@xmath872910 & 0.329 & 10 01 10.7 & @xmath8729 10 08 & ton28 & 0.121 & 0.329 & w + q1007@xmath874147 & 0.611 & 10 07 26.1 & @xmath8741 47 25 & 4c41.21 & 0.359 & 0.611 & w + q1008@xmath871319 & 1.287 & 10 08 29.9 & @xmath8713 19 00 & pg1008@xmath87133 & 0.930 & 1.287 & w + q1010@xmath873606 & 0.070 & 10 10 07.4 & @xmath8736 06 15 & cso251 & 0.004 & 0.070 & + q1017@xmath872759 & 1.928 & 10 17 07.8 & @xmath8727 59 06 & ton34 & 1.470 & 1.693 & + q1026@xmath880045a & 1.437 & 10 26 01.7 & @xmath8800 45 22 & q1026@xmath880045@xmath88a & 1.056 & 1.437 & + q1026@xmath880045b & 1.530 & 10 26 03.6 & @xmath8800 45 06 & q1026@xmath880045@xmath88b & 1.141 & 1.530 & + q1038@xmath870625 & 1.270 & 10 38 41.0 & @xmath8706 25 58 & 4c06.41 & 0.915 & 1.270 & w + q1049@xmath880035 & 0.357 & 10 49 18.1 & @xmath8800 35 21 & pg1049@xmath88005 & 0.327 & 0.357 & w + q1055@xmath872007 & 1.110 & 10 55 37.6 & @xmath8720 07 55 & pks1055@xmath8720 & 0.830 & 1.110 & + q1100@xmath877715 & 0.311 & 11 00 27.5 & @xmath8777 15 08 & 3c249@xmath881 & 0.106 & 0.311 & w + q1104@xmath871644 & 0.634 & 11 04 36.7 & @xmath8716 44 16 & mc1104@xmath87167 & 0.379 & 0.634 & w + q1114@xmath874429 & 0.144 & 11 14 20.0 & @xmath8744 29 57 & pg1114@xmath87445 & 0.002 & 0.144 & + q1115@xmath870802a1 & 1.722 & 11 15 41.5 & @xmath8708 02 23 & pg1115@xmath87080a1 & 1.309 & 1.694 & + q1115@xmath874042 & 0.154 & 11 15 45.9 & @xmath8740 42 19 & pg1115@xmath87407 & 0.002 & 0.154 & + q1116@xmath872135 & 0.177 & 11 16 30.2 & @xmath8721 35 43 & pg1116@xmath87215 & 0.004 & 0.177 & w + q1118@xmath871252 & 0.685 & 11 18 53.5 & @xmath8712 52 43 & mc1118@xmath8712 & 0.422 & 0.685 & + q1130@xmath871108 & 0.510 & 11 30 55.0 & @xmath8711 08 57 & 1130@xmath87106y & 0.337 & 0.510 & w + q1136@xmath881334 & 0.557 & 11 36 38.6 & @xmath8813 34 05 & pks1136@xmath88135 & 0.334 & 0.557 & w + q1137@xmath876604 & 0.652 & 11 37 09.4 & @xmath8766 04 27 & 3c263.0 & 0.394 & 0.652 & uw + q1138@xmath870204 & 0.383 & 11 38 47.8 & @xmath8702 04 41 & q1138@xmath870204 & 0.351 & 0.383 & + q1146@xmath871104b & 1.010 & 11 46 09.8 & @xmath8711 04 37 & 1146@xmath87111b & 0.696 & 0.899 & + q1146@xmath871106c & 1.010 & 11 46 04.9 & @xmath8711 06 57 & 1146@xmath87111c & 0.696 & 0.891 & + q1148@xmath875454 & 0.969 & 11 48 42.6 & @xmath8754 54 13 & 1148@xmath875454 & 0.661 & 0.969 & + q1150@xmath874947 & 0.334 & 11 50 48.1 & @xmath8749 47 50 & lb2136 & 0.295 & 0.334 & + q1156@xmath872123 & 0.349 & 11 56 52.3 & @xmath8721 23 38 & tex1156@xmath87213 & 0.296 & 0.349 & + q1156@xmath872931 & 0.729 & 11 56 57.9 & @xmath8729 31 25 & 4c29.45 & 0.459 & 0.729 & + q1206@xmath874557 & 1.158 & 12 06 26.6 & @xmath8745 57 17 & pg1206@xmath87459 & 0.821 & 1.158 & + q1211@xmath871419 & 0.085 & 12 11 45.0 & @xmath8714 19 52 & pg1211@xmath871431 & 0.012 & 0.085 & + q1214@xmath871804 & 0.375 & 12 14 16.8 & @xmath8718 04 44 & q1214@xmath871804 & 0.344 & 0.373 & + q1215@xmath876423 & 1.288 & 12 15 17.1 & @xmath8764 23 46 & 4c64@xmath8815 & 0.931 & 1.288 & + q1216@xmath870655 & 0.334 & 12 16 47.8 & @xmath8706 55 17 & pg1216@xmath87069 & 0.126 & 0.334 & w + q1216@xmath875032a & 1.450 & 12 16 13.5 & @xmath8750 32 15 & hs1216@xmath875032a & 1.067 & 1.450 & + q1219@xmath870447 & 0.094 & 12 19 04.7 & @xmath8704 47 03 & 1219@xmath87047 & 0.006 & 0.094 & + q1226@xmath870219 & 0.158 & 12 26 33.2 & @xmath8702 19 42 & 3c273 & 0.002 & 0.158 & uw + q1229@xmath880207 & 1.045 & 12 29 26.0 & @xmath8802 07 32 & pks1229@xmath8802 & 0.725 & 1.045 & + q1241@xmath871737 & 1.273 & 12 41 41.0 & @xmath8717 37 28 & pg1241@xmath87176 & 0.918 & 1.273 & w + q1247@xmath872647 & 2.043 & 12 47 39.1 & @xmath8726 47 26 & pg1247@xmath87267 & 1.568 & 1.694 & + q1248@xmath873032 & 1.061 & 12 48 00.2 & @xmath8730 32 58 & b21248@xmath8730 & 0.831 & 1.061 & v + q1248@xmath873142 & 1.020 & 12 48 25.4 & @xmath8731 42 11 & cso173 & 0.704 & 0.899 & v + q1248@xmath874007 & 1.030 & 12 48 26.6 & @xmath8740 07 58 & pg1248@xmath87401 & 0.713 & 1.030 & w + q1249@xmath872929 & 0.820 & 12 49 59.6 & @xmath8729 29 38 & cso176 & 0.536 & 0.820 & v + q1250@xmath873122 & 0.780 & 12 50 52.9 & @xmath8731 22 06 & cso179 & 0.502 & 0.780 & v + q1252@xmath871157 & 0.871 & 12 52 07.7 & @xmath8711 57 21 & pks1252@xmath8711 & 0.579 & 0.871 & w + q1257@xmath873439 & 1.375 & 12 57 26.6 & @xmath8734 39 31 & b2011257@xmath8734 & 1.004 & 1.375 & vw + q1258@xmath872835 & 1.355 & 12 58 36.6 & @xmath8728 35 52 & qso1258@xmath87285 & 0.987 & 1.355 & + q1259@xmath875918 & 0.472 & 12 59 08.3 & @xmath8759 18 14 & pg1259@xmath87593 & 0.242 & 0.472 & uw + q1302@xmath881017 & 0.286 & 13 02 55.9 & @xmath8810 17 17 & pks1302@xmath88102 & 0.085 & 0.286 & w + q1305@xmath870658 & 0.602 & 13 05 22.6 & @xmath8706 58 14 & 3c281 & 0.352 & 0.602 & + q1309@xmath873531 & 0.184 & 13 09 58.4 & @xmath8735 31 15 & pg1309@xmath87355 & 0.007 & 0.184 & v + q1317@xmath872743 & 1.022 & 13 17 34.4 & @xmath8727 43 51 & ton153 & 0.706 & 1.022 & uvw + q1318@xmath872903 & 0.549 & 13 18 54.7 & @xmath8729 03 01 & ton156 & 0.319 & 0.549 & v + q1320@xmath872925 & 0.960 & 13 20 59.9 & @xmath8729 25 45 & ton157 & 0.654 & 0.958 & v + q1322@xmath876557 & 0.168 & 13 22 08.5 & @xmath8765 57 24 & pg1322@xmath87659 & 0.002 & 0.168 & + q1323@xmath876530 & 1.618 & 13 23 48.6 & @xmath8765 30 47 & 4c65.15 & 1.209 & 1.618 & + q1327@xmath882040 & 1.169 & 13 27 24.3 & @xmath8820 40 48 & pks1327@xmath88206 & 0.831 & 1.169 & + q1328@xmath873045 & 0.849 & 13 28 49.7 & @xmath8730 45 58 & 3c286.0 & 0.560 & 0.849 & v + q1329@xmath874117 & 1.930 & 13 29 29.8 & @xmath8741 17 23 & pg1329@xmath87412 & 1.472 & 1.694 & + q1333@xmath871740 & 0.554 & 13 33 36.8 & @xmath8717 40 30 & pg1333@xmath87176 & 0.347 & 0.554 & w + q1338@xmath874138 & 1.219 & 13 38 52.1 & @xmath8741 38 22 & pg1338@xmath87416 & 0.872 & 1.213 & w + q1351@xmath873153 & 1.326 & 13 51 51.3 & @xmath8731 53 45 & b21351@xmath8731 & 0.963 & 1.326 & + q1351@xmath876400 & 0.088 & 13 51 46.4 & @xmath8764 00 29 & pg1351@xmath8764 & 0.008 & 0.088 & + q1352@xmath870106 & 1.117 & 13 52 25.6 & @xmath8701 06 51 & pg1352@xmath87011 & 0.790 & 1.117 & uw + q1352@xmath871819 & 0.152 & 13 52 12.6 & @xmath8718 19 58 & pg1352@xmath87183 & 0.002 & 0.152 & + q1354@xmath871933 & 0.719 & 13 54 42.2 & @xmath8719 33 43 & pks1354@xmath8719 & 0.450 & 0.719 & w + q1404@xmath872238 & 0.098 & 14 04 02.5 & @xmath8722 38 03 & pg1404@xmath87226 & 0.002 & 0.098 & + q1407@xmath872632 & 0.944 & 14 07 07.8 & @xmath8726 32 30 & pg1407@xmath87265 & 0.640 & 0.944 & w + q1415@xmath874509 & 0.114 & 14 15 04.7 & @xmath8745 09 56 & pg1415@xmath87451 & 0.002 & 0.114 & + q1416@xmath870642 & 1.436 & 14 16 38.8 & @xmath8706 42 20 & 3c298 & 1.055 & 1.436 & + q1424@xmath881150 & 0.806 & 14 24 56.0 & @xmath8811 50 25 & pks1424@xmath88118 & 0.524 & 0.806 & w + q1425@xmath872645 & 0.366 & 14 25 21.9 & @xmath8726 45 39 & b21425@xmath8726 & 0.295 & 0.366 & + q1427@xmath874800 & 0.221 & 14 27 54.0 & @xmath8748 00 44 & pg1427@xmath87480 & 0.030 & 0.221 & + q1435@xmath880134 & 1.310 & 14 35 13.3 & @xmath8801 34 13 & q1435@xmath880134 & 0.949 & 1.304 & + q1435@xmath876349 & 2.068 & 14 35 37.3 & @xmath8763 49 36 & s41435@xmath87638 & 1.589 & 1.693 & + q1440@xmath873539 & 0.077 & 14 40 04.6 & @xmath8735 39 07 & 1440@xmath873539 & 0.004 & 0.077 & + q1444@xmath874047 & 0.267 & 14 44 50.2 & @xmath8740 47 38 & pg1444@xmath87407 & 0.069 & 0.267 & + q1512@xmath873701 & 0.371 & 15 12 47.4 & @xmath8737 01 55 & b21512@xmath8737 & 0.330 & 0.371 & w + q1517@xmath872356 & 1.903 & 15 17 08.3 & @xmath8723 56 53 & lb9612 & 1.449 & 1.694 & + q1517@xmath872357 & 1.834 & 15 17 02.2 & @xmath8723 57 44 & lb9605 & 1.391 & 1.694 & + q1521@xmath871009 & 1.324 & 15 21 59.9 & @xmath8710 09 02 & pg1522@xmath87101 & 0.962 & 1.324 & + q1538@xmath874745 & 0.770 & 15 38 01.0 & @xmath8747 45 10 & pg1538@xmath87477 & 0.493 & 0.770 & w + q1544@xmath874855 & 0.400 & 15 44 00.2 & @xmath8748 55 25 & 1543@xmath874855 & 0.182 & 0.400 & + q1618@xmath871743 & 0.555 & 16 18 07.4 & @xmath8717 43 30 & 3c334.0 & 0.315 & 0.555 & w + q1622@xmath872352 & 0.927 & 16 22 32.3 & @xmath8723 52 01 & 3c336.0 & 0.627 & 0.927 & + q1626@xmath875529 & 0.133 & 16 26 51.5 & @xmath8755 29 04 & pg1626@xmath87554 & 0.002 & 0.133 & + q1630@xmath873744 & 1.400 & 16 30 15.2 & @xmath8737 44 08 & 1630@xmath87377 & 1.094 & 1.478 & + q1634@xmath877037 & 1.337 & 16 34 51.8 & @xmath8770 37 37 & pg1634@xmath87706 & 0.978 & 1.337 & uw + q1641@xmath873954 & 0.595 & 16 41 17.7 & @xmath8739 54 10 & 3c345 & 0.346 & 0.595 & + q1704@xmath876048 & 0.371 & 17 04 03.5 & @xmath8760 48 30 & 3c351.0 & 0.160 & 0.371 & uw + q1715@xmath875331 & 1.940 & 17 15 30.3 & @xmath8753 31 27 & pg1715@xmath87535 & 1.481 & 1.694 & + q1718@xmath874807 & 1.084 & 17 18 17.9 & @xmath8748 07 10 & pg1718@xmath87481 & 0.758 & 1.084 & + q1803@xmath877827 & 0.680 & 18 03 39.4 & @xmath8778 27 54 & s51803@xmath8778 & 0.417 & 0.680 & + q1821@xmath876419 & 0.297 & 18 21 44.1 & @xmath8764 19 32 & h1821@xmath87643 & 0.094 & 0.297 & uw + q1845@xmath877943 & 0.056 & 18 45 37.5 & @xmath8779 43 06 & 3c390.3 & 0.002 & 0.056 & + q2112@xmath870556 & 0.457 & 21 12 47.6 & @xmath8705 56 10 & pg2112@xmath87059 & 0.309 & 0.457 & + q2128@xmath881220 & 0.501 & 21 28 52.9 & @xmath8812 20 21 & pks2128@xmath8812 & 0.310 & 0.501 & w + q2135@xmath881446 & 0.200 & 21 35 01.2 & @xmath8814 46 27 & pks2135@xmath88147 & 0.013 & 0.200 & + q2141@xmath871730 & 0.213 & 21 41 13.9 & @xmath8717 30 02 & ox1692141@xmath87175 & 0.023 & 0.213 & + q2145@xmath870643 & 0.990 & 21 45 36.2 & @xmath8706 43 41 & pks2145@xmath87067 & 0.679 & 0.990 & uw + q2243@xmath881222 & 0.630 & 22 43 39.9 & @xmath8812 22 40 & pks2243@xmath88123 & 0.376 & 0.630 & w + q2251@xmath881750 & 0.068 & 22 51 25.9 & @xmath8817 50 54 & mr2251@xmath88178 & 0.002 & 0.068 & + q2251@xmath871120 & 0.323 & 22 51 40.7 & @xmath8711 20 40 & pks2251@xmath87113 & 0.116 & 0.323 & w + q2251@xmath871552 & 0.859 & 22 51 29.7 & @xmath8715 52 54 & 3c454.3 & 0.569 & 0.859 & uw + q2300@xmath886823 & 0.512 & 23 00 27.9 & @xmath8868 23 47 & pks2300@xmath88683 & 0.309 & 0.512 & w + q2340@xmath880339 & 0.896 & 23 40 22.6 & @xmath8803 39 05 & pks2340@xmath88036 & 0.600 & 0.896 & w + q2344@xmath870914 & 0.672 & 23 44 03.9 & @xmath8709 14 06 & pks2344@xmath87092 & 0.411 & 0.672 & w + q2352@xmath883414 & 0.706 & 23 52 50.8 & @xmath8834 14 37 & pks2352@xmath88342 & 0.439 & 0.706 & w + cccccccccc a & var & @xmath89 & 0 & 32.96 & 1298 & [email protected] & [email protected] & & 0.85 + a & var & @xmath89 & 3000 & 29.48 & 1147 & [email protected] & [email protected] & & 0.74 + f & var & @xmath89 & 0 & 32.45 & 1157 & [email protected] & [email protected] & & 0.88 + f & var & @xmath89 & 3000 & 29.07 & 1038 & [email protected] & [email protected] & & 0.88 + a & 0.24 & @xmath89 & 3000 & 14.24 & 622 & [email protected] & [email protected] & 32.7 & 0.96 + f & 0.24 & @xmath89 & 3000 & 13.93 & 544 & [email protected] & [email protected] & 26.9 & 0.99 + f & 0.24 & 0.36 & 3000 & 13.93 & 223 & [email protected] & & 12.3 & 0.56 + f & 0.36 & @xmath89 & 3000 & 19.98 & 437 & [email protected] & [email protected] & 25.8 & 0.93 +

we introduce methods to quantify the x - ray morphologies of supernova remnants observed with the _ chandra _ x - ray telescope . these include a power - ratio technique to measure morphological asymmetries , correlation - length analysis to probe chemical segregation and distribution , and wavelet - transform analysis to quantify substructure . we demonstrate the utility and accuracy of these techniques on relevant synthetic data . additionally , we show the methods capabilities by applying them to the 55-ks _ chandra _ acis observation of the galactic supernova remnant w49b . we analyze the images of prominent emission lines in w49b and use the results to discern physical properties . we find that the iron morphology is very distinct from the other elements : it is statistically more asymmetric , more segregated , and has 25% larger emitting substructures than the lighter ions . comparatively , the silicon , sulfur , argon , and calcium are well - mixed , more isotropic , and have smaller , equally - sized emitting substructures . based on fits of _ xmm - newton _ spectra in regions identified as iron rich and iron poor , we determine that the iron in w49b must have been anisotropically ejected . we measure the abundance ratios in many regions , and we find that large , local variations are persistent throughout the remnant . we compare the mean , global abundance ratios to those predicted by spherical and bipolar core - collapse explosions ; the results are consistent with a bipolar origin from a progenitor . we calculate the filling factor of iron from the volume of its emitting substructures , enabling more precise mass estimates than previous studies . overall , this work is a first step toward rigorously describing the physical properties of supernova remnants for comparison within and between sources .

0810.0009

supernova remnants ( snrs ) are a diverse class of objects that play an essential role in the dynamics of the interstellar medium ( ism ) and that produce and distribute most of the metals in the universe @xcite . since its launch , the _ chandra _ x - ray telescope has observed over one - hundred snrs in the milky way galaxy @xcite . the sub - arcsecond spatial resolution and spectro - imaging capabilities of _ chandra _ have facilitated detailed studies of the metal - rich ejecta from supernova explosions as well as the interactions with their environments as they expand ( see weisskopf & hughes 2006 for a review ) . _ chandra _ has revealed that x - ray spectral and spatial morphologies of snrs are extremely complex . however , no systematic , quantitative methods for comparison within and between snrs have been applied to develop a global picture of these varied sources . rigorous mathematical techniques are necessary to maximally utilize these _ chandra _ datasets . toward this end , we apply well - estalished tools to characterize the x - ray morphologies of snrs . we present three methods especially suited for application to _ chandra _ advanced ccd imaging spectrometer ( acis ) observations : a multipole power - ratio technique to probe morphological symmetry , correlation - length analysis ( an adaptation of two - point correlation ) to measure chemical segregation and distribution , and wavelet - transform analysis to quantify substructure . these methods have been used previously in other astronomical contexts , and we have extended and further developed them for snr science . in @xmath22 , we describe the methods in detail and demonstrate their application to synthetic data of extended sources . in @xmath23 , we apply these techniques to the 55-ks _ chandra _ acis observation of w49b , a snr with a particularly complex spatial distribution . in @xmath24 , we discuss how the results can be used to discern between the heating and explosion properties of the source . we present our concluding remarks in @xmath25 . w49b ( [email protected] ) is a principal example of a galactic snr with complex x - ray morphology . it is the brightest ejecta - dominated snr in x - rays ( @xmath4 erg s@xmath5 ; immler & kuntz 2005 ) , and it has one of the highest surface brightnesses at 1 ghz of all galactic sources @xcite . radio maps of the source show w49b has a prominent shell structure @xmath64 in diameter , yet it has a center - filled morphology in the x - rays . hi absorption measurements indicate the remnant is 8@xmath72 kpc away @xcite , and due to its distance through the galactic plane , no optical emission is detected from the source . a recent _ spitzer _ study of snrs in the galactic plane observed filamentary structure of infrared line emission in w49b from both shocked ionic and molecular gas @xcite . w49b was first detected in x - rays by the _ einstein observatory _ @xcite . an _ exosat _ observation soon after revealed intense iron line emission from the source . _ asca _ spectra confirmed the high abundances of several metals in w49b @xcite , and the emission was described best by two distinct thermal components of optically - thin plasma . given the remnant s relatively young age ( pye et al . 1984 ; smith et al . 1985 ) , the line emission was attributed primarily to the ejecta from the progenitor star . recent high - resolution x - ray images ( see figure [ fig : three ] ) and spectra from _ chandra _ and _ xmm - newton _ @xcite have shown that the metals have remarkable spatial distributions . the emission is elongated in a centrally - bright bar that is enriched in iron and nickel and has two plumes at its eastern and western edges that are perpendicular to the axis of this bar . the eastern part of the remnant terminates sharply , while the western portion of w49b is remarkably diffuse . additionally , the western regions have weak iron emission , whereas the lower-@xmath8 elements have significant emission in these areas @xcite . two potential scenarios may account for the complex morphology of w49b ( miceli et al . 2006 , 2008 ; keohane et al . the high - metallicity and barrel - shaped morphology could be signatures of a bipolar explosion ( possibly a hypernova ) of a massive star . conversely , a complex environment around a typical , symmetric supernova explosion could give rise to inhomogeneous heating to x - ray emitting temperatures . the latter is an appealing model as it is a commonly observed phenomenon ( e.g. , snr g292.08 + 1.8 : park et al . however , this scenario would require unrealistic ejecta masses @xcite . the bipolar origin of w49b is problematic as well because the predicted rate of such events is only 1 per @xmath9 years per galaxy @xcite . thus , it is uncertain presently which scenario is the correct interpretation of w49b s exotic shape . in order to break the degeneracy of environment versus explosion effects , quantitative methods can be applied to the existing data . the _ chandra _ images of w49b contain a wealth of physical information that can be extracted using appropriate mathematical anaylses . measurement of elemental segregation and mixing are important clues to explosion and evolution histories . additionally , a rigorous description of x - ray substructure enables much more reliable estimates of elemental abundances and temperature variations than any previous work . together , these calculations can distinguish the correct interpretation of asymmetries ( anisotropic ejection or heating inhomogeneities ) , one of the biggest outstanding questions in snr studies today @xcite . the anomalous spatial distribution and substructure of w49b make it an ideal source to test our methods . we aim to define quantitatively the spatial and spectral features of w49b as a means to extract physical insights about this enigmatic source . in this section , we describe the techniques that can be used to quantify x - ray morphology of extended sources . we present their mathematical formalism as well as examples of their application to synthetic data . the power - ratio method ( prm ) was developed initially to quantify the x - ray morphology of galaxy clusters observed with _ rosat _ as a probe of their dynamical states ( buote & tsai , 1995 , 1996 ) . the technique was extended to _ chandra _ observations to measure the evolution of cluster morphology as a function of redshift @xcite . the prm measures sensitively asymmetries in an image via calculation of the multipole moments of the x - ray surface brightness in a circular aperture . we present an overview of this technique below , and we refer the reader to the above papers for the rigorous details and examples of its application to synthetic data and x - ray observations . the prm is derived similarly to the multipole expansion of the two - dimensional gravitational potential interior to radius @xmath10 : @xmath11 where the moments @xmath12 and @xmath13 are @xmath14 @xmath15 , and @xmath16 is the surface mass density . for our imaging analyses , the x - ray surface brightness replaces surface mass density in the power ratio calculation . the powers of the multipole expansion are obtained by integrating the magnitude of @xmath17 ( the _ _ m__th term in the multipole expansion of the potential ) over a circle of radius @xmath10 , @xmath18 ignoring the factor of @xmath19 , this equation reduces to @xmath20 ^ 2 \nonumber\ ] ] @xmath21 the moments @xmath22 and @xmath23 ( and consequently , the powers @xmath24 ) are sensitive to asymmetry in the x - ray surface brightness distribution , and the higher - order terms measure asymmetries at successively smaller scales relative to the remnant size . to normalize with respect to flux , we divide the powers by @xmath25 to form the power ratios , @xmath26 . @xmath27 approaches zero when the aperture is centered on the centroid of an image , so we have set the origin in all analyses to the centroid of each individual image . in this case , morphological information is given by the higher - order terms @xmath28 , @xmath29 , and @xmath30 . the quadrupole power @xmath31 is sensitive to the ellipticity of a source . the moment @xmath32 measures deviations from mirror symmetry ( such as triangular morphology ) , and it is not affected by mirror - symmetrical features . the power @xmath33 is correlated with @xmath31 but is more sensitive to smaller - scale structure than @xmath31 . thus , the power ratios provide complementary measures of x - ray morphology in extended sources . two - point correlation has been applied to measure galaxy clustering ( e.g. , davis & peebles 1983 ) as well as to map the anisotropies of the cosmic microwave background ( e.g. , kogut et al . we have adapted this method to characterize spatial distribution and segregation between elements in an x - ray source , and we call our modified technique correlation - length analysis ( cla ) . cla measures essentially whether two images have statistically similar spatial distributions . we define the correlation length of each pixel in one image as the minimum distance ( @xmath340 pixels ) to another pixel of equal intensity in a second image . the two images are normalized to have the same maximum intensity , and pixel values are rounded to give integer counts throughout the images . we calculate the correlation length for each pixel , and we create maps of these values to identify regions with and without particular emission features . additionally , we determine whether two images have similar overall spatial distributions by calculating the difference between the cumulative distributions of their correlation lengths to the cumulative distributions of each of their auto - correlations . if the two images have distinct spatial distributions , the correlation lengths between them will be disparate from those of their auto - correlations . thus , the difference in the trends of correlation length values provides a measurement of the similarity of two images . we perform auto - correlations on mock images produced via a monte carlo approach to estimate the uncertainty of the calculation . we test the utility of cla by applying the method to three groups of ten synthetic images ( figure [ fig : examplecl ] has examples and a mathematical description of the datasets ) . groups 1 and 2 have similar distributions and represent auto - correlations , while group 3 reflects comparison of disjoint distributions . we performed cla on every combination ( ninety different pairings ) of the ten images within each group and produced the three mean correlation - length cdfs and their 1-@xmath35 ranges shown in figure [ fig : cl ] . to determine whether the size of substructure influences cla results , groups 1 and 2 have clumps at fixed location but with scales selected from gaussian distributions of different variance : the clump - size distribution of group 2 has @xmath36 larger variance than that of group 1 . comparing their resulting cdfs , we find that the confidence ranges overlap for these datasets , with an average separation of 0.5@xmath35 . thus , images in groups 1 and 2 have statistically similar spatial distributions , indicating that clump - size variation does not hinder cla from distinguishing homogeneous populations . to ascertain how clump position affects cla results , clump locations of group 3 datasets were selected randomly while the clump sizes did not vary much ( with 11% of the group 1 variance ) . in this case , many pixels have large correlation lengths , so the group 3 cdf is very disparate from those of groups 1 and 2 with average separations of 3.1@xmath35 and 2.7@xmath35 , respectively . thus , group 3 has a statistically distinct spatial distribution from groups 1 and 2 , and we conclude that cla is a good measure of whether substructures have similar locations . therefore , cla is well - suited to compare morphologies within and between extended sources . wavelet - transform analysis ( wta ) was applied successfully to _ rosat _ and _ einstein _ data to extract small - scale x - ray structure of galaxy clusters @xcite . with the superior spatial resolution of _ chandra _ , this technique can be extended to characterize precisely the size and distribution of emitting regions in x - ray sources . a wavelet basis enables scale as well as spatial localization , thus wta can identify substructure , its extent , and its position . a wavelet - transformed image is a decomposed image of a signal s intensity ( from herein , power ) measured at the scale of a filter size . mathematically , a wavelet transform @xmath37 is the correlation of a signal @xmath38 in an image with the analyzing wavelet function @xmath39 : @xmath40 where @xmath41 is the scale ( or width ) of the wavelet transform . for astronomical images , a radial mexican - hat function ( the normalized second derivative of a gaussian function ) is the optimal wavelet to use because of its similar shape to gaussian signals and it removes flat features ( like diffuse emission ) since it has a zero mean @xcite . in the analysis that follows , we use a radial mexican - hat wavelet @xmath42 of the form @xmath43 wavelet - transformed images are produced by calculating @xmath44 for each pixel @xmath45 in a raw ] where @xmath46 is the number of counts in the ( @xmath47 ) pixel . if we assume the number of counts in a given pixel is independent of the counts in nearby pixels , we can use poisson statistics to calculate the variance @xmath48 of @xmath49 : @xmath50 where @xmath51 is the variance of @xmath46 in the @xmath52 pixel of a raw image . essentially , @xmath37 measures the summed intensity enclosed by the area of the mexican hat . thus , the size of an individual source can be characterized by the scale where the convolution of the wavelet and a signal reaches a maximum . to exemplify this point mathematically , we assume that a source has an isotropic gaussian shape with signal @xmath53 given by @xmath54,\ ] ] where @xmath55 is the intensity of the source located at @xmath56 . using eq . 5 and eq . 6 , the wavelet coefficient @xmath57 at this position is @xmath58 if we divide eq . 10 by @xmath41 , we find that @xmath59 eq . 11 has a global maximum at @xmath60 = @xmath35 . consequently , by identifying the peak @xmath61 value at the central pixel of an emitting region , we measure its characteristic size . to demonstrate the utility of wta for astronomical images , we apply the method to relevant synthetic datasets . we begin by using wta to extract the size of a single two - dimensional gaussian function ( from herein , a clump ) with a width @xmath35 . figure [ fig : gauss]a shows this synthetic data , and figure [ fig : gauss]b gives the plot of @xmath61 versus @xmath41 ( intensity per unit area as a function of scale ; from herein , wta plot ) for the central pixel of the clump . larger @xmath61 values correspond to more power emitted at the given scales , so local and absolute maxima indicate the size of emitting regions . indeed , the peak @xmath60 of the wta plot occurs at the exact value of @xmath35 . therefore , the method can measure accurately the size of an individual emitting region or substructure . by applying wta to all the pixels in an image and summing their results for each scale , the method can be utilized to calculate the intensity profile of an entire extended source . as an example , we have applied wta to all pixels in three synthetic datasets with randomly - distributed clumps of equal width ( figure [ fig : nsigma ] , left ) . the number of clumps in these images are varied to demonstrate the effects of increased emitting surface area , or filling factor . figure [ fig : nsigma ] ( right ; solid lines ) displays the averaged wta plot ( @xmath62 versus @xmath41 ) for the three datasets . the curves have similar shape as figure [ fig : gauss ] , and increased @xmath62 values correspond to more power emitted at a particular scale over an entire image . as the number of clumps increases and they overlap , larger agglomerates are formed . this effect increases the surface area of contiguous emitting regions , and the amount of power measured at large scales to rise . thus , increased numbers of clumps shift the peak @xmath60 to higher values , and the turnover in @xmath62 becomes less prominent . in images with many clumps , we can measure the scale of isolated structures using the wta plot for their central pixels . however , larger filling factors cause individual clumps to agglomerate , and the number of segregated structures decrease as a result . in figure [ fig : nsigma ] ( right ; dashed lines ) , we plot the averaged wta plot for the individually identified clumps in the three synthetic datasets of randomly - distributed clumps of equal width ( figure [ fig : nsigma ] , left ) . we define the centroid of individual clumps as the locations with local maxima in @xmath61 separated at least three pixels from another local maximum . in all cases , the algorithm successfully reproduces the scale of the isolated structures . generally , our ability to identify individual clumps decreases with greater emitting surface area ( see figure [ fig : nsigma ] ) . consequently , wta probes an extended source s filling factor as well as the size of its individual emitting regions . real astronomical images have noise that will naturally influence these wta results . to investigate the effects of noise , first we apply our techniques to synthetic data of pure noise ( figure [ fig : purenoise ] , left ) . the noise intensity in each pixel was selected from a poisson distribution . figure [ fig : purenoise ] ( right ) shows the resulting averaged wta plot ; the absolute maximum of the wta curve is at @xmath41 = 1 pixel . since the noise is distributed on a pixel - by - pixel basis , it is natural that the most power is emitted at the single - pixel scale . we then add the pure noise ( figure [ fig : purenoise ] , left ) to synthetic data of randomly - distributed clumps to simulate realistic images of an extended astronomical source . clump sizes were selected from gaussian distributions , and images were produced using different variances in the clump - size distribution to investigate the effects of substructure on wta results . we calculate the averaged wta plot for the datasets ; two example images and their results are shown in figure [ fig : npd ] . all curves have global maxima at @xmath63 pixel since the noise is dominant . however , we can reduce its contribution by ignoring all pixels with global maxima at the smallest scales . using this approach , the global maxima shift toward the scale of mean clump size since the most power resides there when noise is excluded . we determine that the variance in clump size did not alter the wta results either : the intensity profile is similar regardless of how divergent individual clumps are from the average scale . instrumental effects may alter wta results : in particular , an image s resolution is limited by the point - spread function ( psf ) of the detector . to investigate how an instrument s psf changes the scale where most power is emitted , we convolved the data in figure [ fig : npd]a with gaussian functions and then add noise . the resulting four images are shown in figure [ fig : convolve ] , along with their averaged wta plot . generally , the convolution did not influence the intensity profile measurement . we find that the results are unaffected for all psfs 60% or less of their scale , and the maxima in the wta plot increase with psfs larger than this fraction . based on the results from the synthetic data presented above , we conclude that wta is a useful technique to extract the individual sizes of emitting regions and to measure a source s filling factor , an important geometric parameter . noise increases the power extracted at the smallest scales , but we can limit this contribution by ignoring pixels with global maxima of a single pixel . generally , large variation in clump sizes does not reduce the accuracy in determining their intensity profiles , and wta measures the various emitting regions scales necessary to produce a given surface brightness . thus , wta is a useful tool to probe the substructure of extended sources , such as supernova remnants . although the methods are successful on these synthetic data , the techniques have uncertainties and limitations that should be considered . the power - ratio method necessitates an accurate determination of an image s centroid ; this calculation requires a source to have a minimum flux and minimum signal - to - noise ratio ( s / n ) . as shown by @xcite , circular apertures with greater than 500 net counts and s / n greater than 3.0 are sufficient to locate the centroid for unbinned _ chandra _ data . furthermore , to obtain accurate net moments of extended sources , a few criteria should be satisfied : point sources must be removed , the background moment must be much less than the source moment , and the source must fit entirely on the detector . @xcite also notes that low s / n images with highly symmetric / circular sources will have systematically higher power ratios due to noise . a monte carlo approach can be employed to estimate the uncertainty associated with noise ( as described in @xmath23.2 ) . in the case of the cla , the primary limitation is associated with the normalization of the two images to have maximum equal intensity . if the images have very disparate s / n , the normalization procedure will alter the relative quality of the images . namely , if the maximum intensity of the high s / n image is scaled down to match that of the low s / n image , weaker features in the high s / n image would be removed artificially and thus ignored in the cla results . therefore , the method is best applied to images with similar intensity prior to normalization . as with the prm , a monte carlo approach can be used to measure the the effects of noise on the resulting cumulative - distribution function ( as described in @xmath23.3 ) . unlike the other methods , noise is not a significant source of uncertainty in wta provided that certain criteria are met . we have shown in @xmath22.3.2 that we can limit the affects of noise by ignoring pixels with global maxima of a single pixel . generally , we find that with this procedure , the averaged wta plot is roughly independent of s / n for s / n @xmath34 2.0 as long as @xmath64 is much greater than 1 pixel ( @xmath64 @xmath34 8 pixels ) . if emitting regions have s / n and @xmath64 below these values , a monte carlo approach can be employed to estimate the uncertainty introduced by noise . for wta , the principal limitation arises from increased filling factor . in @xmath22.3.2 , we demonstrated that our ability to identify individual clumps diminishes with greater filling factor ( see figure [ fig : nsigma ] ) . the reason for this difficulty is that when clumps agglomerate into larger structures , it becomes challenging to distinguish the central pixel of that emitting region . thus , isolated clumps in sources with low filling factors are the best candidates for wta to measure indiviudual clump sizes . large filling factors negatively influence the averaged wta results as well : the turnover in the averaged wta plot becomes less prominent with increased emitting surface area in an extended source . generally , we find that images with filling factors above @xmath030% do not produce a statistically significant @xmath64 . in the subsequent sections , we demonstrate the application of the three techniques of @xmath22 to a particular snr , the galactic remnant w49b . we analyzed a 55-ks archival _ chandra _ acis observation of w49b ( obsid = 117 ) . the remnant was observed with the backside - illuminated s3 chip in the timed - exposure faint mode . data reduction and analysis was performed using the _ chandra _ interactive analysis of observations ( ciao ) version 3.4 . we followed the ciao data preparation thread to reprocess the level 2 x - ray data : we adjusted the charge transfer inefficiency and time - dependent gain , and we randomized the pha distribution and the event positions . _ asca _ grade 0 , 2 , 3 , 4 , and 6 events were used in the analysis , and good time intervals were applied . data below and above were ignored . throughout the paper , we assume the distance @xmath65 to w49b is @xmath65 = 8 kpc ; this value implies an acis pixel size of 0.492 @xmath6 0.019 pc . we extracted the global _ chandra _ acis x - ray spectrum of w49b ( shown in figure [ fig : spectrum ] ) using the ciao command _ specextract _ , and @xmath62.4@xmath6610@xmath67 total counts were detected from the remnant . the x - ray spectrum is quite complex : it has a thermal bremsstrahlung continuum and numerous emission features . prominent blends include both he - like ions and the h - like ions of silicon , sulfur , argon , and calcium . additionally , a strong iron line is detected . modeling of _ xmm - newton _ spectra from w49b suggests at least two plasma components are necessary to adequately fit these diverse emission features since they are produced at different electron temperatures @xcite . exposure - corrected images of the strong emission features were produced by filtering data to the narrow energy ranges of the prominent emission lines ( as labeled in figure [ fig : spectrum ] ) : si xiii ( 1.741.93 kev ) , si xiv ( 1.942.05 kev ) , s xv ( 2.252.50 kev ) , s xvi ( 2.522.71 kev ) , ar xvii ( 2.983.21 kev ) , ca xix ( 3.754.0 kev ) , and fe xxv ( 6.46.9 kev ) . we multiplied all the exposure - corrected images by the maxima of their exposure maps so that every image had units of counts . the spatial distribution of the elements that produce these lines is inhomogeneous ( figure [ fig : ions ] gives examples ) . the iron is much more prominent in the eastern portion of the remnant , while silicon and sulfur are comparatively more symmetric . additionally , all three ions have discernible x - ray substructure , particularly in the central and western regions where clumping is evident . the line emission in these images will have contamination from the underlying thermal continuum , and it is necessary to remove this component prior to imaging analyses . we estimate the flux of the bremsstrahlung and lines by phenomenologically modeling the x - ray spectra from different regions of the remnant . we find that the relative flux of the thermal emission in the narrow - band images varies by only @xmath05% over the entire source . for example , bremsstrahlung in the energy range of the si xiii line contributes 3540% of the flux for all regions . thus , we continuum - subtract the elemental images by reducing the counts in each pixel by the average fraction of flux from the thermal component in that spectral band . we calculate the power ratios @xmath28 , @xmath29 , and @xmath30 of the continuum - subtracted , exposure - corrected images of the five strongest emission features in the w49b spectrum ( si xiii , s xv , ar xvii , ca xix , and fe xxv ) using the method outlined in @xmath22.1 . these five images have @xmath0750025000 counts each ; thus , they satisfy the criteria outlined in @xmath22.4 for accurate net moment determinations . the results are listed in table 1 . for each ion image , we determined its centroid and calculated the power ratios using a aperture radius that encloses the entire remnant , @xmath68 = 150 . to account for the x - ray background , we measured the counts per pixel in a circular annulus 155 190 from the centroid of each ion image . assuming a constant background , we produced background images for each ion with the same dimensions as the source images by setting every pixel to the extracted count - per - pixel values . we calculated the moments of the source and of its background separately , and we subtracted the background moments from the source moments to obtain the powers using these net moments . to estimate the uncertainty in the power ratios , we follow the monte carlo approach employed in the x - ray cluster studies @xcite . exposure - corrected images ( normalized to have units of counts ) are adaptively - binned using the program _ adaptivebin _ @xcite such that all zero pixels are removed ( corresponding to a minimum of @xmath62 ) to smooth out noise . then , noise is added back in by taking each pixel intensity as the mean of a poisson distribution and selecting randomly a new intensity from that distribution . this process was repeated 100 times for each of the five elements , creating 100 mock images per ion . the 90% confidence limits listed in table 1 were obtained using the fifth highest and fifth lowest power ratios from the 100 mock images of each element . we find that the power ratios of fe xxv are statistically different than the other elements . the ratio @xmath28 ( a measure of ellipticity ) of iron is roughly half the values obtained for si xiii , s xv , ar xvii , and ca xix . similarly , @xmath30 ( a measure of ellipticity on smaller scales than @xmath28 ) of fe xxv is 24 times smaller than the lighter ions . however , @xmath29 ( a measure of deviations from mirror symmetry ) of fe xxv is a factor of 45 higher than the values of the other ions . the surface brightness contours of these ions demonstrate the reason for these results ( figure [ fig : contours ] ) . the large @xmath28 and @xmath30 of the lighter ions reflect their comparatively elongated distribution relative to fe xxv . the large @xmath29 of fe xxv arises because of its absence from the western part of the remnant while it has bright substructure in the central and eastern regions of w49b . the other ions have much smaller @xmath29 values since their distributions are comparatively more balanced in the eastern and western portions of the remnant . therefore , the morphology of fe xxv is quantitatively less elongated and more asymmetric than the lighter ions . we perform the correlation - length analysis ( cla ) outlined in @xmath22.2 on images of the five strongest emission lines in the _ chandra _ acis spectrum . all the ion images have roughly the same maximum pixel intensity ( @xmath05 counts ) prior to normalization , so the limitations of cla discussed in @xmath22.4 do not apply to the w49b data . we analyzed each combination of the five images and produced their correlation - length maps , with examples shown of the three most prominent lines ( fe xxv , si xiii , and s xv ) in figure [ fig : ccmaps ] . we find that all ions are well - mixed throughout the remnant , except for fe xxv which is largely absent from the plumes of the remnant relative to the other ions . additionally , we produced the correlation - length cumulative - distribution functions shown in figure [ fig : cdf ] . 1-@xmath35 confidence ranges of these cdfs were calculated by performing cla on every pair of ten mock images ( the same mock images as in @xmath23.2 ) of each ion . to determine whether the ions have similar spatial distributions , we compare the cdfs to those produced using auto - correlation ( figure [ fig : cdf ] ) . we find that the silicon - to - sulfur cdf is statistically similar to the silicon auto - correlation cdf ( with an average difference of 0.007@xmath35 ) and to the sulfur auto - correlation cdf ( with an average difference of 0.060@xmath35 ) . this result indicates that si xiii and s xv have the same spatial distribution and are well mixed . conversely , fe xxv has a very different spatial distribution than si xiii and s xv : the cdfs are very disparate from those of the auto - correlations , with average differences of 4.6@xmath35 from si xiii and 3.7@xmath35 from s xv for correlation lengths under 10 . thus , the iron is more segregated and disjoint from the other ions , whereas all except fe xxv seem to be well - mixed and similarly distributed throughout the remnant . we apply the discrete wavelet - transform analysis outlined in @xmath22.3 for the seven strong emission lines in the _ chandra _ acis spectrum using matlab version 7.4 . figure [ fig : montage ] shows all the wavelet - transformed images for five different scales . as per the discussion in @xmath22.2 , the fields display the x - ray power at the given sizes . at small scales , noise and random fluctuations dominate , and with increasing mexican - hat sizes , the distribution and substructure of each ion becomes more evident . consistent with the previous analysis , the fe xxv emission is absent from the western region of the fields , whereas other ions are more symmetrically distributed . as in the procedures outlined in @xmath22.4.2 , we determine the scale of individual x - ray emitting regions as well as the intensity profile in our acis ion images . figure [ fig : exampleclumps ] displays examples of identified clumps in the wavelet - transformed images of fe xxv and their wta plots . the method distinguished successfully the location and scale of the substructure , and many different sizes of x - ray emitting regions are evident . figure [ fig : wa ] shows the averaged wta plot for the five strongest ions in the _ chandra _ acis spectrum , fe xxv , si xiii , s xv , ar xvii , and ca xix . we have ignored the global maxima at the smallest scales since these are due only to noise and have no physical interpretation . all the ions except fe xxv have the same global maximum , @xmath60 = 17.5 = 0.67 pc . the fe xxv curve peaks at about 25% larger value , with @xmath60 = 22.5 = 0.86 pc and has 4050% more power than the other ions at scales of @xmath03060 . thus , the characteristic sizes of x - ray emitting regions of the low-@xmath8 ions are equal , whereas the scale of fe xxv clumps is greater . fe xxv contributes much less power at small scales relative to these ions as well : below scales of 15 , fe xxv emits 3060% less power than the other ions . these results indicate that the x - ray emitting plasma has a characteristic size , and this scale is similar for all ions except for fe xxv in w49b . we note that the @xmath64 of all ions is sufficiently high to avoid being limited by noise . we measure the counts from individual clumps using the ciao command _ dmextract _ for a region of area equal to the clump size . we account for the background counts using a circular region with diameter 40 located 200 southwest of the remnant s center . we calculate the x - ray flux from individual clumps by multiplying the extracted counts by the centroid energy of its emission line and dividing by the duration of the observation . in figure [ fig : flux ] , we plot the resulting clump sizes as a function of x - ray flux for fe xxv , si xiii , and s xv . the points at the largest scale reflect the size and flux of these emission lines over the entire remnant . on the small scale , the x - ray fluxes go roughly as the cube of clump size ( i.e. , volume ) . however , when extrapolated to the full size of the remnant , such a relationship would severely overpredict the flux of the source . the line fluxes over the entire remnant are @xmath665100 times less than if they were proportional to volume . the inverse of these values gives the filling factor of the emission in the remnant : 11.5% . this small value underscores the importance of considering substructure when using x - ray flux to estimate remnant properties ( such as abundance ratios or elemental masses ) . from the above analyses , it is obvious that fe xxv has a distinct x - ray morphology . the distribution of fe xxv is more asymmetric than and disjoint from the other ions , and emitting regions of fe xxv are larger than those of the lighter ions . two possible scenarios can account for these properties : the iron is insufficiently heated over large areas of the remnant , or it was ejected anisotropically during the supernova explosion . from the wavelet - transform analysis , we have identified quantitatively the regions with and without strong fe xxv emission , and we can model spectra in these locations to probe plasma properties and ultimately , to discern heating versus explosion effects . we extracted _ xmm - newton _ spectra at seventeen locations identified to have fe xxv clumps ( white circles a q in figure [ fig : locations ] ) . w49b was observed three times by _ xmm _ : 2004-apr-03 for @xmath020 ks ( obsid 0084100401 ) ; 2004-apr-05 for @xmath020 ks ( obsid 0084100501 ) ; and 2004-apr-13 for 2.5 ks ( ) . we downloaded reprocessed versions of the first two observations from the _ xmm - newton _ science archive . both observations were taken with the pn and both mos in the full window imaging mode and with the medium filter . after filtering for intervals of high background flaring , the first observation was left with 14.5 ks of exposure for the pn and 18 ks for each mos , and the second observation was left with 14.8 ks for the pn and 18 ks for each mos . for each of the locations circled in figure [ fig : locations ] , we extracted spectra from the pn and both mos detectors using the _ evselect _ tool . our extraction regions were 10@xmath69 in radius . we selected events with pattern in the 04 range for the pn and 012 range for the mos and with @xmath70 for both instruments . response files were generated for each spectrum using the _ rmfgen _ and _ arfgen _ tools to take into account the selection criteria , extraction region size , and off - axis angle . our background spectra were made from a source - free 250@xmath69 diameter circle located @xmath010 northwest of the center of the remnant . spectra were fit using xspec version 12.4.0 . data were grouped such that a minimum of five counts were in each energy bin , and the six mos and pn spectra from each region were fit simulateneously to improve statistics . standard weighting was used in calculating chi - squared values . we modeled the spectra with two components , one cool plasma with fixed solar abundances and one hotter plasma with varying supersolar abundances ( similar to miceli et al . 2006 ) , in collisional ionization equilibrium ( cie ) using the xspec model mekal ( mewe et al . 1985 , 1986 ; liedahl et al . 1995 ) . example spectra and fits from one region ( circle a in figure [ fig : locations ] ) are plotted in figure [ fig : examplespectra ] . cie occurs when the electrons and ions reach equipartition , and the electrons are heated only by coulumb collisions with the ions . the timescale @xmath71 to reach cie is related to the electron density @xmath72 ( in @xmath73 ) and the electron temperature @xmath74 ( in units of @xmath75 k ; spitzer 1978 ) : @xmath76 for a young supernova remnant like w49b , with an estimated age of @xmath01000 years , cie is only realistic with sufficiently high @xmath72 . from radio and infrared studies of w49b , @xmath72 is estimated to be quite high ( @xmath010008000 @xmath73 ; keohane et al . 2007 ) , making cie plasmas a plausible scenario . in our fitting of the _ xmm _ spectra , cie models gave chi - squared values @xmath01015% lower than non - equilibrium ionization ( nei ) models . all seventeen produced good statistical descriptions of the data ; table 2 lists the chi - squared values and degrees of freedom associated with each model . the best - fit temperatures for the regions with iron emission are given in : the range in the cold plasma temperature is ( corresponding to ) and in the hot plasma temperature is ( corresponding to ) . generally , the temperatures increase toward the eastern side of the remnant . @xcite found that the strong si xiii line arises from the cold plasma , while the other emission lines require the hot plasma . to test this result , figure [ fig : si ] shows how the silicon abundance ( denoted by color ) changes as a function of the temperatures of the cool and hot components , @xmath77 and @xmath78 . confidence contours of @xmath77 and @xmath78 are overlaid to highlight their statistically viable values . for constant @xmath77 , the silicon abundance varies less than 10% ; by contrast , for constant @xmath78 , the silicon abundance changes by a factor of @xmath634 . thus , we confirm the conclusions of @xcite that the si xiii arises mostly from the cool plasma . we extracted _ xmm - newton _ spectra at six locations identified by wta as weak in fe xxv emission and yet enhanced abundances of lighter ions ( the red circles r w in figure [ fig : locations ] ) . we fit these spectra using a similar model as the regions with strong fe xxv emission : two cie plasmas , one cool component with fixed solar abundances and one hot component with varying supersolar abundances . the resulting chi - squared values and degrees of freedom associated with each fit are given in table 2 ; all six gave statistically good descriptions of the data . to verify the lack of fe xxv emission in these regions , we measured the reduced chi - squared of the fits as a function of iron abundance and compared the results to those from regions with strong iron emission . figure [ fig : chisq ] plots the resulting reduced chi - squared values for iron abundances relative to calcium for one region with strong iron emission ( circle a in figure [ fig : locations ] ) and for one region with weak iron emission ( circle v in figure [ fig : locations ] ) . since fe xxv and ca xix form at the same electron temperatures , we plot the fe / ca ratio to establish whether iron is comparatively depleted the iron - weak region has roughly constant reduced chi - squared values for iron - to - calcium abundances ranging @xmath79 to 0.3 and rises sharply at larger ratios . by contrast , fits to the iron - strong emitting regions gave a local minimum in reduced chi - squared values at unity . consequently , the results support the wta identification of these regions as depleted in fe xxv emission . to test whether two plasmas ( and not one ) are necessary in these areas with depleted iron emission , we varied the electron temperatures of the cool component ( @xmath80 ) and of the hot component ( @xmath81 ) over a grid of values while letting all other parameters float . this process produced the confidence contours plotted in figure [ fig : cc ] . as the 99% confidence region does not intersect the axes ( i.e. , electron temperatures of zero ) , plasmas with two temperatures are statistically necessary throughout the remnant . we model the spectra with weak fe xxv emission with two cie plasmas to estimate the electron temperatures there . the six regions gave best - fit hot plasma temperatures ( corresponding to ) , very close to the values obtained for regions with strong fe xxv emission . at these electron temperatures , fe xxv should have a prominent line ( see figure [ fig : fe ] ) if iron is abundant there . however , little - to - no iron is x - ray illuminated in these regions , indicating that the iron is depleted from these parts of the remnant ( with an abundance [email protected] relative to solar ) . thus , insufficient heating is not a viable explanation for observed asymmetries , and the anomalous fe xxv distribution must arise from an anistropic ejection of iron . since we have determined that the asymmetries in w49b result from an anisotropic ejection , we now explore supernova explosions that could lead to the observed fe xxv morphology . bipolar explosions , a special class of core - collapse supernovae , are highly asymmetric explosions that would distribute heavy metals similarly to those in w49b . core - collapse supernovae occur when the iron cores of massive stars ( @xmath82 ) collapse to form neutron stars or black holes . these explosions are categorized based on whether the progenitor lost its hydrogen envelope ( type ib or type ic ) or not ( type ii ) . a core - collapse supernova could produce a bipolar explosion if the progenitor s core is undergoing sufficiently large rotation . numerous mechanisms could be responsible for the resulting bipolar nature of the event @xcite . typically , core - collapse sne have kinetic energies around @xmath83 ergs . however , in the last ten years , numerous detections of hypernovae supernovae with high ejecta velocities and sometimes large kinetic energies ( e.g. , nomoto et al 2001 ) have challenged this paradigm . some of these hypernovae may originate from bipolar explosions ( maeda & nomoto 2003 ) . with the number of bipolar explosion / hypernovae candidates growing , the distinguishing properties of these extreme sources are being revealed . in the case of sn 1998bw , the [ fe ii ] lines were broader than the [ o i ] @xmath84 doublet , indicating that the @xmath85ni ( which decays into iron ) was ejected at larger velocities with respect to the oxygen @xcite . a spherically symmetric model of the collapse could not account for this observation , and @xcite successfully reproduced the result using an axisymmetric model . in this scenario , the kinetic energy is assumed to decrease as a function of solid angle from the polar axis , causing the nickel to be more efficiently synthesized and ejected at high velocity in that direction . perpendicular to the rotational axis , mechanical heating decreases ; thus , this phenomenon predicts that lower-@xmath8 elements are ejected more isotropically and at smaller velocities than the nickel ( e.g. , mazzali et al . 2005 ) . generally , these asymmetries alter the chemical yield of bipolar explosions relative to typical cc sne . while the typical core - collapse event ejects @xmath86 of @xmath85ni , bipolar explosions can have much greater nickel yields that also increase with asphericity , progenitor mass and explosion energy @xcite . examples of nickel masses derived for bipolar explosion candidates include @xmath00.250.45 @xmath1 ( 2003dh : mazzali et al . 2003 ) , ( 2003lw : mazzali et al . 2006 ) , and @xmath00.20.7 @xmath1 ( 1998bw : woosley et al . 1999 ; kaneko et al . 2007 ) . the abundance ratios obtained in our spectral fits can be used to constain whether w49b may be the remnant of a core - collapse , bipolar explosion . @xcite performed similar analyses for the bright , central region of w49b ( circle i in fig . [ fig : locations ] ) and compared its abundance ratios to aspherical explosion models of extremely metal - poor stars by @xcite . based on low silicon and sulfur abundances relative to iron , these authors exclude massive progenitors ( @xmath87 25 @xmath1 ) as well as large explosion energies ( @xmath88 erg ) . this conclusion assumes that the abundances in the central region are reflective of the entire remnant . however , the iron abundance is greatly enhanced in that location : as listed in table 2 , this area ( circle i ) has the lowest si / fe and s / fe abundance ratios of all twenty - three regions in figure [ fig : locations ] . additionally , we have demonstrated in @xmath23 that the iron has statistically different symmetry , distribution , and substructure than the other elements . this attribute makes it imperative to average over many regions of the remnant to get an accurate representation of the abundance ratios . however , it is not adequate to estimate abundances from fits to the total spectrum because the central bar dominates the emission , giving uncharacteristicly low values for the lighter elements relative to iron . for example , the total spectrum gives a si / fe ratio of @xmath00.18 and a s / fe ratio is @xmath00.07 , close to the circle i values . we also note that the miceli ratios are derived solely from the hot plasma , which affects the accuracy of the silicon abundance determination since it comes mostly from the cool plasma ( as shown in figure [ fig : si ] ) . consequently , the miceli estimate of the silicon abundance is low since it does not include the primary heating source for that element . to obtain an accurate measurement of the global abundance ratios in w49b , we fit the _ xmm _ spectra from the twenty - three regions in figure [ fig : locations ] , seventeen with strong fe xxv emission and six with depleted fe xxv emission . we plot the mean abundance ratios of the hot plasma ( since the cool plasma is fixed to solar values ) relative to iron for silicon , sulfur , argon , and calcium in figure [ fig : abundances ] as well as list their values and 1-@xmath35 errors in table 2 . the ratios have fairly large dispersions , with all spanning factors of 58 . this result demonstrates that relative abundances of the elements change throughout the remnant , underscoring the importance of averaging over many regions in this calculation . we increase the silicon abundances plotted in figure [ fig : abundances ] by 30% from their table 2 values to account for the contribution of the cool component . we estimated this factor by re - fitting the regions spectra and fixing all elements of the cool plasma to solar metallicity except the silicon abundance , which was allowed to vary . following this procedure , the combined silicon abundance of the two plasmas was 30% greater than the hot component alone . in figure [ fig : abundances ] , we compare our abundance ratios to those predicted by six core - collapse , type ic models : four energetic , spherical explosions @xcite and two aspherical , bipolar explosions ( 25a , 25b : maeda & nomoto 2003 ) . the aspherical cases have the best agreement with our measured abundance ratios ; both models values are within the 1-@xmath35 range for all elements in the w49b spectrum . the primary difference between the two aspherical models is their jet opening angles @xmath89 : 25a has @xmath89 = 15@xmath90 and 25b has @xmath89 = 45@xmath90 . for the fixed kinetic energy , a larger @xmath89 means less mechanical energy is released per unit solid angle in 25b , causing the explosion to appear less luminous and altering the nucleosynthetic products . despite these distinctions , the dispersion in measured abundance ratios for w49b prevents us from discriminating between these two alternatives . nonetheless , we can exclude the spherical explosion cases based on the silicon and sulfur ratios . therefore , we find that a bipolar origin with typical or slightly larger kinetic energies of w49b is consistent with its global abundance ratios . the explosion mechanism can also be constrained by estimating the mass of metals in w49b . we perform this calculation using the elements filling factor ( as determined by the wta ) and the strength of the emission lines . appendix b of @xcite summarizes a similar method to find masses of individual regions of a source ; we adapt this technique to the full w49b remnant and account for substructure using wta . masses are determined based on parameters given in the x - ray spectral fits . for a given electron temperature , the flux of an emission line is proportional to its emission measure , @xmath91 , where @xmath92 is the ion number density and @xmath93 is the emitting volume . if @xmath94 is defined as the ratio of electron and ion number densities , @xmath95 , then @xmath92 is given by @xmath96 where we have set @xmath97 , the volume of a chosen emitting region . from there , it is simple to derive the ion mass @xmath98 over the remnant : @xmath99 where @xmath100 is the mass per ion in grams and @xmath101 is the filling factor of the ion s emission over the remnant . the primary uncertainty in the above mass calculation is the electron number density @xmath72 . however , since the plasma is in a collisional equilibrium state , we can set a physically motivated lower limit on @xmath72 . by setting @xmath71 to the age of the remnant in equation [ eq : spitzer ] and setting the electron temperature to the greatest value obtained by our spectral fits , , we find . additionally , two reasonable scenarios define a strict range for @xmath102 ( see @xmath23 of hughes et al . 2003 for a discussion ) . in the extreme case of a pure metal plasma , there are 50 electrons per fe ion ( @xmath103 ) . conversely , if the plasma has solar abundance ( a large hydrogen content ) , there are 3@xmath6610@xmath104 electrons per fe ion . an intermediate case , one where hydrogen mass is equal to that of metals ( where ) , is also plausible . our uncertainties are defined by the range of possible @xmath72 , and our values are consistent with the results of @xcite . we can determine the filling factor @xmath101 using our wta results and direct observables . the filling factor is defined as the ratio of the summed volume of individual clumps @xmath105 to the total volume of the entire source , @xmath106 . assuming the distance @xmath107 kpc , we estimate the radius of x - ray emission in w49b is @xmath108 138 = 5.3 pc = , giving . we find the summed volume of individual clumps by multiplying the mean volume of clumps from our wta , @xmath109 , by the number of clumps , @xmath110 . the number of clumps is determined simply by computing how many clumps with @xmath111 and a flux @xmath112 are necessary to account for the total flux @xmath113 of the source ( as calculated using the procedure in @xmath23.3 : @xmath114 we opt to measure @xmath110 this way instead of counting all the clumps in the wta images because the latter would require us to define a @xmath61 threshold for an individual clump . putting all of these relations together , the filling factor @xmath101 is @xmath115 thus , @xmath116 we use equation [ eq : all ] to calculate the total mass of iron in w49b . from figure [ fig : wa ] , = 0.86 pc for iron , and from figure [ fig : flux ] , @xmath117 for iron . these values give a filling factor @xmath101 = 0.016 , an order of magnitude smaller than previous estimates that were based on assumptions about the extension of the emitting plasma @xcite . thus , a rigorous measurement of @xmath101 demonstrates that the value is actually much smaller , and iron especially must emit its power in concentrated locations of the remnant . we find the emission measure of iron @xmath118 by the following procedure . using the model from @xmath24.1 , we fit the global chandra _ acis _ x - ray spectrum of w49b ( figure [ fig : spectrum ] ) to obtain the emission measure of the hot - plasma component : @xmath119 = 1.805 @xmath66 10@xmath120 @xmath73 . to compute the emission measure of an individual line @xmath121 , we reduce @xmath119 by the ratio of flux in the line to that of the entire spectrum . x - ray fluxes are computed using the xspec command _ flux _ over the energy ranges of interest for a phenomenological model of the lines and continuum.based on this procedure , iron contributes 23% of the flux to the hot component , so . using the equations above , we find an iron mass of @xmath122 @xmath6 0.12 @xmath1 assuming a solar abundance plasma . for the case of hydrogen mass equal to that of metals , we obtain an iron mass of @xmath122 @xmath6 0.80 @xmath1 . the condition of a plasma in collisional equilibrium defines the upper - limit iron mass in w49b ( since it gives the minimum @xmath72 necessary to reach cie by the age of the remnant ) : @xmath122 @xmath6 1.29 @xmath1 . while this range of iron masses is large ( reflecting the uncertainty in electron density ) , masses above 1.29 @xmath1 and below 0.12 @xmath1 can be excluded because it would require physically implausible metal abundances . furthermore , the large iron mass can not be attributed to a thermonuclear explosion since the abundance ratios relative to iron in figure [ fig : abundances ] are much greater than those expected from type ia events . although the derived iron masses are larger than those usually predicted for core - collapse supernovae , the lower limit is close to those values for typical bipolar explosions ( @xmath123 ) . additionally , as noted in @xmath24.2 , recent models of bipolar explosions demonstrate that nickel yields ( which decays into iron ) can actually be much greater than the typical core - collapse scenario ( up to 4 @xmath1 ) with increased asphericity , explosion energy , or progenitor mass ( umeda & nomoto 2008 ) . the anisotropic ejection of iron ( as shown in @xmath24.1 ) and abundance ratios ( given in @xmath24.2 ) in w49b are consistent with such an asymmetric explosion that would overproduce nickel . generally , we emphasize that this calculation is an order - of - magnitude estimate , and precise measurement of the filling factor is an important step in obtaining a plausible range of masses . a few factors contribute to the uncertainty in our analyses . as noted previously , the largest source of error is the electron number density since we do not know precisely the plasma s composition . we attempt to limit this uncertainty by defining a plausible range in its value . another source of error is the physical modeling of the x - ray spectrum . while the cie model produces a chi - squared value lower than an nei plasma , it is barely a statistically significant difference ( @xmath010% ) . however , nei models give abundance ratios within the errors of our cie results , so we expect this uncertainty is minimal . other sources of error are instrumental in nature : the off - axis point - spread function ( psf ) effects with _ chandra _ and the possible clump contamination with _ xmm - newton _ from its limited spatial resolution . the psf of _ chandra _ acis increases as a function of energy and of off - axis angle . at its aimpoint , the psf is @xmath01 at 1.49 kev and @xmath02 at 6.4 kev ; at a distance 6 off - axis , the psf increases to @xmath03 at 1.49 kev and @xmath05 at 6.4 kev ( where we have characterized the psf as the scale where the encircled energy fraction is 90% ) . we anticipate that these effects do not alter our results much since the overall extent of w49b is small ( @xmath04 in diameter ) and the characteristic clump sizes of all ions are many factors greater than these values . we minimize the uncertainty associated with _ xmm - newton _ clump contamination by selecting emitting regions that were sufficiently isolated from other substructure . the snr mass estimate in this paper is the first with the contribution of substructure rigorously defined . measurement of ejecta masses requires detailed knowledge about the scale of the emitting regions since emission measure increases linearly with volume . previous snr studies assume generally that a source s emission is homogeneous ( e.g. , hughes & singh 1994 ) or estimate the filling factor from basic arguments about the plasma s extension ( e.g. , miceli et al . 2006 ) when calculating physical parameters . however , the complex ion distribution and low filling factor in w49b demonstrates the necessity for more rigorous approaches . consequently , it is vital to probe accurately the role of substructure when determining global properties of a supernova remnant . in this paper , we have explored three methods that quantify the morphologies of extended sources . we have demonstrated the accuracy of these techniques on relevant synthetic data , and we have exemplified the utility of these tools by applying them to the chandra _ acis _ observation of the snr w49b . using a multipole power - ratio technique , we measured the asymmetry of the ion distributions in w49b . we discovered that the morphology of fe xxv is less elongated and more asymmetric than the lighter ions . we applied an adapted two - point correlation method called correlation - length analysis to compare the ions spatial distributions . we determined that the fe xxv emission is more segregated and disjoint from that of the lower-@xmath8 ions , which are well - mixed and similarly distributed throughout the remnant . additionally , we measured properties of the x - ray ions substructure in w49b using wavelet - transform analysis . we found that the mean clump scale of fe xxv is roughly 25% larger than those of the lighter ions . together , these results demonstrate quantitatively that fe xxv has a distinct x - ray morphology . two potential scenarios have been proposed to explain the anomalous properties of iron in w49b ( miceli et al . 2006 ; 2008 ; keohane et al . 2007 ) : insufficient heating of iron over large areas of the remnant or anisotropic ejection of iron during the supernova explosion . by fitting _ xmm - newton _ spectra in twenty - three locations with strong and weak fe xxv emission ( as identified by our wavelet - transform analysis ) , we determined that the electron temperatures throughout the remnant should heat any iron there to radiate at x - ray temperatures . thus , iron is strongly depleted from the identified regions , and the distinct x - ray morphology must arise from an anisotropic ejection . therefore , any models used to describe the explosion mechanism of w49b must include a means for the iron to be ejected an asymmetric way . we further constrained the nature of the explosion by measuring the mean abundance ratios of elements in our twenty - three x - ray spectral fits . the ratios varied tremendously in different locations , demonstrating the importance of averaging the ratios over many regions to obtain the global abundances in w49b . we found that the mean ratios are consistent with bipolar explosion models of massive stars . this type of explosion could also account for the anistropic ejection of nickel since heavy elements are preferentially ejected along the polar axis of the progenitor in this scenario . we calculated the total iron mass in w49b by determining the filling factor ( using our wavelet - transform analysis ) and the emission measure . we found that the iron mass and abundance ratios are broadly consistent with a core - collapse supernova . this work is a first attempt to rigorously describe the physical properties of supernova remnants . snrs play a vital role in our galaxy , and the wealth of data available on these sources is tremendous . robust methods for comparison within and between these complex targets can advance snr science as well as probe many poorly understood phenomena . we plan to apply the techniques presented here to other sources to constrain further processes as particle acceleration , interactions with environments , and explosion mechanisms . we thank k. c. schlaufman and s. e. woosley for helpful discussions and contributions . this work is supported by doe scidac de - fc02 - 01er41176 ( lal and er - r ) and a national science foundation graduate research fellowship ( lal ) . green , d. a. 2006 , ` a catalogue of galactic supernova remnants ( 2006 april version ) ' , astrophysics group , cavendish laboratory , cambridge , united kingdom ( available at `` http://www.mrao.cam.ac.uk/surveys/snrs/ '' ) cccc fe xxv & 72.68@xmath124 & 15.88@xmath125 & 0.38@xmath126 + si xiii & 137.54@xmath127 & 2.87@xmath128 & 0.92@xmath129 + s xv & 141.64@xmath130 & 4.15@xmath131 & 1.45@xmath132 + ar xvii & 141.29@xmath133 & 3.45@xmath134 & 1.51@xmath135 + ca xix & 136.19@xmath136 & 3.42@xmath137 & 1.76@xmath138 + cccccccc a & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 939.7 / 918 + b & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 482.5 / 514 + c & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 694.3 / 701 + d & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 877.9 / 934 + e & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 832.7 / 855 + f & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 461.5 / 480 + g & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 387.9 / 416 + h & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1358 / 1405 + i & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1366 / 1410 + j & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 351.3 / 388 + k & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 663.9 / 676 + l & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 617.6 / 657 + m & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 644.7 / 738 + n & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 819.3 / 868 + o & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 863.9 / 941 + p & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 820.0 / 818 + q & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 207.9 / 245 + r & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 217.0 / 271 + s & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 302.0 / 339 + t & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 327.7 / 376 + u & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 558.1 / 627 + v & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 276.9 / 319 + w & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 242.5 / 283 +

we have adapted coupled escape probability , a new exact method of solving radiative transfer problems , for use in asymmetrical spherical situations . our model is intended specifically for use in modeling optically thick cometary comae , although not limited to such use . this method enables the accurate modeling of comets spectra even in the potentially optically thick regions nearest the nucleus , such as those seen in deep impact observations of 9p / tempel 1 and epoxi observations of 103p / hartley 2 .

1212.1226

comets are understood to be frozen remnants from the formation of our solar system . as such , their chemical composition is of great significance to understanding the origin of the planets and the distribution of important molecules , including water , throughout the solar system . this was and is a major goal of the deep impact and epoxi missions , among others , as well as ground - based observations of comets . recent observations , in particular those of the deep impact and epoxi missions ( see e.g. feaga , et al . 2007 or ahearn , et al . 2011 ) , have provided better spectra of a cometary coma than were , in general , previously available . these observations include spectra with high spatial resolution very near to the nucleus . no previous observations had as much spatially well - resolved spectral data , and thus there was little observationally driven need to pay special or close attention to the densest part of the coma . ground - based observations could only see optically thick regions of comae for the brightest and/or most active of comets . ( e.g. hale bopp ; see disanti et al . 2001 . ) therefore many earlier studies that modeled spectra of comae , in keeping with the available observations of the time , did not attempt to calculate optical depth effects on spectra . optically thin comae were assumed , since the field of view in those observations being modeled would be dominated by the majority of the coma far from the nucleus , which is optically thin ( see e.g. chin & weaver 1984 ; crovisier & le bourlot 1983 ) . however , with the proliferation of space missions to comets , as well as much better instruments for ground - based observations ( see , e.g. disanti et al . 1999 ) , this is no longer a truly tenable approach . our goal is to better understand the abundances , distributions and creation mechanisms of various gases observed in comae , in particular of comet 9p / tempel 1 , the target of the deep impact mission , and 103p / hartley 2 , the subject of the epoxi mission . in order to do so , we have built a computer model of the spectrum of the comet s coma which includes the difficult and often ignored problem of accurately including radiative transfer to account for the potentially optically thick coma ( or regions of the coma ) near the nucleus . this model will facilitate analyzing the actual spectral data from the deep impact and epoxi missions to better determine abundances of key species , including co , co@xmath0 , and h@xmath0o , as well as remote sensing data on active comets . we begin our modeling of ir ro - vibrational spectra of a coma by initially following the method used by chin & weaver 1984 , crovisier 1987 , and others with some minor improvements . as in those models , we assume a constant expansion velocity , thus linearly relating any radial distance to a specific time since a `` parcel '' of gas was released from the surface of the nucleus . therefore we numerically integrate over time the linear differential equations defined by the einstein coefficients and collisional rate coefficients to get fractional molecular enegry level populations for each distance from the nucleus , from which we calculate emission spectra . @xmath1 here @xmath2 and @xmath3 indicate energy levels of a molecule with the @xmath4 s with those indices being the corresponding level populations . the einstein coefficients between levels x and y are @xmath5 and @xmath6 , and c is a similar collisional coefficient . we use @xmath7 for the mean intensity of radiation at the frequency corresponding to the transition between x and y , @xmath8 is the energy difference between levels x and y , and @xmath9 and t have their usual meanings . the summations are over all levels i , l or j which have a transition into or out of level k. for collisional coefficients @xmath10 where @xmath11 is the number density of h@xmath0o ( assumed to be the dominant collisional partner ) @xmath12 is the collisional cross section for a given transition and @xmath13 is the mean ( thermal ) molecular speed . this allows us to include a time variable production rate . we use such a coma integration as the initial basis for our coma model before including potential optical depth effects . our primary improvement is the inclusion of radiative transfer calculations using our spherical adaptation of the coupled escape probability method ( hereafter , `` cep '' ; see elitzur & asensio ramos 2006 , hereafter , `` cep06 '' ) to more correctly model optically thick ( or potentially thick ) regions of cometary comae . this is described in detail below , and is the main part of this paper . we use the coma integration results to provide the `` initial guess '' values for populations used in the subsequent radiative transfer calculations using cep . for the purposes of the initial coma model , we treat the comet as spherically symmetric , and as having a uniform and constant gas production rate over its entire surface . the outward speed of the gas is also assumed to be constant , as per chin & weaver 1984 . while this is not strictly physically accurate ( see e.g. combi 1996 ) the variation over the majority of the coma is relatively small . we use a temperature profile that varies with radial distance from the nucleus , having closely followed combi s 1989 model ( see fig . [ combi200 ] . ) these approximations make integration over time equivalent to calculating these values over increasing distances from the comet nucleus for a `` shell '' of gas expanding outwards from the nucleus . we ignore the photodestruction of co in our coma model , due to its long lifetime ( see crovisier 1994 . ) the lifetime is @xmath14 s , and we are integrating out to @xmath15 km with an expansion velocity of 0.8 km s@xmath16 . ( note that others , e.g. morgenthaler , et al . 2011 , find a shorter lifetime , but still @xmath17 s , which is still large enough that it can be neglected in our modeling out to 10@xmath18 km . ) the model can include coma morphology features as well , each modeled with its own coma integration using separate conditions . such features , as seen in the deep impact and epoxi encounters , are a main motivation for creating this model to better understand possible optical depth effects in the near nucleus regions of the coma . ( see e.g. feaga , et al . 2007 or ahearn , et al . 2011 ) after describing our method in section 2 , we present ( in section 3 ) some results demonstrating its use in better understanding possible optically thick spectra for the carbon monoxide 1 - 0 ( x@xmath19 ) band in spherically symmetric comae . these may be useful for ground - based observers ( or space telescope observations of comets ) to better fathom the depths of cometary spectra . forthcoming model results for co@xmath0 , h@xmath0o and near - nucleus morphological features will follow ( in other papers ) . this section describes our adaption of the coupled escape probability radiative transfer technique to spherical cases in which the plane parallel approximation is not appropriate , including most cometary problems . ( note that yun , et al . 2009 , previously adapted cep for _ purely symmetric _ spherical cases . ) we have developed this model specifically for use in studying cometary comae , but it could be applied to many other astrophysical phenomena as well ( e.g. planetary atmospheres , molecular clouds , etc . ) . we derive expressions analogous to the relevant cep06 equations , adapted from the plane parallel situation to spherically symmetric cases . then we include asymmetries of two different types : radiation from an outside source ( here , the sun ) and non - uniform conditions due to morphology . in brief , the cep method ( see cep06 ) divides up a plane - parallel `` slab '' into `` zones '' ( each of which has uniform properties ) and calculates the net radiative bracket ( `` @xmath20 '' ) that is multiplied by the einstein a coefficient in the equations of statistical equilibrium ( see equation [ bigeqn ] ) for each radiative transition for each zone . ( note that the `` p '' term effectively combines the `` b '' terms into the `` a '' term . see cep06 for more detail . ) the innovation of cep is that the net radiative brackets for each zone can accurately represent the contributions of _ all _ zones emission and absorption to / from other zones . ( as opposed to `` plain '' escape probability where a similar factor added to the statistical equilibrium equations is only a _ local _ approximation of a photon s likelihood of escaping the entire slab . see , bockelee - morvan 1987 , zakharov , et al . 2007 and litvak & kuiper 1982 . ) the statistical equilibrium equations for all zones , with the inclusion of the net radiative bracket , form a single non - linear matrix . this matrix can be solved using an algorithm for non - linear matrix solving such as newton s method . we use functions from numerical recipes in c ( see press , et al . 1992 ) for the newton based matrix solver , as well as other calculations such as numerical integration . further computational details can be found in the appendix . this solution yields the fractional populations of molecular energy levels for each zone . from these , the flux emitted by the slab can be calculated . see cep06 for more details and the derivations of the original plane - parallel equations to which we make reference below . in cep06 equation 7 , elitzur & asensio ramos derive a purely analytical expression for the net radiative bracket in a plane parallel slab , based on the formal solution of the radiative transfer equation and the definition of the net radiative bracket : @xmath21 @xmath22 where @xmath23 and @xmath24 refer to optical depths for a specific wavenumber , with @xmath25 being the overall optical thickness of the slab , @xmath26 or @xmath27 to the source function for a given line , ( @xmath28 ) @xmath29 is a dimensionless line profile , and @xmath30 cos@xmath31 , where @xmath31 is the angle of a given ray measured from the normal to the plane . we use a spherical analog to this theoretical expression ( i.e. as opposed to the discretized expression they introduce later involving a number of `` zones '' ) for the net radiative bracket : @xmath32 @xmath33 d\omega \;\;\;\;\;\;\end{aligned}\ ] ] where @xmath34 refers to the net radiative bracket at any point labeled by the coordinates @xmath35 in spherical coordinates with the origin at the center of some sphere of interest . ( in our study , the sphere is centered on the comet nucleus ; it could be the center of any spherical astronomical object for any given case . ) as in cep06 , @xmath36 is a dimensionless line profile , normalized so that @xmath37 , where @xmath38 is the dimensionless line shift from line center @xmath39 and @xmath40 is the doppler line width , @xmath41 . in the present work , we neglect doppler offsets between different locations in the coma with relative velocities . this is due to our work being primarily motivated by the deep impact observations of the near - nucleus coma ( see feaga , et al . 2007 ) , for which gas tends to be released , broadly speaking , towards one direction / side of the nucleus . for this situation , doppler shifts are likely to be less than line broadening . in future work , we plan to include a more generally accurate treatment . both @xmath42 and @xmath43 refer to optical depths and @xmath44 to the source function _ as viewed from the coordinates _ @xmath35 along a `` pencil '' of solid angle @xmath45 . ( see fig . [ fig1 ] . ) the optical depth along a `` pencil '' ( or cone ) of solid angle @xmath45 is , of course , highly dependent on the particular direction of a given solid angle element @xmath45 , i.e. dependent on @xmath46 , the direction of a vector centered on @xmath47 pointing along the centerline of @xmath45 . ( see fig . [ fig1 ] . ) the above cep06 expression for @xmath48 ( equation [ cepeqn7 ] ) is turned into a discrete expression by dividing the slab into multiple `` zones '' ( labeled with indices @xmath49 & @xmath50 ) . this yields cep06 equation 14 : @xmath51 this expression for @xmath20 can be calculated for each zone ( and each wavenumber / frequency ) and the resultant @xmath20 s included in the equations of statistical equilibrium , which are then solved . ( see also cep06 equations 32 - 36 , and accompanying cep06 text , for a more complete discussion . ) similar to cep06 equation 14 , for the purposes of our integration of a discretized @xmath52 the sphere is divided into spherical shells ( analogous to the plane - parallel zones of the original cep ) where @xmath49 ( or @xmath50 ) is a shell index . ( see fig . [ fig2 ] . ) the integration is broken down into a sum of integrals along different `` cones '' of solid angle ( the aforementioned @xmath45 s ) . note that although @xmath45 may seem somewhat conceptually analogous to @xmath53 in the plane - parallel case , we can not use the convenient exponential integrals as in the plane - parallel situation , but instead must integrate along each @xmath45 separately . @xmath54 d\omega \qquad \nonumber \\\end{aligned}\ ] ] here we have dropped the coordinate subscripts @xmath55 for clarity / simplicity and use shell indices instead ( where some shell @xmath49 contains the point defined by @xmath55 ) . note that the `` z '' in the summation , the maximum number of shells along a given @xmath45 , will be different for each @xmath45 . for each `` cone '' of solid angle we will sum @xmath56 from `` z '' ( the outermost shell ) to @xmath57 , where @xmath57 is the shell adjacent to shell @xmath49 . in actual practice ( i.e. computer implementation ) , the integration of the source function of @xmath45 over @xmath58 steradians will also be done by a discrete summation . along any particular element of @xmath45 originating in shell @xmath49 , this sum will be : @xmath59 note that , unlike the plane - parallel case , each @xmath60 must be calculated explicitly from the geometry and local molecular energy level populations for each shell along the `` pencil '' : @xmath61 where @xmath62 is the absorption coefficient in region @xmath50 and @xmath63 is the distance through that region . note that this distance may vary over the width of a given @xmath45 but we can either approximate using the centerline of @xmath45 or derive a @xmath64-dependent expression and calculate a proper integration to get a mean value over the region . ( in our implementation we use the former option , for the sake of simplicity . ) this integration over solid angle is more tedious than in the plane - parallel case ( which was able to use exponential integrals over a zone s @xmath65 ) and more computationally costly , but straightforward enough to be feasible . turning this all into a fully discrete expression for @xmath52 ( for shell `` @xmath49 '' ) we get @xmath66 \quad\end{aligned}\ ] ] where @xmath67 is the maximum numbered spherical shell and all the quantities indicated by the subscript @xmath68 are dependent on a particular direction viewed from a given point ( or shell @xmath49 , in a spherically symmetric case ) . our motivating interest in this study is comets comae , which are _ not _ spherically symmetric cases . to adapt cep to this asymmetry we divide each shell further into `` regions . '' in the case of comets , one source of asymmetry is incident solar radiation coming from one direction ( outside the outermost shell ) . therefore , the natural way to further divide shells into regions is along lines parallel to the direction of solar radiation ( i.e. the center - of - comet - to - center - of - sun line , which we will arbitrarily label as the @xmath69-axis . ) thus we superimpose a set of co - axial cylinders ( with radii equal to corresponding shell radii , for the sake of simplicity ) on the shells to divide the coma into regions bounded by two cylinders and two spherical shells . ( note that some regions , specifically those along the @xmath70 plane perpendicular to the solar radiation , are only bounded by an inner cylinder and outer sphere . [ fig3 ] & [ fig4 ] . ) these regions form annuli or rings of unusual , but easily envisioned , cross - sections . ( see fig . [ fig4 ] . ) incident solar radiation is parallel to the @xmath69-axis ( due to our choice of the @xmath69 direction ) . hence each ray of sunlight travels along the axial direction of a specific cylinder . for each region , the solar radiation absorbed is calculated based on the relevant optical depths along that direction ( the @xmath60 s ) of those regions in the same cylinder that are closer to the sun than the given region . this is similar to the case of external radiation described in cep06 appendix a , equations a3 and a4 , in which we simply set @xmath71 , due to the above constraint of cylinders being co - axial with the incident solar radiation : @xmath72 \\ \nonumber \\ & where \nonumber \\ \nonumber \\ & \gamma ( \tau ) = \int_{-\infty}^{\infty } [ 1 - e^{-\tau\phi(x ) } ] dx \nonumber\end{aligned}\ ] ] where @xmath73 is the average over a region @xmath49 of @xmath74 , the contribution of external radiation , @xmath75 , to the mean intensity of the region , which is to be included in the equations of statistical equilibrium ( equation [ bigeqn ] ) by addition to the `` b '' term . from a purely radiative standpoint , assuming that within each region there exists uniform density , temperature and other physical conditions , the radiative excitation of molecules ( hence , the emission and absorption ) in each region / annulus should be equal throughout the region . in our expanded cep implementation , these regions are analogous to zones in the plane - parallel cep . each region s radiative effect or contribution to each other region , i.e. the net radiative bracket , must be calculated . note that self - irradiation from around an annulus must also be taken into account , as well as irradiation from other regions . once this calculation is done , the entire region will be equal with respect to radiative processes ( i.e. there is symmetry around the @xmath69-axis ) . if models of distantly observed comets were all we needed , this might be sufficient . but we are motivated by the desire to better understand deep impact & epoxi spectral observations that have very high spatial resolution around the comets nuclei . ( see e.g. feaga et al . 2007 and feaga , et al . therefore the above radiative treatment alone is insufficiently asymmetric to _ fully _ model a cometary coma , when coma morphology is included in the model . the inclusion of morphology undoes the aforementioned symmetry around the @xmath69-axis within each annulus / region . these observations are one of the primary motives for this study , and therefore these morphological asymmetries must also be dealt with appropriately in this model . to model morphological features , we use a cone shape superimposed over the aforementioned divisions into regions . ( other geometric shapes could also have been used . we chose to implement a cone due to its similarity in shape to many observed coma features . ) a cone of arbitrary orientation and size with its vertex at the center of the sphere creates intersections with the above - described regions . each of these is then added as a sub - region , which may have different properties from the surrounding ( or subsumed / replaced ) region . each sub - region can posess different initial conditions from the surrounding region . thus morphological features , which by their nature tend not to be axisymmetric around our @xmath69-axis , can be included in the model . it should be noted that these sub - regions are only included as necessary . thus for those annuli that _ do _ have constant axisymmetric conditions ( i.e. no interesting morphological features impinging on them ) , we can save time and memory computationally by leaving them undivided , as they would have been originally . given the above geometric divisions , we have implemented asymmetric spherical cep as follows . for each region ( or sub - region , as applicable ) we take representative population values from the coma integration and use these as the basis of an `` initial guess . '' we then make an immediate improvement to the initial guess values by recalculating each region s populations ( individually ) taking into account the attenuation of incident solar radiation by intervening regions in the solar direction ( as per equation [ externalradn ] ) . these recalculated populations are the values we then use as the initial guess ( required by the implementation of newton s method in press , et al . 1992 ) for cep calculations . based on these populations we calculate the necessary source functions , delta - taus , and net radiative brackets , `` @xmath20 , '' as above , for each wavenumber ( or line , transition , etc . ) following the above discretized equation ( [ fullydiscreteeqn ] ) for each region , which in this context we will call the `` recipient '' region , we iterate over all other regions to calculate their contribution to the recipient s @xmath20 . each other region s contribution is essentially its own source function attenuated over the optical depth of all intervening regions along the line of sight between itself and the recipient region , integrated over ( or , to simplify , multiplied by ) the solid angle subtended by one region from the other . this is then divided by the recipient region s source function and optical depth ( along the given line of sight ) . to implement this in a practical algorithm of manageable complexity , we make several simplifying approximations . due to the @xmath69-axis symmetry that exists ( before adding sub - regions ) , we can _ partially _ simplify to a two - dimensional diagram in which a region is represented by the cross - section of the annulus in the ( arbitrarily chosen ) @xmath76 plane . we calculate a region s `` centroid '' , i.e. the centroid of its 2-d projected area in this plane which ( in our approximation ) corresponds to a point @xmath35 in spherical coordinates . for ( cone shaped ) sub - regions , which in general do not have their centerline on the x - z plane , we must use a different centroid . we use the midpoint along the cone s centerline ( within region boundaries ) . we also choose a series of points distributed evenly along circles parallel to the x - y plane around each region , which will be the `` starting points '' for calculation of lines of sight ( which will terminate at the centroids ) . these are chosen by rotating a region s centroid around the @xmath69-axis by multiples of some angle that depends on the size ( radius ) of the region . the choice of angle is such that the larger a region s size , the more starting points it will have , and thus the region will be divided into more elements of solid angle . we use the line of sight between the `` centroids '' of regions and this series of starting points to calculate the contributions of every other region ( or the region to itself ) to a given recipient region s @xmath20 . we calculate the optical depth of each intervening region , along the line of sight , based on the molecular population levels of the intervening regions . ( see fig . [ fig5 ] . ) these `` integration lines '' encapsulate the main part ( within the square brackets in equation [ fullydiscreteeqn ] ) of the calculation of @xmath20 . we approximate the solid angle subtended by the integration lines from another region s centroid by integrating @xmath77 sin@xmath78 from zero up to the mean value of the angles between the starting point of that region , the centroid of the recipient region and the multiple `` corners '' of that region around the starting point . ( see fig . [ fig6 ] . ) effectively , this gives a solid angle between regions of @xmath79 cos@xmath80 . note that due to these approximations , the sum of solid angles over all integration lines between a region and all regions in a given shell exterior to the region is _ not _ necessarily constrained to exactly equal @xmath58 , as it should be in reality . therefore , in calculating a `` @xmath20 '' value , we sum the solid angles involved and average over that sum instead of @xmath58 steradians ( as eqns . [ analyticaleqn ] & [ partlydiscreteeqn ] dictate we should do ) . used to calculate the solid angle subtended by one region as viewed from the other s centroid . note that not all regions will have eight `` corner points.''[fig6],scaledwidth=50.0% ] in the limit of arbitrarily small ( and numerous ) regions , these approximations _ would _ approach a physical situation of arbitrarily precise accuracy . thus we maintain the `` exactness '' of the cep method . unlike the plane - parallel situation , the flux exiting the surface of the coma ( or other sphere of interest ) is not simply a single value ( per wavenumber ) that has been integrated over angle . in the spherical situation , the resultant intensities form a two - dimensional mapping ( in a plane perpendicular to the observer s line of sight . [ fig7 ] . ) above , with observer plane also shown , aligned perpendicular to @xmath81-axis . ( a color image is included in the online version . ) [ fig7],scaledwidth=50.0% ] in our implementation , this plane is specified by rotation angles , @xmath31 , @xmath82 and @xmath83 with the comet s center at the origin , and is assumed to be at a distance @xmath84 , the maximum radius of the comet s coma . we can also specify the density of and interval between points on this plane for which the output intensities will be calculated . each point in this planar mapping shows the intensity ( or surface brightness ) integrated along a specific line of sight , perpendicular to the plane , through the coma from one side to the other ( again , for each wavenumber ) . @xmath85 where @xmath86 is the source function of a region @xmath49 , @xmath87 is the line width of wavenumber / frequency @xmath88 in region @xmath49 , and @xmath89 , or @xmath90 , represents the optical depth of wavenumber @xmath88 in region @xmath49 or @xmath50 along the relevant line of sight . indices @xmath49 and @xmath50 run from 1 to @xmath69 , where @xmath69 equals the number of regions along a given line of sight . thus the spherical cep algorithm produces results that could be described as a four - dimensional data `` hypercube '' : for each point in the above 2-d mapping , there is a complete ( 2-d flux vs. wavenumber ) spectrum . this data can then be presented in multiple formats . several forms of data presentation for simulating observations are described in the following section . this is also precisely the output needed to compare with the deep impact and epoxi observations that have been displayed as two - dimensional brightness maps for specific wavelengths or bands . we present here examples of model results for three different production rates of carbon monoxide which could be potentially useful for distant ( e.g. ground based or orbital telescope ) observers of comets . these are modeled using a spherical coma , without any morphological features but including optical depth effects both with respect to incident solar radiation within the coma and with respect to emergent `` observed '' radiation . the output data from the cep model can be presented in various ways . here , we show an example of a band total brightness map ( analogous to feaga , et al . 2007 , but for the entire coma ) , radial profiles of brightness , column density , and g - factors for various azimuthal angles . ( these could also be done for individual spectral lines , but in the interests of space and avoiding complexity we have not presented such results here . ) we also present spectra integrated over different `` aperture '' sizes . the band total brightness is more likely to be similar to actual observations , but high resolution spectra are possible , even from ground based telescopes ( see e.g. disanti , et al . 1999 and disanti , et al . 2001 ) , in particular for comets close to earth , which might more closely resemble the latter form of model results . we also demonstrate the potential usefulness to observers of the ratio of the total brightness of the p branch to that of the r branch of co to determine whether observations include significant optical depth effects . this may be measurable even with relatively poor spectral resolution . many of the model `` input values '' ( see table [ theoreticalinputparams ] ) have been chosen so as to facilitate comparisons ( of our optically thin cases ) with other earlier models . model parameters that are the same for all the following examples are : solar distance = 1 au . solar flux ( over the co band ) is @xmath91 , as per labs & neckel 1968 ( and as used by chin & weaver 1984 and weaver & mumma 1984 ) . gas expansion speed is a constant 0.8 km s@xmath16 and the initial gas temperature at the surface is 200 k. @xmath92 , as in chin & weaver 1984 . as mentioned above , the radial temperature profile closely follows combi s 1989 model ( see fig . [ combi200 ] ) but scaled to the initial gas temperature at the surface , t@xmath93 . [ cols= " < , < " , ] + total cross section is always @xmath94 = @xmath95 . + excitation is derived from de - excitation using + detailed balance . + [ cwcollisionstable1 ] we present here one example of a brightness map of a modeled coma ( see fig . [ brightnessmapq28 ] ) . this format of output presentation is most similar to the radiance maps of feaga , et al . 2007 and ahearn , et al . 2011 . for a spherical coma with no morphological features , it is rather uninteresting . it is nevertheless included here as a demonstation and used to illustrate the azimuthal angles of the radial profiles presented below with the addition of overlaid lines . note that for the @xmath96 s@xmath16 case , there is some difference in brightness noticeable to the eye between the sunward side ( azimuthal angles nearer to zero ) and the anti - sunward side , especially in the near - nucleus portion of the image . this is due to the optical depth along the solar direction reducing the excitation and emission from one side of the coma to the other . 600 km near the nucleus of a theoretical comet with @xmath97 , viewed from phase angle = 90@xmath98 . overlaid radial lines indicate the orientation of azimuthal angles in radial profiles below . the sunward direction is up , i.e. azimuthal angle = 0@xmath98 . note that within the @xmath99600 km field of view , the brightness never reaches zero ( even where it appears to be totally dark ) . this image is in color in the online version of the article and the colors of the azimuthal angles correspond to those in subsequent radial profiles . [ brightnessmapq28 ] , scaledwidth=50.0% ] abundances of cometary species are frequently calculated from observed fluxes using fluorescence efficiencies , or g - factors . in an optically thin case , the brightness of a given line or band is proportional to the column density of the relevant molecule . in such cases @xmath100 and @xmath101 where @xmath102 is the band total flux ( or brightness ) , @xmath103 the band g - factor , @xmath104 the total column density , @xmath105 the einstein a coefficient for the relevant transition originating in upper level `` u '' and @xmath106 is the column density of the population of a specific upper level `` u '' ( which in our model is numerically approximated as the sum over all regions along a line of sight of the fractional population of level `` u '' times each region s column density ) . however , large optical depths will spoil this simple linear relation between column density and brightness . with radiative transfer modeling , it is possible to get a calculated g - factor ( @xmath101 ) from the model and the `` effective g - factor '' , @xmath107 , which is the actual ratio of brightness to column density . in figs . [ fluxvsrphase90multiazq26 ] , [ fluxvsrphase90multiazq27 ] , and [ fluxvsrphase90multiazq28 ] we present all these values together as radial profiles , for theoretical comets of three different production rates , @xmath108 , and @xmath109 . ( all observed at 1 au , at a phase angle of 90@xmath98and multiple azimuthal angles . in all cases @xmath110 . ) in our results , we typically see that @xmath111 does tend towards the calculated g - factor values at larger impact parameters , where optical depth effects are negligible , as would be expected . ( the actual numerical values of the `` asymptotic '' band g - factors produced by our model , @xmath112 per molecule for co at 1 au , also agree well with other published values such as those calculated by chin & weaver 1984 , crovisier & le bourlot 1983 , and weaver & mumma 1984 . ) the actual radii at which this convergence occurs depends primarily on the production rate of a comet . we can use the distance at which @xmath113 as a very rough measure of the point where a coma can be considered to transition from optically thick to thin . for the `` thin '' and `` intermediate '' coma models ( @xmath114 and @xmath115 ) the convergence is fairly close to the nucleus , within @xmath116100 - 200 km . but for the `` thick '' model ( @xmath97 ) with its high production rate , the `` optically thick regime '' can extend as far as @xmath117 km , which can be spatially resolved even in some remote observations . note that at radial distances very near the nucleus , worrying about optical depth effects may be relevant even for lower production rates . . upper frame : radial profile of band total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98and for multiple azimuthal angles . profiles of azimuthal angles ( indicated by color coding in the online version ) show no variation for this case and overlap , appearing indistinguishable . column density is included as the bold solid line ( red in the online version ) using a different y - axis scale . lower frame : g - factors . both the calculated g - factor ( the higher group of lines ) and @xmath111 , the observed brightness over column density , plotted with matching styles ( colors in the online version ) for each azimuthal angle ( which also match those in the upper frame ) . profiles for 0@xmath98 , @xmath9945@xmath98 , and 90@xmath98overlap each other almost entirely . [ fluxvsrphase90multiazq26],scaledwidth=50.0% ] . upper frame : radial profile of band total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98and for multiple azimuthal angles . profiles of azimuthal angles ( indicated by color coding in the online version ) for 0@xmath98 , @xmath9945@xmath98 , and 90@xmath98overlap each other almost entirely and other brightness profiles are almost indistinguishable . column density is included as the bold solid line ( red in the online version ) using a different y - axis scale . lower frame : g - factors . both the calculated g - factor ( the higher group of lines ) and @xmath111 , the observed brightness over column density , plotted with matching styles ( colors in the online version ) for each azimuthal angle ( which also match those in the upper frame ) . profiles of azimuthal angles 0@xmath98 , @xmath9945@xmath98 , and 90@xmath98overlap each other almost entirely and are almost indistinguishable . [ fluxvsrphase90multiazq27],scaledwidth=50.0% ] . upper frame : radial profile of band total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98and for multiple azimuthal angles . profiles of azimuthal angles ( indicated by color coding in the online version ) for 0@xmath98 , @xmath9945@xmath98 , and 90@xmath98overlap each other almost entirely . column density is included as the bold solid line ( red in the online version ) using a different y - axis scale . lower frame : g - factors . both the calculated g - factor ( the higher group of lines ) and @xmath111 , the observed brightness over column density , plotted with matching styles ( colors in the online version ) for each azimuthal angle ( which also match those in the upper frame ) . [ fluxvsrphase90multiazq28],scaledwidth=50.0% ] another optical depth effect is variation of brightness ( and corresponding g - factor ) with varying angles , both phase angle ( of observer ) and azimuthal angle within any given observation . the radial profiles in figs . [ fluxvsrphase90multiazq26]-[fluxvsrphase90multiazq28 ] demonstrate the azimuthal variation for a single phase ( observing ) angle . not surprisingly , the farther a radial profile line is from from the sunward direction , the less bright it is and the lower its g - factors for a given distance from the nucleus . however , the degree to which the actual g - factor varies is of note . even for the only moderately thick case of @xmath118 for radii @xmath119 100 km , there is a difference of as much as @xmath11610% between sunward and anti - sunward directions . the effect is even more pronounced for the thicker case of @xmath97 . in figs . [ fluxvsrphase90multiazq28]-[fluxvsrphase180multiazq28 ] we present profiles of brightness ( and column density ) for model results observed from different phase angles , ranging from 0@xmath98to 180@xmath98 , for the optically thicker case of @xmath97 . each plot includes multiple azimuthal angles , which would all be visible simultaneously in a wide field observation ( i.e. including the entire coma ) from each given phase angle . ( the above plot for 90@xmath98phase angle for @xmath97 , fig . [ fluxvsrphase90multiazq28 ] , should be considered part of this series as well . ) observations with a slit spectrometer with sufficient spatial resolution ( see , e.g. disanti , et al . 1999 ) might observe along one specific azimuthal angle and thus see possible variations along the slit with sufficiently high spatial resolution . , @xmath97 . upper frame : radial profile of band total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98for multiple azimuthal angles . profiles of azimuthal angles ( indicated by color coding in the online version ) show no variation for this case ( as should be expected ) and overlap , appearing indistinguishable . column density is included as the bold solid line ( red in the online version ) using a different y - axis scale . lower frame : g - factors . both the calculated g - factor ( the higher group of lines ) and @xmath111 , the observed brightness over column density , plotted with matching styles ( colors in the online version ) for each azimuthal angle ( which also match those in the upper frame ) . [ fluxvsrphase0multiazq28],scaledwidth=50.0% ] , @xmath97 . upper frame : radial profile of band total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98for multiple azimuthal angles . profiles of azimuthal angles ( indicated by color coding in the online version ) are slightly displaced for viewing purposes . column density is included as the bold solid line ( red in the online version ) using a different y - axis scale . lower frame : g - factors . both the calculated g - factor ( the higher group of lines ) and @xmath111 , the observed brightness over column density , plotted with matching styles ( colors in the online version ) for each azimuthal angle ( which also match those in the upper frame ) . [ fluxvsrphase45multiazq28],scaledwidth=50.0% ] , @xmath97 . upper frame : radial profile of band total brightness vs. r ( impact paramater ) for phase angle = 135@xmath98for multiple azimuthal angles . profiles of azimuthal angles ( indicated by color coding in the online version ) are slightly displaced for viewing purposes . column density is included as the bold solid line ( red in the online version ) using a different y - axis scale . lower frame : g - factors . both the calculated g - factor ( the higher group of lines ) and @xmath111 , the observed brightness over column density , plotted with matching styles ( colors in the online version ) for each azimuthal angle ( which also match those in the upper frame ) . [ fluxvsrphase135multiazq28],scaledwidth=50.0% ] , @xmath97 . upper frame : radial profile of band total brightness vs. r ( impact paramater ) for phase angle = 180@xmath98for multiple azimuthal angles . profiles of azimuthal angles ( indicated by color coding in the online version ) show no variation for this case ( as should be expected ) and overlap , appearing indistinguishable . column density is included as the bold solid line ( red in the online version ) using a different y - axis scale . lower frame : g - factors . both the calculated g - factor ( the higher group of lines ) and @xmath111 , the observed brightness over column density , plotted with matching styles ( colors in the online version ) for each azimuthal angle ( which also match those in the upper frame ) . [ fluxvsrphase180multiazq28],scaledwidth=50.0% ] the most obvious effect seen at a glance in these figures is the spread among azimuthal angles for a given phase angle . as would be expected , the 0@xmath98and 180@xmath98phase angles ( sunward and anti - sunward ) have no real azimuthal variation . from phase angle 45@xmath98to 90@xmath98to 135@xmath98there is a progression : the azimuthal lines get spread out farther from each other , as well as noticably dropping in brightness for those lines farther from the sunward side . with respect to total brightness , the phase angles 0@xmath98 , 45@xmath98and 90@xmath98 , produce roughly equal peak brightness for their _ strongest _ azimuthal profiles ( the more sunward directions , at the nucleus grazing radius ) of about @xmath120 photons s@xmath16 @xmath121 sr@xmath16 for the 135@xmath98phase angle , the peak values are about @xmath122 . most significantly , for the 180@xmath98phase angle , the peak values are about @xmath123 - only about half as bright as at other phase angles . with respect to `` effective '' ( or `` observed '' ) g - factors , the minimum values in the most optically thick regions ( again , at the nucleus - grazing radii ) also show a trend from sunward to anti - sunward . for the 0@xmath98 , 45@xmath98 , and 90@xmath98phase angles , the minimum values of the ratio of the band total flux over the column density are roughly @xmath124 photons s@xmath16 molecule@xmath16 . from 90@xmath98 , through 135@xmath98and down to 180@xmath98the values go down monotonically to a minimum of about @xmath125 . this trend is due to a combination of two optical depth effects . the first is attenuation of incident solar light from the sunward to anti - sunward sides of the coma , leading to less fluorescent pumping , and thus less emission , towards the anti - sunward direction . second , whatever emission there is is more likely to `` escape '' the coma over shorter optical depths - i.e. closer to where it is emitted . thus the already greater emission of the sunward regions is also more likely to be observed along azimuthal directions closer to sunward . however , the similarly located values for `` actual '' calculated g - factors do _ not _ follow a similar simple monotonic trend . the greatest values for a given phase angle rise from 0@xmath98through 45@xmath98and peak for phase angle 90@xmath98 . from 90@xmath98 , through 135@xmath98down to 180@xmath98they fall through the same values , creating a symmetric peak around 90@xmath98 . for 90@xmath98(see fig . [ fluxvsrphase90multiazq28 ] ) there are two `` clusters '' of lines . the higher one corresponding to more sunward azimuthal angles ( between @xmath126 ) and the lower one to other angles . the upper cluster s ( minimum ) values are actually the highest values in this comparison , about @xmath127 . the lower cluster of values at 90@xmath98is still greater than either angle 0@xmath98or 180@xmath98 , which are the lowest for any phase angle , notwithstanding the azimuthal spread for the other phase angles . ( the calculated values also show considerably more spread among azimuthal angles than the profiles of flux over column density . ) this symmetrical and non - monotonic pattern is less intuitive than the trend of @xmath111 above . yet it is clearly understood in light of the fact that these values are based only on the actual population distributions in different regions and do _ not _ include optical depth effects on the emergent radiation . thus observing from azimuthal angles 0@xmath98and 180@xmath98are sampling exactly the same lines of sight and regions populations , including the darkest ( i.e. least excited populations ) of the anti - sunward side of the coma . the same is essentially true for 45@xmath98and 135@xmath98(due to symmetry around the z - axis ) but they do not sample the darkest parts of the anti - sunward directions ( and the differences in populations are less between outermost regions on the sunward side among azimuthal angles between @xmath9945 they are all experiencing direct solar illumination ) . for 90@xmath98the higher cluster is sampling from more excited and higher emmision populations than the lower cluster , and consistently so all the way along their lines of sight ( which is not true for 45@xmath98and 135@xmath98 ) . thus the sunward cluster of lines for 90@xmath98is the brightest seen , and the anti - sunward cluster values falls between values of the 0@xmath98or 180@xmath98and the 45@xmath98or 135@xmath98lines . lastly , the profiles for @xmath128 have a `` bump '' in brightness in the vicinity of @xmath116 100 @xmath116 1000 km , most easily visible for the 180@xmath98plot , but also present for the other phase angles ( and growing in size from 0@xmath98up to 180@xmath98 ) . this is mostly due to the temperature profile reaching its minimal values at these radii in conjunction with the higher density of the @xmath128 case . the higher density leads to this still being a collsionally dominated regime , and the low temperatures lead to the lowest population levels being most highly populated . these levels also have the highest einstein a values , thus leading to higher overall number of photons emitted for the same number of molecules . the effect is greatest for the 180@xmath98view due to a cumulative effect the lines of sight all sample the most dense and cold regions at these radii . the same extreme effect is not seen for the 0@xmath98phase angle due to the overall greater fluorescent excitation of the sunward regions dominating it . ( note that the difference in total brightness between the two azimuthal angles at these radii is about a factor of two . ) if one is observing with high spectral resolution but low spatial resolution , the spectra observed will be the sum of as much of the coma as fills the field of view . to model this , we have simulated `` aperture averaged '' spectra , where the `` aperture '' controls the area of the coma sampled . our apertures are square boxes and are all centered exactly on the center of the comet , and sample a nucleus centered area equal to the square of the `` aperture size '' over which we average the brightness . we present a series of apertures from @xmath129 km ( very near the nucleus ) through @xmath130 km ( the whole coma ) for each of the three production rates . ( all these example spectra are modeled at a heliocentric distance of 1 au and phase angle 90@xmath98 . @xmath92 , as in chin & weaver 1984 . ) band shape for apertures including the outer coma ( approximately @xmath131 km ) does not change significantly for different production rates . the total brightness for this regime increases approximately in linear proportion to production rate . this is due to spectra with such large aperture sizes being dominated by the fluorescence dominated optically thin outer coma with optical depth effects playing a minimal role . ( but not entirely non - existent : note the small , @xmath132 reduction in g - factor with higher production rate for @xmath133 km . ) in the inner coma , however , optical depth effects can be very striking . the `` thickest '' spectra ( for higher production rates and smaller @xmath134 ) have remarkably altered band shapes from the optically thin spectra . first , however , a word about changes that are _ not _ specifically caused by optical depth effects . it is clear that there is much variation , even within a given production rate , from large to small aperture sizes . not all of this is due to optical depth effects . even in our optically thin case , the band shape changes noticably in breadth this occurs primarily due to the temperature profile . we have used a fairly simplified profile , which can be scaled to a surface temperature parameter , but does not vary much otherwise between different model cases . this provides a straightforward `` control '' for this aspect of spectral change with aperture size . in the innermost coma near the nucleus , the temperature is quite warm ( @xmath116100 - 200k ) , which leads to a broader band in the 10 km spectrum . the coma gas cools to a minimum ( @xmath11620k ) around 100 - 200 km out from the nucleus , which produces a much narrowed band . since the einstein a coefficient for the lowest j lines is higher than for the lines in the `` wings '' of the band , the cold temperature also increases the g - factor , even in optically thick regions . at larger radii , the temperature rises again , but becomes less significant since the coma gets less dense and tends towards fluorescent equilibrium . between these regimes , in a `` transition region , '' there are still optical depth effects , which can be more easily isolated as g - factors are less temperature controlled . temperature is also a factor in determining doppler broadening and line width , which is proprtional to @xmath135 , so the ratio between line widths for the coldest and the warmest regions of the coma are about 2 - 3 , for a given wavenumber . this may lead to temperature playing a significant role in the optical thickness of the coma to incident solar radiation . temperature effects notwithstanding , the spectra from the denser near - nucleus regions of a coma show optical depth effects in several aspects . in addition to the total brightness no longer increasing linearly with production rate ( and a corresponding reduction of g - factors ) , energy is also dramatically shifted between lines within the band . the notable shifting of flux from r branch to p branch ( evident in many of the thicker spectra ) , and to lower wavenumbers in both branches ( as is most evident in the @xmath136100 & 200 km spectra for @xmath137 s@xmath16 ) , are very noticeable optical depth effects . ( see sahai & wannier 1985 , for an analytical discussion of similar effects . ) this effect appears due to the branching ratio of a given pair of p and r branch lines originating in the same upper level , which generally ( slightly ) favors emission in the p branch line . in optically thick cases , repeated absorption and emission of photons leads to a cumulative effect which favors the p branch over the r branch much more than in optically thin conditions ( where it is probable that any emitted photon will not be re - absorbed before escaping the coma ) . similarly , flux is `` pushed '' outwards in the branches , and more so in the p branch due to combination with the above effect . this is due to the lines closer to the center of the band becoming optically thick before those in the wings ( both due to their higher einstein coefficients and generally being more populated . ) flux initially emitted in lines that are optically thick will through repeated absorption and emission be forced out into lines that are less thick . as seen above , the p branch total brightness and the r branch total brightness vary with respect to each other over different optical depths ( as well as other factors , such as temperature distribution . ) the ratio of the sums of p / r branches brightnesses can be useful to alert an observer ( or anyone analyzing observations ) that they must in a given case beware of , and if possible account for , optical depth effects . this alone , would not be sufficient , as temperatures along a given line of sight are also a significant factor in controlling the p / r ratio , in the collisionally dominated inner coma . colder population distributions will emit more in the lower lines ( in both branches ) for which the ratio of p / r for each pair of lines originating in the same upper state is greater . also , @xmath60 will effectively vary inversely with linewidth , other factors being equal . use of a model like ours can show where the p / r ratio is large due to temperature and where ( its excess beyond that value is ) due to optical depth . in our optically thin , @xmath114 s@xmath16 , model the p / r ratio does not exceed @xmath1161.44 , even for aperture sizes dominated by the coldest portion of the coma . ( note , however , that this is an aperture _ averaged _ value . in fig . [ pr_q26 ] below , the peak value is slightly higher , @xmath1161.5 . ) however , the ratio for corresponding aperture sizes in the @xmath138 s@xmath16 and @xmath96 s@xmath16 cases is @xmath1161.8 and @xmath1162.4 , respectively . furthermore , in the thickest case modeled , even the spectrum with aperture size of 2000 km has a ratio of @xmath1161.4 . note that in all cases the @xmath139 km aperture , which is dominated by the outer coma in fluorescent equilibrium , has a ratio of @xmath1161.12 . all of this indicates that a p / r ratio in excess of @[email protected] is a warning sign of optical depth effects involved . while it would be ideal to be able to derive a simple correction factor from the p / r ratio in such cases , alas , it is not exactly possible . however , a rough estimate of the degree of optical depth effects can be derived . to do so , we create radial profiles of the p / r ratio , for both the observed emergent flux / brightness and the calculated value based on underlying populations without attenuation of emergent light , as shown in figs . [ pr_q26 ] , [ pr_q27 ] , and [ pr_q28 ] . by cross - referencing the observed p / r ratio for a given radial distance with the corresponding g - factor in figs . [ fluxvsrphase90multiazq26 ] , [ fluxvsrphase90multiazq27 ] , and [ fluxvsrphase90multiazq28 ] one can ascertain the `` real '' g - factor to use to calculate a correct column density from the observed flux . . ratio of p branch vs. r branch total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98and for multiple azimuthal angles . profiles of azimuthal angles show negligible variation for this case and overlap , appearing indistinguishable . ( profiles are indicated by color coding in the online version ) . [ pr_q26],scaledwidth=40.0% ] . ratio of p branch vs. r branch total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98and for multiple azimuthal angles . profiles of azimuthal angles show minimal variation for this case and overlap , appearing _ nearly _ indistinguishable , except inwards of @xmath11640 - 50 km . ( profiles are indicated by color coding in the online version ) . [ pr_q27],scaledwidth=40.0% ] . ratio of p branch vs. r branch total brightness vs. r ( impact paramater ) for phase angle = 90@xmath98and for multiple azimuthal angles . profiles of azimuthal angles show some variation in this case , but those for 0@xmath98 , @xmath9945@xmath98 , and 90@xmath98overlap each other entirely in the calculated profiles and almost entirely in the observed . ( profiles are indicated by color coding in the online version ) . [ pr_q28],scaledwidth=40.0% ] this heuristic is , of course , limited in use to to the carbon monoxide @xmath140 band . other spectra with p and r branches will have their own ratios , which can be derived by similar modeling . more complicated spectra may also , but such ratios would be more complicated to find than for cases with a simple two - branch structure . we plan to provide further useful results in forthcoming papers ( currently in preparation ) dealing with similar modeling to that presented here for h@xmath141o and co@xmath141 , as well as a more in depth treatment of co alongside those species . we will present model results including morphology and comparison with the in - situ spectral observations of the deep impact and epoxi missions . the model and code have already been implemented to produce those results . further work , not yet implemented , will include several planned improvements to our model . a more accurate treatment of radial velocites and doppler shifts in lines due to them is a highly desirable improvement . ( at present , only a thermal doppler profile is used for calculating absorption of solar radiation . ) we neglected a precise treatment until now primarily due to the nature of our originally intended problem . the near nucleus morphology which we intended to model is represented by cones expanding radially in one direction , a geometry that is likely to reduce doppler effects . however , when modeling the whole coma as a sphere , as in the results presented here , this assumption is no longer valid . furthermore , in as much as we are primarily looking at the overall _ band _ shape as opposed to individual line shapes , the effect of doppler shifts on the spectra is expected to be less than if one were modeling specific lines ( as is often the case for other spectral regimes ) . however , the effect of this neglect would be to increase the optical thickness , and therefore our results may be over - estimating optical depth effects . if so , the effects we describe would still be observable , but for higher production rates than those modeled . in addition to more accurate radial velocities , more flexible radial temperature and density profiles are planned , so as to be able to model deviations from a very simple haser model . ( e.g. volatiles produced from icy grains or large chunks and not solely from the nucleus surface , as per ahearn , et al . 2011 , and/or photodissociation of molecules , and corresponding creation of daughter species . ) computational limits are currently a limiting factor in how optically thick and how refined ( in terms of granularity of conditions , in that more variation requires more regions ) the modeled cases can be . as of now , the maximum production rates we can deal with are on the order of @xmath137 s@xmath16 , and somewhat less for co@xmath141 than the other molecules . we are planning to address these concerns with algorithmic improvements . running the code on faster and more powerful computers is also a possibility . we have demonstrated our model s usefulness in understanding emission spectra of cometary comae . there are several possible effects of optical depth that could lead observers to mistaken conclusions regarding the calculated abundances , or other characteristics , of species of interest . the moral of the story : ignore radiative transfer and optical depth effects at your own peril ! although designed specifically with comets in mind , our model and code are versatile enough to be used in other radiative transfer problems as well . parameters that define a specific comet model or other problem , including molecule of interest , size of nucleus and radial shells , production rate , morphology ( if any ) , incident radiation , etc . are all fairly flexible . thus our adaptation of coupled escape probability to an asymmetric spherical situation has created a very useful tool for modeling cometary spectra , as well as other spherical astrophysical phenomena . we have coded the aglorithm described above in the c++ language , using numerous functions from press , et al . 1992 , primarily to implement numerical integration of functions ( with * odeint * and * stifbs * and associated functions ) and solution of n - dimensional non - linear matrices with newton s method ( using * newt * , and associated functions ) . the bulk of the coding , which implements the radiative transfer algorithm in spherical geometry is our own . a major practical limitation of our algorithm is the matrix size ; since newton s method requires ( repeated ) @xmath142 matrix solving operations ( for which * newt * uses the brute force approach of * ludcmp * and * lubksb * ) , the algorithm can get prohibitively slow for large matrices . for example , on an intel core computer ( with cpu speed of 2.9 ghz and 7.6 gb of memory ) running scientific linux 6.3 ( carbon ) , when n @xmath143 15,000 , a solution may take one or more days . for greater sizes , even a week . the matrix size equals the number of molecular levels used times the number of regions . the molecule ( and band / s ) being modeled determines the first value , with the latter value demanding to be increased with greater optical depths and production rates . depending on the species of interest , the maximum practical production rates we can currently manage on such a system are of @xmath144 . the most simple workaround is to use a more powerful computer . algorithms for solving sparse matrices faster than the above functions can also be used , and we have begun to explore this option . in implementing the cep method , and our adaptation to spherical geometry , we have created c++ classes for diatomic and triatomic molecules . the object oriented programming style of c++ lends itself ideally to being able to switch molecules easily . the * diatomic * molecule class and its subclasses ( currently implemented for * co * and * sio * ) calculate energies and einstein coefficients based on constants ( taken primarily from krupenie 1966 , for co ) that are included in the code . the * triatomic * class , which can actually be used for other polyatomic molecules as well ( or the aforementioned diatomic molecules themselves , for that matter ) , must be provided with energies and coefficients from some outside source in formatted input files . we used the hitran database ( see rothman , et al . 1998 ) to supply these values for co@xmath141 and h@xmath141o . this approach has the versatility to handle many other molecules with a minimal effort of `` data massaging '' to get the data into the proper format . we have also created a large number of classes which encapsulate the spherical geometry and the attendant calculations . these are all `` controlled '' by the * comet * class that reads parameters for a given case from a `` comet definition file '' ( a text file of key / value pairs , essentially ) and , using the above classes , sets up and runs the model for a given case . although designed specifically with comets in mind , our code is versatile enough to be used in other spherical radiative transfer problems as well . straightforward input files describe all the required and optional parameters for a specific comet model or other problem , including molecule of interest , size of nucleus and radial shells , production rate , morphology ( if any ) , incident radiation , etc . the c++ code outputs a file containing a point - by - point line - by - line spectral mapping , as described above , for one or more specified viewing orientations . this data can then be presented in multiple formats as described above in section 3 . we have implemented this data presentation portion of the model using various short idl programs which we have developed , each specifically for the purpose of producing a given type of graphical output . * acknowledgements * we gratefully thank profs . j. p. harrington and d. c. richardson and dr . l. feaga for their invaluable advice in the course of this work . we would also like to thank prof . m. elitzur for his encouragement . this work was supported by nasa s discovery program contract nnm07aa99c to the university of maryland .

for integers and , we study the eigenvalue problemu(z)=\lambda u(z)$ ] with the boundary conditions that decays to zero as tends to infinity along the rays in the complex plane , where is a polynomial . we provide asymptotic expansions of the eigenvalue counting function and the eigenvalues . then we apply these to the inverse spectral problem , reconstructing some coefficients of polynomial potentials from asymptotic expansions of the eigenvalues . also , we show for arbitrary-symmetric polynomial potentials of degree and all symmetric decaying boundary conditions that the eigenvalues are all real and positive , with only finitely many exceptions . _ preprint . _ = 18pt

math0411143

in this paper , we study schrdinger operators with any polynomial potential of degree @xmath0 with complex coefficients , under decaying boundary conditions along two rays to infinity in the complex plane , and provide asymptotic expansions of the eigenvalue counting functions and the eigenvalues . then we will use these to reconstruct some coefficients of polynomial potentials from asymptotic expansions of the eigenvalues , and to show that all but finitely many eigenvalues of every @xmath8-symmetric oscillator with a polynomial potential are real and positive . for integers @xmath0 fixed and @xmath9 , we are considering the non - standard eigenvalue problems @xmath10u(z,\lambda)=\lambda u(z,\lambda),\quad\text{for some $ \lambda\in{\mathbb c}$},\ ] ] with the boundary condition that @xmath11 where @xmath12 is a polynomial of degree at most @xmath13 of the form @xmath14 the anharmonic oscillators @xmath15 with the various boundary conditions are considered in @xcite . the most studied case is when @xmath16 is even and @xmath17 , for example , see @xcite . in this case , @xmath15 is an schrdinger operator in @xmath18 . this is self - adjoint if all the coefficients of the polynomial @xmath19 in @xmath4 are real , and non - self - adjoint if a coefficient of @xmath19 is non - real . the case when @xmath20 has been studied extensively in recent years in the context of @xmath8-symmetric theory @xcite . throughout this paper , we use the integer @xmath0 for the degree of the polynomial potential and integer @xmath21 with @xmath1 . we will use @xmath22 , depending on the potential and the boundary condition , to denote the eigenvalues of @xmath15 , without explicitly indicating their dependence on the potential and the boundary condition . also , we let @xmath23 be the coefficient vector of @xmath24 . if a nonconstant function @xmath25 satisfies with some @xmath26 and the boundary condition , then we call @xmath27 an _ eigenvalue _ of @xmath15 and @xmath25 an _ eigenfunction of @xmath15 associated with the eigenvalue @xmath27_. also , the _ geometric multiplicity of an eigenvalue @xmath27 _ is the number of linearly independent eigenfunctions associated with the eigenvalue @xmath27 . for each @xmath21 fixed , we number the eigenvalues @xmath28 in the order of nondecreasing magnitudes , counting their `` algebraic multiplicities '' , where the integer @xmath29 , depending on the potential and the boundary condition , is due to our method of proof of theorem [ main_thm1 ] . in theorem [ main_thm1 ] we show that for every large @xmath30 , there exists @xmath22 satisfying below . however , we do not know the number of eigenvalues `` near '' zero , and this is why we need to have @xmath29 in numbering the eigenvalues . before we state our main theorems , we first introduce some known facts by sibuya @xcite about the eigenvalues @xmath22 of @xmath15 . [ main2 ] the eigenvalues @xmath7 of @xmath15 have the following properties . 1 . the set of all eigenvalues is a discrete set in @xmath31 . the geometric multiplicity of every eigenvalue is one . infinitely many eigenvalues , accumulating at infinity , exist . when @xmath20 the eigenvalues have the following asymptotic expansion @xmath32\quad\text{as $ n$ tends to infinity},\quad n \in { \mathbb n}.\ ] ] this paper contains results on direct and inverse spectral probelms , and their applications to @xmath8-symmetric potential problems . theorem [ main_thm1 ] below is the main result , regarding asymptotic expansions of `` eigenvalue counting functions '' . the other results below in the introduction are deduced from theorem [ main_thm1 ] . here , we first introduce the following theorem , regarding asymptotic expansions of a kind of eigenvalue counting functions , where we use multi - index notations with @xmath33 also , we use @xmath34 , @xmath35 and @xmath36 . [ main_thm1 ] for @xmath37 , the eigenvalues @xmath22 of @xmath15 satisfy @xmath38 where @xmath39 is the largest integer that is less than or equal to @xmath40 , and where the error @xmath41 term is uniform on any compact set of @xmath37 , and @xmath42 where @xmath43 @xmath44 and @xmath45 where @xmath46 is the beta function . we obtain by investigating the asymptotic expansions of an entire function whose zeros are the eigenvalues . sibuya @xcite got by using the first order asymptotic expansion of the entire function . next , we let @xmath47 , @xmath48 , be the eigenvalue counting function , that is , @xmath47 is the number of eigenvalues @xmath27 of @xmath15 such that @xmath49 . then the following theorem on an asymptotic expansion of the eigenalue counting function is a consequence of theorem [ main_thm1 ] . [ main_thm2 ] let @xmath37 be fixed . suppose that @xmath50 for @xmath51 . then @xmath47 has the asymptotic expansion @xmath52 where the error @xmath53 is uniform for any compact set of @xmath37 . in theorem [ ineq_eq ] below , we show that @xmath54 for all large @xmath30 . suppose that @xmath55 . then since @xmath56 we see from theorem [ eigen_asy ] below that @xmath57 . thus , @xmath58 hence , replacing @xmath59 in by @xmath60 , and solving the resulting equation for @xmath61 complete the proof . next , we improve the asymptotic expansion of the eigenvalues @xmath7 of @xmath62 . in particular , we will prove the following , which essentially invert to get @xmath22 in terms of @xmath61 . [ eigen_asy ] for each @xmath37 , there exist some constants @xmath63 , @xmath64 , such that @xmath65 where the error term is uniform for any compact set of @xmath37 , and where @xmath66 and @xmath67 , @xmath68 , are defined recurrently by @xmath69 and @xmath70 where @xmath71 . we note for the first summation in the definition of @xmath72 that @xmath73 implies @xmath74 whenever @xmath75 . also , for the second summation , we point out that @xmath76 implies @xmath74 whenever @xmath77 . the asymptotic expansions of the eigenvalues of @xmath15 with @xmath78 and @xmath20 have been studied by a number of people . for example , maslov @xcite computed the first three terms of asymptotic expansions of @xmath79 , where @xmath22 are the eigenvalues of @xmath80 helffer and robert @xcite considered @xmath81 where @xmath82 are positive integers and where @xmath83 is a _ real _ polynomial of degree at most @xmath84 . they obtained existence of asymptotic expansions of eigenvalues to all orders , and suggested an explicit way of computing the coefficients of the asymptotic expansion . in particular , for the case when the potential is @xmath85 , @xmath86 , helffer and robert @xcite computed the first nine terms of the asymptotic expansion of @xmath79 . fedoryuk @xcite considered with complex polynomial potentials and with for @xmath78 , and showed the existence of asymptotic expansions of the eigenvalues to all orders . note that there appear to be typographical errors in @xcite . for example , when @xmath16 is even ( and @xmath17 ) the leading coefficient of the asymptotic expansion of @xmath22 in @xcite is @xmath87 , which is different from @xmath88 found in @xcite and again in theorem [ eigen_asy ] above . next , we point out some differences between work of fedoryuk @xcite and the present work . fedoryuk @xcite showed the existence of asymptotic expansion of the eigenvalues to all orders while we do not . on the other hand , we treat all decaying boundary conditions with @xmath1 , while fedoryuk @xcite studied the case @xmath78 only . moreover , we computed more coefficients @xmath67 explicitly , and our methods are different from fedoryuk s . here , we introduce results on inverse spectral problems , but first the following corollary is an easy consequence of theorems [ main_thm1 ] and [ eigen_asy ] , regarding how the coefficients of the asymptotic expansions depend on @xmath37 . [ corollary_1 ] let @xmath68 be a fixed integer . then we have the following . 1 . @xmath89 and @xmath72 are polynomials in @xmath90 . 2 . @xmath89 and @xmath72 do not depend on @xmath91 . 3 . if @xmath92 is a multiple of @xmath16 , then @xmath93 , and @xmath72 does not depend on @xmath94 . 4 . if @xmath92 is not a multiple of @xmath16 , then @xmath89 and @xmath72 depend linearly on @xmath94 . statements on @xmath89 are direct consequences of the definition of @xmath89 in theorem [ main_thm1 ] . one can use statements on @xmath89 and induction on @xmath95 to prove statements on @xmath72 . next , one can reconstruct some coefficients of the polynomial potential from the asymptotic expansion of the eigenvalues . let @xmath96 be a fixed integer . suppose that @xmath97 is known whenever @xmath98 and @xmath99 is a multiple of @xmath16 . if @xmath92 is a multiple of @xmath16 , then the asymptotic expansions of the eigenvalues @xmath7 of @xmath15 of type with an error term @xmath100 uniquely and explicitly determine @xmath94 . from the asymptotic expansion of the eigenvalues , one gets @xmath101 that are explicit polynomials in @xmath102 . then since we know @xmath97 if @xmath99 is a multiple of @xmath16 , corollary [ corollary_1 ] says that we can find all @xmath102 . one says that @xmath15 is @xmath8-symmetric if the potential @xmath103 satisfies @xmath104 , @xmath105 , that is equivalent to @xmath106 . here , we prove the partial reality of the eigenvalues @xmath22 of @xmath8-symmetric @xmath15 . but first , we show the following theorem , regarding monotonicity of modulus of @xmath22 for all large @xmath30 . [ ineq_eq ] for each @xmath37 there exists @xmath107 such that @xmath108 if @xmath109 . see theorem 3 in @xcite for the proof of the case when @xmath20 . one can see that proof of theorem 3 in @xcite can be easily adapted for the cases when @xmath110 . now we are ready to prove the following theorem on the partial reality of the eigenvalues @xmath22 of @xmath15 . [ main_theorem ] suppose that @xmath106 . then all but finitely many eigenvalues of @xmath15 are real and positive . hence @xmath50 for all @xmath51 , so that the counting function formula in theorem [ main_thm2 ] is valid . when @xmath15 is @xmath8-symmetric ( i.e. , @xmath106 ) , @xmath111 is an eigenfunction associated with an eigenvalue @xmath27 if and only if @xmath112 is an eigenfunction associated with the eigenvalue @xmath113 . thus , the eigenvalues either appear in complex conjugate pairs , or else are real . so theorem [ ineq_eq ] implies theorem [ main_theorem ] . in recent years , these @xmath8-symmetric operators have gathered considerable attention , because ample numerical and asymptotic studies suggest that many of such operators have real eigenvalues only even though they are not self - adjoint . in particular , the differential operators @xmath114 with some polynomial potential @xmath103 and with the boundary condition have been considered in @xcite and references therein . the rigorous proof of reality and positivity of the eigenvalues of @xmath8-symmetric operators with certain classes of polynomial potentials and with the boundary condition for @xmath20 , was given by dorey , dunning and tateo @xcite in 2001 and by the present author @xcite in 2002 . however , there are some @xmath8-symmetric polynomial potentials that produce a finite number of non - real eigenvalues @xcite for some particular classes of polynomial potentials . so without any further restrictions on the real coefficients @xmath97 , theorem [ main_theorem ] is the most general result one can expect about reality of eigenvalues of @xmath8-symmetric operators with polynomial potentials . this paper is organized as follows . in section [ prop_sect ] , we will introduce work of hille @xcite and sibuya @xcite , regarding properties of solutions of . also , we introduce entire functions @xmath115 whose zeros are closely related with the eigenvalues of @xmath15 , due to sibuya @xcite ( c. f. , section [ asymp_eigen ] ) . in section [ sec_4 ] , we then provide asymptotics of the entire function @xmath116 as @xmath117 in the complex plane @xcite , improving the asymptotics of @xmath116 in @xcite . in section [ sec_5 ] , we provide asymptotic expansions of @xmath115 as @xmath117 in @xmath31 . in section [ asymp_eigen ] , we investigate how the zeros of @xmath118 are related with the eigenvalues of @xmath15 . in sections [ sec_7 ] and [ sec_8 ] , we prove theorem [ main_thm1 ] . in section [ sec_9 ] , we prove theorem [ eigen_asy ] . finally , in the appendix we compute @xmath119 in theorem [ main_thm1 ] , that is originally given in terms of certain integrals . in this section , we introduce work of hille @xcite and sibuya @xcite about properties of the solutions of . first , we scale equation because many facts that we need later are stated for the scaled equation . let @xmath25 be a solution of ( [ ptsym ] ) and let @xmath120 . then @xmath121 solves @xmath122v(z,\lambda)=0,\ ] ] where @xmath0 and @xmath12 is a polynomial ( possibly , @xmath123 ) of the form @xmath124 when @xmath21 is odd , becomes @xmath125v(z,\lambda)=0.\ ] ] later we will treat the case when @xmath21 is even . since we scaled the argument of @xmath126 , we must rotate the boundary condition . we state them in a more general context by using the following definition . _ the stokes sectors _ @xmath127 of the equation ( [ rotated ] ) are @xmath128 see figure [ f : graph1 ] . . the dashed rays represent @xmath129 it is known from hille @xcite that every nonconstant solution of ( [ rotated ] ) either decays to zero or blows up exponentially , in each stokes sector @xmath127 . that is , one has the following result . [ gen_pro ] @xmath130 * for each @xmath131 , every solution @xmath121 of ( [ rotated ] ) is asymptotic to @xmath132^{\frac{1}{2}}\,d\xi\right]\ ] ] as @xmath133 in every closed subsector of @xmath127 . * if a nonconstant solution @xmath121 of decays in @xmath127 , it must blow up in @xmath134 . however , when @xmath121 blows up in @xmath127 , @xmath121 need not be decaying in @xmath135 or in @xmath136 . lemma [ gen_pro ] ( i ) implies that if @xmath121 decays along one ray in @xmath127 , then it decays along all rays in @xmath127 . also , if @xmath121 blows up along one ray in @xmath127 , then it blows up along all rays in @xmath127 . this is essentially why we said in the introduction that the boundary conditions with @xmath1 represent all decaying boundary conditions . still with @xmath21 odd , the two rays in map , by @xmath137 , to the rays @xmath138 which are the center rays of the stokes sectors @xmath139 and @xmath140 . thus , the boundary conditions on @xmath25 become @xmath141 when @xmath21 is even , we let @xmath142 so that becomes @xmath143y(z,\lambda)=0,\ ] ] where @xmath144\ ] ] and hence , @xmath145 . for these cases , the boundary conditions become @xmath146 next we will introduce sibuya s results , but first we define a sequence of complex numbers @xmath147 in terms of the @xmath97 and @xmath27 , as follows . for @xmath148 fixed , we expand @xmath149 note that @xmath150 do not depend on @xmath27 , so we write @xmath151 for @xmath152 . so the above expansion without the @xmath153 term still gives @xmath147 for @xmath154 . we further define @xmath155 if @xmath16 is odd , and @xmath156 if @xmath16 is even . the following theorem is a special case of theorems 6.1 , 7.2 , 19.1 and 20.1 of sibuya @xcite that is the main ingredient of the proofs of the main results in this paper . [ prop ] equation ( [ rotated ] ) , with @xmath157 , admits a solution @xmath158 with the following properties . 1 . @xmath158 is an entire function of @xmath159 and @xmath27 . 2 . @xmath158 and @xmath160 admit the following asymptotic expansions . let @xmath86 . then @xmath161,\nonumber\\ f^{\prime}(z , a,\lambda)=&-z^{r_m+\frac{m}{2}}(1+o(z^{-1/2}))\exp\left[-f(z , a,\lambda ) \right],\nonumber\end{aligned}\ ] ] as @xmath4 tends to infinity in the sector @xmath162 , uniformly on each compact set of @xmath163-values . here @xmath164 3 . properties and uniquely determine the solution @xmath158 of ( [ rotated ] ) . 4 . for each fixed @xmath37 and @xmath165 , @xmath166 and @xmath167 also admit the asymptotic expansions , @xmath168\lambda^{-1/4}\exp\left[l(a,\lambda)\right],\label{eq1}\\ f^{\prime}(0,a,\lambda)=&-[1+o(1)]\lambda^{1/4}\exp\left[l(a,\lambda)\right],\label{eq2}\end{aligned}\ ] ] as @xmath117 in the sector @xmath169 , uniformly on each compact set of @xmath37 , where @xmath170 5 . the entire functions @xmath171 and @xmath172 have orders @xmath173 . in sibuya s book @xcite , see theorem 6.1 for a proof of ( i ) and ( ii ) ; theorem 7.2 for a proof of ( iii ) ; and theorem 19.1 for a proof of ( iv ) . moreover , ( v ) is a consequence of ( iv ) along with theorem 20.1 in @xcite . note that properties ( i ) , ( ii ) and ( iv ) are summarized on pages 112113 of sibuya @xcite . \(i ) uniformness of the error term in theorem [ main_thm1 ] is essentially due to uniformness of error terms in and . one can check this by carefully following our proofs . in this paper , we omit this part of the proof . \(ii ) throughout this paper , we will deal with numbers like @xmath174 for some @xmath175 , and @xmath176 . as usual , we will use @xmath177\ ] ] and if @xmath178 is specified , then @xmath179=s\left[{{\rm re}\,}(\nu)\frac{2\pi}{m+2}+\arg(\lambda)\right],\quad s\in{\mathbb r}.\ ] ] if @xmath180 then the branch of @xmath181 is chosen to be the negative real axis . next , we provide an asymptotic expansion of @xmath182 in @xcite . but first , we recall that for @xmath183 , @xmath184 then @xmath185 also , we define @xmath186 and define @xmath187 see appendix for @xmath119 in terms of some gamma functions as in theorem [ main_thm1 ] . [ asy_lemma ] let @xmath0 and @xmath37 be fixed . then there exist constants @xmath188 , @xmath189 , such that @xmath190 as @xmath117 in the sector @xmath191 , uniformly on each compact set of @xmath37 . see @xcite for a proof . sibuya @xcite proved the following corollary , directly from theorem [ prop ] , that will be used later in sections [ sec_4 ] and [ sec_5 ] . [ lemma_decay ] let @xmath37 be fixed . then @xmath192 as @xmath27 tends to infinity in the sector @xmath193 , and hence @xmath194 as @xmath117 in the sector @xmath195 . in particular , @xmath196 as @xmath117 in any closed subsector of the sector @xmath197 . in addition , @xmath198 as @xmath117 in any closed subsector of the sectors @xmath199 . based on corollary [ lemma_decay ] , sibuya ( * ? ? ? * theorem 29.1 ) also proved the asymptotic expansion of the eigenvalues of @xmath15 for @xmath20 . also , sibuya @xcite constructed solutions of that decays in @xmath127 , @xmath131 . before we introduce this , we let @xmath200 then we have the following lemma , regarding properties of @xmath201 . [ lemma_25 ] for @xmath37 fixed , and @xmath202 , @xmath203 , and @xmath204 if @xmath205 then @xmath206 next , recall that the function @xmath158 in theorem [ prop ] solves ( [ rotated ] ) and decays to zero exponentially as @xmath207 in @xmath208 , and blows up in @xmath209 . one can check that the function @xmath210 which is obtained by scaling @xmath211 in the @xmath4-variable , also solves ( [ rotated ] ) . it is clear that @xmath212 , and that @xmath213 decays in @xmath127 and blows up in @xmath134 since @xmath211 decays in @xmath208 . since no nonconstant solution decays in two consecutive stokes sectors ( see lemma [ gen_pro ] ( ii ) ) , @xmath214 and @xmath215 are linearly independent and hence any solution of ( [ rotated ] ) can be expressed as a linear combination of these two . especially , for each @xmath131 there exist some coefficients @xmath216 and @xmath217 such that @xmath218 we then see that @xmath219 where @xmath220 is the wronskian of @xmath221 and @xmath222 . since both @xmath223 are solutions of the same linear equation ( [ rotated ] ) , we know that the wronskians are constant functions of @xmath4 . also , @xmath224 and @xmath215 are linearly independent , and hence @xmath225 for all @xmath226 . also , the following is an easy consequence of and . for each @xmath227 we have @xmath228 moreover , we have the following lemma that is useful later on . [ shift_lemma ] suppose @xmath229 . then @xmath230 and @xmath231 , where @xmath232 see sibuya @xcite . we let @xmath233 , that is , @xmath234 in this section , we introduce asymptotic expansions of @xmath116 as @xmath117 along the rays in the complex plane @xcite . first , we provide an asymptotic expansion of the wronskian @xmath235 of @xmath236 and @xmath221 that will be frequently used later . [ lemma7 ] suppose that @xmath237 . then for each @xmath37 , @xmath238\exp\left[l(g^{j}(a),\omega^{2j - m-2}\lambda)+l(a,\lambda)\right],\ ] ] as @xmath117 in the sector @xmath239 next , we provide an asymptotic expansion of @xmath116 as @xmath117 in the sector near the negative real axis . [ thm_neg ] let @xmath0 , @xmath37 and @xmath240 be fixed . then @xmath241\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)+l(g(a),\omega^{-m}\lambda)\right],\ ] ] as @xmath242 along the rays in the sector @xmath243 this is an easy consequence of lemma [ lemma7 ] with @xmath244 and . also , for integers @xmath245 we provide an asymptotic expansion of @xmath116 as @xmath117 in the sector @xmath191 . [ zero_thm ] let @xmath37 and @xmath246 be fixed . if @xmath245 then @xmath247\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)-l(a,\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{1}{2}+\mu(a)+2\nu(a)}+o(1)]\exp\left[l(g(a),\omega^{2}\lambda)-l(a,\lambda)\right],\label{tot_asy}\end{aligned}\ ] ] as @xmath117 in the sector @xmath248 we provide an asymptotic expansion of @xmath116 as @xmath117 along the rays in the upper half plane . [ lemma_up ] let @xmath245 , @xmath37 and @xmath240 be fixed . then @xmath249\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)-l(a,\lambda)\right],\ ] ] as @xmath117 in the sector @xmath250 . also , @xmath251\exp\left[l(g(a),\omega^{2}\lambda)-l(a,\lambda)\right],\ ] ] as @xmath117 in the sector @xmath252 . we will determine which term in dominates in the upper and lower half planes . since , by , @xmath253 we have @xmath254-\left[{{\rm re}\,}(l(g(a),\omega^{2}\lambda))-{{\rm re}\,}(l(a,\lambda))\right]\nonumber\\ & = k_m\left[\cos\left(-\frac{2\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)-\cos\left(\frac{2\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)\right]|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1))\nonumber\\ & = 2k_m\sin\left(\frac{2\pi}{m}\right)\sin\left(\frac{m+2}{2m}\arg(\lambda)\right)|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1)).\nonumber\end{aligned}\ ] ] thus , the first term in dominates as @xmath117 along the rays in the upper half plane , and the second term dominates in the lower half plane . this completes the proof . in @xcite , @xmath255 is used for @xmath256 and asymptotics of @xmath255 are provided . notice that @xmath257 . theorem 13 in @xcite implies for the sector @xmath258 theorem 14 in @xcite implies that @xmath247\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)-l(a,\lambda)\right]\nonumber\\ & + [ 2\omega^{1 + 2\mu(a)+4\nu(a)}+o(1)]\exp\left[-l(g^2(a),\omega^{2-m}\lambda)-l(a,\lambda)\right],\nonumber\end{aligned}\ ] ] as @xmath117 in the sector @xmath259 . one can check that the first term dominates in this sector , by using an argument similar to that in the proof of corollary [ lemma_up ] . also , theorem 15 in @xcite implies that @xmath260\exp\left[-l(a,\omega^{-m-2}\lambda)-l(^{-2}(a),\omega^{-4}\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{1}{2}+\mu(a)+2\nu(a)}+o(1)]\exp\left[l(g(a),\omega^{-m}\lambda)-l(a,\omega^{-m-2}\lambda)\right],\nonumber\end{aligned}\ ] ] as @xmath117 in the sector @xmath261 . one can check that the second term dominates in this sector . then we replace @xmath27 by @xmath262 to convert the sector here to @xmath263 . this completes the proof . [ thm_sector2 ] let @xmath264 and let @xmath37 and @xmath240 be fixed . then @xmath265\exp\left[l(g^{4}(a),\omega^{-2}\lambda)-l(a,\lambda)\right]\nonumber\\ & -[2i\omega^{\frac{5}{2}}+o(1)]\exp\left[-l(g^2(a),\omega^{-1}\lambda)-l(a,\lambda)\right],\nonumber\end{aligned}\ ] ] as @xmath117 in the sector @xmath266 . also , @xmath267\exp\left[-l(a,\omega^{-5}\lambda)-l(g^{-2}(a),\omega^{-4}\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{15}{4}}+o(1)]\exp\left[l(g(a),\omega^{-3}\lambda)-l(a,\omega^{-5}\lambda)\right],\nonumber\end{aligned}\ ] ] as @xmath117 in the sector @xmath268 . see theorems 14 and 15 in @xcite for a proof . in this section , we will provide asymptotic expansions of @xmath270 , zeros of which will be closely related with the eigenvalues of @xmath271 . first , we treat the cases when @xmath272 . let @xmath272 be an integer . then @xmath270 admits the following asymptotic expansion @xmath273\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)-l(g^{n-1}(a),\omega^{2(n-1)}\lambda)\right],\ ] ] as @xmath117 in the sector @xmath274 also , @xmath275\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)-l(g^{n-1}(a),\omega^{2(n-1)}\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{2-n}{2}+\mu(g^{-1}(a))}+o(1)]\exp\left[l(g^n(a),\omega^{2n}\lambda)-l(a,\lambda)\right],\label{zero_asy}\end{aligned}\ ] ] as @xmath117 in the sector @xmath276 first we will prove for the sector @xmath277 and the second part of the theorem by induction on @xmath61 . the case when @xmath278 is trivially satisfied by theorem [ zero_thm ] and corollary [ lemma_up ] since @xmath279 . suppose that holds in the sector for @xmath280 . from this induction hypothesis we have @xmath281\exp\left[l(a,\lambda)-l(g^{n-1}(a),\omega^{2(n-1)}\lambda)\right],\label{1steq}\end{aligned}\ ] ] as @xmath117 in the sector @xmath282 that is , @xmath283 also , from lemma [ lemma7 ] if @xmath284 , then we have @xmath285\exp\left[l(g^{j}(a),\omega^{2j - m-2}\lambda)+l(a,\lambda)\right],\ ] ] as @xmath117 in the sector @xmath286 we solve for @xmath287 and set @xmath288 to get @xmath289 set @xmath290 . then since @xmath291 , using , @xmath292\exp\left[l(g^{n}(a),\omega^{2k - m-2}\lambda)+l(g^n(a),\omega^{2n}\lambda)\right]}{\omega^{n-1}[2i\omega^{-\frac{k}{2}}+o(1)]\exp\left[l(g^{k}(a),\omega^{2k - m-2}\lambda)+l(a,\lambda)\right]}\nonumber\\ & + \frac{[2\omega^{-\frac{n-1}{2}+\mu(g^{n-1}(a))}+o(1)]\exp\left[l(a,\lambda)-l(g^{n-1}(a),\omega^{2(n-1)}\lambda)\right]}{\omega^{-1}[2i\omega^{-\frac{k}{2}}+o(1)]\exp\left[l(g^{k}(a),\omega^{2k - m-2}\lambda)+l(a,\lambda)\right]}\nonumber\\ & \times [ 2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[l(g^{k}(a),\omega^{2k - m-2}\lambda)+l(g^{-1}(a),\omega^{-2}\lambda)\right]\nonumber\\ & = [ 2\omega^{\frac{2-n}{2}+\mu(g^{-1}(a))}+o(1)]\exp\left[l(g^n(a),\omega^{2n}\lambda)-l(a,\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{2-n}{2}+\mu(g^{n-1}(a))}+o(1)]\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)-l(g^{n-1}(a),\omega^{2(n-1)}\lambda)\right],\label{asy_sector}\end{aligned}\ ] ] where we used for @xmath293 and for everything else , provided that @xmath27 lies in and that @xmath294 that is , @xmath295 thus , the second part of the theorem is proved by induction . next in order to prove the first part of the theorem for the sector , we will determine which term in dominates as @xmath117 . to do that , we look at @xmath296|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1))\nonumber\\ & = 2k_m\sin\left(\frac{n\pi}{m}\right)\left[\sin\left(\frac{(n-2)\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)\right.\nonumber\\ & \qquad\qquad\qquad\qquad\qquad\qquad\left.+\sin\left(\frac{n\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)\right]|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1))\nonumber\\ & = 4k_m\sin\left(\frac{n\pi}{m}\right)\cos\left(\frac{\pi}{m}\right)\sin\left(\frac{(n-1)\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1)),\label{dom_set}\end{aligned}\ ] ] that tends to positive infinity as @xmath117 ( and hence the second term in dominates ) if @xmath297 . we still need to prove for the sector @xmath298 for which we use induction on @xmath61 again . when @xmath278 , holds by lemma [ lemma_up ] . suppose that in the sector for @xmath280 with @xmath299 . then and with @xmath300 yield @xmath301 if @xmath302 , then @xmath303 . so @xmath304\exp\left[l(g^{n+1}(a),\omega^{2n - m}\lambda)+l(a,\lambda)\right]}\nonumber\\ & + \frac{[2\omega^{\frac{3-n}{2}+\mu(g^{n-1}(a))}+o(1)]\exp\left[l(a,\lambda)-l(g^{n-1}(a),\omega^{2(n-1)}\lambda)\right]}{[2i\omega^{-\frac{n+1}{2}}+o(1)]\exp\left[l(g^{n+1}(a),\omega^{2n - m}\lambda)+l(a,\lambda)\right]}\nonumber\\ & \times [ 2i\omega^{-\frac{n+2}{2}}+o(1)]\exp\left[l(g^{n+1}(a),\omega^{2n - m}\lambda)+l(g^{-1}(a),\omega^{-2}\lambda)\right]\nonumber\\ & = [ -2i\omega^{\frac{3-n}{2}+\mu(g^{-1}(a))+\mu(g^n(a))}+o(1)]\exp\left[-l(g^{n+1}(a),\omega^{2n - m}\lambda)-l(a,\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{2-n}{2}+\mu(g^{n-1}(a))}+o(1)]\exp\left[l(g^{-1}(a),\omega^{-2}\lambda)-l(g^{n-1}(a),\omega^{2(n-1)}\lambda)\right],\nonumber\end{aligned}\ ] ] where we use the induction hypothesis for @xmath305 and use for @xmath306 and @xmath307 . next , we use an argument similar to complete the induction step . thus , the theorem is proved . next we investigate @xmath308 . [ bod_thm ] if @xmath245 is an even integer , then @xmath309\exp\left[-l(g^{\frac{m+2}{2}}(a),\lambda)-l(a,\lambda)\right]\nonumber\\ & -[2\omega^{2+\mu(a)+\mu(g^{\frac{m-2}{2}}(a))}+o(1)]\exp\left[-l(g^{\frac{m-2}{2}}(a),\omega^{m-2}\lambda)-l(g^{m}(a),\omega^{m-2}\lambda)\right].\label{thm_eq1}\end{aligned}\ ] ] as @xmath117 in the sector @xmath310 if @xmath245 is an odd integer , then @xmath311\exp\left[-l(g^{\frac{m+1}{2}}(a),\omega^{-1}\lambda)-l(a,\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{5}{4}}+o(1)]\exp\left[l(g^{m+1}(a),\omega^{-2}\lambda)-l(g^{\frac{m-3}{2}}(a),\omega^{m-3}\lambda)\right].\label{thm_eq2}\end{aligned}\ ] ] as @xmath117 in the sector @xmath312 we will use with @xmath313 , that is , @xmath314 when @xmath16 is even , say @xmath315 , @xmath316 so @xmath317 since @xmath27 lies in , @xmath318 @xmath319\exp\left[l(g^{k+1}(a),\omega^{2k - m}\lambda)+l(a,\lambda)\right]}\nonumber\\ & + \frac{[2\omega^{\frac{3-k}{2}+\mu(g^{k-1}(a))}+o(1)]\exp\left[l(a,\lambda)-l(g^{k-1}(a),\omega^{2(k-1)}\lambda)\right]}{\omega^2 [ 2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[l(g^{k+1}(a),\omega^{2k - m}\lambda)+l(a,\lambda)\right]}\nonumber\\ & \times[2\omega^{\frac{3-k}{2}+\mu(g^{2k}(a))}+o(1)]\exp\left[l(g^{k+1}(a),\lambda)-l(g^{2k}(a),\omega^{2(k-1)}\lambda)\right]\nonumber\\ & = [ -2i\omega^{\frac{3-k}{2}+\mu(g^{-1}(a))+\mu(g^{k}(a))}+o(1)]\exp\left[-l(g^{k+1}(a),\lambda)-l(a,\lambda)\right]\nonumber\\ & -[2i\omega^{\frac{3-k}{2}+\mu(g^{k-1}(a))+\mu(g^{2k}(a))}+o(1)]\exp\left[-l(g^{k-1}(a),\omega^{2(k-1)}\lambda)-l(g^{2k}(a),\omega^{2(k-1)}\lambda)\right]\nonumber\\ & = -[2i\omega^{\frac{3-k}{2}+\mu(g^{-1}(a))+\mu(g^{k}(a))}+o(1)]\exp\left[-l(g^{k+1}(a),\lambda)-l(a,\lambda)\right]\nonumber\\ & -[2i\omega^{\frac{3-k}{2}+\mu(a)+\mu(g^{k-1}(a))}+o(1)]\exp\left[-l(g^{k-1}(a),\omega^{2(k-1)}\lambda)-l(g^{2k}(a),\omega^{2(k-1)}\lambda)\right],\nonumber\end{aligned}\ ] ] where we used for @xmath320 and @xmath321 , and for @xmath322 . finally , we use @xmath323 , to get the desired asymptotic expansion of @xmath324 . next we investigate the case when @xmath16 is odd , say @xmath325 ( so @xmath326 ) . @xmath327 and @xmath328 similarly to the proof of the theorem for @xmath16 even , @xmath329\exp\left[l(g^{k+1}(a),\omega^{2k - m}\lambda)+l(a,\lambda)\right]}\nonumber\\ & + \frac{[2\omega^{\frac{3-k}{2}+\mu(g^{k-1}(a))}+o(1)]\exp\left[l(a,\lambda)-l(g^{k-1}(a),\omega^{2(k-1)}\lambda)\right]}{\omega^{\frac{1}{2 } } [ 2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[l(g^{k+1}(a),\omega^{2k - m}\lambda)+l(a,\lambda)\right]}\nonumber\\ & \times[2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[l(g^{2k+2}(a),\omega^{2k - m-1}\lambda)+l(g^{k+1}(a),\omega^{-1}\lambda)\right]\nonumber\\ & = [ -2i\omega^{\frac{3-k}{2}+\mu(g^{-1}(a))+\mu(g^{k}(a))}+o(1)]\exp\left[-l(g^{k+1}(a),\omega^{-1}\lambda)-l(a,\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{2-k}{2}+\mu(g^{k-1}(a))}+o(1)]\exp\left[l(g^{2k+2}(a),\omega^{-2}\lambda)-l(g^{k-1}(a),\omega^{2(k-1)}\lambda)\right]\nonumber\\ & = -[2i\omega^{\frac{3-k}{2}+\frac{m}{2}}+o(1)]\exp\left[-l(g^{k+1}(a),\omega^{-1}\lambda)-l(a,\lambda)\right]\nonumber\\ & + [ 2\omega^{\frac{2-k}{2}+\frac{m}{4}}+o(1)]\exp\left[l(g^{2k+2}(a),\omega^{-2}\lambda)-l(g^{k-1}(a),\omega^{2(k-1)}\lambda)\right],\nonumber\end{aligned}\ ] ] where we use for @xmath320 , and use for @xmath322 and @xmath330 . finally , we use @xmath331 , to get the asymptotic expansion of @xmath324 . this completes the proof . the _ order of an entire function _ @xmath332 is defined by @xmath333 where @xmath334 for @xmath335 . if for some positive real numbers @xmath336 , we have @xmath337\leq m(r , g)\leq \exp[c_2 r^{\sigma}]$ ] for all large @xmath338 , then the order of @xmath332 is @xmath339 . [ zero_free ] let @xmath340 . then the entire functions @xmath270 are of order @xmath173 , and hence they have infinitely many zeros in the complex plane . moreover , @xmath270 have at most finitely many zeros in the sector @xmath341 in this section , we will relate the eigenvalues of @xmath15 with zeros of some entire function @xmath270 . suppose @xmath342 is odd with @xmath343 . then becomes by the scaling @xmath120 , and @xmath344 decays in the stokes sectors @xmath345 and @xmath346 . since @xmath347 and @xmath214 are linearly independent , for some @xmath348 and @xmath349 one can write @xmath350 then one finds @xmath351 also it is easy to see that @xmath27 is an eigenvalue of @xmath15 if and only if @xmath352 if and only if @xmath353 . since @xmath354 , by corollary [ zero_free ] @xmath355 has at most finitely many zeros in the sector @xmath356 that is , @xmath357 this is true for all @xmath37 fixed . next , by symmetry one can show that @xmath355 has at most finitely many zeros in the sector @xmath358 . we look at @xmath15 with @xmath24 replaced by @xmath359 whose coefficient vector is @xmath360 . then one sees that @xmath353 if and only if @xmath361 . the latter has at most finitely many zeros in the sector @xmath362 thus , @xmath355 has at most finitely many zeros in the sector @xmath358 , and has infinitely many zeros in the sector @xmath363 since it is an entire function of order @xmath364 . suppose that @xmath365 is even with @xmath366 . then becomes by the scaling @xmath367 , and @xmath368 decays in the stokes sectors @xmath345 and @xmath136 . we then see that the coefficient vector @xmath369 of the polynomial @xmath370 becomes @xmath371 now one can express @xmath372 as a linear combination of @xmath214 and @xmath215 as follows . @xmath373 thus , @xmath27 is an eigenvalue of @xmath15 if and only if @xmath374 . since @xmath375 by corollary [ zero_free ] , @xmath376 has at most finitely many zeros in the sector @xmath377 that is , @xmath357 this is true for each @xmath37 . so one can show that @xmath376 has at most finitely many zeros in the sector @xmath358 by symmetry , similar to the cases when @xmath21 is odd . thus , @xmath378 has infinitely many zeros in the sector @xmath363 . in this section , we prove theorem [ main_thm1 ] for @xmath379 . we first treat the case when @xmath21 is odd . suppose that @xmath380 . recall that when @xmath21 is odd , @xmath27 is an eigenvalue of @xmath15 if and only if @xmath353 . then since @xmath355 has at most finitely many zeros outside the sector @xmath363 , all the eigenvalues @xmath27 of @xmath15 lie in the sector @xmath363 if @xmath381 is large enough . since @xmath382 we will use to investigate asymptotics of large eigenvalues . suppose that @xmath353 and @xmath381 is large enough . then from with @xmath383 , and with @xmath384 and @xmath27 replaced by @xmath385 and @xmath386 , respectively , we have @xmath387&\exp\left[l(g^{k}(a),\omega^{2k}\lambda)-l(g^{-k}(a),\omega^{-2k}\lambda)\right]\nonumber\\ & \times\exp\left[l(g^{k-1}(a),\omega^{2k-2}\lambda)-l(g^{-k+1}(a),\omega^{-2k+2}\lambda)\right]=-\omega^{2\nu(g^k(a))}.\nonumber\end{aligned}\ ] ] next the term @xmath388 $ ] can be absorbed into the exponential function so that we get @xmath389\nonumber\\ & \times\exp\left[l(g^{k-1}(a),\omega^{2k-2}\lambda)-l(g^{-k+1}(a),\omega^{-2k+2}\lambda)+o(1)\right]=-\omega^{2\nu(g^k(a))}.\label{sim_eq1}\end{aligned}\ ] ] for each odd integer @xmath390 , we define @xmath391 where the error term @xmath41 is the same as that in . then by corollary [ lemma_decay ] , @xmath392 in the sector @xmath363 . since @xmath393 and @xmath394 , the function @xmath395 maps the region @xmath396 for some large @xmath397 and @xmath363 into a region containing @xmath398 for some large @xmath399 and @xmath400 for some @xmath401 . so there exists a sequence of @xmath7 in the sector @xmath363 such that @xmath402= -\omega^{2\nu(g^k(a))}$ ] for all large @xmath30 . so , from , we have @xmath403 thus , since @xmath404 from lemma [ asy_lemma ] , @xmath405 where for @xmath406 , the coefficients @xmath89 are given by @xmath407 since @xmath342 , for @xmath51 , @xmath408 where we used lemma [ lemma_25 ] along with . notice that shows @xmath409 if @xmath16 is even , then we redefine @xmath410 as the sum of @xmath411 above and @xmath412 ( c. f. , equation ) . finally , note that @xmath413 this completes the proof . next we prove theorem [ main_thm1 ] for @xmath379 is even . proof is very similar to the case when @xmath379 is even . let @xmath365 for some @xmath414 . recall that @xmath27 is an eigenvalue of @xmath15 if and only if @xmath415 . then from , we have @xmath416\nonumber\\ & \times\exp\left[l(g^{k}(\widetilde{a}),\omega^{2k-1}\lambda)-l(g^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)+o(1)\right]=-1,\label{sim_eq2}\end{aligned}\ ] ] where we used @xmath388=\exp[o(1)]$ ] again . like in the case when @xmath379 is odd , from lemma [ asy_lemma ] , @xmath417 where for @xmath406 , @xmath418 where we used @xmath419 and @xmath365 as well . in this section , we prove theorem [ main_thm1 ] for @xmath78 . we first prove the theorem when @xmath16 is even , and later , we will treat the cases when @xmath16 is odd . then at the end of the section , we will prove the theorem when @xmath420 , by scaling . we further divide the case into when @xmath21 is odd and when @xmath21 is even . suppose that @xmath16 is even and @xmath421 for some @xmath414 . since @xmath27 is an eigenvalue of @xmath15 if and only if @xmath353 if and only if @xmath422 . suppose that @xmath422 . if @xmath381 is large enough , then from , @xmath423\nonumber\\ & \times\exp\left[l(g^{k-1}(a),\omega^{2k-2}\lambda)-l(g^{-k+1}(a),\omega^{-2k+2}\lambda)+o(1)\right]=-\omega^{4\nu(g^{k}(a))}.\nonumber\end{aligned}\ ] ] from lemma [ asy_lemma ] , @xmath424 where for @xmath406 , the coefficients @xmath89 are given by @xmath425 where we used lemma [ lemma_25 ] as well as some other thing as before . in addition , we used @xmath426 let @xmath427 for some @xmath414 . then @xmath27 is an eigenvalue of @xmath15 if and only if @xmath415 . if @xmath415 and @xmath381 large enough , then from , @xmath428\nonumber\\ & \times\exp\left[l(g^{k}(\widetilde{a}),\omega^{2k-1}\lambda)-l(g^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)+o(1)\right]=-1.\nonumber\end{aligned}\ ] ] then like before , @xmath429 where for @xmath406 , @xmath430 we divide the case into when @xmath21 is odd and when @xmath21 is even . let @xmath16 and @xmath431 be odd . suppose that @xmath27 is an eigenvalue of @xmath15 . since @xmath27 is an eigenvalue of @xmath15 if and only if @xmath422 , if @xmath381 is large enough , then from , @xmath432\nonumber\\ & \times\exp\left[-l(g^{k+1}(a),\omega^{-2k+1}\lambda)-l(g^{-k+1}(a),\omega^{-2k+2}\lambda)+o(1)\right]=-1.\nonumber\end{aligned}\ ] ] then , from lemma [ asy_lemma ] , @xmath433 where for @xmath406 , @xmath434 where we used @xmath435 let @xmath436 for some @xmath414 . then @xmath27 is an eigenvalue of @xmath15 if and only if @xmath415 . if @xmath415 and @xmath381 large enough , then from , @xmath437\nonumber\\ & \times\exp\left[-l(g^{3k+3}(\widetilde{a}),\omega^{-2k-1}\lambda)-l(g^{k+2}(\widetilde{a}),\omega^{-2k}\lambda)+o(1)\right]=-1.\nonumber\end{aligned}\ ] ] then , from lemma [ asy_lemma ] , @xmath438 where for @xmath406 , @xmath439 theorem [ main_thm1 ] for @xmath440 has been proved . next we prove theorem [ main_thm1 ] for @xmath420 , by scaling @xmath441 . suppose that @xmath442 . if @xmath126 is an eigenfunction of @xmath15 , then @xmath443 solves @xmath444v(z,\lambda ) = \lambda v(z,\lambda),\ ] ] and @xmath445 the coefficient vector of @xmath446 is @xmath447 certainly , @xmath448 also , one can find from that for @xmath449 , @xmath450 moreover , @xmath451 this completes proof of theorem [ main_thm1 ] . in this section , we prove theorem [ eigen_asy ] . we begin with @xmath452 @xmath453 let @xmath454 introduce the decomposition @xmath455 where @xmath456 then we have @xmath457 thus , @xmath458 and hence @xmath459 thus , one concludes @xmath460 where @xmath461 notice that @xmath462 since @xmath463 . hence , from along with we have @xmath464 suppose that @xmath465 where @xmath466 and @xmath467 for some @xmath468 . then from @xmath469 hence , @xmath470 next , @xmath471 we use this in to see that the left hand side of is a power series in @xmath472 . then comparing coefficients of @xmath473 , @xmath474 , we have @xmath475 moreover , if @xmath476 ( i. e. , @xmath477 ) then there exists some constant @xmath478 such that @xmath479 now we let @xmath480 where @xmath481 and @xmath482 . if @xmath483 then @xmath484 could be smaller than the error term @xmath485 in , and hence we can not deduce existence of @xmath486 like we do in . this completes proof of theorem [ eigen_asy ] . a first few @xmath72 are as follows . let @xmath0 be an integer . then @xmath488 suppose that @xmath496 , @xmath497 . then since @xmath498 we have @xmath499 next , we use the substitution @xmath489 to show @xmath500 finally , we use equations in to complete the proof for @xmath496 , @xmath497 . the author thanks mark ashbaugh , fritz gesztesy , richard laugesen , boris mityagin , grigori rozenblioum and alexander turbiner for helpful discussions and references . he also thanks richard laugesen for reading a part of this manuscript and suggestions for improving its presentation .

we present r - band images covering more than 11 square degrees of sky that were obtained in preparation for the spitzer space telescope first look survey ( fls ) . the fls was designed to characterize the mid - infrared sky at depths 2 orders of magnitude deeper than previous surveys . the extragalactic component is the first cosmological survey done with _ spitzer_. source catalogs extracted from the r - band images are also presented . the r - band images were obtained using the mosaic-1 camera on the mayall 4 m telescope of the kitt peak national observatory . two relatively large regions of the sky were observed to modest depth : the main fls extra galactic field ( j2000 ; l , b ) and elais - n1 field ( j2000 ; l , b ) . while both of these fields were in early plans for the fls , only a single deep pointing test observation was made at the elais - n1 location . the larger legacy program swire ( lonsdale et al . , 2003 ) will include this region among its surveyed areas . the data products of our kpno imaging ( images and object catalogs ) are made available to the community through the world wide web ( via the spitzer science center and noao science archives ) . the overall quality of the images is high . the measured positions of sources detected in the images have rms uncertainties in their absolute positions of order 0.35 arc - seconds with possible systematic offsets of order 0.1 arc - seconds , depending on the reference frame of comparison . the relative astrometric accuracy is much better than 0.1 of an arc - second . typical delivered image quality in the images is 1.1 arc - seconds full width at half maximum . images are relatively deep since they reach a median 5 depth limiting magnitude of r=25.5 ( vega ) , as measured within a 1.35 fwhm aperture for which the s / n ratio is maximal . catalogs have been extracted using sextractor using thresholds in area and flux for which the number of false detections is below 1% at r=25 . only sources with s / n greater than 3 have been retained in the final catalogs . comparing the galaxy number counts from our images with those of deeper r - band surveys , we estimate that our observations are 50% complete at r=24.5 . these limits in depth are sufficient to identify a substantial fraction of the infrared sources which will be detected by _ spitzer_.

astro-ph0403490

one of the main advantages of the spitzer space telescope ( formerly known as the space infrared infrared telescope facility ; sirtf , gallagher et al . 2003 ) is the possibility to make extragalactic surveys of large regions of the sky in a relatively short time covering wavelengths from the near - ir to the far - ir with the instruments irac ( fazio et al . 1998 ) and mips ( rieke et al . 1996 ) . compared to _ s predecessors ( e.g. iras , soifer et al . 1983 , and iso , kessler et al . 1996 ) , there are improvements in the detectors ( number of pixels and better responsivity ) , the collecting area of the primary mirror ( 85 cm diameter ) , and sun - earth - moon avoidance constraints due _ spitzer _ s heliocentric orbit . _ spitzer _ can also make observations simultaneously in multiple bands ( with irac3.6 and 4.5 or 5.8 and 8 @xmath10 m ; with mips24 , 70 and 160 @xmath10 m ) . many extragalactic surveys are already scheduled with _ spitzer _ as legacy programs ( swire , lonsdale et al . 2003 ; goods , dickinson & giavalisco 2001 ) or as observations by the instrument teams ( wide , deep and ultra - deep _ spitzer _ surveys which will cover regions like the botes field of the noao deep wide - field survey , the groth strip , lockman hole , xmm - deep and so on , see e.g. dole et al . the first look survey utilizes 112 hours of director s discretionary time on _ spitzer _ and includes extragalactic , galactic , and ecliptic components . these data will be available to all observers when the spitzer science archive opens in may , 2004 . the purpose of the fls is to characterize the mid - infrared sky at previously unexplored depths and make these data rapidly available to the astronomical community . the extragalactic component is comprised of a 4 square degree survey with irac and mips near the north ecliptic pole centered at j1718 + 5930 . these observations were executed on 2003 december 1 - 11 . to fully exploit the _ spitzer _ fls data we have obtained ancillary surveys at optical ( this paper ) and radio wavelengths ( condon et al . 2003 ) . given the modest spatial resolution of the _ spitzer _ imagers ( the point spread function is large , especially in the mid and far - ir , e.g. 5.7 arcseconds full width at half maximum , fwhm , for the 24 @xmath10 m channel ) , the first problem to solve for the infrared sources detected by _ spitzer _ will be to associate these sources with an optical counterpart , when possible . this will then allow the higher spatial resolution of the available optical images to assist with the source classification ( e.g. as stars , galaxies or qso ) and enable targeting of subsets of the sources for spectroscopy with optical or near - ir spectrographs . since many of the infrared sources that will be detected by _ spitzer _ will be dust - obscured galaxies with faint optical counterparts , the complementary optical imaging must be relatively deep . although a deep multi - wavelength optical survey would be more useful , allowing one to compute photometric redshifts ( e.g. the noao deep wide - field survey , jannuzi and dey 1999 , brown et al . 2003 ) , the task of deeply covering a large region of sky in a homogeneous manner is quite time - consuming . therefore , for the initial optical ancillary survey we chose to observe the entire field in the r - band . noao provided 4 nights of director s discretionary time on the kpno 4 m in may , 2000 , for this survey . we have limited multi - wavelength optical observations to the central portion of the fls field . the sloan digital sky survey included the fls field in their early release observations ( stoughton et al . 2002 ) and mosaics and catalogs for the region are also now available ( hogg et al . 2004 ) . in this paper , we present the r - band optical observations made with the mosaic-1 camera on the mayall 4 m telescope of kitt peak national observatory . centered on the main fls field , a region 9.4 square degrees in area was imaged . in addition , 2.3 square degrees covering the elais - n1 field was also observed . although originally the elais - n1 field was planned to be part of the fls program , the fls observations of the elais - n1 field have now been revised to a very deep 10@xmath910 pointing to evaluate the confusion limits of the mips instrument . the remainder of the elais - n1 field will now be imaged as part of a larger survey in this region , a portion of the swire _ spitzer _ legacy survey ( lonsdale et al . , 2003 ) . in section 2 we review the overall observing strategy and describe the mosaic-1 observations . in section 3 we discuss the techniques used in the data reduction including the astrometric and photometric calibration of the images . we describe in section 4 the data products made publicly available . we detail in section 5 the criteria used to detect , classify , and photometrically measure objects in the images . section 5 also includes a description of the information available in our catalogs . in section 6 we examine the quality of the imaging data by comparing them with other available data sets . finally , a brief summary is given in section 7 . the optical observations of the fls region ( centered at @xmath0 @xmath11 , j2000 ) and of the elais - n1 region ( centered at @xmath4 @xmath12 , j2000 , see oliver et al . 2000 ) were carried out using the mosaic-1 camera on the mayall 4 m telescope at the kitt peak national observatory . the mosaic-1 camera is comprised of eight thinned , back - illuminated site 2048 @xmath9 4098 ccds with a projected pixel size of 0.258 arcsec ( muller et al . the 8 ccds are physically separated by gaps with widths of approximately 14 and 15.5 arcsec along the right ascension and declination directions , respectively . the full field of view of the camera is therefore 36 @xmath9 36 square arcmin , with a filling factor of 97% . observations were performed using the harris set kron - cousins r - band filter whose main features are summarized in table [ tbl : filter ] . the transmission curve of the filter is shown in figure [ fig : transcurve ] together with the resulting modifications that would be introduced by the corrector and camera optics , the ccd quantum efficiency , and atmospheric extinction . for organizational purposes we chose to divide the proposed fls survey region into 30 subfields each roughly the size of an individual mosaic-1 pointing . the coordinates of these subfields are listed in table [ tbl : coordinates ] . during our observing run we were able to complete observations for 26 of these subfields . we similarly divided the elais - n1 field into 12 subfields , but only completed observations for 5 of these fields . each subfield was observed for a minimum of three 10 min exposures . in practice , some images were not suitable ( poor seeing , flat fielding problem , or some other defect ) and were not included in the final coadded or `` stacked '' images we are providing to the community . table [ tbl : coordinates ] lists in col . 4 the number of exposures which were obtained and included in the coadded or `` stacked '' images . in order to provide some coverage in the regions of the sky that would fall in the inter - chip gaps , the positions of successive exposures of a given field were offset on order of an arcminute relative to each other . in general the first exposure was at the nominal ( tabulated ) position , the second shifted by @xmath13 in @xmath14 and @xmath15 in @xmath16 , and the third with a shift of @xmath17 and 62.3@xmath18 in @xmath14 and @xmath16 . in a few cases an additional position with @xmath19 and @xmath20 has been observed . for a few fields , observations have been repeated due to the bad seeing or pointing errors . the kpno imaging observations were made on 2000 may 4 - 7,9 ut . a log of the observations is summarized in table [ tbl : log ] , which lists for each group of observations : the date of the observations ( col . 1 ) ; the subfield name ( col.2 ) ; the integration time and the number of exposures ( col . 3 ) ; and the seeing ( delivered image quality expressed as fwhm in arcseconds of bright unsaturated stars ) range of each exposure ( col . 4 ) . during a portion of the observing run , the pointing of the telescope was incorrectly initialized , resulting in approximately a @xmath21 error in the pointing for some fields . these are noted in the observing log . since the @xmath21 offset did not map exactly on to our subfield grid , we chose to make stacked or combined images for several of the subfields . this was done for subfields # 11 and # 17 in the fls region and subfields ( # 6 , # 9 and # 10 ) of elais - n1 . in figures [ fig : flsfields ] and [ fig : elaisfields ] we show the positions of each subfield on the sky with respect to the fields which have been covered with the irac and mips instruments on - board _ spitzer_. in figure [ fig : elaisfields ] we display the position of the subfields in the elais - n1 region and the region observed with isocam as part of the european large area iso survey ( elais , oliver et al . the shaded square indicates the area which has been observed to test the confusion mips 24 micron confusion limits , as part of the fls observations . m ( oliver et al . the grey shaded square is the field observed with spitzer to test the mips 24 micron confusion limit . the swire planned field ( lonsdale et al . 2003 ) is marked with a dash - dotted line . , scaledwidth=50.0% ] the processing of the raw mosaic-1 exposures followed the steps outlined in version 7.01 of `` the noao deep wide - field survey mosaic data reductions guide '' and discussed by jannuzi et al . ( in preparation ) . the bulk of the software used to process the images and generate combined images for each subfield from the individual 10 minute exposures is described by valdes ( 2002 ) and contained as part of the mscred software package ( v4.7 ) , which is part of iraf . the image quality of the final stacks is variable , as was the seeing during the run . users of the images should be aware that the detailed shape of the point spread function in a given image stack could be variable across the field , not only because of residual distortions in the camera , but because a given position in the field might be the average of different input images , each with their own point spread function . no attempt was made to match the psfs of the individual images before combining the images . in a survey area this large there will be fields with very bright stars . this can cause some regions of the survey area to be impacted by scattered light from these stars . some effort was made to minimize this impact during the observations ( by shifting pointings ) and reduction of the images ( through masking of affected regions in some images to allow unaffected images to be the sole contribution to the stacked image ) , but users of the images should be aware that some scattered light will have made it into some of the stacks . an example of a subfield with significant scattered light is fls_15 . flat fielding of the images was accomplished through the application of calibration files generated first from observations of a flat - field screen inside the dome at the 4 m and from a `` super - sky '' flat constructed by combining the majority of the fls and elais - n1 images ( with objects masked and rejected ) . the result is generally excellent flat fielding of the sky ( with some fields with slightly different sky color , due to moon light or twilight worse ) , but users should be aware that there are likely to be some color - dependent variations between and within the 8 ccds , meaning that uncorrected errors in the photometry of a given object , attributable to the flat fielding correction not being derived from a source with color matched to the color of that object , on the order of 1 to 3 percent could still be present in the data even though the sky is generally quite `` flat '' . the astrometric calibration of the mosaic-1 images is accomplished in two steps . first , the high - order distortions in the field , which will in general be common to all our exposures ( those anticipated to be the result of the optics and/or ccd placement and as a result stable over the course of an observing run or season ) are calibrated using observations of an astrometric standard field . these distortions can be wavelength dependent , so the calibration is made for the specific filter being used . these images are analyzed to produce a default correction for each of the ccds . low - order corrections ( translational offset , small rotation , and/or scale adjustments needed to compensate for pointing errors , instrument mounting variations between runs , and atmospheric effects ) are corrected on an exposure by exposure basis using the many catalog objects in each exposure , the previously mentioned knowledge of the high - order distortions as a function of ccd position , and the software _ msccmatch _ in iraf . the astrometric calibration has been performed using the reduction protocol developed for the reduction of the noao deep wide - field survey ( ndwfs ) data which assumes as astrometric reference the usno - a2.0 catalog ( monet et al . 1998 ) . adopting the gsc ii catalog , which became available after the development of the ndwfs reduction protocol , might improve the quality of the astrometric solutions and this might be done if a rereduction of the data set is done in the future . as we discuss further below , the anticipated improvement would be slight . for the images described in this paper our solutions for the mapping of the mosaic-1 pixels into a world coordinate frame had an rms scatter of between 0.3 to 0.45 arcseconds , depending on the particular field . after each image was provided with an improved astrometric calibration , it was tangent - plane projected with respect to the fls and elais - n1 field center positions listed above . following projection , the individual exposures ( typically three for a given region of the sky ) could be combined into a final , stacked , image . these are the images currently being made available in the _ spitzer _ and noao archives . during the final stages of the data reduction , when the individual images were being combined to make the final image stacks , it was determined that there was an error in the high - order astrometric correction file for the harris r - band filter affecting the adjacent edges of ccds 7 and 8 in the mosaic-1 camera . the original solutions for the high order distortion terms were determined on a chip by chip basis with no requirement that the solution be continuous across the entire field . in general , while this requirement was not imposed , it was met by the solutions provided by noao . however , for the r - band the solution available at the time we were reducing the data was discontinuous at the ccd7 and ccd8 boundary . since our image stacks are made from the combination of three or more images that are offset by 30 arcseconds to an arcminute from each other , this difference between the solutions for the two ccds can result in a mis - mapping , into ra and dec space , of a region of the sky imaged first on ccd7 and then on ccd8 . the error is small , and introduces at most an additional @xmath22 uncertainty to the positions of sources in the affected region ( which is located in the se corner of each field , about 25% of the field up from the southern edge ) , but will result in a degradation of the psf for objects affected in this region . the size of the affected region in each subfield is approximately 8.5@xmath18 east - west and 2.5@xmath18 north - south , or a bit less than 2% of the surveyed area . [ b ! ] since not all the subfields were observed during nights with photometric conditions ( see table [ tbl : log ] ) , we derived a coherent photometric system in two steps . first , we computed the relative photometric zero - points between the different stacked images of the subfields . since each subfield overlaps its neighbor by approximately two arcminutes , we can estimate the extinction difference between them using a set of common sources . this can be expressed in terms of zero - point that , by definition , includes the effects of airmass and extinction . because each frame has multiple overlaps , the number of frame - to - frame magnitude differences is over - determined with respect to the number of frames . therefore , one can derive the relative zero - point for each frame simultaneously using a least squares estimator . we extracted the sources from each subfield using as a first guess for the zero - point that we computed for a central subfield ( # 18 and # 5 for the fls and elais - n1 fields , respectively ) . for each overlap between contiguous fields , we selected the pairs of stars with magnitude @xmath23 and computed the median of the difference in magnitudes by using a 3@xmath8 clipping procedure . the magnitudes considered were the auto magnitudes ( mag_auto ) from sextractor , which are fairly robust with respect to seeing variations . to estimate the relative zero - points in the sense of the least squares , we minimized the sum : @xmath24 where @xmath25 is the variation with respect to the initial guess of the zero - point of the subfield @xmath26 , and @xmath27 is the median of the differences of magnitude for the set of the @xmath28 source pairs in the overlapping region between the subfields @xmath26 and @xmath29 . solving the linear system obtained by imposing that the derivatives of ( 1 ) with respect each @xmath25 are equal to zero , we corrected the initial guesses for the zero - points . the procedure was then iterated until the number of pairs @xmath28 became stable . the second step was to make use of those fields observed under photometric conditions in order to converge on the best zero - point . standards were measured several times over a broad range of airmasses on the first two nights of the observations , which had the best photometric conditions ( see table [ tbl : log ] ) . only the second night ( may 5 ) was really photometric since all the measurements are coherent ( see figure [ fig : zeropoint ] ) . magnitudes were calibrated to the kron - cousins system using landolt standards taken from landolt ( 1992 ) . they have been obtained using an aperture of 6 arcsec in diameter , large enough to obtain accurate measurements according to the growth curves of all the measured stars . we have fitted the linear relationship : @xmath30 with @xmath31 magnitude of landolt standard , @xmath32 airmass during the observation , @xmath33 counts during the exposure time @xmath34 , @xmath35 and @xmath36 the zero and extinction terms , respectively . we used an iterative 3@xmath8 clipping in order to discard deviant measurements , finding the best fit for @xmath37 and @xmath38 . the standard deviation of the residuals for the best fit is 0.03 . the analysis of the survey data produced a set of intermediate and final products , images and catalogs , which are publicly available at the spitzer science center and noao science archives . in particular , we provide the astronomical community with : * coadded ( stacked ) images of the subfields * : sky - subtracted , fully processed coadded frames for each subfield . the subfields are mapped using a tangential projection . the size of each fits compressed image is 220 mb . for each image , a bad pixel mask and an exposure map is given ( in the pixel - list iraf format ) . the photometric zero - point of each subfield after the absolute photometric calibration of the frame appears in the header of the image . a full description of the header keywords is available at the http sites . * low - resolution field image * : sky - subtracted , fully processed coadded images of the whole field . the images have been created to give a general overview of the fls and elais - n1 fields and have been produced using a 1.3 arcsec pixel size ( corresponding to five times the original size of the pixel ) . users are discouraged from using them to extract sources and compute photometry . these images have a size of approximatively 380 mb and 80 mb . * single subfield catalogs * : object catalogs associated with each single subfield . a full description of the parameters available is described in the following section . the catalogs are in ascii format . the main goal of this survey is to catalog galaxies and faint stars and make a first distinction between stars and galaxies on the basis of their intensity profiles . several bright objects ( mainly stars ) are saturated and excluded from the catalog , but can be found in catalogs from shallower surveys ( like the sloan digital sky survey in the case of the fls region , see e.g. stoughton et al . 2002 and hogg et al . the source extraction was performed with the sextractor package ( bertin & arnouts 1996 ; ver . 2.3 ) which is well suited for surveys with low to moderate source density as is the case of our surveys . several parameters have to be fixed to achieve an efficient source extraction with sextractor . the first problem is the evaluation of the background . sextractor proceeds computing a _ mini - background _ on a scale large enough to contain several faint objects and filtering it with a box - car to avoid the contamination by isolated , extended objects . finally , a full - resolution background map is obtained by interpolation and it is subtracted from the science image . in our case , many bright stars populate the fls field since it has a moderate galactic latitude ( 34.9 degrees ) while the problem is less important in the case of the elais - n1 field ( gal . latitude : 44.9 degrees ) . to evaluate a background which is not locally dominated by bright stars we adopted meshes of 128@xmath9128 pixels for the _ mini - background _ corresponding to 33.0 arcsec and used a 9@xmath99 box - car for the median - filtering . in order to improve the detection of faint sources , the image is filtered to enhance the spatial frequency typical of the sources with respect to those of the background noise . a gaussian filter with an fwhm similar to the seeing of the image ( in our case 4 pixels , since the overall seeing is 1.1 ) has been used . although the choice of a convolution kernel with a constant fwhm may not always be optimal since the seeing is varying in the different images , the impact on detectability is fairly small ( irwin 1985 ) . moreover , it has the advantage of requiring no changes of the relative detection threshold . finally , the detection is made on the background - subtracted and filtered image looking for groups of connected pixel above the detection threshold . thresholding is in fact the most efficient way to detect low surface brightness objects . in our case , we fixed the minimum number of connected pixels to 15 and the detection threshold to 0.8 ( in units of the standard deviation of the background noise ) , which corresponds to a typical limiting surface brightness @xmath39 mag arcsec@xmath40 . for the detection we made use also of the exposure map as a weight considered to set the noise level for each pixel . some pixels have a null weight since they correspond to saturated objects , trails of bright objects and other artifacts . these pixels are also marked in the bad pixel mask and the false detections around these image artifacts are flagged and easily excluded from our final catalogs . only objects detected with a signal - to - noise greater than 3 ( based on the total magnitude errors ) are accepted in our final catalog . although not well suited to detect objects in crowded fields , sextractor allows one also to deblend close objects using a multiple isophotal analysis technique . two parameters affect the deblending : the number of thresholds used to split a set of connected pixels according to their luminosity peaks and the minimal contrast ( light in a peak divided by the total light in the object ) used to decide if deblending a sub - object from the rest of the object . in our analysis we used a high number of thresholds ( 64 ) and a very low minimal contrast ( 1.5 @xmath41 ) . nevertheless , a few blended objects still remain in the catalogs . visual inspection or other extraction algorithms more efficient in crowded fields ( e.g , daophot ) are needed to treat these particular cases . the photometry has been performed on the stacked images . several measurements were made : aperture and isophotal magnitudes and an estimate of the total magnitudes . we measured the aperture magnitude within a diameter of 3 arcsec , roughly corresponding to three times the overall seeing . total magnitudes ( _ mag_auto _ ) are estimated using an elliptical aperture with an approach similar to that proposed by kron ( 1980 ) . since these fields have been selected in sky regions with low galactic extinction to observe extragalactic infrared sources , the corrections for galactic extinction ( schlegel et al . 1998 ) are small : 0.06 and 0.01 on average for the fls and elais - n1 fields , respectively . the total magnitude of sources close to bright objects are usually inaccurate since the local background is affected by the halo of the bright objects and the kron radius is not correctly computed . to improve the photometry for these sources we have subtracted bright saturated stars from the images and excluded from the catalogs the sources detected in square boxes around these stars where the subtraction is not correct . moreover , we have considered bright extended galaxies and excluded from the catalogs all the sources inside the kron ellipses of the galaxies . in fact , most of these sources are bright regions of the galaxies or their photometry is highly affected by the diffuse luminosity of the galaxies . to subtract bright saturated stars from the images we have computed radial density profiles on concentric annuli around the stars . then , after subtracting these profiles , we have removed the diffraction spikes by fitting their profiles along the radius at different angles with chebyshev polynomials . as visible in figures [ fig : removestar6 ] and [ fig : removestar13 ] , the background is much more uniform and the spikes become shorter . this improves the photometry for the objects surrounding the stars and avoid the detection of faint false sources on the diffraction spikes . the correction works well for most of the stars , although we exclude from the catalogs the immediate neighborhood . in the case of very bright stars ( fig [ fig : removestar13 ] ) , multiple reflections between the ccd and the optics make difficult the subraction of a median radial profile and a faint halo is still visible after the correction . sextractor uses a neural network to separate star - like from extended sources returning a stellarity index ( _ class_star _ ) with values between 1 ( a perfect star - like object ) and 0 . the distribution of this index as a function of the magnitude in one of our fields ( fls18 ) is shown in figure [ fig : stellarity ] . the standard neural network of sextractor has been trained for seeing fwhm values between 0.02 and 5.5 arcsec and for images that have 1.5 @xmath42 fwhm @xmath42 5 pixels . it is therefore perfectly suited for our images . although at bright magnitudes two sequences can be easily distinguished ( see figure [ fig : stellarity ] ) , for fainter magnitudes it becomes more difficult to separate extended from point objects . in order to select stars in an efficient way , we followed the technique of groenewegen et al . ( 2002 ) choosing a threshold which is a function of the magnitude . we consider that our objects are stars if : @xmath43 a drawback of this technique is that at faint magnitudes qsos are also classified as stars . however , using multi - band catalogs one can address the issue of separating stars and qsos on the basis of their spectral energy distributions . as an illustration , the tabulation of the first 30 entries in the fls_2 source catalog is presented in table [ tbl : catalog ] . all magnitudes are given in the vega system . the table lists : * column 1 : the full iau designation of the source ; * columns 2 - 3 : right ascension and declination ( j2000 ) ; * columns 4 - 7 : aperture ( 3 arcsec diameter ) and total magnitudes and respective errors . the total magnitude corresponds to the _ mag_auto _ magnitude measured by sextractor . the magnitudes have been not corrected for galactic extinction . the errors are those estimated by sextractor and include only the shot - noise of the measured source and background counts . only objects detected with signal to noise s / n @xmath44 3 ( based on the total magnitude errors ) and without saturated pixels are included . * column 8 : an estimate of the s / n of the detection , from the errors estimated for the total magnitude ; * column 9 : the stellarity index computed by sextractor ; * column 10 : the galactic extinction taken from schlegel et al . ( 1998 ) . all the catalogs in the archive are in ascii format . each catalog has been verified with several tests to check the reliability of the measured quantities with other sets of data . the results of some of these tests are described in the next section . the characteristics of the data obtained by the present survey are summarized in table [ tbl : summary ] which lists : in col . 1 the name of the subfield , in cols . 2 - 3 the center of the subfield , in col . 4 the seeing of the combined image , in cols . 5 - 6 the 3@xmath8 and 5@xmath8 limiting r cousin magnitudes , in cols . 7 - 8 the number of galaxies and stars with s / n @xmath44 3@xmath8 which are not saturated . we measured the limiting r cousin magnitude inside the aperture with the highest signal - to - noise ratio in case of a gaussian psf dominated by the sky . considering a gaussian profile , the s / n ratio inside an aperture @xmath45 can be written as : @xmath46 with @xmath47 , fwhm , @xmath33 , central intensity of the source , @xmath48 , pixel size , and @xmath49 , the sky noise . the @xmath45 for which s / n is maximal corresponds to the @xmath45 for which @xmath50 . this condition is realized at @xmath51 , i.e. an aperture of 1.35 @xmath9 fhwm . since we are considering in our catalogs an aperture of 3 arcsec and our typical fhwm is 1.0 arcsec , these values are slightly deeper than what one can find in our catalogs . the fraction of spurious objects was estimated by creating catalogs from the survey images multiplied by -1 . since ideally the noise is symmetric , we can use these images to produce a catalog of spurious sources by applying the same criteria of extraction which have been used with the real images . analyzing the central square degree , false - positive detections occur only at faint magnitudes ( [email protected] ) . considering all the 3@xmath8 objects , in the magnitude interval 23.5@[email protected] there are 150 false - positive detection per sq . degree corresponding to 0.5% of the total number of sources . in the magnitude interval 24.5@[email protected] , there are 119 false - positive detections per sq . degree which correspond to 17% of the total of number of sources detected in this magnitude range . to assess the accuracy of the astrometric calibration , we compared the positions of the stars inside the fls field with those available from the sloan survey ( stoughton et al . 2002 , data release 1 ) and with radio sources from the vla survey in the fls field ( condon et al . 2003 ) . in the comparison with the sloan sources , we considered only good objects ( according to the flags ) classified as stars in the sloan catalog with r magnitude between 18 and 21 . table [ tbl : astrometry ] reports the number of stars used in the comparison and the offsets between our and sloan positions . typical offsets are 0.1 arcsec in right ascension and declination with an rms of 0.1 arcsec . the offsets between our catalog and the sloan stars in the case of the field fls-18 are shown in figure [ fig : cfr_astro ] . the comparison with the vla sources has been made considering all the optical counterparts down to r=24 of non - extended radio sources . the offsets from the vla positions are also of order 0.1 arcsec . in the various fls fields . the average difference is @xmath53 . , scaledwidth=50.0% ] the systematic offset between ours and sloan positions comes from the fact that our astrometry is based on the usno - a2 catalog , while sloan takes tycho2 stars as reference . comparing the sloan and usno - a2 positions in the whole fls field , we have an offset of @xmath54 and @xmath55 arcsec which is in complete agreement with the offsets found in table [ tbl : astrometry ] . finally , we compared our r - band catalog to the gsc ii finding an offset of @xmath56 and @xmath57 arcsec . while not large , this offset should be noted when making comparison to data sets that used the gsc ii for reference . nevertheless , the remarkably small rms in both directions obtained comparing our and sloan positions suggests an intrinsic accuracy of @xmath58 arcsec for each catalog which is well within the requirements for slit / fiber positioning , an essential requirement for public surveys . the sloan survey also allows us to compare the photometric calibration . although the magnitude system used in this survey is different from ours , we can obtain a relationship between the r cousin magnitude and the sloan magnitudes using the 92 landolt stars which have been observed by the sloan group to calibrate their observations ( smith et al . 2002 ) . before making the comparison , we have converted the r and i magnitudes to the sdss 2.5 m natural system , using the equations : @xmath59 as explained at the sloan web site . then , we have obtained a relationship between r and the sloan colors with a least square fit : @xmath60 using a biweight estimator , the difference between the real r and the value estimated with the relationship ( [ eqt : landolt_sloan ] ) is on average of 0.0001 ( with an rms of 0.007 ) . in spite of the large scatter , the relationship is useful from a statistical point of view since we are interested only in confirming our magnitude zero - point . we have therefore compared the magnitudes of the stars in the fls fields with the values deduced from the sloan survey with the equation ( [ eqt : landolt_sloan ] ) . table [ tbl : astrometry ] summarizes the median differences in magnitude between our measurements ( _ auto - magnitudes _ ) and the _ model _ complete magnitude as computed by sloan for stars with magnitude @xmath23 in the different fls fields . in figure [ fig : magoff18 ] we show the distribution of the magnitude offsets in the case of the central field fls-18 , while in figure [ fig : magoff ] we illustrate the offsets for the various fields . our calibration agrees on average with the sloan one , since the average difference between our and sloan measurements is @xmath53 . the biggest differences are found in the external subfields where the relative zero is not well constrained due to the low number of stars in common between adjacent subfields ( see figure [ fig : magoff ] ) . counting galaxies and stars as a function of the magnitude allows one to evaluate the overall characteristics of a catalog as depth and homogeneity . in figures [ fig : starcounts ] and [ fig : galscounts ] we show star and galaxy counts in the fls region and compare them with analogous counts using the sloan digital sky survey ( stoughton et al . 2002 ) . to compare the two distributions , we have transformed the sdss magnitudes into the r vega magnitudes using the relationship [ eqt : landolt_sloan ] . in the case of star counts ( fig . [ fig : starcounts ] ) , the counts from our survey and the sdss agree very well between r=18 and r=22 . for magnitudes brighter than r=18 , most of the stars detected in our survey are saturated and do not appear in our catalogs . star counts drop very rapidly for magnitudes fainter than r=24 since the profile criterion used for the star / galaxy separation fails for faint objects . for comparison , we show in figure [ fig : starcounts ] the star counts in the elais field and those in the chandra field ( groenewegen et al . , 2002 ) . these fields , which lie at higher galactic latitudes are , as expected , less populated by stars . to evaluate the variation in the number counts due to the varying observing conditions , we have computed the counts for each of the subfield in the fls field . in figure [ fig : fieldscounts ] we show these counts as well as the median counts with error bars corresponding to the standard deviation as measured from the observed scatter in the counts of the different subfields . one can easily see that a few subfields ( # 4 , # 7 , # 8 , # 15 and # 26 ) are less deep than the other ones , as expected from the quantities measured in table [ tbl : summary ] . fortunately , these fields are external and have been only partially covered by spitzer observations . the other 20 fields in the fls are quite homogeneous . median counts are then reported in figure [ fig : galscounts ] to compare them with the results from other surveys . the dotted line corresponds the counts from sdss in the fls field ( stoughton et al . 2002 ) computed transforming the sdss magnitudes into the r vega magnitudes using the relationship in equation ( [ eqt : landolt_sloan ] ) . the points from the general sdss counts ( yasuda et al . , 2001 ) have been approximatively transformed using the relationship by fukugita et al . ( 1995 ) assuming that galaxies have @xmath61 colors typical of spirals at redshift of 0.2 0.5 . at the faint end of the counts , results from several deep surveys are reported . as an estimate of the completeness of our images we have compared our counts to the sdss counts in the fls field for r@xmath4221 and to median counts from the other deep surveys at magnitudes fainter than @xmath62 . in figure [ fig : completeness ] we show the completeness of the single subfields in the fls field ( thin lines ) and those of the global field and the sdss survey ( continuous and dashed thick lines , respectively ) . our survey is deeper than the sdss data by almost three magnitudes . it is 50% complete around r=24.5 . this estimate is conservative , since the number of spurious detection at r=24.5 is still relatively small ( less than 0.5% ) . finally , for magnitudes brighter than r=18 , the galaxy catalogs are slightly incomplete since a few extended objects are saturated . and that of median counts from several deep surveys at @xmath63 . dotted thin lines show the completeness of single fls subfields , while the dashed line shows the completeness of the sdss survey in the fls field . , a deep noao / kpno survey in the r band has been carried out to observe a field of more than 9 square degrees centered at 17:18:00 + 59:30:00 ( j2000 ) aimed to find optical counterparts for the _ first look survey _ which surveys 7 different infrared wavelengths with the instruments irac and mips using the spitzer space telescope . another 2.3 square degrees have been surveyed in the elais - n1 region which will be observed in the _ spitzer _ swire legacy survey . this paper describes the observation strategy , the data reduction and the products which are publicly available to the astronomical community on the world wide web at the spitzer science center and the noao science archives . the overall quality of the data is good and homogeneous : the average seeing is 1.1 and typically varies between 0.9 and 1.2 . the limiting magnitude of the images , measured inside an aperture of 1.35 @xmath9 fwhm for which the s / n ratio is maximal , is around r=25.5 at 5@xmath8 , deep enough to detect optical counterparts for a substantial fraction of the new _ spitzer _ selected objects . an average number of 35000 extragalactic sources are detected in each subfield ( 40 @xmath9 40 , approximatively ) with a 50% completeness limit of r=24.5 as deduced by comparing the counts with other deeper surveys . images and catalogs are available to the astronomical community along with the first release of the fls infrared data to exploit in the best way the wealth of extragalactic data expected from the new infrared observatory _ we are grateful to those that built , maintain , and operate the mosaic-1 camera and mayall 4 m telescope at kitt peak national observatory , a part of the national optical astronomy observatory , which is operated by the association of universities for research in astronomy , inc . ( aura ) , under a cooperative agreement with the national science foundation . we thank the former director of noao , dr . sidney wolff , for allocating the directors discretionary time used to gather the optical data we present in this paper , m. j. i. brown for assistance with software used to prepare the images for the data release , and t. lauer for useful discussions about photometric depth measurements . we wish to thank also the anonymous referee for his comments about the sloan photometric calibration . we acknowledge the support of the extragalactic first look survey team members at the spitzer science center ( b.t . soifer , p. appleton , l. armus , s. chapman , p. choi , f. fang , d. frayer , i. heinrichsen , g. helou , m. i m . , m. lacy , s. laine , f. marleau , d. shupe , g. squires j. surace , h. teplitz , g. wilson , l. yan , j. colbert , and i. drozdovsky ) jannuzi , b. t. , & dey , a. 1999 , in asp conf . ser . 191 , photometric redshifts and high redshift galaxies , ed . r. j. weymann , l. j. storrie - lombardi , m. sawicki , & r. j. brunner ( san francisco : asp ) , p. 111 soifer , b.t . , neugebauer , g. , beichman , c.a . , houck , j.r . , rowan - robinson , m. 1983 , in infrared technology ix ; proceedings of the ninth annual meeting , san diego , ca , august 23 - 25 , 1983 . bellingham , wa , spie - the international society for optical engineering , 297 cccc fls 2 & 17:17:48.12 & + 61:06:00.0 & 3 + fls 4 & 17:08:32.33 & + 60:32:00.0 & 3 + fls 5 & 17:13:13.17 & + 60:32:00.0 & 3 + fls 6 & 17:17:45.01 & + 60:32:00.0 & 3 + fls 7 & 17:22:25.85 & + 60:32:00.0 & 3 + fls 8 & 17:27:06.69 & + 60:32:00.0 & 3 + fls 9 & 17:05:44.94 & + 59:57:00.0 & 3 + fls 10 & 17:10:12.89 & + 59:57:00.0 & 5 + fls 11 & 17:14:48.65 & + 59:57:00.0 & 5 + fls 12 & 17:19:24.61 & + 59:57:00.0 & 4 + fls 13 & 17:23:59.47 & + 59:57:00.0 & 5 + fls 14 & 17:28:29.33 & + 59:57:00.0 & 3 + fls 15 & 17:05:56.30 & + 59:22:00.0 & 3 + fls 16 & 17:10:25.14 & + 59:22:00.0 & 3 + fls 17 & 17:14:54.99 & + 59:22:00.0 & 4 + fls 18 & 17:19:20.82 & + 59:22:00.0 & 4 + fls 19 & 17:23:53.67 & + 59:22:00.0 & 3 + fls 20 & 17:28:22.51 & + 59:22:00.0 & 3 + fls 22 & 17:10:18.66 & + 58:47:00.0 & 3 + fls 23 & 17:14:46.17 & + 58:47:00.0 & 3 + fls 24 & 17:19:12.00 & + 58:47:00.0 & 3 + fls 25 & 17:23:37.32 & + 58:47:00.0 & 3 + fls 26 & 17:28:02.83 & + 58:47:00.0 & 3 + fls 28 & 17:12:58.88 & + 58:12:30.0 & 3 + fls 29 & 17:17:19.44 & + 58:12:00.0 & 3 + fls 30 & 17:21:40.00 & + 58:12:30.0 & 3 + elais 2 & 16:10:01.00 & + 55:23:06.0 & 3 + elais 5 & 16:08:00.00 & + 54:48:06.0 & 5 + elais 6 & 16:12:06.76 & + 54:48:06.0 & 7 + elais 9 & 16:08:02.04 & + 54:13:06.2 & 9 + elais 10 & 16:12:06.76 & + 54:13:06.2 & 4 + ccccc 2000 may 4 & fls04 & 3 @xmath9 600 . & 1.4 - 1.5 & light cirri on sunset + & fls05 & 3 @xmath9 600 . & 1.4 - 1.45 & photometric + & fls06 & 3 @xmath9 600 . & 1.5 - 1.6 & + & fls07 & 3 @xmath9 600 . & 1.2 - 1.5 & + & fls08 & 3 @xmath9 600 . & 1.2 - 1.3 & + & fls10 & 5 @xmath9 600 . & 0.98 - 1.15 & + & fls11 & 3 @xmath9 600 . & 0.97 - 1.15 & + & fls12 & 4 @xmath9 600 . & 0.79 - 1.3 & + & fls17 & 4 @xmath9 600 . & 0.8 - 0.95 & + 2000 may 5 & elais05 & 3 @xmath9 600 . & 1.07 - 1.25 & photometric + & fls15 & 3 @xmath9 600 . & 1.05 - 1.12 & + & fls16 & 3 @xmath9 600 . & 0.98 - 1.04 & + & fls20 & 3 @xmath9 600 . & 0.88 - 0.9 & + & fls19 & 3 @xmath9 600 . & 0.85 - 0.87 & + & fls18 & 4 @xmath9 600 . & 0.77 - 0.88 & + & fls24 & 3 @xmath9 600 . & 0.87 - 0.95 & + & fls23 & 3 @xmath9 600 . & 0.9 - 0.94 & + & fls22 & 3 @xmath9 600 . & 0.81 - 0.86 & + & fls25 & 3 @xmath9 600 . & 0.85 - 0.9 & + & fls13 & 1 @xmath9 600 . & 0.9 & + 2000 may 6 & fls13 & 4 @xmath9 600 . & 1.05 - 1.25 & light cirri + & fls11 * & 2 @xmath9 600 . & 0.95 - 1.1 & non - photometric + & elais06 * & 4 @xmath9 600 . & 0.88 - 0.95 & + & elais09 * & 4 @xmath9 600 . & 0.86 - 0.95 & + & elais10 * & 4 @xmath9 600 . & 0.87 - 1.0 & + & fls14 * & 4 @xmath9 600 . & 0.84 - 0.85 & + 2000 may 7 & elais02 & 3 @xmath9 600 . & 1.09 - 1.2 & cirri + & fls26 & 3 @xmath9 600 . & 1.0 - 1.2 & non - photometric + & fls28 & 3 @xmath9 600 . & 0.96 - 1.0 & + & fls29 & 3 @xmath9 600 . & 0.95 - 1.2 & + & fls30 & 3 @xmath9 600 . & 1.02 - 1.07 & + & fls09 & 3 @xmath9 600 . & 1.05 - 1.1 & + & fls02 & 3 @xmath9 600 . & 0.94 - 0.98 & + & fls05 & 3 @xmath9 600 . & 0.9 - 1.06 & + & fls06 & 3 @xmath9 600 . & 0.9 - 0.91 & + & elais05 & 2 @xmath9 600 . & 0.9 - 1.1 & + & elais06 & 3 @xmath9 600 . & 0.9 - 1.1 & + & fls14 & 3 @xmath9 600 . & 0.86 - 0.92 & + 2000 may 9 & elais09 & 5 @xmath9 600 . & 1.0 - 1.2 & non - photometric + cccccccrcc fls_r_j171814.3 + 604643 & 17:18:14.303 & + 60:46:43.93 & 21.44 & 0.02 & 21.44 & 0.03 & 38.8 & 0.997 & 0.071 + fls_r_j171804.5 + 604644 & 17:18:04.511 & + 60:46:44.00 & 23.60 & 0.12 & 23.66 & 0.12 & 8.8 & 0.753 & 0.071 + fls_r_j171847.4 + 604644 & 17:18:47.496 & + 60:46:44.22 & 24.12 & 0.20 & 24.19 & 0.21 & 5.2 & 0.735 & 0.073 + fls_r_j171822.5 + 604644 & 17:18:22.536 & + 60:46:44.07 & 22.99 & 0.07 & 22.95 & 0.09 & 12.2 & 0.975 & 0.071 + fls_r_j171840.3 + 604644 & 17:18:40.368 & + 60:46:44.40 & 23.61 & 0.12 & 23.60 & 0.12 & 9.1 & 0.936 & 0.072 + fls_r_j171729.6 + 604645 & 17:17:29.639 & + 60:46:45.47 & 24.08 & 0.19 & 23.53 & 0.18 & 6.0 & 0.784 & 0.069 + fls_r_j171739.6 + 604645 & 17:17:39.672 & + 60:46:45.69 & 24.65 & 0.32 & 24.36 & 0.17 & 6.4 & 0.802 & 0.070 + fls_r_j171808.7 + 604643 & 17:18:08.784 & + 60:46:43.96 & 20.72 & 0.01 & 20.69 & 0.01 & 104.4 & 0.983 & 0.071 + fls_r_j171824.5 + 604644 & 17:18:24.575 & + 60:46:44.11 & 20.84 & 0.01 & 20.73 & 0.02 & 67.4 & 0.983 & 0.071 + fls_r_j171817.5 + 604644 & 17:18:17.567 & + 60:46:44.65 & 22.51 & 0.04 & 22.49 & 0.06 & 18.8 & 0.960 & 0.071 + fls_r_j171851.9 + 604644 & 17:18:51.984 & + 60:46:44.61 & 20.75 & 0.01 & 20.71 & 0.01 & 94.4 & 0.984 & 0.073 + fls_r_j171914.2 + 604645 & 17:19:14.232 & + 60:46:45.44 & 22.62 & 0.05 & 22.58 & 0.06 & 18.3 & 0.973 & 0.074 + fls_r_j171734.8 + 604646 & 17:17:34.872 & + 60:46:46.05 & 23.22 & 0.09 & 23.12 & 0.09 & 12.0 & 0.851 & 0.070 + fls_r_j171912.1 + 604647 & 17:19:12.191 & + 60:46:47.13 & 24.28 & 0.23 & 24.29 & 0.23 & 4.7 & 0.652 & 0.074 + fls_r_j171841.0 + 604646 & 17:18:41.087 & + 60:46:46.91 & 23.66 & 0.13 & 23.64 & 0.13 & 8.6 & 0.944 & 0.072 + fls_r_j171747.7 + 604648 & 17:17:47.712 & + 60:46:48.53 & 24.61 & 0.31 & 24.45 & 0.18 & 6.1 & 0.758 & 0.070 + fls_r_j171726.8 + 604646 & 17:17:26.807 & + 60:46:46.12 & 23.29 & 0.09 & 22.58 & 0.11 & 10.3 & 0.851 & 0.069 + fls_r_j171827.8 + 604646 & 17:18:27.887 & + 60:46:46.99 & 23.30 & 0.09 & 22.86 & 0.11 & 9.6 & 0.873 & 0.072 + fls_r_j171834.5 + 604648 & 17:18:34.560 & + 60:46:48.28 & 23.94 & 0.17 & 23.89 & 0.14 & 7.6 & 0.945 & 0.072 + fls_r_j171838.2 + 604644 & 17:18:38.279 & + 60:46:44.40 & 20.22 & 0.01 & 20.19 & 0.01 & 155.1 & 0.985 & 0.072 + fls_r_j171631.4 + 604647 & 17:16:31.487 & + 60:46:47.20 & 23.47 & 0.11 & 21.60 & 0.06 & 18.8 & 0.947 & 0.064 + fls_r_j171731.9 + 604646 & 17:17:31.967 & + 60:46:46.99 & 23.09 & 0.08 & 22.29 & 0.09 & 11.6 & 0.749 & 0.069 + fls_r_j171821.9 + 604645 & 17:18:21.911 & + 60:46:45.91 & 23.17 & 0.08 & 22.98 & 0.11 & 9.6 & 0.434 & 0.071 + fls_r_j171719.7 + 604649 & 17:17:19.704 & + 60:46:49.00 & 23.81 & 0.15 & 23.74 & 0.21 & 5.1 & 0.928 & 0.068 + fls_r_j171618.3 + 604647 & 17:16:18.312 & + 60:46:47.63 & 24.48 & 0.27 & 24.22 & 0.18 & 6.0 & 0.671 & 0.063 + fls_r_j171843.1 + 604648 & 17:18:43.128 & + 60:46:48.39 & 23.74 & 0.14 & 23.84 & 0.14 & 7.6 & 0.955 & 0.073 + fls_r_j171733.5 + 604649 & 17:17:33.503 & + 60:46:49.40 & 23.62 & 0.12 & 23.33 & 0.16 & 6.7 & 0.974 & 0.069 + fls_r_j171730.6 + 604649 & 17:17:30.671 & + 60:46:49.58 & 24.10 & 0.19 & 23.08 & 0.14 & 7.8 & 0.879 & 0.069 + fls_r_j171851.1 + 604645 & 17:18:51.143 & + 60:46:45.26 & 19.53 & 0.00 & 19.49 & 0.00 & 258.5 & 0.985 & 0.073 + fls_r_j171646.7 + 604647 & 17:16:46.751 & + 60:46:47.28 & 22.53 & 0.05 & 22.05 & 0.07 & 16.4 & 0.043 & 0.065 + cccccccc fls 2 & 17:17:45.70 & 61:05:44.12 & 1.08 & 25.59 & 26.15 & 34766 & 6787 + fls 4 & 17:08:27.23 & 60:32:10.74 & 1.57 & 24.72 & 25.28 & 18016 & 2962 + fls 5 & 17:13:10.74 & 60:31:44.01 & 1.06 & 25.64 & 26.20 & 36372 & 5012 + fls 6 & 17:17:42.13 & 60:31:42.33 & 1.02 & 25.78 & 26.34 & 36988 & 5050 + fls 7 & 17:22:21.27 & 60:32:06.70 & 1.54 & 24.86 & 25.42 & 19959 & 1316 + fls 8 & 17:27:02.18 & 60:32:07.10 & 1.36 & 25.09 & 25.64 & 22392 & 3049 + fls 9 & 17:05:42.63 & 59:56:46.44 & 1.14 & 25.58 & 26.13 & 34985 & 5254 + fls 10 & 17:10:07.30 & 59:57:07.22 & 1.24 & 25.63 & 26.19 & 32253 & 3309 + fls 11 - 17 & 17:14:47.10 & 59:39:33.96 & 1.04 & 25.86 & 26.41 & 72709 & 6865 + fls 12 & 17:19:19.96 & 59:57:03.33 & 1.00 & 25.86 & 26.42 & 35905 & 4352 + fls 13 & 17:23:54.12 & 59:57:09.50 & 1.16 & 25.41 & 25.96 & 30797 & 5496 + fls 14 & 17:28:26.67 & 59:56:39.57 & 0.98 & 25.51 & 26.06 & 34672 & 5114 + fls 15 & 17:05:50.01 & 59:22:06.83 & 1.22 & 25.14 & 25.69 & 25218 & 4450 + fls 16 & 17:10:19.12 & 59:22:07.62 & 1.17 & 25.25 & 25.80 & 26996 & 4313 + fls 18 & 17:19:14.84 & 59:22:07.54 & 0.92 & 25.91 & 26.46 & 37922 & 4228 + fls 19 & 17:23:48.01 & 59:22:07.27 & 0.97 & 25.64 & 26.19 & 34771 & 4409 + fls 20 & 17:28:17.04 & 59:22:08.13 & 1.00 & 25.59 & 26.15 & 35297 & 4278 + fls 22 & 17:10:12.41 & 58:47:07.63 & 0.94 & 25.77 & 26.33 & 35201 & 4259 + fls 23 & 17:14:40.18 & 58:47:07.50 & 1.06 & 25.67 & 26.22 & 34624 & 4313 + fls 24 & 17:19:06.04 & 58:47:04.74 & 1.03 & 25.71 & 26.26 & 31151 & 4277 + fls 25 & 17:23:30.85 & 58:47:10.81 & 0.99 & 25.76 & 26.31 & 33847 & 4682 + fls 26 & 17:28:00.70 & 58:46:43.81 & 1.22 & 25.21 & 25.77 & 22636 & 5538 + fls 28 & 17:12:56.31 & 58:12:15.54 & 1.12 & 25.37 & 25.93 & 30609 & 5830 + fls 29 & 17:17:16.96 & 58:11:44.23 & 1.12 & 25.52 & 26.07 & 34524 & 6264 + fls 30 & 17:21:37.91 & 58:12:14.09 & 1.20 & 25.49 & 26.05 & 38493&10099 + elais 2 & 16:09:58.00 & 55:22:51.47 & 1.20 & 25.18 & 25.74 & 26960 & 3802 + elais 5 & 16:07:55.71 & 54:47:58.39 & 1.22 & 25.59 & 26.14 & 33876 & 3240 + elais 6 & 16:11:59.38 & 54:47:43.34 & 1.11 & 25.49 & 26.05 & 34736 & 3406 + elais 6s & 16:11:58.89 & 54:23:29.02 & 0.98 & 25.82 & 26.38 & 41506 & 3160 + elais 9 & 16:07:59.66 & 54:12:59.29 & 1.11 & 25.77 & 26.32 & 38231 & 2926 + elais 9s & 16:07:58.67 & 53:48:27.43 & 1.01 & 25.78 & 26.33 & 38167 & 3538 + elais 10s & 16:11:56.45 & 53:48:23.99 & 1.02 & 25.72 & 26.28 & 36497 & 3496 + fls 2 & 2004&[email protected] & [email protected] & [email protected] & 23 & [email protected] & [email protected] + fls 4 & 1523 & [email protected] & [email protected]&[email protected] & & & + fls 5 & 1523&[email protected] & [email protected] & [email protected] & 84 & [email protected] & [email protected] + fls 6 & 1967&[email protected] & [email protected] & [email protected]&117 & [email protected] & [email protected] + fls 7 & 1501 & [email protected] & [email protected] & [email protected] & 76 & [email protected] & [email protected] + fls 8 & 1571 & [email protected] & [email protected] & [email protected] & & & + fls 9 & 1319&[email protected] & [email protected] & [email protected] & & & + fls 10 & 1509 & [email protected] & [email protected] & [email protected]&125 & [email protected] & [email protected] + fls 11 - 17&2810&[email protected] & [email protected] & [email protected]&269 & [email protected] & [email protected] + fls 12 & 2132&[email protected] & [email protected] & [email protected]&121 & [email protected] & [email protected] + fls 13 & 1412 & [email protected] & [email protected] & [email protected]&128 & [email protected] & [email protected] + fls 14 & 1590&[email protected] & [email protected] & [email protected] & 29 & [email protected] & [email protected] + fls 15 & 1335 & [email protected] & [email protected] & [email protected] & & & + fls 16 & 1379 & [email protected] & [email protected] & [email protected]&106 & [email protected] & [email protected] + fls 18 & 2294&[email protected] & [email protected] & [email protected]&144 & [email protected] & [email protected] + fls 19 & 1614&[email protected] & [email protected] & [email protected]&142 & [email protected] & [email protected] + fls 20 & 1608&[email protected] & [email protected] & [email protected] & 58 & [email protected] & [email protected] + fls 22 & 1461&[email protected] & [email protected]&[email protected] & 61 & [email protected]&[email protected] + fls 23 & 1510 & [email protected] & [email protected] & [email protected]&121 & [email protected] & [email protected] + fls 24 & 2044&[email protected] & [email protected] & [email protected]&143 & [email protected] & [email protected] + fls 25 & 2022&[email protected] & [email protected] & [email protected]&117 & [email protected] & [email protected] + fls 26 & 1539&[email protected] & [email protected] & [email protected] & 34 & [email protected] & [email protected] + fls 28 & 1273&[email protected]&[email protected] & [email protected] & & & + fls 29 & 1769&[email protected]&[email protected] & [email protected] & 57 & [email protected] & [email protected] + fls 30 & 3198&[email protected] & [email protected] & [email protected] & 23 & [email protected] & [email protected] + average & & [email protected] & [email protected] & [email protected] & & [email protected] & [email protected] +

for any regular courant algebroid , we construct a characteristic class _ la _ chern - weil . this intrinsic invariant of the courant algebroid is a degree- class in its naive cohomology . when the courant algebroid is exact , it reduces to the evera class ( in ) . on the other hand , when the courant algebroid is a quadratic lie algebra , it coincides with the class of the cartan-form ( in ) . we also give a complete classification of regular courant algebroids and discuss its relation to the characteristic class .

0909.0319

courant algebroids were introduced in @xcite as a way to merge the concept of lie bialgebra and the bracket on @xmath4 first discovered by courant @xcite here @xmath5 is a smooth manifold . roytenberg gave an equivalent definition phrased in terms of the dorfman bracket @xcite , which highlighted the relation of courant algebroids to @xmath6-algebras @xcite observed by roytenberg & weinstein @xcite . despite its importance , the subject suffered from the lack of examples for a long time . one important class of courant algebroids , called exact courant algebroids , was discovered by evera @xcite . a courant algebroid @xmath7 is said to be exact if its underlying vector bundle fits into an exact sequence @xmath8 , where @xmath9 is the anchor map . evera proved that the isomorphism classes of exact courant algebroids are classified by a degree-@xmath0 class in the de rham cohomology of @xmath5 : the evera class . furthermore , the structure of transitive courant algebroids courant algebroids with surjective anchors was described independently by vaisman @xcite , evera ( in a private correspondence with weinstein @xcite ) and bressler @xcite . evera and bressler also classified transitive courant algebroids as extensions of transitive lie algebroids @xcite . indeed evera also outlined some nice ideas of classification of transitive courant algebroids in @xcite . in @xcite , the second and third authors introduced the modular class of courant algebroids , a degree 1 characteristic class in the naive cohomology @xmath10 of the courant algebroid @xmath7 . it is natural to ask whether there is a degree 3 characteristic class for a general courant algebroid resembling the severa class , and if so , what role is played by such a class in the classification of courant algebroids . the main purpose of this paper is to answer these questions for regular courant algebroids , that is , courant algebroids with a constant rank anchor . such courant algebroids @xmath7 are particularly easier to handle , for their associated characteristic distributions @xmath11 do not have any singularity . note that characteristic classes of lie algebroids were studied by evens - lu - weinstein @xcite , crainic and fernandes @xcite . it would be interesting to explore if there is any intrinsic connection between these constructions . to any regular courant algebroid , we will construct a degree 3 class , called the characteristic class , in the naive cohomology @xmath12 of the courant algebroid , an analogue of evera s class . note that when @xmath7 is an exact courant algebroid , @xmath12 is isomorphic to the de rham cohomology @xmath1 . however , unlike the exact case , @xmath12 does not classify regular courant algebroids . the classification problem is much subtler in this case . given a regular courant algebroid @xmath7 with anchor @xmath9 , @xmath13 is clearly a regular lie algebroid , i.e. a lie algebroid with a constant rank anchor . it is called the ample lie algebroid of @xmath7 . the kernel @xmath14 of the anchor of @xmath15 is a bundle of quadratic lie algebras , which satisfies certain compatibility conditions . therefore , the lie algebroid @xmath15 is quadratic ( see definition [ def : quad ] ) . the first natural question is whether every quadratic lie algebroid @xmath15 arises in this way . it turns out that there is an obstruction : the first pontryagin class , an element in @xmath16 naturally associated to any quadratic lie algebroid . here @xmath17 is the image of the anchor ( an integrable subbundle of @xmath18 ) and @xmath16 stands for the leafwise de rham cohomology of @xmath17 . this obstruction is similar to the one described by evera and bressler in the transitive case @xcite . to recover a courant algebroid from a given a quadratic lie algebroid @xmath15 with vanishing first pontryagin class , one needs an extra piece of data : a coherent form @xmath19 . this is a closed @xmath0-form on the lie algebroid @xmath15 satisfying certain compatibility conditions . in this case @xmath20 is called a characteristic pair . an equivalence is introduced on characteristic pairs . let @xmath21 denote the cohomology groups of the subcomplex @xmath22 of @xmath23 , where @xmath24 is made of the sections of @xmath25 which are annihilated by all sections of @xmath26 . roughly speaking , two coherent forms @xmath27 and @xmath28 on @xmath15 are equivalent if and only if the class of @xmath29 in @xmath30 is zero . we prove that there is a one - to - one correspondence between regular courant algebroids up to isomorphisms and equivalence classes of characteristic pairs . the inclusion @xmath31 is a lie algebroid morphism . therefore , it induces a morphism @xmath32 . we denote the class of the well - known cartan @xmath0-form of lie theory in @xmath33 by @xmath34 . the main result of the paper can be summarized as the following 1 . there is a natural map from regular courant algebroids to quadratic lie algebroids with vanishing first pontryagin class . 2 . for any quadratic lie algebroid @xmath15 with vanishing first pontryagin class , the isomorphism classes of courant algebroids whose ample lie algebroids are isomorphic to @xmath15 are parametrized by @xmath35 . here @xmath36 is a certain abelian subgroup of @xmath37 and @xmath38 denotes the anchor of @xmath15 . 3 . for any regular courant algebroid @xmath7 , there is a degree 3 characteristic class , which is a cohomology class in @xmath39 . here @xmath15 is the ample lie algebroid of @xmath7 , and @xmath40 is the cartan @xmath0-form . hence , we have the `` exact sequence '' @xmath41 @xmath42 . to this day , little is known about courant algebroid cohomology @xcite : roytenberg computed it for @xmath43 and ginot & grtzmann for transitive and some very special regular courant algebroids @xcite . our result should be useful for computing the courant algebroid cohomology of arbitrary regular courant algebroids . * acknowledgments * the authors thank camille laurent - gengoux , zhang - ju liu , jim stasheff and alan weinstein for useful discussions and comments . in particular , we are indebted to pavol evera , who provides example [ rmk : severa ] , and called their attention to @xcite , which led to the spotting of an error in an early version of the paper . a courant algebroid consists of a vector bundle @xmath44 , a fiberwise nondegenerate pseudo - metric @xmath45 , a bundle map @xmath46 called anchor and a @xmath47-bilinear operation @xmath48 on @xmath49 called dorfman bracket , which , for all @xmath50 and @xmath51 satisfy the following relations : @xmath52 } ; \\ \label{algian : caaxioms3 } & { e_1\circ ( fe_2)}=\big(\rho(e_1)f\big)e_2+f({e_1\circ e_2 } ) ; \\ \label{algian : caaxioms4 } & { e_1\circ e_2}+{e_2\circ e_1}=2{\mathcal{d}}{(e_1,e_2 ) } ; \\ \label{algian : caaxioms5 } & { { \mathcal{d}}f\circ e_1}=0 ; \\ \label{algian : caaxioms6 } & \rho(e_1){(e_2,e_3)}={({e_1\circ e_2},e_3)}+{(e_2,{e_1\circ e_3})},\end{aligned}\ ] ] where @xmath53 is the @xmath47-linear map defined by @xmath54 the symmetric part of the dorfman bracket is given by eq . . the courant bracket is defined as the skew - symmetric part @xmath55 of the dorfman bracket . thus we have the relation @xmath56 . using the identification @xmath57 induced by the pseudo - metric @xmath45 : @xmath58 we can rewrite eq . as @xmath59 it is easy to see that @xmath60 , the subbundle of @xmath7 orthogonal to @xmath61 w.r.t . the pseudo - metric , coincides with @xmath62 , the subbundle of @xmath7 generated by the image of @xmath53 . from eq . , it follows that @xmath63 for any @xmath50 . therefore , the kernel of the anchor is coisotropic : @xmath64 the spaces of sections of @xmath61 and @xmath60 are two - sided ideals of @xmath49 w.r.t . the dorfman bracket . a courant algebroid @xmath7 is said to be * regular * if @xmath65 has constant rank , in which case @xmath17 is an integrable distribution on the base manifold @xmath5 . moreover , if @xmath7 is regular , then @xmath61 and @xmath60 are smooth ( constant rank ) subbundles of @xmath7 and the quotients @xmath66 and @xmath67 are lie algebroids . obviously , @xmath66 and @xmath17 are canonically isomorphic . we call @xmath67 the * ample * lie algebroid associated to @xmath7 . it will be denoted by the symbol @xmath68 . the inclusions @xmath69 yield four exact sequences : @xmath70 \ar@(ur , ul)[rr ] & & ( \ker\rho){^{\perp}}\ar[dr ] \ar@(ur , ul)[rr ] & & e \ar[dr]^q \ar@(ur , ul)[rr]^{\rho } & & \frac{e}{\ker\rho } \ar[dr ] \ar@(ur , ul)[rr ] & & 0 \\ & 0 \ar[ur ] \ar[dr ] & & \ker\rho \ar[ur ] \ar[dr]_{\pi } & & \frac{e}{(\ker\rho){^{\perp } } } \ar[ur ] \ar[dr ] & & 0 \ar[ur ] \ar[dr ] & \\ 0 \ar[ur ] \ar@(dr , dl)[rr ] & & 0 \ar[ur ] \ar@(dr , dl)[rr ] & & { \mathcal{g}}\ar[ur ] \ar@(dr , dl)[rr ] & & 0 \ar[ur ] \ar@(dr , dl)[rr ] & & 0 } \ ] ] here @xmath71 . we use the symbol @xmath72 ( resp . @xmath73 ) to denote the projection @xmath74 ( resp . @xmath75 ) . a @xmath76-bilinear and skew - symmetric bracket on @xmath77 determined by the dorfman bracket of @xmath7 through the relation @xmath78}^{{\mathcal{g}}}=\pi({r\circ s } ) , \qquad\forall r , s\in{\gamma(\ker\rho ) } , \ ] ] turns @xmath14 into a bundle of lie algebras . moreover , the map @xmath79 is a well - defined nondegenerate symmetric and ad - invariant pseudo - metric on @xmath14 , which we will denote by the symbol @xmath80 . hence @xmath14 is a bundle of quadratic lie algebras . note that @xmath14 is also a module over @xmath68 ; the representation is given by @xmath81 } , \qquad \forall\ ; x\in{\gamma({\mathcal{a}}_e)},\ ; r\in{\gamma(\ker\rho ) } .\ ] ] this representation is compatible with the pseudo - metric on @xmath14 : @xmath82 let @xmath7 be a regular courant algebroid with characteristic distribution @xmath17 and bundle of quadratic lie algebras @xmath14 . a * dissection * of @xmath7 is an isomorphism of vector bundles @xmath83 such that @xmath84 for all @xmath85 , @xmath86 and @xmath87 . such an isomorphism transports the courant algebroid structure of @xmath7 to @xmath88 . each dissection of @xmath7 induces three canonical maps : 1 . @xmath89 : @xmath90 2 . @xmath91 : @xmath92 3 . @xmath93 : @xmath94 [ prop : courantbracketintermsofdata ] 1 . the map @xmath95 satisfies @xmath96 for all @xmath97 , @xmath98 and @xmath50 . the map @xmath99 is skew - symmetric and @xmath76-bilinear . it can thus be regarded as a bundle map @xmath100 . 3 . the map @xmath101 is skew - symmetric and @xmath76-bilinear . it can thus be regarded as a section of @xmath102 . the following lemma shows that dissections always exist . [ lem : existenceofsplitting ] let @xmath103 be a regular courant algebroid . and set @xmath104 . 1 . there exists a splitting @xmath105 of the short exact sequence @xmath106 whose image @xmath107 is isotropic in @xmath7 . given such a splitting @xmath108 , there exists a unique splitting @xmath109 of the short exact sequence @xmath110 with image @xmath111 orthogonal to @xmath107 in @xmath7 . 3 . given a pair of splittings @xmath108 and @xmath112 as above , the map @xmath113 defined by @xmath114 is a dissection of @xmath7 . \(a ) take any section @xmath115 of @xmath116 . consider the bundle map @xmath117 defined by @xmath118 where @xmath87 . then @xmath119 is a section of @xmath116 such that @xmath120 ( b ) take any section @xmath121 of @xmath122 . consider the bundle map @xmath123 defined by @xmath124 where @xmath125 and @xmath126 . then @xmath127 is the only section of @xmath122 such that @xmath128 ( c ) this is obvious . let us first recall the notion of @xmath7-connections @xcite . vaisman s metric connections are a related notion @xcite . an @xmath7-connection ( or @xmath7-covariant derivative ) on @xmath7 is an @xmath47-bilinear map @xmath129 such that @xmath130 for all @xmath50 and @xmath131 . the next lemma shows that each regular courant algebroid @xmath7 admits @xmath7-connections . if @xmath7 is a regular courant algebroid , there always exists a torsion - free connection @xmath132 on the integrable distribution @xmath133 . indeed , it suffices to consider the restriction to @xmath17 of the levi - civit connection of some riemanian metric on @xmath5 . in the sequel , the symbol @xmath132 will be used to denote a chosen torsion - free connection on @xmath17 and the dual connection on @xmath134 as well . 1 . given a dissection @xmath135 of a regular courant algebroid @xmath7 ( identifying @xmath7 with @xmath88 ) and the data @xmath95 , @xmath99 and @xmath101 it induces , each torsion - free connection @xmath132 on @xmath17 determines an @xmath7-connection @xmath136 on @xmath7 through the defining relation @xmath137^\mathcal{g } } ) + { \nabla^f}_{x}{y},\ ] ] where @xmath87 , @xmath138 and @xmath86 . this covariant derivative @xmath136 preserves the pseudo - metric : @xmath139 3 . moreover , we have @xmath140 } .\ ] ] we omit the proof as it is straightforward . let @xmath7 be a courant algebroid . the nondegenerate pseudo - metric on @xmath7 induces a nondegenerate pseudo - metric on @xmath141 : @xmath142 the isomorphism of vector bundles @xmath143 coming from this nondegenerate pseudo - metric will be denoted by the same symbol @xmath144 . in the sequel , @xmath141 and @xmath145 are sometimes identified with each other when it is clear from the context . the sections of @xmath146 are called naive @xmath147-cochains . the operator @xmath148 defined by the relation @xmath149 where @xmath150 and @xmath151 , makes @xmath152 a cochain complex . its cohomology @xmath153 is called the * naive cohomology * of the courant algebroid @xmath7 @xcite . [ rmk : naiveidformsonfat ] if @xmath7 is regular , then @xmath154 is a regular lie algebroid . and it is easy to see that the cochain complexes @xmath152 and @xmath155 are isomorphic . indeed , we have @xmath156 and @xmath157 hence the naive cohomology of @xmath7 is isomorphic to the lie algebroid cohomology of @xmath158 . the map @xmath159 defined by @xmath160 where @xmath51 , is _ not _ a @xmath0-form on @xmath7 as it is not @xmath76-linear . however , it can be modified using an @xmath7-connection so that the result is @xmath76-linear . if @xmath136 is a covariant derivative on a courant algebroid @xmath7 , then @xmath161 where @xmath162 , defines a @xmath0-form on @xmath7 . the following is an analogue of the chern - weil construction . [ thm : main1 ] let @xmath7 be a regular courant algebroid . 1 . if @xmath136 is the @xmath7-connection on @xmath7 given by , then the @xmath0-form @xmath163 does not depend on the chosen torsion - free connection @xmath132 . it only depends on the chosen dissection @xmath135 . it will henceforth be denoted by the symbol @xmath164 . 2 . the @xmath0-form @xmath164 is a naive @xmath0-cocycle . 3 . the cohomology class of @xmath164 does not depend on the chosen dissection @xmath135 . the class of @xmath164 is called the * characteristic class * of the courant algebroid @xmath7 . the proof of this theorem is deferred to section [ sec : proofs ] . we fix a regular lie algebroid @xmath165_{{\mathcal{a}}}},{a})$ ] over the base manifold @xmath5 and with cohomology differential operator @xmath166 . we have the short exact sequence of vector bundles @xmath167 where @xmath168 and @xmath169 . it is clear that @xmath14 is a bundle of lie algebras , called the lie algebra bundle of @xmath15 . the fiberwise bracket is denoted by @xmath170^\mathcal{g}}$ ] . [ def : quad ] 1 . a regular lie algebroid @xmath15 is said to be a * quadratic * lie algebroid if the kernel @xmath14 of its anchor @xmath171 is equipped with a fiberwise nondegenerate @xmath172-invariant symmetric bilinear form @xmath80 satisfying : @xmath173_{{\mathcal{a}}}},{\boldsymbol{s}})^\mathcal{g}}+{({\boldsymbol{r}},{[x,{\boldsymbol{s}}]_{{\mathcal{a}}}})^\mathcal{g } } , \qquad \forall x\in { \gamma({\mathcal{a } } ) } , { \boldsymbol{r}},{\boldsymbol{s}}\in{\gamma({\mathcal{g } } ) } .\ ] ] 2 . two regular quadratic lie algebroids @xmath174 and @xmath175 are said to be isomorphic if there is a lie algebroid isomorphism @xmath176 , whose restriction @xmath177 to the kernels of the anchors is an isomorphism of quadratic lie algebras . for example , the ample lie algebroid @xmath154 of any regular courant algebroid @xmath103 is a quadratic lie algebroid . now a natural question arises as to whether every quadratic lie algebroid can be realized as the ample lie algebroid of a courant algebroid . it turns out that there is an obstruction . to see this , let us introduce some notations . a * hoist * of the lie algebroid @xmath15 is a section @xmath178 of the anchor @xmath171 . the bundle map @xmath179 given by @xmath180_{{\mathcal{a}}}}-{\kappa}([x , y ] ) , \qquad\forall x , y\in{\gamma(f)}\ ] ] is its * curvature*. given a quadratic lie algebroid endowed with a hoist , we can consider the @xmath181-form @xmath182 given by @xmath183 for all @xmath184 . the @xmath181-form @xmath185 is closed and its cohomology class in @xmath16 does not depend on the choice of @xmath186 . following @xcite , this cohomology class is called the * first pontryagin class * of the quadratic lie algebroid @xmath15 . [ thm : bresslersthm ] a regular quadratic lie algebroid is isomorphic to the ample lie algebroid associated to some regular courant algebroid if and only if its first pontryagin class vanishes . the proof of this theorem is deferred to section [ sec : standard3form ] . [ defn : good ] 1 . let @xmath15 be a regular quadratic lie algebroid and let @xmath186 be a hoist of @xmath15 . a closed @xmath0-form @xmath187 is said to be * @xmath186-coherent * if @xmath188^\mathcal{g}},{\boldsymbol{t}})^\mathcal{g } } ; \\ \label{align : goodformscdt2 } c({\boldsymbol{r}},{\boldsymbol{s}},{\kappa}(x))=0 ; \\ \label{align : goodformscdt3 } c({\boldsymbol{r}},{\kappa}(x),{\kappa}(y))={({\boldsymbol{r}},{r}^{{\kappa}}(x , y))^\mathcal{g}},\end{gathered}\ ] ] for all @xmath189 and @xmath190 . we also say that @xmath191 is a coherent pair . a closed @xmath0-form @xmath187 on a regular quadratic lie algebroid is called * coherent * if there exists a hoist @xmath186 such that @xmath19 is @xmath186-coherent . a * characteristic pair * is a couple @xmath20 made of a regular quadratic lie algebroid @xmath15 and a coherent @xmath0-form @xmath19 on it . if @xmath192 is a quadratic lie algebra , then the only coherent @xmath0-form is the cartan @xmath0-form defined by eq . . if @xmath193 is the product lie algebroid with the obvious quadratic lie algebroid structure and @xmath101 is a closed @xmath0-form on @xmath5 , then the @xmath0-form @xmath194 defined by @xmath195^\mathcal{g}},{\boldsymbol{t}})^\mathcal{g}}+{\mathcal{h}}(x , y , z ) , \ ] ] for all @xmath196 and @xmath197 , is coherent . [ prop : pontryagincoherentextensionallequivalent ] let @xmath15 be a regular quadratic lie algebroid . the following statements are equivalent . 1 . the first pontryagin class of @xmath15 vanishes . 2 . there exists a coherent @xmath0-form on @xmath15 . there exists a courant algebroid whose ample lie algebroid is @xmath15 . consider the subcomplex @xmath22 of @xmath23 , where @xmath24 is made of the sections of @xmath25 which are annihilated by all sections of @xmath26 . its cohomology groups are denoted @xmath21 . given two coherent @xmath0-forms @xmath27 and @xmath28 on @xmath15 , we have @xmath198 . two characteristic pairs @xmath199 and @xmath200 are said to be * equivalent * if there exists an isomorphism of quadratic lie algebroids @xmath201 such that @xmath202=0\in h^3_{\leftrightarrow}({\mathcal{a}}_1)$ ] . the equivalence class of a characteristic pair @xmath20 will be denoted @xmath203}$ ] . our main result is the following : [ thm : onetoone2 ] 1 . there is a one - to - one correspondence between regular courant algebroids up to isomorphisms and equivalence classes of characteristic pairs . 2 . if @xmath7 is a courant algebroid corresponding to the characteristic pair @xmath204 , then the characteristic class of @xmath7 is equal to @xmath205\in h^3({\mathcal{a}})$ ] . the proof is postponed to section [ sec : proofs ] . let @xmath17 be an integrable subbundle of @xmath18 and @xmath14 be a bundle of quadratic lie algebras over @xmath5 . in this section , we are only interested in those courant algebroid structures on @xmath207 whose anchor map is @xmath208 whose pseudo - metric is @xmath209 and whose dorfman bracket satisfies @xmath210^\mathcal{g } } , \ ] ] where @xmath211 , @xmath212 , and @xmath213 . we call them * standard * courant algebroid structures on @xmath207 . given such a standard courant algebroid structure on @xmath207 , we define @xmath95 , @xmath99 , and @xmath101 as in eqs . - . the following lemma shows that the dorfman bracket on @xmath214 can be recovered from @xmath95 , @xmath99 , and @xmath101 . [ prop : courantbracketintermsofdata2 ] let @xmath215 and @xmath216 be the maps defined by @xmath217 and @xmath218 then we have @xmath219 } , \\ \label{fsgfg1g2 } & { { \boldsymbol{r}}_1\circ { \boldsymbol{r}}_2}={\mathcal{p}}({\boldsymbol{r}}_1,{\boldsymbol{r}}_2)+{[{\boldsymbol{r}}_1,{\boldsymbol{r}}_2]^\mathcal{g } } , \\ \label{fsgfxi1g2 } & { \xi_1\circ { \boldsymbol{r}}_2}={{\boldsymbol{r}}_1\circ \xi_2}={\xi_1\circ \xi_2}=0 , \\ \label{fsgfx1xi2 } & { { x}_1\circ \xi_2}={l}_{{x}_1}\xi_2 , \\ \label{fsgfxi1x2 } & { \xi_1\circ { x}_2}=-{l}_{{x}_2}\xi_1 + { d^f}{\langle \xi_1| { x}_2\rangle } , \\ \label{fsgfx1g2 } & { { x}_1\circ { \boldsymbol{r}}_2}=-{{\boldsymbol{r}}_2\circ { x}_1}=-2\mathcal{q}({x}_1,{\boldsymbol{r}}_2)+ { \nabla}_{{x}_1}{\boldsymbol{r}}_2,\end{aligned}\ ] ] for all @xmath220 , @xmath221 , @xmath222 . here @xmath223 denotes the leafwise de rham differential . eqs . and are quite obvious . to get eq . , we observe that , for all @xmath224 , @xmath225 eq . can be proved similarly . to prove eq . , we only need to show that @xmath226 . in fact , for all @xmath98 , @xmath227 eq . follows directly from eq . . [ prop : courantbracketintermsofdata3 ] the following identities hold : @xmath228^\mathcal{g}}={[{\nabla}_x{\boldsymbol{r}},{\boldsymbol{s}}]^\mathcal{g}}+{[{\boldsymbol{r}},{\nabla}_x{\boldsymbol{s}}]^\mathcal{g } } , \\ \label{dcurvautre0 } \big({\nabla}_x{r}(y , z)-{r}({[x , y]},z)\big)+{\text{c.p.}}=0 , \\ \label{curvatureconnectionder } { \nabla}_x{\nabla}_y{\boldsymbol{r}}-{\nabla}_y{\nabla}_x{\boldsymbol{r}}-{\nabla}_{{[x , y]}}{\boldsymbol{r}}={[{r}(x , y),{\boldsymbol{r}}]^\mathcal{g}},\end{gathered}\ ] ] for all @xmath229 and @xmath86 . moreover , we have @xmath230 where @xmath231 denotes the @xmath181-form on @xmath17 given by @xmath232 where @xmath233 . applying the expressions found for the dorfman bracket in lemma [ prop : courantbracketintermsofdata2 ] , eq . follows from @xmath234 eq . from @xmath235 eqs . and from @xmath236 and eq . from @xmath237 in summary , we have proved the following [ thm : standardcourantstructure ] a courant algebroid structure on @xmath207 , with pseudo - metric and anchor map , and satisfying eq . , is completely determined by an @xmath17-connection @xmath95 on @xmath14 , a bundle map @xmath238 , and a @xmath0-form @xmath239 satisfying the compatibility conditions - . the dorfman bracket on @xmath207 is then given by eqs . - . this courant algebroid structure will be called the standard courant algebroid determined by the quintuple @xmath240 . we assume again that @xmath17 is an integrable subbundle of @xmath18 and @xmath14 is a bundle of quadratic lie algebras over @xmath5 . [ prop : standardquadraticliealgebroidstructure ] let @xmath15 be a quadratic lie algebroid with anchor @xmath241 such that @xmath242 and @xmath243 is isomorphic , as a bundle of quadratic lie algebras , to @xmath14 . then any hoist of @xmath15 determines an isomorphism @xmath244 . the transported lie bracket on @xmath245 is completely determined by an @xmath17-connection @xmath95 on @xmath14 and a bundle map @xmath238 satisfying the conditions - , through the following relations : @xmath246_{{\mathcal{a}}}}={[{\boldsymbol{r}},{\boldsymbol{s}}]^\mathcal{g } } , \qquad { [ x , y]_{{\mathcal{a}}}}={r}({x},{y})+{[x , y ] } , \qquad { [ x,{\boldsymbol{r}}]_{{\mathcal{a}}}}={\nabla}_x{\boldsymbol{r}},\ ] ] for all @xmath247 , @xmath248 , @xmath249 , @xmath250 . this quadratic lie algebroid @xmath251 will be called the standard quadratic lie algebroid determined by the quadruple @xmath252 . although it is similar to the proof of corollary 3.2 . in @xcite , we now sketch a proof of theorem [ thm : bresslersthm ] for completeness . given a regular courant algebroid @xmath7 , choose a dissection to identify it to @xmath207 . then , according to eq . , the pontryagin 4-cocycle of the ample lie algebroid @xmath158 associated to @xmath7 is a coboundary . conversely , given a quadratic lie algebroid @xmath15 with anchor @xmath253 such that @xmath254 and @xmath243 is isomorphic to @xmath14 as a bundle of quadratic lie algebras , assume there is a hoist @xmath186 of the lie algebroid @xmath15 such that the pontryagin @xmath181-cocycle @xmath185 is a coboundary , say @xmath255 for some @xmath239 . we define an @xmath17-connection @xmath256 on @xmath14 by setting @xmath257_{{\mathcal{a } } } } , \qquad\forall\;x\in{\gamma(f)},\;{\boldsymbol{r}}\in{\gamma({\mathcal{g}})}.\ ] ] the quintuple @xmath258 satisfies all conditions in theorem [ thm : standardcourantstructure ] and it is clear that @xmath15 is the ample lie algebroid associated to the courant algebroid determined by this quintuple . let @xmath251 denote the standard quadratic lie algebroid determined by the quadruple @xmath260 . given any @xmath239 , we define a @xmath0-form @xmath261 by @xmath262^\mathcal{g}},{\boldsymbol{t}})^\mathcal{g } } + { ( { r}({x},{y}),{\boldsymbol{t}})^\mathcal{g}}+{({r}({y},{z}),{\boldsymbol{r}})^\mathcal{g}}+{({r}({z},{x}),{\boldsymbol{s}})^\mathcal{g}},\ ] ] for all @xmath196 and @xmath263 . [ lem : csclosediff ] the following statements are equivalent : 1 . @xmath101 satisfies eq . ; @xmath264 is a closed @xmath0-form ; 3 . @xmath264 is a coherent @xmath0-form . straightforward calculations lead to @xmath265},z)\big ) + { \text{c.p.}},{\boldsymbol{r}})^\mathcal{g } } , \\ \label{align:3 } ( { \check{d}}c^s)(x , y,{\boldsymbol{r}},{\boldsymbol{s } } ) = & 0 , \\ \label{align:4 } ( { \check{d}}c^s)(x,{\boldsymbol{r}},{\boldsymbol{s}},{\boldsymbol{t } } ) = & -{l}_x{({[{\boldsymbol{r}},{\boldsymbol{s}}]^\mathcal{g}},{\boldsymbol{t}})^\mathcal{g } } + { ( { [ { \nabla}_x{\boldsymbol{r}},{\boldsymbol{s}}]^\mathcal{g } } + { [ { \boldsymbol{r}},{\nabla}_x{\boldsymbol{s}}]^\mathcal{g}},{\boldsymbol{t}})^\mathcal{g } } + { ( { [ { \boldsymbol{r}},{\boldsymbol{s}}]^\mathcal{g}},{\nabla}_x{\boldsymbol{t}})^\mathcal{g } } , \\ \label{align:5 } ( { \check{d}}c^s)({\boldsymbol{q}},{\boldsymbol{r}},{\boldsymbol{s}},{\boldsymbol{t } } ) = & 0,\end{aligned}\ ] ] for all @xmath266 and @xmath267 . the right hand side of eqs . and vanish due to eqs . - . thus @xmath268 if and only if the right hand side of eq . the latter is equivalent to eq . . this proves the equivalence between the first two statements . the equivalence between the last two statements follows from definition [ defn : good ] and eq . . let @xmath259 be the standard courant algebroid determined by the quintuple @xmath240 . clearly its associated ample quadratic lie algebroid is the standard quadratic lie algebroid @xmath251 determined by the quadruple @xmath252 . it follows from proposition [ lem : csclosediff ] and the compatibility conditions satisfied by @xmath269 that the form @xmath264 defined by formula is a closed and thus coherent @xmath0-form on @xmath251 . since the cocycles of the lie algebroid @xmath251 can be identified to the naive cocycles on @xmath259 , we obtain a naive @xmath0-cocycle on @xmath259 called the * standard @xmath0-form * on @xmath259 . at this point , proposition [ prop : pontryagincoherentextensionallequivalent ] and the following result are quite obvious . [ pro:3formclass ] let @xmath259 be the standard courant algebroid determined by the quintuple @xmath240 , and @xmath264 the @xmath0-form on @xmath251 defined by . then the characteristic class of @xmath259 is equal to @xmath270\in h^3 ( { \mathcal{a}}^s)$ ] . [ generalisomorphism ] given two different standard courant algebroid structures @xmath271 and @xmath272 on @xmath207 , any isomorphism of courant algebroids @xmath273 is of the form @xmath274 for @xmath275 , @xmath125 and @xmath126 . here @xmath276 is an automorphism of the bundle of quadratic lie algebras @xmath14 and @xmath277 and @xmath278 are bundle maps satisfying the following compatibility conditions : @xmath279^\mathcal{g } } , \\ \label{eqn : iso3 } { r}^2({x},{y})-\tau{r}^1({x},{y } ) = \tau({\nabla}^1_{{y}}{\tau^{-1}}\varphi({x } ) -{\nabla}^1_{{x}}{\tau^{-1}}\varphi({y } ) ) + \varphi{[{x},{y}]}+{[\varphi({x}),\varphi({y})]^\mathcal{g } } , \\ \label{eqn : iso4 } \begin{split } & \quad { \mathcal{h}}^2({x},{y},{z})-{\mathcal{h}}^1({x},{y},{z } ) + 2{(\varphi({x}),{[\varphi({y}),\varphi({z})]^\mathcal{g}})^\mathcal{g } } \\ & = \big(2{(\varphi({x}),\tau{\nabla}^1_{{y}}{\tau^{-1}}\varphi({z } ) -\tau{r}^1({y},{z}))^\mathcal{g } } + { l}_{{x}}{\langle { y}| \beta({z})\rangle}+{\langle { x}| \beta{[{y},{z}]}\rangle}\big)+{\text{c.p . } } , \end{split}\end{gathered}\ ] ] for all @xmath263 and @xmath98 . consider the bundle maps @xmath280 where @xmath281 stands for @xmath207 . it is clear that @xmath282 , for all @xmath126 . since @xmath283 respects the pseudo - metric , we have @xmath284 , for all @xmath285 and eq . follows . since @xmath286 for all @xmath125 and @xmath287 , we have @xmath288 . hence @xmath289 . from eq . , it is easy to get that @xmath290 , for all @xmath291 . thus @xmath283 must be of the form . since @xmath283 takes the dorfman bracket of @xmath271 to @xmath272 , eqs . - follow from lemma [ prop : courantbracketintermsofdata2 ] . the following two technical lemmas follow from direct verifications . [ lem : dphijj ] given a bundle map @xmath292 , define @xmath293 by @xmath294 for all @xmath285 and @xmath295 . then the differential of @xmath296 is given by : @xmath297})^\mathcal{g } } \\ -{(j({x}_1),{r}({x}_2,{x}_3)+{[{\boldsymbol{r}}_2,{\boldsymbol{r}}_3]^\mathcal{g}})^\mathcal{g}}+{\text{c.p.}},\end{gathered}\ ] ] for all @xmath298 and @xmath299 . [ lem : dphijjbis ] given a bundle map @xmath300 , define @xmath301 by @xmath302 for all @xmath285 and @xmath295 . then the differential of @xmath303 is given by : @xmath304})\rangle } + { \langle { [ x , y]}| k(z)\rangle}\big)+{\text{c.p.}},\ ] ] for all @xmath298 and @xmath299 . [ lem : ipullbackdiffcoboundary ] the isomorphism @xmath273 of proposition [ generalisomorphism ] yields an isomorphism @xmath305 between the associated ample lie algebroid structures on @xmath306 : @xmath307 if @xmath308 and @xmath309 denote the standard coherent @xmath0-forms ( defined in section [ sec : standard3formabdses ] ) on @xmath310 and @xmath311 , respectively , then @xmath312 . the proof is a direct but cumbersome calculation using proposition [ generalisomorphism ] , lemmas [ lem : dphijj ] and [ lem : dphijjbis ] , and eq . . for @xmath163 defined in , it is easy to see the following equivalent formula @xmath313 using the expressions of the dorfman brackets given by proposition [ prop : courantbracketintermsofdata2 ] , one obtains @xmath314^\mathcal{g}},{\boldsymbol{t}})^\mathcal{g}}+{\mathcal{h}}(x , y , z ) , \end{gathered}\ ] ] for all @xmath315 , @xmath196 , @xmath316 . this shows that @xmath163 coincides with the standard @xmath0-form @xmath264 on @xmath259 ( defined by eq . ) . hence it is naive , closed , does not depend on the choice of @xmath317 and we get statements ( 1 ) , ( 2 ) . finally , we prove statement ( 3 ) . for any two dissections @xmath318 and @xmath319 of @xmath7 , they induce two different courant algebroid structures on @xmath320 , but @xmath271 and @xmath272 are isomorphic . on the other hand , the @xmath0-form @xmath321 on @xmath7 is the pull back of a standard naive @xmath0-form @xmath322 on @xmath323 . therefore , by proposition [ lem : ipullbackdiffcoboundary ] and remark [ rmk : naiveidformsonfat ] , @xmath324 must be a coboundary @xmath325 , for some naive @xmath326-form @xmath327 on @xmath7 . we first show how to construct a courant algebroid out of a characteristic pair . let @xmath15 be a regular quadratic lie algebroid and let @xmath328 be a coherent pair . set @xmath168 and @xmath329 . together , @xmath15 and @xmath191 induce the @xmath17-connection @xmath330 on @xmath14 ; the bundle map @xmath331 ; and the @xmath0-form @xmath332 defined by @xmath333 , for @xmath229 . by proposition [ lem : csclosediff ] , @xmath19 being a @xmath186-coherent @xmath0-form implies that @xmath334 satisfies eq . and thus the quintuple @xmath335 satisfies the requirements to construct a standard courant algebroid structure on the bundle @xmath336 described by theorem [ thm : standardcourantstructure ] . we shall denote such a courant algebroid by @xmath337 . if @xmath15 is understood , we just write @xmath338 . in the sequel , the isomorphism class of a regular courant algebroid @xmath7 will be denoted by @xmath339}$ ] . similarly , the equivalence class of a characteristic pair @xmath340 is denoted by @xmath203}$ ] . define a map @xmath341 from the set of equivalence classes of characteristic pairs to the set of isomorphism classes of regular courant algebroids sending @xmath203}$ ] to @xmath342}$ ] . here @xmath186 is any hoist of @xmath15 such that @xmath191 is a coherent pair on @xmath15 . before we can prove it , let us establish some useful facts . first , we study the change of hoists . it is obvious that for any two hoists of @xmath15 : @xmath186 and @xmath344 , there exists a unique map @xmath345 such that @xmath346 , @xmath347 . recall the notation @xmath256 defined by formula . it is easy to check that @xmath348^\mathcal{g } } , \\ { r}^{{\kappa}'}(x , y)-{r}^{{\kappa}}(x , y ) = { [ { j}(x),{j}(y)]}-{\nabla}^{{\kappa}}_x { j}(y)+{\nabla}^{{\kappa}}_y { j}(x)+{j}[x , y],\end{gathered}\ ] ] for all @xmath349 , @xmath98 . write @xmath353 , @xmath354 , @xmath355 , @xmath356 and let @xmath357 . clearly @xmath358 is determined by the quintuple @xmath359 , where @xmath360 , @xmath361 and @xmath362 . we have the following equalities : @xmath363^\mathcal{g } } , \\ { r}^2(x , y)-{r}^1(x , y)={[{j}(x),{j}(y)]^\mathcal{g } } -{\nabla}^1_x{j}(y)+{\nabla}^1_y{j}(x)+{j}{[x , y ] } , \\ \begin{split } { \mathcal{h}}^2(x , y , z)-{\mathcal{h}}^1(x , y , z ) & = { ( { j}(x),{\nabla}^1_y{j}(z ) -{\nabla}^1_z{j}(y)-{j}{[y , z]}-2{r}^1(y , z))^\mathcal{g } } \\ & \quad -2{({[{j}(x),{j}(y)]^\mathcal{g}},{j}(z))^\mathcal{g } } + { \text{c.p . } } , \end{split}\end{gathered}\ ] ] for all @xmath316 , @xmath98 . define a map @xmath364 by @xmath365 where @xmath275 , @xmath125 , @xmath366 , and @xmath278 denotes the map @xmath367 . according to proposition [ generalisomorphism ] and the above equalities , @xmath368 is a courant algebroid isomorphism . let @xmath371 , @xmath372 . clearly , @xmath358 is the standard courant algebroid determined by the data @xmath373 , @xmath374 and @xmath375 , where @xmath376 , @xmath377 . since @xmath191 and @xmath369 are both coherent pairs , we have @xmath378 , for all @xmath125 . by lemma [ lem : dphijj ] , one can check that this is equivalent to two conditions : @xmath379}=0 , \quad\forall x , y\in{\gamma(f ) } .\ ] ] using these , one has @xmath380 for all @xmath229 . then by proposition [ generalisomorphism ] , the map @xmath364 defined by @xmath381 where @xmath275 ; @xmath125 ; @xmath366 ; @xmath382 ; and @xmath383 , is an isomorphism of courant algebroids . suppose that @xmath386 , for some bundle map @xmath345 , then according to lemma [ lem : goodpairstandardchange ] , the pair @xmath369 is also coherent . by lemma [ lem : almostidenticalgoodpairsiso ] and [ lem : cplusdphijproducesameasc ] , we have isomorphisms @xmath387 . it is clear that @xmath391 , @xmath392 . thus one can easily check that @xmath388 is a coherent pair . let @xmath358 be the standard courant algebroid determined by the quintuple @xmath393 , where @xmath376 , @xmath394 . obviously , @xmath395 . by proposition [ generalisomorphism ] , we can construct an isomorphism @xmath364 by setting @xmath396 , for @xmath275 , @xmath125 , and @xmath366 . here @xmath397 is defined by @xmath398 , @xmath399 . choose @xmath345 so that @xmath406 . lemma [ lem : almostidenticalgoodpairsiso ] implies that @xmath407 the hoist @xmath408 produces a decomposition @xmath409 . hence , there exist @xmath410 and @xmath411 such that @xmath412 . by lemmas [ lem : cplusanchorupsdfomega ] and [ lem : cplusdphijproducesameasc ] , we have @xmath413 first we show that @xmath341 is well defined . suppose that two characteristic pairs @xmath204 and @xmath414 are equivalent , i.e. there is an isomorphism @xmath415 such that @xmath416 , for some @xmath403 . we need to show that there is an isomorphism of courant algebroids @xmath417 . it is quite obvious that @xmath418 is also a coherent pair on @xmath15 and @xmath419 since @xmath416 , lemma [ lem : c2c1homologic0producesame ] gives the isomorphism @xmath420 thus we conclude that @xmath417 and @xmath341 is well defined . \(1 ) let us denote by @xmath421 the set of isomorphism classes of regular courant algebroids , and by @xmath422 the set of equivalence classes of characteristic pairs . we need to establish a one - to - one correspondence between @xmath421 and @xmath422 . for any regular courant algebroid @xmath7 , one may choose a dissection so as to identify @xmath7 with @xmath423 . in section [ sec : standard3form ] , we showed that there is a standard naive @xmath0-cocycle @xmath264 on @xmath259 and it corresponds to a coherent @xmath0-form on @xmath424 , also denoted by @xmath264 . so we can define a map @xmath425 , @xmath339}\mapsto { [ { ( { \mathcal{a}}^s , c^s)}]}$ ] . by proposition [ lem : ipullbackdiffcoboundary ] , this map is well defined and does not depend on the choice of the dissection . on the other hand , proposition [ prop : bijectivemapinversewelldefined ] indicates that there is a well defined map @xmath426 . an easy verification shows that it is indeed the inverse of @xmath427 . this completes the proof . [ pro : allcoherentforms ] let @xmath187 be a coherent @xmath0-form . then any other coherent @xmath0-form is of the form @xmath428 , where @xmath429 is a closed @xmath0-form on @xmath17 and @xmath296 is defined by . assume that @xmath19 ( resp . @xmath430 ) is @xmath186 ( resp . @xmath344)-coherent , for some hoist @xmath186 ( resp . @xmath344 ) of @xmath15 . the difference between @xmath186 and @xmath344 determines a bundle map @xmath345 such that @xmath431 . a direct calculation shows that the @xmath0-form @xmath432 must be the pull back of some @xmath0-form @xmath433 by the anchor map @xmath38 . conversely , it is easy to verify that @xmath428 is coherent w.r.t . @xmath431 . a @xmath0-form @xmath434 of a quadratic lie algebroid @xmath15 is said to be * intrinsic * , if there exists a coherent @xmath0-form @xmath19 and an automorphism @xmath435 over the identity such that @xmath436 . it is simple to see that every intrinsic @xmath0-form is closed , and therefore defines a class in @xmath30 . it turns out that this is the obstruction class to lift the automorphism @xmath437 to a courant algebroid automorphism . [ lem : i ] let @xmath68 be the ample lie algebroid of a regular courant algebroid @xmath7 and @xmath436 an intrinsic form of @xmath68 . then @xmath438}=0 $ ] if and only if @xmath437 can be lifted to an automorphism @xmath439 of the courant algebroid @xmath7 . let @xmath443 be the natural map from the isomorphism classes of regular courant algebroids to the isomorphism classes of quadratic lie algebroids with vanishing first pontryagin class . we are now ready to prove the last main theorem of this paper . by theorem [ thm : onetoone2 ] , we understand that every isomorphism class of courant algebroid in @xmath444)$ ] can be characterized by an equivalence class of characteristic pairs @xmath445 , where @xmath19 is a coherent @xmath0-form . lemma [ pro : allcoherentforms ] implies that @xmath446 acts transitively on @xmath444)$ ] . now we determine the isotropy group of this action . assume that @xmath429 is a closed @xmath0-form such that the characteristic pairs @xmath20 and @xmath447 are equivalent . by definition , there exists an automorphism @xmath437 of @xmath15 such that @xmath448=0 $ ] in @xmath30 . hence it follows that @xmath449}\in\mathbb{i}$ ] . conversely , it is simple to see that @xmath36 acts trivially on @xmath444)$ ] . this concludes the proof of the theorem . @xmath450 is called the intrinsic group of @xmath15 . note that the classification problem investigated here differs from that studied by evera and bressler even in the transitive case . evera and bressler classified the isomorphism classes of transitive courant algebroid extensions of a given lie algebroid @xcite , while we are interested in isomorphism classes of courant algebroids themselves . let @xmath5 be a contact manifold with @xmath452 being the contact @xmath453-form . consider the transitive lie algebroid @xmath454 , which is a central extension of lie algebroid @xmath18 by the exact @xmath326-form @xmath455 considered as a 2-cocycle of the lie algebroid @xmath18 . equip the standard fiberwise metric on @xmath456 . it is clear that @xmath15 is a quadratic lie algebroid . it is easy to check that @xmath457=0\in h^3_{\leftrightarrow}({\mathcal{a}})$ ] . however @xmath458 $ ] is a nontrivial class in @xmath1 . for instance , if @xmath459 with the canonical contact structure , @xmath460 and @xmath461 . as a consequence , there is a unique courant algebroid ( up to isomorphisms ) whose induced quadratic lie algebroid is isomorphic to @xmath462 . on the other hand , according to @xcite , courant algebroid extensions of @xmath454 are parametrized by @xmath47 . in general , it is hard to describe explicitly the intrinsic group @xmath450 . the following example , due to evera @xcite , gives a nice description of the intrinsic group in a special case . we are grateful to evera , who pointed out this example to us , which led us to find an error in an earlier version of the draft . [ rmk : severa ] let @xmath463 be a compact lie algebra such that both @xmath464 and @xmath465 are trivial . assume that @xmath5 is a connected and simply connected manifold . choose an ad - invariant non - degenerate bilinear form on @xmath463 . consider the canonical quadratic lie algebroid @xmath466 . the cartan @xmath0-form induces a coherent @xmath0-form @xmath194 : @xmath467^{{{\mathfrak{g}}}}},{\boldsymbol{t } } ) } , \ ] ] @xmath468 and @xmath197 . let @xmath469 be a connected and simply connected lie group whose lie algebra is @xmath463 . then any automorphism @xmath437 of @xmath15 is determined by a smooth map @xmath470 , where @xmath437 can be written as @xmath471 here @xmath472 is the bundle map defined by : @xmath473 to see this , one can consider @xmath15 as the atiyah lie algebroid of the trivial principal bundle @xmath474 . then @xmath437 is induced by the principal bundle automorphism defined by the gauge group element @xmath470 . using lemma [ lem : dphijj ] , one obtains @xmath475^{{{\mathfrak{g}}}}},j(z))}.\ ] ] it thus follows that @xmath476}=-(j\circ{a})^ * { [ { \alpha}]}=-{{a}^*}(j^ * { [ { \alpha}]}),\ ] ] where @xmath477 , \cdot ) \in \wedge^3 { { \mathfrak{g}}}^*$ ] . it is simple to see that @xmath478 , where @xmath479 is the corresponding cartan @xmath0-form on @xmath469 . as a consequence , the intrinsic group @xmath450 is generated by elements of the form @xmath480}$ ] , where @xmath481 is any smooth map . note that @xmath450 is discrete , because when @xmath482 and @xmath483 are homotopic , then @xmath484}=\tau_2^ * { [ { \tilde{\alpha}}]}$ ] . here is one example . let @xmath485 . then @xmath486 and the intrinsic group @xmath450 is isomorphic to @xmath487 with the generator being @xmath488}$ ] . hence by theorem [ thm : classification ] , the isomorphism classes of courant algebroids with ample lie algebroid isomorphic to the quadratic lie algebroid @xmath489 is parametrized by @xmath490 .

we give the generalized triangle inequalities which determine the possible-valued side lengths of-gons in thick euclidean buildings of rank .

1009.1316

let @xmath3 be a symmetric space of noncompact type or a thick euclidean building . we are interested in the following geometric question : _ which are the possible side lengths of polygons in @xmath3 ? _ in this context the appropriate notion of _ length _ of an oriented geodesic segment is given by a vector in the euclidean weyl chamber @xmath4 associated to @xmath3 . if @xmath5 is a symmetric space , the full invariant of a segment modulo the action of @xmath6 is precisely this vector - valued length since we can identify @xmath7 ( cf . @xcite ) . for @xmath3 a euclidean building the same notion of vector - valued length can be defined ( cf . we denote by @xmath8 the set of all possible @xmath4-valued side lengths of @xmath1-gons in @xmath3 . an algebraic question ( the so - called _ eigenvalue problem _ ) , which goes back to 1912 when it was already studied by h. weyl , is closely related to a special case of the geometric question above , namely , for the symmetric space @xmath9 . it is one of the motivations for considering this geometric problem . the eigenvalue problem asks : _ how are the eigenvalues of two hermitian matrices related to the eigenvalues of their sum ? _ we refer to @xcite for more information on the relation between these two questions and @xcite for more history on this problem . in @xcite and @xcite it is shown that the set @xmath10 depends only on the spherical coxeter complex associated to @xmath3 ( i.e. on the spherical weyl chamber @xmath11 ) . we will therefore sometimes refer to @xmath12 as the set of side lengths of @xmath1-gons in @xmath3 a symmetric space or a euclidean building with @xmath11 as spherical weyl chamber . for a symmetric space @xmath5 the set of possible side lengths has been completely determined in @xcite : @xmath10 is a finite sided convex polyhedral cone and it can be described as the solution set of a finite set of homogeneous linear inequalities in terms of the schubert calculus in the homology of the _ generalized grassmannian manifolds _ associated to the symmetric space @xmath13 . it follows , that for a euclidean building @xmath14 with the same associated spherical weyl chamber @xmath11 as @xmath3 , the set @xmath15 is also a finite sided convex polyhedral cone determined by the same inequalities as @xmath16 . as already pointed out in @xcite for the case of _ exotic _ spherical coxeter complexes ( i.e. when it is the coxeter complex of a euclidean building but it does not occur for a symmetric space ) the structure of the set @xmath12 can not be described with this method , since we do not have a schubert calculus for these coxeter complexes . thus , the structure of @xmath12 for these coxeter complexes and even its convexity were unknown . it is clear that we can restrict our attention to irreducible coxeter complexes . by a result of tits @xcite , exotic irreducible coxeter complexes occur only in rank 2 . our main result is the description of @xmath10 in this case ( compare with theorem [ thm : mainthm ] ) . [ thm : intro ] for a euclidean building @xmath3 of rank 2 , the space @xmath10 is a finite sided convex polyhedral cone . the set of inequalities defining @xmath10 can be given in terms of the combinatorics of the spherical coxeter complex associated to @xmath3 . the inequalities given in our main theorem coincide with the so - called _ weak triangle inequalities _ * sec . 3.8 ) ) . moreover , our arguments also work ( see remark [ rem : weakineq2 ] ) to prove the weak triangle inequalities for buildings of arbitrary rank ( cf . * thm . 3.34 ) ) . for symmetric spaces , these inequalities correspond to specially simple intersections of schubert cells in the description of @xmath10 given in @xcite . their description depend only in the weyl group of @xmath3 and therefore , they can be defined for arbitrary coxeter complexes . consider the _ side length map _ @xmath17 . the set @xmath10 which we are interested in is nothing else than the image of @xmath18 . we use a direct geometric approach to describe this image . our main idea is to study the singular values of @xmath18 by deforming the sides of a given polygon in @xmath3 . this strategy was already used for the case of symmetric spaces by b. leeb in @xcite to give a simple proof of the thompson conjecture ( cf . * theorem 1.1 ) ) . in this paper we adapt this _ variational _ method to the case of euclidean buildings and use it to describe the space @xmath19 . throughout this paper we state the results , whenever possible , in such a way that they apply to euclidean buildings of arbitrary rank . in particular , sections [ sec : polygons ] , [ sec : criticalvalues ] and [ sec : crosswalls ] ( except lemma [ lem : crossablewallsrank2 ] and proposition [ prop : unionofcones ] ) do not use the assumption on the rank of the building . and when we do use the assumption , we indicate it explicitly in the statement of the corresponding result . the set of inequalities obtained in theorem [ thm : intro ] constitute an irredundant system defining the polyhedral cone @xmath19 . the inequalities given by schubert calculus in @xcite are known to be irredundant for the cases of type @xmath20 ( see @xcite ) , however , these seem to be the only cases . a smaller set of inequalities is given in @xcite by defining a new product in the cohomology of flag varieties . the irredundancy of this set has been recently shown in @xcite . after a first version of this paper was written , the author learned about a recent related paper of berenstein and kapovich @xcite , where the generalized triangle inequalities for rank 2 are also determined by a different approach . _ acknowledgments . _ i would like to thank bernhard leeb for bringing this problem to my attention and sharing his ideas in the case of symmetric spaces with me . a very good introduction to the concepts used in this paper is the work ( * ? ? ? * ch . 2 - 4 ) . we refer also to @xcite for more information on metric spaces with upper curvature bounds and to ( * ? ? ? * ch . 2 - 3 ) for the different concepts of _ length _ in euclidean buildings . recall that a complete geodesic metric space @xmath3 is said to be @xmath21 if the geodesic triangles in @xmath3 are not _ thicker _ that the corresponding triangles in the euclidean space . for two points @xmath22 we denote with @xmath23 the geodesic segment between them . the link @xmath24 is the completion of the space of directions at @xmath25 with the angle metric . @xmath26 denotes the direction of the segment @xmath23 at @xmath25 . two complete geodesic lines @xmath27 are said to be _ parallel _ if they have finite hausdorff distance , or equivalently , if the functions @xmath28 are constant . parallel set _ @xmath29 is defined as the union of all geodesic lines parallel to @xmath30 . it is a closed convex set that splits as a metric product @xmath31 , where @xmath32 is also a @xmath21 space . for a _ polygon _ @xmath33 , or more precisely , an _ @xmath1-gon _ in @xmath3 we mean the union of @xmath1 oriented geodesic segments @xmath34 with @xmath35 . since geodesic segments in @xmath21 spaces between two given points are unique , we can also describe @xmath33 by its _ vertices_. we write @xmath36 . the union @xmath37 of @xmath1 oriented geodesic segments @xmath34 where @xmath38 will be called a _ polyhedral path _ and we write @xmath39 . a _ spherical coxeter complex _ is a pair @xmath40 consisting of a unit sphere @xmath41 with its usual metric and a finite group @xmath42 of isometries , the _ weyl group _ , generated by reflections on total geodesic spheres of codimension one . a _ weyl chamber _ in @xmath41 is a fundamental domain of the action @xmath43 . the _ model weyl chamber _ is defined as @xmath44 . we say that two points in @xmath41 have the same _ @xmath42-type _ ( or just _ type _ ) if they belong to the same @xmath42-orbit . a _ euclidean coxeter complex _ is a pair @xmath45 consisting of a euclidean space @xmath46 and a group of isometries @xmath47 , the _ affine weyl group _ , generated by reflections on hyperplanes and such that its rotational part @xmath48 is finite . the set of fixed points of reflections in @xmath47 are called _ walls _ of @xmath45 . we define the _ @xmath47-type _ of a point in @xmath46 as above . to @xmath45 , we can associate the spherical coxeter complex @xmath49 , where @xmath50 is the tits boundary of @xmath46 . euclidean model weyl chamber _ @xmath4 is the complete euclidean cone over @xmath51 . the link @xmath52 of a point @xmath53 is naturally a spherical coxeter complex with weyl group @xmath54 . we will also use another structure on @xmath52 as a coxeter complex with weyl group @xmath42 . this will be given by the natural identification @xmath55 . the _ refined length _ of the oriented geodesic segment @xmath56 is defined as the image of @xmath57 under the projection @xmath58 . the _ @xmath0-valued length _ , or just _ length _ , is the image of the refined length under the natural _ forgetful _ map @xmath59 . we denote with @xmath18 the _ length map _ assigning to a segment its @xmath0-valued length . we can also define the _ refined length _ of an oriented segment @xmath23 in the spherical coxeter complex @xmath40 analogously as the image of @xmath57 under the projection @xmath60 . for an introduction to spherical and euclidean buildings from the point of view of metric geometry , we refer to @xcite . let @xmath3 be a thick euclidean building modelled in the euclidean coxeter complex @xmath45 . the concepts of refined length and @xmath0-valued length of an oriented geodesic segment @xmath61 can be also defined naturally by identifying an apartment containing @xmath23 with the coxeter complex @xmath45 . for a polygon @xmath36 in @xmath3 , we write @xmath62 and call @xmath63 the _ side length map_. the space @xmath64 is the set of possible @xmath0-valued side lengths of @xmath1-gons in @xmath3 . we say that a polygon in @xmath3 is _ regular _ if all its sides are regular , that is if their @xmath0-valued lengths lie in the interior of @xmath4 . the space of regular polygons is an open dense subset of @xmath65 . we will use following result from @xcite concerning the refined side lengths of polygons in @xmath3 . we reproduce here its statement for the convenience of the reader . [ thm : transfer ] let @xmath3 and @xmath14 be thick euclidean buildings modelled on the same euclidean coxeter complex @xmath45 . let @xmath36 be a polygon in @xmath3 and let @xmath66 be a segment in @xmath14 with the same refined length as @xmath67 . then there exists a polygon @xmath68 in @xmath14 with the same refined side lengths as of @xmath33 . we fix a vertex @xmath70 of @xmath45 with @xmath71 . we obtain in this way an identification @xmath72 . by fixing @xmath70 we get an embedding @xmath73 and also the ( coarser ) structure @xmath74 as euclidean coxeter complex . we will think of the euclidean weyl chamber @xmath75 as embedded in @xmath46 , such that @xmath4 is a fundamental domain of the action @xmath76 . hence , the cone point of @xmath4 corresponds to @xmath70 . let @xmath77 be a maximal singular unit vector , i.e. @xmath78 is a vertex of @xmath79 . we define the following linear functional : @xmath80 where @xmath81 denotes the standard scalar product on @xmath82 . we denote with @xmath69 the finite set of functionals on @xmath83 of the form @xmath84 where all the @xmath85 have the same @xmath42-type . we write @xmath86 for such a functional . let @xmath87 denote the hyperplane @xmath88 for @xmath89 . we call @xmath87 a _ wall _ in @xmath83 . the set of walls @xmath87 divide @xmath83 in finitely many convex polyhedral cones . we denote with @xmath90 the family of the interiors of these cones , i.e. @xmath90 is the set of the connected components of @xmath91 . let @xmath92 be an @xmath1-gon in @xmath3 . we say that a @xmath1-tuple @xmath93 of apartments in @xmath3 _ supports the polygon _ @xmath33 if @xmath94 and the convex set @xmath95 is top dimensional and contains @xmath96 in its interior . if @xmath33 is a regular polygon then there always exists an @xmath1-tuple @xmath97 supporting @xmath33 . @xmath97 can be constructed as follows : let @xmath98 be an apartment containing @xmath99 and @xmath100 and take @xmath101 antipodal to @xmath99 . extend the segment @xmath67 a little further than @xmath102 in direction of @xmath103 to a segment @xmath104 . inductively for @xmath105 choose @xmath106 to be an apartment containing @xmath107 and an initial part of @xmath108 and extend @xmath108 in @xmath109 a little further than @xmath96 to a segment @xmath110 . finally choose @xmath111 to contain @xmath112 and an initial part of @xmath113 and @xmath114 . this last step is possible because of our first choice of @xmath115 . the polyhedron @xmath95 contains a regular segment with @xmath96 in its interior . in particular @xmath95 is top dimensional . let now @xmath33 be a polygon and @xmath97 an @xmath1-tuple supporting it . notice that the convex set @xmath95 is a neighborhood of @xmath96 in @xmath109 and @xmath116 . therefore we have : @xmath117 so we have a natural map @xmath118 ( just take _ parallel transport _ in @xmath116 along the side @xmath119 ) and an associated _ holonomy map _ @xmath120 defined as the composition @xmath121 . we introduce also the following notation : @xmath122 if we identify @xmath123 with @xmath124 in the natural way , we obtain a structure of spherical coxeter complex on @xmath123 with weyl group @xmath42 . with this structure the maps @xmath125 are isomorphisms of coxeter complexes and the holonomy map @xmath126 is an element of the weyl group @xmath42 . in particular the set of fixed points of @xmath126 is a singular sphere in @xmath127 . notice that the holonomy map ( and therefore also its fixed points set ) depends on the choice of the @xmath1-tuple @xmath97 supporting @xmath33 . we will make use of this flexibility later . let @xmath92 be a @xmath1-gon in @xmath3 and let @xmath97 be an @xmath1-tuple supporting it . we construct points @xmath128 , @xmath129 inductively as follows : for @xmath130 just set @xmath131 and @xmath132 and suppose we have already constructed @xmath133 . for each @xmath134 we can identify naturally @xmath135 with @xmath136 thus giving it a structure of spherical coxeter complex with weyl group @xmath42 . let @xmath137 be a isomorphism of spherical coxeter complexes such that @xmath138 . notice that if @xmath33 is regular such an isomorphism is unique let now @xmath139 be the point in @xmath140 such that @xmath141 and @xmath142 ( see fig . [ fig : openpolygon ] ) . we remark that in general @xmath143 and @xmath144 is a polygonal path hence the expression `` opening a polygon '' . we can continue this process and define @xmath145 for @xmath146 . the isomorphisms @xmath147 can be chosen ( and we do so ) so that the induced automorphisms of @xmath148 @xmath149 are just the identity map . this construction was first considered in ( * ? ? ? 6.1 ) . for simplicity on the notation , suppose @xmath150 is a triangle in @xmath3 . there is a partition @xmath151 of the segment @xmath152 such that the triangles @xmath153 for @xmath154 are contained in an apartment @xmath155 . we define points @xmath156 in the apartment @xmath157 inductively as follows : for @xmath158 set @xmath159 and suppose we have already defined @xmath156 . let @xmath160 be an isomorphism of euclidean coxeter complexes , such that @xmath161 . we define @xmath162 . we say that the polygon @xmath163 is the result of _ folding _ the triangle @xmath33 into @xmath157 . we say that the points @xmath156 for @xmath164 are the _ break points _ of the folded polygon @xmath165 . notice that the segments @xmath67 and @xmath166 have the same refined side lengths as the segments @xmath167 and @xmath168 respectively . write @xmath169 and define @xmath170 and @xmath171 , analogously @xmath172 and @xmath173 . a _ billiard triangle _ is a polygon @xmath163 in an apartment @xmath157 such that for @xmath164 the directions @xmath174 and @xmath175 are antipodal in the spherical coxeter complex @xmath176 modulo the action of the weyl group @xmath177 . clearly , a folded triangle is a billiard triangle . conversely , the next condition is necessary and sufficient for a billiard triangle to be a folded triangle . _ for @xmath164 there is a triangle @xmath178 in the spherical building @xmath179 such that @xmath180 and the refined lengths of @xmath181 and @xmath182 are the same as of @xmath183 and @xmath184 respectively . _ we investigate now the relation between the constructions of opening and folding a polygon . let @xmath150 be a triangle in @xmath3 and let @xmath97 be a triple supporting @xmath33 . observe that we can choose @xmath185 . let @xmath163 be the folded triangle . again we identify naturally @xmath186 with @xmath187 for each @xmath188 and give the structure of spherical coxeter complex with weyl group @xmath42 . for @xmath154 . let @xmath189 be an isomorphism of spherical coxeter complexes so that @xmath190 . notice that for @xmath158 we just have @xmath191 . analogously , let @xmath192 be an isomorphism so that @xmath193 and let @xmath194 be so that @xmath195 . observe that if @xmath33 is regular , then the @xmath196 are unique . since @xmath163 is a billiard triangle , there are isometries @xmath197 of @xmath140 in the affine weyl group @xmath47 for @xmath198 such that @xmath199 has the same refined length as @xmath200 . in particular for @xmath164 the points @xmath201 , @xmath202 , @xmath203 lie on a geodesic segment . hence , we call the @xmath197 the _ straightening _ isometries . it holds : @xmath204 where @xmath205 is constructed as in section [ sec : openpolygon ] . consider the natural action of @xmath197 on @xmath206 . the straightening isometries can be chosen ( if @xmath33 is regular then they are unique ) such that [ cols= " < , < " , ] it follows that @xmath207 is the holonomy map at @xmath208 . the constructions for @xmath1-gons ( @xmath209 ) are analogous . for a _ regular _ value of the side length map @xmath18 we mean a value @xmath210 for which there is a polygon @xmath33 with @xmath211 and such that @xmath18 is an open map at @xmath33 . first we give a sufficient condition in terms of the holonomy map for @xmath212 being a _ regular _ value of @xmath18 . [ prop : regularvalue ] let @xmath33 be an @xmath1-gon in @xmath3 and @xmath97 an @xmath1-tuple supporting @xmath33 . suppose that the holonomy map @xmath126 has no fixed points , then the space @xmath10 is a neighborhood of @xmath212 in @xmath83 . choose @xmath213 so that @xmath214 for all @xmath215 . for @xmath216 and for @xmath217 we write @xmath218 to denote the point @xmath219 with @xmath220 and @xmath221 . we want to _ vary _ the polygon @xmath33 _ along _ @xmath216 to a polygon @xmath222 with side lengths @xmath223 for @xmath224 . for this , let @xmath225 and define @xmath226 for @xmath227 where the subindices are considered modulo @xmath1 . notice that for @xmath224 the segment @xmath228 is just a translation in the apartment @xmath229 of the segment @xmath230 . hence the condition on the side lengths above is clearly fulfilled . but since @xmath231 we get ( see fig . [ fig : variationofside ] ) @xmath232 ] since @xmath126 has no fixed points the set @xmath233 is a neighborhood of @xmath234 in @xmath4 . this means that we can deform every side length of @xmath33 independently , thus @xmath10 is a neighborhood of @xmath212 in @xmath83 . the next proposition says that for a building with only one vertex the critical values of @xmath18 must lie in the walls @xmath87 . [ prop : criticalvalue ] let @xmath33 be an @xmath1-gon in a thick euclidean building @xmath3 which has only one vertex . let @xmath97 be an @xmath1-tuple supporting @xmath33 . suppose that the holonomy map @xmath126 fixes a maximal singular direction . then there exists a functional @xmath235 , such that @xmath236 . first observe that we have a natural identification of any apartment with @xmath237 since we assumed that @xmath3 has only one vertex . this gives us also an identification @xmath238 . let @xmath239 be a maximal singular direction fixed by @xmath240 . let @xmath241 be a unit vector with direction @xmath239 . now open the polygon @xmath242 in the apartment @xmath140 to the polygonal path @xmath243 . we can also fold @xmath33 into @xmath140 and obtain the straightening isometry @xmath244 . recall that @xmath245 and @xmath246 since @xmath247 induces the holonomy map . it then follows that @xmath248 . now let @xmath249 be a maximal singular unit vector of the same @xmath42-type as @xmath250 , such that @xmath251 . set @xmath86 , then @xmath252 we use next the result in @xcite that @xmath10 depends only on the spherical coxeter complex to transfer the result above to arbitrary buildings . [ cor : imageisopen ] let @xmath253 and suppose that @xmath254 for all functionals @xmath89 . then @xmath10 is a neighborhood of @xmath255 in @xmath83 . by @xcite we may assume that @xmath3 has only one vertex . let @xmath33 be a regular polygon with @xmath211 and let @xmath97 be an @xmath1-tuple supporting @xmath33 . by proposition [ prop : criticalvalue ] the holonomy map has no fixed points . the result now follows from proposition [ prop : regularvalue ] . [ lem : imageisclosed ] let @xmath256 be a sequence of regular @xmath1-gons in @xmath3 such that @xmath257 in @xmath83 , then there exists an @xmath1-gon @xmath33 in @xmath3 such that @xmath211 . we assume again that @xmath3 has only one vertex . let @xmath258 and let @xmath259 be @xmath1-tuples supporting @xmath260 . after transferring the polygons @xmath260 ( cf . theorem [ thm : transfer ] ) we may assume that the sides @xmath261 lie in the same apartment @xmath262 and that @xmath263 lie in the same euclidean weyl chamber @xmath264 . after a small perturbation of the polygons we may also suppose that @xmath263 lie in the interior of @xmath4 . we open now the polygons @xmath260 in the apartment @xmath262 to polygonal paths @xmath265 . if @xmath266 in @xmath262 , then for @xmath267 big enough @xmath268 must be completely contained in the interior of @xmath4 . in particular , folding the polygon @xmath260 into @xmath262 can not have break points . this implies that @xmath260 is contained in the apartment @xmath262 for @xmath267 big enough . since @xmath257 , then it is clear that the polygons @xmath260 subconverge in @xmath262 modulo translations in @xmath262 to a polygon @xmath33 with @xmath211 . suppose now that @xmath269 stay in a bounded region . then after taking a subsequence we can assume that the polygonal paths @xmath268 converge to a polygonal path @xmath270 with @xmath0-valued side lengths @xmath255 . we want now to lift this polygonal path near the polygons @xmath260 . let @xmath271 be the isomorphisms of euclidean coxeter complexes that send @xmath272 to @xmath273 . so we have @xmath274 . hence , for @xmath267 big enough we have @xmath275 and we can define @xmath276 . then @xmath277 is a polygonal path with the same side lengths as @xmath278 , i.e. @xmath279 . however @xmath280 may still not be a closed polygon . notice that @xmath281 and @xmath282 , thus @xmath283 . on the other hand , observe that @xmath269 and @xmath284 have the same @xmath47-type and therefore also @xmath285 and @xmath286 have the same type . but @xmath47 is finite , so @xmath287 can only take finitely many values . it follows that for @xmath267 big enough @xmath288 and @xmath280 is a closed polygon with @xmath0-valued side lengths @xmath255 . [ cor : opencones ] for any open cone @xmath289 the intersection @xmath290 is empty or @xmath291 . moreover , if @xmath292 , then @xmath293 . the intersection @xmath290 is open by corollary [ cor : imageisopen ] and closed by lemma [ lem : imageisclosed ] . suppose @xmath33 is a polygon in @xmath3 with @xmath294 for some functional @xmath89 . considering corollary [ cor : opencones ] the natural question is if there is a cone @xmath295 such that @xmath296 . we would also like to describe all cones in @xmath90 with this property . with this in mind we investigate in this section following question . when can we find polygons @xmath278 with @xmath0-valued side lengths near @xmath255 and such that @xmath297 ( or @xmath298 ) ? for this we might try to study the side lengths of small perturbations of @xmath33 . however since a euclidean building has dimension equal to his rank , we do not have much flexibility to perturbate the polygon . thus we must be more compliant with the variations of @xmath33 that we want to admit . therefore we will often have to translate the polygon to other place in @xmath3 where we can perform the perturbations . let @xmath97 be a @xmath1-tuple of apartments supporting @xmath33 . let @xmath300 be maximal singular directions ( in the structure coming from @xmath301 with weyl group @xmath42 ) such that if @xmath302 , @xmath303 are unit vectors with base point @xmath96 and directions @xmath304 and @xmath305 respectively , then @xmath306 and @xmath307 . observe that @xmath308 are of the same @xmath42-type as @xmath309 we will therefore sometimes write @xmath310 and @xmath311 . notice that @xmath312 is just the parallel translation along @xmath313 in @xmath109 of @xmath314 , that is @xmath315 . [ lem : wallbyaccident ] if in the notation above @xmath316 for some @xmath215 , then for any neighborhood @xmath317 of @xmath212 in @xmath83 there exist @xmath1-gons @xmath318 in @xmath3 with @xmath319 and @xmath320 . assume now that @xmath326 for all @xmath215 . in particular , the holonomy map @xmath327 has the fixed point @xmath304 . let @xmath328 ( resp . @xmath329 ) be the line ( i.e. complete geodesic ) in @xmath109 ( resp . @xmath116 ) with @xmath330 and @xmath331 . if @xmath332 for all @xmath215 , then the polygon @xmath33 is contained in a parallel set , namely the set @xmath333 of all lines parallel to @xmath334 . [ lem : localparallelset ] suppose @xmath33 is not contained in any parallel set @xmath335 , where @xmath30 is a geodesic line with @xmath336 such that @xmath337 for all @xmath215 . then for any neighborhood @xmath317 of @xmath212 in @xmath83 there exist @xmath1-gons @xmath318 in @xmath3 with @xmath319 and @xmath320 . let @xmath338 be an @xmath1-tuple of geodesic segments @xmath339\rightarrow x$ ] with @xmath340 , @xmath341 . and such that the convex hull @xmath342 is a ( 2-dimensional ) flat quadrilateral . such a @xmath343 exists , just take the initial parts of the geodesics @xmath344 . suppose now that @xmath343 is maximal , i.e. the segments @xmath345 can not be extended . if @xmath346 , then the @xmath345 are parallel geodesic lines and @xmath347 . hence at least one of @xmath348 or @xmath349 must be @xmath350 . suppose @xmath351 ( the other case is analogous ) . now we want to displace @xmath33 along @xmath345 to the region , where it does not look locally like a parallel set anymore : set @xmath352 . then @xmath278 is an @xmath1-gon with @xmath353 . choose apartments @xmath155 containing the convex sets @xmath354 . let @xmath355 and let @xmath356 , @xmath357 be the antipodes of @xmath358 in @xmath359 and @xmath360 respectively . moreover , if it holds for all @xmath215 that @xmath365 , then @xmath366 is a geodesic segment in @xmath367 of length @xmath368 . let @xmath369 be a point near @xmath139 with @xmath370 . we can choose @xmath371 close enough to @xmath139 , so that @xmath372 is a regular point in the same weyl chamber as @xmath373 . it follows that @xmath372 lies in the intersection of the segments @xmath366 and @xmath374 . thus @xmath375 is a geodesic segment of length @xmath368 . let now @xmath376 be a point with @xmath377 and so that @xmath378 is a flat triangle . it follows that the union of the ( 2-dimensional ) flat convex sets @xmath379 , @xmath380 and @xmath381 is a flat convex quadrilateral . ( see figure [ fig : localparallelset ] . ) notice also that @xmath382 are extensions of the geodesic segments @xmath383 . thus this contradicts as well the maximality of @xmath343 . hence , there is a @xmath363 such that @xmath384 . let @xmath385 in @xmath386 for some small @xmath387 . then @xmath388 for some unit vector @xmath389 of the same type as @xmath390 . by the above consideration we must have @xmath391 , otherwise @xmath392 . in particular @xmath393 . on the other hand , @xmath394 . thus , for @xmath395 we have @xmath396 . the next question is what happens when @xmath33 is contained in such a parallel set @xmath335 . in this last situation we can not always get the same conclusion as in lemmata [ lem : wallbyaccident ] and [ lem : localparallelset ] . for instance , if the wall @xmath87 lies in the boundary of @xmath10 , then we can cross @xmath87 in one direction but not in the opposite one . suppose that @xmath33 is contained in @xmath29 . let @xmath398 be a busemann function associated to @xmath399 ( see e.g. ( * ? ? ? 2.2 ) for a definition ) . then by considering an apartment parallel to @xmath30 containing the side @xmath400 , we see that @xmath401 . in particular @xmath402 thus , if @xmath278 is the result of a small variation of the polygon @xmath33 within the parallel set @xmath29 , it still holds @xmath403 . suppose @xmath33 is contained in @xmath335 . assume also that there are vertices @xmath404 of @xmath33 with the following property . let @xmath405 be apartments in @xmath29 containing the segment @xmath406 and an initial part of the segment @xmath407 and @xmath408 respectively . let @xmath409 for @xmath410 be points in the initial parts of the segments @xmath407 and @xmath408 respectively . thus @xmath411 are flat triangles in @xmath412 . suppose that for some @xmath410 there is a singular hyperplane @xmath413 such that the directions @xmath336 , @xmath414 and @xmath415 lie in the same open half space determined by @xmath416 ( after the natural identification of @xmath417 and @xmath418 for @xmath419 ) . ( see figure [ fig : samehalfspace ] . ) denote @xmath425 the open half space of @xmath426 determined by @xmath427 containing the direction @xmath428 . let @xmath429 be small . first we displace the polygon @xmath33 along @xmath30 such that @xmath430 lies in @xmath431 and @xmath432 . let @xmath433 be an apartment in @xmath3 such that @xmath434 . let @xmath435 be the point such that @xmath436 and @xmath437 . let @xmath438 be the reflection of @xmath430 in the hyperplane @xmath427 . notice that the refined length of @xmath443 is the same as of @xmath407 . hence , by theorem [ thm : transfer ] we can transfer the polygon @xmath444 to a polygon @xmath445 with the same @xmath0-valued side lengths . the @xmath1-gon @xmath446 satisfy the conclusion of the lemma . [ cor : polygoninapt ] suppose @xmath33 is contained in an apartment @xmath447 . suppose there are two sides @xmath448 , @xmath406 of @xmath33 and a singular hyperplane @xmath449 , such that the directions @xmath336 , @xmath450 and @xmath414 lie in the same open half space determined by @xmath451 . then for any neighborhood @xmath317 of @xmath212 in @xmath83 there is an @xmath1-gon @xmath420 in @xmath3 with @xmath421 and @xmath422 . consider the segments @xmath452 and @xmath453 . after a small variation of the polygon @xmath33 inside of the apartment @xmath454 , we may assume that @xmath455 ( and therefore also @xmath456 ) is regular . then for one @xmath457 , it must hold , that @xmath458 and @xmath250 lie in the same open half space determined by @xmath451 . suppose w.l.o.g . . then lemma [ lem : crossablewalls ] applies for the vertices @xmath404 . let @xmath150 be a regular triangle contained in @xmath29 but not contained in any apartment . it is easy to see , that when we fold @xmath33 into an apartment @xmath454 , it has exactly one break point . after relabeling the vertices we can assume that the break point @xmath459 lies in the side @xmath152 and that the sides of the folded triangle @xmath460 do not intersect in their interiors ( see figure [ fig : variationrank2a ] ) . after displacing @xmath165 along @xmath30 we can assume that @xmath459 is a vertex of @xmath3 . we can take @xmath30 to be contained in @xmath454 and go through @xmath459 . [ lem : crossablewallsrank2 ] we use the setting above ( in particular , @xmath461 ) . suppose that the weyl chamber containing @xmath462 is not adjacent to @xmath463 . then for any neighborhood @xmath317 of @xmath212 in @xmath464 there are triangles @xmath318 in @xmath3 with @xmath319 and @xmath320 . we identify @xmath454 with @xmath465 by taking @xmath459 to the origin . for a unit vector @xmath466 we write @xmath467 . let @xmath468 be the singular line through @xmath459 such that @xmath469 is adjacent to the simplicial convex hull of @xmath470 and the directions @xmath336 , @xmath462 and @xmath471 are in the same open half plane determined by @xmath472 . it exists by the assumptions of the lemma . let @xmath473 be the reflection of @xmath472 in @xmath30 . let @xmath474 be unit vectors orthogonal to @xmath30 , @xmath472 and @xmath475 respectively and such that @xmath476 and @xmath477 . then the simplicial convex hull of @xmath470 is @xmath478 . ( see figure [ fig : variationrank2a ] . ) let @xmath479 be an apartment in @xmath3 such that @xmath480 . let @xmath481 be the point so that @xmath482 and @xmath483 . notice that @xmath484 , thus , @xmath485 . observe also that @xmath486 , hence , @xmath487 . let @xmath488 . the concatenation of the segments @xmath489 and @xmath490 gives a segment in @xmath491 of length @xmath368 ( see figure [ fig : variationrank2b ] ) . therefore @xmath492 is a geodesic segment and the triangle @xmath493 has the same side lengths as @xmath33 . set @xmath494 and let @xmath495 be an apartment in @xmath3 containing the segment @xmath496 . let @xmath345 be the geodesic rays with @xmath497 and @xmath498 . then @xmath342 are ( 2-dimensional ) flat stripes . we want to see that the @xmath345 can not be extended to parallel geodesic lines . suppose then the contrary : there are parallel geodesic lines @xmath499 containing @xmath345 . set @xmath500 . then @xmath501 and in particular , @xmath502 . since @xmath503 are antipodal regular points , the apartment containing them is unique . therefore @xmath504 and in particular , @xmath505 . let @xmath506 so that the weyl chamber containing @xmath507 is adjacent to @xmath508 . let @xmath509 be the weyl chamber containing @xmath510 and let @xmath511 be the antipodal chamber to @xmath512 . ( see figure [ fig : variationrank2b ] for @xmath513 . ) notice that @xmath514 intersects @xmath515 in its interior . in particular @xmath516 . it follows that the unique apartment containing @xmath512 and @xmath515 is contained in @xmath517 , i.e. @xmath518 . let @xmath519 be the weyl chamber adjacent to @xmath472 . the weyl chamber containing @xmath520 is antipodal to @xmath18 . hence , the unique apartment containing @xmath18 and @xmath520 is contained in @xmath517 , i.e. @xmath521 . we have conclude that @xmath522 , but this is not possible because of the construction of @xmath479 . therefore the geodesic rays @xmath345 can not be extended to complete parallel geodesic lines . the lemma now follows from lemma [ lem : localparallelset ] and its proof . we have already seen in corollary [ cor : opencones ] that if for @xmath289 holds @xmath523 , then @xmath524 . now let @xmath36 be a polygon in @xmath3 . we want to show that @xmath212 is contained in @xmath525 for some @xmath289 . since any polygon can be approximated by regular polygons , we may assume that @xmath33 is regular . suppose now @xmath526 . if for any neighborhood @xmath317 of @xmath255 we can find polygons with side lengths in @xmath527 , then we are done . indeed , in this case , there is an open cone @xmath289 such that @xmath523 and @xmath528 . suppose then that for some neighborhood @xmath317 of @xmath212 we can not find polygons @xmath278 with side lengths in @xmath317 and @xmath529 . lemmata [ lem : wallbyaccident ] and [ lem : localparallelset ] implies that @xmath33 lies in a parallel set @xmath29 and the functional @xmath530 is given in @xmath33 by taking scalar product with the direction of @xmath336 . suppose first that the triangle @xmath531 lies in an apartment parallel to @xmath30 . then it is easy to see that lemma [ lem : crossablewalls ] must apply for one of the functionals @xmath532 or @xmath533 , where @xmath534 is so that @xmath535 . if @xmath536 is not contained in an apartment , then we fold it into an apartment as in the setting of lemma [ lem : crossablewallsrank2 ] . then , either lemma [ lem : crossablewallsrank2 ] applies or the weyl chamber containing the direction @xmath450 of the side of @xmath536 with the break point must be adjacent to @xmath30 . if the last occurs , it is again easy to see , that lemma [ lem : crossablewalls ] must apply for @xmath537 or @xmath533 . in either case , we find a triangle @xmath538 with @xmath539 and such that ( modulo displacement along @xmath30 ) the refined side lengths of @xmath540 are as near as we want to the ones of @xmath536 . after a small variation of the polygon @xmath541 inside the parallel set @xmath29 and displacing it along @xmath30 , we obtain a polygon @xmath542 so that the refined side length of @xmath543 is the same as of @xmath544 . then by the transfer theorem [ thm : transfer ] we can glue @xmath536 and @xmath37 along @xmath543 and @xmath544 to a polygon @xmath278 with @xmath0-valued side lengths near @xmath255 and @xmath545 . we have seen in the previous section different methods which allows to cross certain walls @xmath87 within the space @xmath10 . we will show in this section that for the case of buildings of rank 2 the walls where this method can not be applied are precisely the walls that determine the boundary of @xmath10 . that is , if a wall can not be crossed with the methods of section [ sec : crosswalls ] , it is because that wall can not be crossed at all . first we characterize the walls @xmath87 that can not be crossed with the methods above in terms of the combinatorics of the associated spherical coxeter complex @xmath40 . let @xmath547 be a maximal singular unit vector ( we use the same notation as in section [ sec : functional ] ) . we define the following set of singular hyperplanes of @xmath46 through @xmath70 ( i.e. walls of @xmath74 ) : @xmath548 for each element @xmath549 we define the subset of @xmath550 @xmath551 finally define @xmath552 as the set of @xmath1-tuples @xmath553 such that for @xmath129 there are @xmath554 with @xmath555 and with the following properties : we will see in lemma [ lem : notpassablewalls ] below that the walls @xmath87 that can not be crossed with our previous methods are precisely the ones of the form @xmath564 with @xmath565 satisfying the property ( @xmath566 ) . a motivation for this property ( @xmath566 ) can already be seen in corollary [ cor : polygoninapt ] . the property ( @xmath567 ) is introduced to avoid later obvious redundancies in the set of generalized triangle inequalities . this can be seen in the proposition [ prop : weakineq ] . [ lem : ineq = weakineq ] if @xmath74 has rank 2 , then @xmath568 if and only if for @xmath129 we can find @xmath554 with @xmath555 such that there exist @xmath569 , @xmath570 with @xmath571 antipodal to @xmath572 and @xmath573 adjacent to @xmath574 for @xmath575 . ( @xmath579 ) . by property ( @xmath567 ) , there is a @xmath363 with @xmath580 . if @xmath581 , then @xmath582 and the assertion is clear . otherwise let @xmath583 be the singular line adjacent to @xmath571 . let @xmath584 be the other singular line adjacent to @xmath571 . then @xmath585 . let @xmath586 be such that @xmath587 , it follows that @xmath588 and @xmath572 must be antipodal to @xmath571 . the rest follows as in the first part . [ rem : weakineq ] the @xmath576 direction in lemma [ lem : ineq = weakineq ] holds for arbitrary rank . let @xmath589 be the set of functionals @xmath564 for @xmath590 with this property ( the assumption in the @xmath576 direction ) . the inequalities @xmath591 for @xmath592 are the so - called _ weak triangle inequalities _ * section 3.8 ) ) . thus , lemma [ lem : ineq = weakineq ] states that @xmath593 and for rank 2 also holds @xmath594 . [ prop : weakineq ] suppose @xmath3 has rank 2 . for any @xmath1-gon @xmath33 in @xmath3 and any functional @xmath595 holds @xmath596 . that is , @xmath597 moreover , if @xmath598 satisfies the property ( @xmath566 ) but not the property ( @xmath567 ) , then there is a @xmath599 so that @xmath600 for all @xmath1-gons @xmath33 in @xmath3 . if @xmath33 is regular , then the strict inequality holds . let @xmath36 be an @xmath1-gon in @xmath3 . for the functional @xmath601 , let @xmath554 and @xmath602 be as in lemma [ lem : ineq = weakineq ] . notice that since for @xmath575 , @xmath573 is adjacent to @xmath574 and @xmath555 then we have @xmath603 in @xmath4 for all @xmath534 of the same type as @xmath250 . that is , @xmath604 is the smallest functional of the same type as @xmath250 . after shifting the subindices of the polygon and the functional we can assume that @xmath605 . suppose first that @xmath606 , that is @xmath607 . fold the polygon @xmath33 into an apartment @xmath454 , so that the broken sides are @xmath608 . let @xmath609 be an isometry that sends @xmath102 to the vertex of @xmath610 , induces an isomorphism of the coxeter complexes @xmath611 and @xmath74 and so that @xmath612 . notice that @xmath613 is not necessarily an isomorphism of coxeter complexes with the weyl group @xmath47 . denote with @xmath37 the image under @xmath613 of the folded polygon . by folding @xmath46 onto the euclidean weyl chamber @xmath614 with the natural `` accordion '' map , we obtain a further folded polygon @xmath615 where @xmath616 is the vertex of @xmath4 and the @xmath0-valued side lengths of @xmath617 are the same as for @xmath67 and @xmath618 respectively . observe that @xmath619 is not necessarily a billiard polygon in @xmath45 , but if the side @xmath620 of @xmath33 is broken in @xmath619 to the sides @xmath621 , then the vectors @xmath622 are just multiples of @xmath623 . this means , that if @xmath624 is the group generated by @xmath47 and the whole translation group of @xmath46 , then @xmath619 is a billiard polygon in @xmath625 . notice also that for @xmath626 holds @xmath627 because of the observation at the beginning of the proof . it follows that @xmath628 on the other hand , since @xmath617 and @xmath629 is antipodal to @xmath614 it follows that @xmath630 and @xmath631 . hence , @xmath632 . the general case now follows from the special case above by considering the polygons @xmath633 , i.e. @xmath634 is the polygon @xmath33 with the vertices @xmath635 deleted , and @xmath636 with the functionals @xmath637 respectively @xmath638 . indeed , notice that since @xmath571 and @xmath572 are antipodal , it follows @xmath639 . for the second assertion , let @xmath555 satisfy the property ( @xmath566 ) . it is easy to see that in rank 2 at most for two indices @xmath215 can hold @xmath640 . let @xmath570 be so that @xmath578 for all @xmath575 . if @xmath590 does not satisfy the property ( @xmath567 ) , then @xmath572 is not antipodal to @xmath641 . let @xmath642 be so that @xmath643 is antipodal to @xmath641 . from the property ( @xmath566 ) follows that for @xmath644 holds @xmath645 and since @xmath646 the strict inequality holds for regular segments . [ rem : weakineq2 ] the same proof as for the first assertion of proposition [ prop : weakineq ] works for buildings of arbitrary rank to prove the weak triangle inequalities ( see remark [ rem : weakineq ] ) . that is , @xmath647 [ lem : notpassablewalls ] suppose @xmath3 has rank 2 and let @xmath33 be a regular @xmath1-gon in @xmath3 . suppose that @xmath299 for some functional @xmath530 with @xmath648 . then for any neighborhood @xmath317 of @xmath212 in @xmath83 there exist @xmath1-gons @xmath318 in @xmath3 with @xmath319 and @xmath320 . suppose that for a neighborhood @xmath317 of @xmath212 in @xmath83 , we can not find a polygon @xmath634 in @xmath3 with @xmath649 and @xmath650 . ( the other inequality follows considering the functional @xmath651 . ) it follows from lemmata [ lem : wallbyaccident ] and [ lem : localparallelset ] that @xmath33 lies in a parallel set @xmath29 and the functional @xmath530 in @xmath33 is just given by taking scalar product with the direction of @xmath652 . fold the polygon in an apartment @xmath447 so that the broken sides are @xmath608 . let @xmath609 be an isomorphism that sends @xmath336 to the singular direction in @xmath4 of the same type . by abusing the notation , we write also @xmath250 to denote the unit vector in @xmath4 with direction @xmath653 . suppose @xmath3 has only one vertex and @xmath30 goes through it . then the break points of the folded polygon all lie on @xmath30 . we may assume that the folded polygon has only one break point because any two consecutive break points can be simultaneously _ unfolded_. let @xmath267 be so that the break point @xmath459 lies on the side @xmath654 ( if there is no break point we take @xmath655 ) . then the folded polygon has the form @xmath656 . let @xmath554 be so that @xmath573 contains the direction @xmath657 for @xmath658 , @xmath659 for @xmath660 , @xmath661 for @xmath662 , and @xmath663 for @xmath664 , respectively . then the functional @xmath530 is just given by @xmath665 for @xmath666 . after a small variation inside the parallel set @xmath29 we may assume that the segments @xmath667 and @xmath668 are regular . let @xmath669 be so that @xmath670 contains the direction @xmath671 and @xmath672 contains @xmath673 . let @xmath674 be such that @xmath4 and @xmath675 are antipodal . for @xmath676 set @xmath677 . consider the regular polygon @xmath678 and the functional @xmath679 for @xmath680 . that is , @xmath537 is the functional given in @xmath37 by taking scalar product with the direction @xmath250 . hence @xmath681 . set @xmath682 . suppose that there are @xmath683 such that @xmath684 . corollary [ cor : polygoninapt ] and its proof imply that there is a polygon @xmath685 with @xmath686 and with refined side lengths as near as we want to those of @xmath37 modulo displacement along @xmath30 . we can then choose @xmath687 near @xmath688 such that @xmath689 has the same refined side length ( again modulo displacement along @xmath30 ) as @xmath690 . the functional @xmath691 applied to the polygon @xmath692 is 0 because it is contained in the parallel set @xmath29 . after displacing the polygon @xmath692 along @xmath30 we can glue it together to @xmath619 and obtain a polygon @xmath634 with @xmath0-valued side lengths as near as we want to those of @xmath33 and with @xmath693 ( compare with the proof of proposition [ prop : unionofcones ] ) . this contradicts the assumption at the beginning of the proof . thus , @xmath694 for all @xmath683 . since @xmath37 is a regular polygon with @xmath695 , then by the second claim in proposition [ prop : weakineq ] we must also have @xmath696 , or equivalently , @xmath697 . consider now the triangle @xmath702 with the functional @xmath703 . let @xmath704 be so that @xmath705 contains the direction @xmath706 . then @xmath707 . we want to show that @xmath708 or @xmath709 have the property ( @xmath566 ) . by lemma [ lem : crossablewalls ] applied to the side @xmath667 we get @xmath710 . again by lemma [ lem : crossablewalls ] now applied to the side @xmath711 we obtain @xmath712 . therefore if @xmath713 or @xmath714 , then we are done , so suppose both are nonempty . now by lemma [ lem : crossablewallsrank2 ] one of @xmath670 , @xmath672 or @xmath715 must be adjacent to @xmath716 . notice that for @xmath676 , @xmath717 is adjacent to @xmath716 if and only if @xmath718 . this and @xmath719 imply that @xmath715 must be adjacent to @xmath716 . @xmath720 implies that @xmath721 must be empty and we are also done in this case . [ thm : mainthm ] let @xmath3 be a building of rank 2 . @xmath10 is a convex polyhedral cone determined by the inequalities @xmath722 for @xmath595 . that is , @xmath723 this inequalities constitute an irredundant set of inequalities . let @xmath724 be the subset of open cones such that @xmath725 . analogously , let @xmath726 be the subset of open cones such that @xmath727 ( this can be done by proposition [ prop : unionofcones ] ) . we have shown in proposition [ prop : weakineq ] that @xmath728 . let @xmath729 and @xmath730 . take a chain @xmath731 such that @xmath732 is a face of codimension one . we prove now inductively that @xmath733 . suppose then that @xmath733 and take a regular polygon @xmath33 with @xmath212 in the interior of the face @xmath732 . since @xmath732 is not in the boundary of @xmath734 , it lies in a wall @xmath87 with neither @xmath735 in @xmath736 . it follows from lemma [ lem : notpassablewalls ] that @xmath737 is not empty and therefore @xmath738 . thus @xmath739 , and @xmath740 . for @xmath595 it is clear that we can find a regular polygon @xmath33 in an apartment @xmath454 and @xmath741 a maximal singular line , such that the functional @xmath530 in @xmath33 is given by taking scalar product with the direction of @xmath336 . in particular , @xmath236 . it is also clear that we can find a regular polygon @xmath278 in @xmath29 but not contained in any apartment and such that the functional @xmath530 in @xmath278 is also given by taking scalar product with the direction of @xmath250 . it follows from lemmata [ lem : wallbyaccident ] and [ lem : localparallelset ] that @xmath530 is the only functional in @xmath736 for which it can hold @xmath403 . thus the inequalities @xmath722 with @xmath595 are irredundant . p. belkale , s. kumar , _ eigenvalue problem and a new product in cohomology of flag varieties _ , invent . 166 ( 2006 ) , no . 1 , 185 - 228 . a. berenstein , m. kapovich , _ stability inequalities and universal schubert calculus of rank 2 _ , preprint 2010 . m. bridson , a. haefliger , _ metric spaces of non - positive curvature _ , springer 1999 . w. fulton , _ eigenvalues , invariant factors , highest weights and schubert calculus _ , bull . ams 37 ( 2000 ) , no . 3 , 209 - 249 . m. kapovich , b. leeb , j. millson , _ convex functions on symmetric spaces , side lengths of polygons and the stability inequalities for weighted configurations at infinity _ , j. differ . geom . 81 ( 2009 ) , 297 - 354 . m. kapovich , b. leeb , j. millson , _ polygons in buildings and their refined side lengths _ , geom . 19 , no . 4 , 1081 - 1100 ( 2009 ) . b. kleiner , b. leeb , _ rigidity of quasi - isometries for symmetric spaces and euclidean buildings _ , inst . hautes tudes sci . 86 ( 1997 ) , 115197 ( 1998 ) . a. knutson , t. tao and c. woodward , _ the honeycomb model of @xmath742 tensor products . puzzles determine facets of the littlewood - richardson cone _ , j. amer . 17 ( 2004 ) , no . 1 , 19 - 48 . b. leeb , unpublished notes . n. ressayre , _ geometric invariant theory and generalized eigenvalue problem _ , invent . math . 180 ( 2010 ) , no . 2 , 389 - 441 .

by applying a variant of the tqft constructed by blanchet , habegger , masbaum , and vogel , and using a construction of ohtsuki , we define a module endomorphism for each knot by using a tangle obtained from a surgery presentation of . we show that it is strong shift equivalent to the turaev - viro endomorphism associated to . following viro , we consider the endomorphisms that one obtains after coloring the meridian and longitude of the knot . we show that the traces of these endomorphisms encode the same information as the colored jones polynomials of at a root of unity . most of the discussion is carried out in the more general setting of infinite cyclic covers of 3-manifolds .

1202.1815

walker first noticed @xcite that the endomorphism induced in a @xmath1-tqft ( defined over a field ) by the exterior of a closed off seifert surface of a knot in zero - framed surgery along the knot can be used to give lower bounds for the genus of the knot . he did this by showing the number of non - zero eigenvalues of this endomorphism counted with multiplicity is an invariant @xcite , i.e. it does not depend on the choice of the seifert surface . thus the number of such eigenvalues must be less than or equal to the dimension of the vector space that the tqft assigns to a closed surface of this minimal genus . next turaev and viro @xcite , again assuming the tqft is defined over a field , saw that the similarity class of the induced map on the vector space associated to a seifert surface modulo the generalized @xmath2-eigenspace was a stronger invariant . if the tqft is defined over a more general commutative ring , the second author observed that the strong shift equivalence class of the endomorphism is an invariant of the knot @xcite . strong shift equivalence ( abbreviated sse ) is a notion from symbolic dynamics which we will discuss in [ sse ] below . for a tqft defined over a field @xmath3 , the similarity class considered by turaev - viro is a complete invariant of sse . in this case , the vector space modulo the generalized @xmath2-eigenspace together with the induced automorphism , considered as a module over @xmath4 $ ] , is called the turaev - viro module . it should be considered as somewhat analogous to the alexander module . the order of the turaev - viro module is called the turaev - viro polynomial and lies in @xmath4 $ ] . we will refer to the endomorphisms constructed as above ( and those in the same sse class ) as turaev - viro endomorphisms . in @xcite , turaev - viro endomorphisms were studied and methods for computing the endomorphism explicitly were given . these methods adapted rolfsen s surgery technique of studying infinite cyclic covers of knots . this method requires finding a surgery description of the knot ; that is a framed link in the complement of the unknot such that the framed link describes @xmath5 and the unknot represents the original knot . moreover each of the components of the framed link should have linking number zero with the unknot . for this method to work , it is important that the surgery presentation have a nice form . in this paper , we will show that all knots have a surgery presentations of this form ( in fact an even nicer form that we will call standard . ) another explicit method of computation was given by achir , and blanchet @xcite . this method starts with any seifert surface . the second author also considered the further invariant obtained by decorating a knot with a colored meridian ( this was needed to give formulas for the turaev - viro endomorphism of a connected sum , and to use the turaev - viro endomorphism to compute the quantum invariants of branched cyclic covers of the knot ) . ohtsuki @xcite arrived at the same invariant as the turaev - viro polynomial but from a very different point of view . ohtsuki extracts this invariant from a surgery description of a knot ( alternatively of a closed 3-manifold with a primitive one dimensional cohomology class ) and the data of a modular category . his method starts from any surgery description standard or not . this is a significant advantage of his approach . ohtsuki s proof of the invariance of the polynomial in @xcite is only sketched . he stated that his invariant is the same as the turaev - viro polynomial , but does not give an explanation . recently viro has returned to these ideas @xcite . he has studied the turaev - viro endomorphism of a knot after coloring both the meridian and the longitude of the knot . viro observed that a weighted sum of the traces of these endomorphisms is the colored jones polynomial evaluated at a root of unity . in @xcite , turaev - viro endomorphisms were defined more generally for infinite cyclic cover of 3-manifolds . suppose @xmath6 is a closed connected oriented @xmath7-manifold @xmath8 with @xmath9 such that @xmath10 is onto . let @xmath11 be the infinite cyclic cover of @xmath8 corresponding to @xmath12 . choose a surface @xmath13 in @xmath8 dual to @xmath12 . by lifting @xmath13 to @xmath11 , we obtain a fundamental domain @xmath14 with respect to the action of @xmath15 on @xmath11 . @xmath14 is a cobordism from a surface @xmath13 to itself . let @xmath16 be a @xmath1-tqft on the cobordism category of extended @xmath7-manifolds and extended surfaces . applying @xmath16 to @xmath14 and @xmath13 , we can construct an endomorphism @xmath17 . in @xcite , it is proved that the strong shift equivalent class of @xmath17 is an invariant of the pair @xmath6 , i.e. it does not depend on the choice of @xmath13 . we denote this sse class by @xmath18 . we will sometimes refer to a pair @xmath19 as above , informally , as a 3-manifold with an infinite cyclic covering . the knot invariants discussed above can be obtained as special cases of the above invariants of 3-manifolds with an infinite cyclic covering . for any _ oriented _ knot @xmath0 in @xmath5 , we obtain an extended @xmath7-manifold @xmath20 by doing @xmath2-surgery along @xmath0 . we choose @xmath12 to be the integral cohomology class that evaluates to @xmath21 on a positive meridian of @xmath0 . then it is easy to see that the invariant @xmath22 corresponding to @xmath23 only depends on @xmath0 . if our tqft is defined for 3-manifolds with colored links , one may obtain further invariants by coloring the meridian and the longitude ( a little further away ) of the knot . for the knot invariants discussed above , it is required , in general , that @xmath0 be oriented . ] this is so that the exterior of a seifert surface acquires a direction as a cobordism from the seifert surface to itself . however , we decided to delay mentioning this technicality . to avoid issues that arise from phase anomalies in tqft , in this paper , we work with extended manifolds as in walker @xcite and turaev @xcite . in this introduction , we omit mention of the integer weights and lagrangian subspaces of extended manifolds . we discuss extended manifolds carefully in the main text . inspired by ohtsuki , we construct a sse class @xmath24 from a framed ( or banded ) tangle in @xmath25 that arises in a surgery presentation of @xmath6 . we call this the tangle endomorphism . moreover we show that the endomorphism ( or square matrix ) that ohtsuki considers in this situation is well defined up to sse . by relating the definition of the turaev - viro endomorphism to ohtsuki s matrix , we give a different proof of the invariance of ohtsuki s invariant . in fact , we show that ohtsuki s matrix has the same sse class as the turaev - viro endomorphism , i.e. @xmath26 . we do not prove these results in the general case of a tqft arising from a modular category . we only work in the context of the skein approach for tqfts associated to @xmath27 and @xmath28 . we work with a modified blanchet - habegger - masbaum - vogel approach @xcite as outlined in @xcite . this theory is defined over a slightly localized cyclotomic ring of integers . it is worthwhile studying endomorphisms defined up to strong shift equivalence over this ring rather than passing to a field . we show that the traces of the turaev - viro polynomial of knots with the meridian and longitude colored turns out to encode exactly the same information as the colored jones polynomial evaluated at a root of unity . in section [ preliminary ] , we discuss extended manifolds , a variant of the tqft constructed in @xcite , surgery presentations and the definition of sse . in section [ thetanglemorphism ] , we construct an endomorphism for each framed tangle in @xmath25 and apply it to the tangle obtained from a surgery presentation of an infinite cyclic cover of a 3-manifold . we call it the tangle endomorphism . then we state theorem [ main ] which states that the sse class of a tangle endomorphism constructed from a surgery presentation of @xmath19 is an of invariant @xmath19 . in section [ tvm ] , we discuss technical details concerning the turaev - viro endomorphism for @xmath6 , and the method of calculating @xmath29 introduced in @xcite . in section [ relation ] , we relate the tangle endomorphism associated to a nice surgery presentation to the corresponding turaev - viro endomorphism . in section [ pmt ] , we prove theorem [ main ] . in section [ cj ] , we give formulas relating the colored jones polynomial to the traces of turaev - viro endomorphism of a knot whose meridian and longitude are colored . in section [ example ] , we compute two examples to illustrate these ideas . all surfaces and 3-mainifolds are assumed to be oriented . for each integer @xmath30 , blanchet , habegger , masbaum and vogel define a tqft from quantum invariants of @xmath7-manifolds at @xmath31th root of unity over a @xmath1-cobordism category in @xcite . the cobordism category has surfaces with @xmath32-structures as objects and @xmath7-manifolds with @xmath32-structures as morphisms . they introduce @xmath32-structures in order to resolve the framing anomaly . following @xcite , we will adapt the theory by using extended surfaces and extended @xmath7-manifolds in @xcite instead of @xmath32-structures to resolve the framing anomaly . in the following , all homology groups have rational coefficients except otherwise stated . an extended surface @xmath33 is a closed surface @xmath13 with a lagrangian subspace @xmath34 of @xmath35 with respect to its intersection form , which is a symplectic form on @xmath35 . an extended @xmath7-manifold @xmath36 is a @xmath7-manifold with an integer @xmath37 , called its weight , and whose oriented boundary @xmath38 is given an extended surface structure with lagrangian subspace @xmath39 . if @xmath8 is a closed extended 3-manifold , we may denote the extended 3-manifold simply by @xmath40 . suppose we have an extended @xmath7-manifold @xmath36 and @xmath41 is a closed surface . then @xmath42 need not be a lagrangian subspace of @xmath35 . suppose we have an extended @xmath7-manifold @xmath36 and @xmath41 is a closed surface . if @xmath43 is a lagrangian subspace of @xmath35 , we call @xmath13 equipped with this lagrangrian a boundary surface of the extended 3-manifold @xmath36 . if @xmath13 is a surface , we use @xmath44 to denote the surface @xmath13 with the opposite orientation . [ lagrangiansubspace ] suppose @xmath45 and @xmath46 are two symplectic vector spaces . consider the symplectic vector space @xmath47 with symplectic form @xmath48 . we can identify @xmath49 and @xmath50 as symplectic subspaces of @xmath47 . if @xmath51 is a lagrangian subspace such that @xmath52 is a lagrangian subspace of @xmath49 , then @xmath53 is a lagrangian subspace of @xmath50 . since @xmath54 where @xmath55 , we can assume that @xmath56 where @xmath57 . since for any @xmath58 @xmath59 we have @xmath60 . therefore , @xmath61 that means @xmath62 . so @xmath53 is a lagrangian subspace in @xmath50 . [ complement ] suppose we have an extended @xmath7-manifold @xmath36 and @xmath41 is a boundary surface . then @xmath63 , equipped with the lagrangian @xmath64 , is also a boundary surface . this follows from proposition [ lagrangiansubspace ] . in the next three definitions , we describe the morphisms and the composition of morphisms in @xmath65 , a cobordism category whose objects are extended surfaces . let @xmath36 be an extended @xmath7-manifold . suppose @xmath66 and this boundary has been partitioned into two boundary surfaces @xmath44 , called ( minus ) the source , and @xmath67 , called the target . we write @xmath68 and call @xmath36 an extended cobordism . let @xmath13 be a boundary surface of an extended @xmath7-manifold @xmath36 with inclusion map @xmath69 let @xmath67 be @xmath63 with inclusion map @xmath70 then we define @xmath71 we define the composition of morphisms in @xmath65 as the extended gluing of cobordisms . [ extendedgluing ] let @xmath36 and @xmath72 be two extended @xmath7-manifolds . suppose @xmath73 is a boundary surface of @xmath36 and @xmath74 is a boundary surface of @xmath72 . then we can glue @xmath36 and @xmath72 together with the orientation reversing identity from @xmath13 to @xmath44 to form a new extended @xmath7-manifold . the new extended @xmath7-manifold has 1 . base manifold : @xmath75 2 . lagrangian subspace : @xmath76\oplus[\lambda(\partial m')\cap h_1(\partial m'-\bar{\sigma } ) ] , \notag\ ] ] 3 . weight : @xmath77 where @xmath78 is the maslov index as in @xcite . [ selfgluing ] let @xmath36 be an extended @xmath7-manifold with a boundary surface of the form @xmath79 . then we define the extended @xmath7-manifold obtained by gluing @xmath13 and @xmath44 together to be the extended @xmath7-manifold that results from gluing @xmath36 and @xmath80,0,\lambda ( \sigma\cup\bar{\sigma}))$ ] along @xmath81 . in the special case that @xmath82 , we call the resulting extended @xmath7-manifold the closure of @xmath36 . one should think of the weight of an extended @xmath7-manifold @xmath8 as the signature of some background @xmath83-manifold @xcite . see also @xcite . [ closure ] let @xmath84 be a morphism from @xmath33 to @xmath85 and @xmath86 be a morphism from @xmath85 to @xmath73 . then the extended @xmath7-manifold we obtain by gluing @xmath84 to @xmath86 along @xmath67 first and then closing it up along @xmath13 is the same as the one we obtained from gluing @xmath86 to @xmath84 along @xmath13 first and then closing it up along @xmath67 . this can be seen from @xmath83-manifold interpretation of weights in @xcite . extended surfaces may also be equipped with banded points : this is an embedding of the disjoint union of oriented intervals . by a framed link , we will mean what is called a banded link in @xcite , i.e. an embedding of the disjoint union of oriented annuli . framed @xmath21-manifold are defined similarly . extended @xmath7-manifolds are sometimes equipped with framed links , or framed 1-manifolds or more generally trivalent fat graphs . by a trivalent fat graphs , we will mean what is called a banded graph in @xcite . the framed links , framed 1-manifolds and trivalent fat graphs must meet the boundary surfaces of a 3-manifold in banded points with the induced `` banding '' . of course , we could have used the word `` banded '' in all cases , but the other terminology is more common . there is a surgery theory for extended @xmath7-manifolds . we refer the reader to @xcite . here we give extended version of kirby moves @xcite . these moves relate framed links in @xmath5 where @xmath5 is itself equipped with an integer weight . the result of extended surgery of @xmath5 with its given weight along the link is preserved by these moves . moreover ( but we do not use this ) if surgery along two framed links in weighted copies of @xmath5 result in the same extended manifold then there is sequence of extended kirby moves relating them . the extended kirby-@xmath21 move is the regular kirby-@xmath21 move with weight of manifold changed accordingly . more specifically , if we add an @xmath87-framed unknot to the surgery link , then we change the weight of the manifold by @xmath88 , where @xmath89 . if we delete an @xmath87-framed unknot from the surgery link , then we change the weight of the manifold by @xmath87 . the extended kirby-@xmath90 move is the regular kirby-@xmath90 move with the weight remaining the same . suppose a closed connected @xmath7-manifold @xmath8 is obtained from @xmath5 by doing surgery along a framed link @xmath91 , then @xmath40 is obtained from @xmath92 by doing extended surgery along @xmath91 . here @xmath93 is the signature of the linking matrix of @xmath91 . warning this is different than the signature of @xmath91 . the quantum invariant of @xmath40 at a @xmath31th root of unity @xmath94 is then defined as : @xmath95 where @xmath96 we use @xmath97 to denote the kauffman bracket evaluation of a linear combination of colored links in @xmath5 , and @xmath98 to denote the satellization of a framed link @xmath99 by a skein @xmath100 of the solid torus . moreover @xmath101 denotes the skein class in the solid torus obtained by taking the closure of @xmath102 , the jones - wenzl idempotent in the @xmath103-strand temperley - lieb algebra . here @xmath104 denotes the zero framed unknot and @xmath105 is the unknot with framing @xmath106 . the sum is over the colors @xmath107 if @xmath108 is even and @xmath109 with @xmath103 even if @xmath108 is odd . one has that @xmath110 is a square root of @xmath111 . the choice of square root here determines the choice in the square root in the formula of @xmath112 , or vice - versa . see the formula for @xmath113 in @xcite . the closed connected manifold @xmath8 may also have an embedded @xmath108-admissibly colored fat trivalent graph @xmath114 in the complement of the surgery , then @xmath115 by following the exactly the same procedure in @xcite , we can construct a tqft for the category of extended surfaces and extended @xmath7-manifolds from quantum invariants . the tqft assigns to each extended surface @xmath73 , possibly with some banded colored points , a module @xmath116 over @xmath117 $ ] , and assigns to each extended cobordism @xmath8 , with a @xmath108-admissibly colored trivalent fat graph meeting the banded colored points , @xmath118 a @xmath119-module homomorphism : @xmath120 then by using this tqft , we can produce a turaev - viro endomorphism associated to each weighted closed 3-manifold equipped with a choice of infinite cyclic cover using the procedure described in [ introduction ] . we introduce some notations that will be used later . 1 . @xmath121 , 2 . @xmath122 , 3 . @xmath123 is the kauffman bracket of the left diagram in figure [ thetatet ] , 4 . @xmath124 is the kauffman bracket of the right diagram in figure [ thetatet ] . , and on the right is @xmath125.,width=192,height=96 ] the earliest use of surgery presentations , that we are aware of , was by rolfsen @xcite to compute and study the alexander polynomial . in this paper we consider surgery descriptions for extended closed 3-manifolds with an infinite cyclic cover . we will use these descriptions for extended 3-manifolds that contain certain colored trivalent fat graphs . as this involves no added difficulty , we will not always mention these graphs in this discussion . let @xmath126 be a framed link inside @xmath127 where @xmath128 is an oriented @xmath2-framed unknot , and the linking numbers of the components of @xmath91 with @xmath128 are all zero . let @xmath129 be a disk in @xmath5 with boundary @xmath128 which is transverse to @xmath91 . suppose @xmath40 is the result of extended surgery along @xmath126 , then there exists a unique epimorphism @xmath130 which agrees with the linking number with @xmath128 on cycles in @xmath131 . we will call @xmath132 a surgery presentation of @xmath133 . we remark that , in this situation , we will have @xmath134 . if there are graphs @xmath135 in @xmath8 and @xmath114 in @xmath136 ( transverse to @xmath129 ) related by the surgery , we will say @xmath137 a surgery presentation of @xmath138 . if the result of surgery along @xmath91 returns @xmath5 with the image of @xmath128 after surgery becoming a knot oriented knot @xmath0 , and the linking numbers of the components of @xmath91 with @xmath128 , then @xmath126 is a surgery presentation of @xmath0 as in rolfsen . the manifold obtained by surgery along @xmath126 in @xmath5 is the same as @xmath2-framed surgery along @xmath0 in @xmath5 . the following proposition is proved in section 4 of @xcite for non - extended manifolds . the extended version involves no extra difficulty every extended connected @xmath7-manifold with an epimorphism @xmath139 has a surgery presentation . every surgery presentation can be described by diagram as in figure [ almoststandardform ] which we will refer to as a surgery presentation diagram . has zero linking number with @xmath128 . notice the orientation on @xmath128 . , width=115,height=115 ] if a surgery presentation diagram is in the form of figure [ standardform ] , then we say this surgery presentation diagram is in standard form . we will also say that a surgery presentation @xmath137 is standard if it has a surgery presentation diagram in standard form . intersects the flat disc @xmath129 bounded by the trivial knot algebraically @xmath2 times , but geometrically @xmath90 times.,title="fig:",width=115,height=115]0.5 in ohtsuki ( * ? ? ? * bottom of p. 259 ) stated a proposition about surgery presentations of knots which is similar to the following proposition . our proof is similar to the proof that ohtsuki indicated . we will call a kirby-@xmath21 move in a surgery presentation a small kirby-@xmath21 move if a disk which bounds the created or deleted component is in the complement of @xmath129 . we will call a kirby-@xmath90 move in a surgery presentation a small kirby-@xmath90 move if it involves sliding a component other than @xmath128 over another component that is in the complement of @xmath129 . a @xmath129-move is a choice of a new spanning disk @xmath140 with @xmath141 followed by an ambient isotopy that moves @xmath140 to the original position of @xmath129 and moves @xmath91 at the same time . [ standardization ] a surgery presentation described by a surgery presentation diagram can be transformed into a surgery presentation described by a surgery presentation diagram in standard form by a sequence of isotopies of @xmath142 relative to @xmath129 , small kirby-@xmath21 moves , small kirby-@xmath90 moves , and @xmath129-moves . therefore , every extended @xmath7-manifold with an epimorphism @xmath130 has a standard surgery presentation . we need to prove that we can change a surgery presentation described by a surgery diagram as in figure [ almoststandardform ] into surgery presentation described by a diagram as in figure [ standardform ] using the permitted moves . let @xmath143 we will prove the theorem by induction on @xmath144 . since each component @xmath145 has linking number 0 with @xmath128 , it is easy to see that @xmath144 is even . if @xmath146 , then @xmath91 can be taken to be contained in the tangle box . when @xmath147 , we may * first do a @xmath129 move to shift @xmath129 slightly ; * then perform an isotopy relative to the new @xmath129 of @xmath91 so that the points on intersection of the image of the old @xmath129 with each components of @xmath91 are adjacent to each other ; * then do another @xmath129-move to move the old @xmath129 back to its original position . now the arcs emitted from the bottom edge of the tangle are in a correct order . but the diagram in figure [ almoststandardform ] may differ from a standard tangle in the way that the arcs emitted from bottom edge of the tangle box are not in the specified simple form . this means they could be knotted and linked with each other . however we may perform small kirby-@xmath21 and small kirby-@xmath90 moves as in figure [ changecrossing ] to unknot and unlink these arcs so that the resulting diagram has standard form . or @xmath148 surgery on unknot to change the crossing.,title="fig:",width=57,height=96 ] or @xmath148 surgery on unknot to change the crossing.,title="fig:",width=28,height=96 ] or @xmath148 surgery on unknot to change the crossing.,title="fig:",width=76,height=96]0.2 in or @xmath148 surgery on unknot to change the crossing.,title="fig:",width=57,height=96 ] or @xmath148 surgery on unknot to change the crossing.,title="fig:",width=28,height=96 ] or @xmath148 surgery on unknot to change the crossing.,title="fig:",width=76,height=96 ] we now prove that the theorem holds for all links with @xmath149 where @xmath150 assuming it holds for all links with @xmath151 . suppose the component @xmath152 intersects @xmath129 geometrically @xmath153 times . because @xmath152 has linking number @xmath2 with @xmath128 , we have that at least one arc , say @xmath154 of @xmath152 in figure [ almoststandardform ] which joins two points on the bottom of the tangle box , i.e. it is a `` turn - back '' . for each crossing with exactly one arc from @xmath154 , we can make the arc @xmath154 to be the top arc ( in the direction perpendicular to the plane of the diagram ) by using the moves of figure [ changecrossing ] , which just involve some small kirby-@xmath21 and small kirby-@xmath90 moves . then it is only simply linked to other components by some new trivial components with framing @xmath155 . then by using isotopies relative to @xmath129 , we can slide the arc @xmath154 towards bottom of the tangle , with the newly created unknots stretched vertically in the diagram so that they intersect each horizontal cross - section in at most 2-points . see the central illustration figure [ example2 ] where @xmath154 is illustrated by two vertical arcs meeting a small box labeled @xmath156 . this small box contains the rest of @xmath154 . now perform a @xmath129 move which has the effect of pulling the turn - back across @xmath129 . those trivial components will follow the turn back and pass through @xmath129 . but since at the beginning , those components have geometric intersection @xmath2 with @xmath129 , they have geometric intersection @xmath90 with @xmath129 now . after this process , @xmath157 is reduced by @xmath90 . this process does not change the number of intersections with @xmath129 of the other components of the original @xmath91 . with @xmath129 . we perform small k - moves and isotopies to change to the middle picture . we perform a @xmath129-move to change to the right hand picture . , title="fig:",width=76,height=96 ] with @xmath129 . we perform small k - moves and isotopies to change to the middle picture . we perform a @xmath129-move to change to the right hand picture . , title="fig:",width=28,height=96 ] with @xmath129 . we perform small k - moves and isotopies to change to the middle picture . we perform a @xmath129-move to change to the right hand picture . , title="fig:",width=76,height=96 ] with @xmath129 . we perform small k - moves and isotopies to change to the middle picture . we perform a @xmath129-move to change to the right hand picture . , title="fig:",width=28,height=96 ] with @xmath129 . we perform small k - moves and isotopies to change to the middle picture . we perform a @xmath129-move to change to the right hand picture . , title="fig:",width=76,height=96 ] we do this process for all components @xmath158 with @xmath159 . then the new link has @xmath151 . by our induction hypothesis , we can transform @xmath126 into a standard form using the allowed moves . we will discuss sse in the category of free finitely generated modules over a commutative ring with identity . this notion arose in symbolic dynamics . for more information , see @xcite and references therein . suppose @xmath160 are module endomorphisms . we say @xmath156 is elementarily strong shift equivalent to @xmath161 if there are two module morphisms @xmath162 such that @xmath163 we denote this by @xmath164 . suppose @xmath165 are module endomorphisms . we say @xmath156 is strong shift equivalent to @xmath161 if there are finite number of module endomorphisms @xmath166 such that @xmath167 we denote this by @xmath168 . it is easy to see that if @xmath168 , then @xmath169 . [ ker ] let @xmath156 be a module endomorphism of @xmath170 . suppose @xmath171 where @xmath104 and @xmath172 are free finitely generated modules such that @xmath104 in the kernel of @xmath156 , and let @xmath173 be the induced endomorphism of @xmath172 , then @xmath173 is sse to @xmath174 suppose @xmath175 , and @xmath176 . the result follows from the following block matrix equations . @xmath177_{(n+m ) \times n } \cdot \left [ \begin{array}{c|c } 0_{n\times m } & i_n \end{array } \right]_{n \times ( n+m ) } = \left [ \begin{array}{c|c } 0_{m \times m } & v_{m \times n } \\ \hline 0_{n \times m } & \hat x_{n \times n } \end{array}\right]_{(n+m ) \times ( n+m)}\ ] ] @xmath178_{n \times ( n+m ) } \cdot \left [ \begin{array}{c } v_{m \times n } \\ \hline \hat x_{n \times n } \end{array}\right]_{(n+m ) \times n } = \left[\hat x_{n \times n } \right]_{n \times n}\ ] ] if @xmath179 is an endomorphism of a vector space @xmath170 , let @xmath180 denote the generalized @xmath2-eigenspace for @xmath179 , and @xmath181 denote the induced endomorphism on @xmath182 . the next proposition may be deduced from more general statements made in ( * ? ? ? * , prop(2.4 ) ) . for the convenience of the reader , we give direct proof . [ similar ] let @xmath179 and @xmath183 be endomorphisms of vector spaces . @xmath179 and @xmath183 are sse if and only if @xmath181 and @xmath184 are similar . the only if implication is well - known ( * ? ? ? * theorem 7.4.6 ) . the if implication follows from the easy observations that similar transformations are strong shift equivalent and that @xmath179 is strong shift equivalent to @xmath181 . this second fact follows from the repeated use of the following observation : if @xmath185 is in the null space of @xmath179 , @xmath186 denotes the space spanned by @xmath100 , and @xmath187 denotes the induced map on @xmath188 , then @xmath179 and @xmath187 are strong shift equivalent . this follows from proposition [ ker ] with @xmath189 . in this section , we will assign a @xmath119-module homomorphism to any framed tangle in @xmath190 enhanced with an embedded @xmath108-admissibly colored trivalent fat graph in the complement of the tangle . by slicing a surgery presentation for an infinite cyclic cover of an extended @xmath7-manifold and applying the tqft , we obtain such a tangle , and thus a @xmath119-module endomorphism . the idea of constructing this endomorphism is inspired by the work of ohtsuki in @xcite . there is a unique lagrangian for a 2-sphere . thus we can consider any 2-sphere as an extended manifold without specifying a lagrangian . similarly , we let @xmath191 denote the extended manifold @xmath25 with weight @xmath37 , as there is no need to specify a lagrangian . let @xmath192 be a 2-sphere equipped with @xmath144 ordered uncolored banded points , and @xmath193 ordered banded points colored by @xmath194 . we define @xmath195 to be this 2-sphere where the @xmath144 uncolored banded points have been colored by @xmath196 ( and the @xmath193 points already colored remain colored ) . we define @xmath197 here @xmath198 is the module for a extended 2-sphere with @xmath144 uncolored banded points colored by @xmath199 and @xmath193 banded points colored by @xmath200 obtained by applying the tqft that we introduced in [ preliminary ] . by an @xmath201-tangle in @xmath191 , we mean a properly embedded framed 1-manifold in @xmath191 with @xmath144 endpoints on @xmath202 , @xmath203 points on @xmath204 , with possibly some black dots on its components and a ( possibly empty ) colored trivalent fat graph ( in the complement of the 1-manifold ) meeting @xmath205 in @xmath193 colored points @xmath206 and meeting @xmath207 in @xmath208 colored points @xmath209 . thus @xmath205 is a 2-sphere with @xmath144 ordered uncolored banded points and @xmath193 colored banded points . similarly @xmath207 is a 2-sphere with @xmath203 ordered uncolored banded points and @xmath208 colored banded points . for any @xmath201-tangle , we will define a homomorphism from @xmath210 to @xmath211 . before doing that , we introduce some definitions . from now on , we will not explicitly mention the banding on the selected points of a surface or the framing of a tangle , or the fattening of a trivalent graph . each comes equipped with such and the framing / fattening of a link / graph induces the banding on its boundary points . nor will we mention the ordering chosen for uncolored sets of points . [ tangle ] suppose we have a @xmath201-tangle in @xmath191 with a colored trivalent graph with @xmath193 edges colored by @xmath206 meeting @xmath212 and @xmath208 edges colored by @xmath209 meeting @xmath213 . suppose we color the @xmath144 endpoints from the tangle on @xmath202 by @xmath214 and color the @xmath203 endpoints from the tangle on @xmath204 by @xmath215 . we say that the coloring @xmath216 is legal if the two endpoints of the same strand have the same coloring . we denote the tangle with the endpoints so - colored by @xmath217 . for an example , see figure [ colorings ] . . in this example , the colored trivalent graph is empty.,title="fig:",width=96,height=120]0.5 in . in this example , the colored trivalent graph is empty.,title="fig:",width=96,height=120 ] [ componenttanglemap ] suppose we have @xmath217 , a @xmath218 colored @xmath201-tangle as in definition [ tangle ] . we define a homomorphism @xmath219 as follows . * if @xmath218 is a illegal coloring . we take the homomorphism to be the zero homomorphism . * if @xmath218 is a legal coloring . we decorate uncolored components of the tangle by some skeins in @xmath90 cases : 1 . if there are @xmath220 black dots on the component , @xmath221 , and the component has two endpoints with color @xmath103 , @xmath222 , then we decorate the component by @xmath223 . if there are @xmath220 black dots on the component , @xmath221 , and the component lies entirely in @xmath224 , then we decorate the component by @xmath225 . + then we apply @xmath226 to @xmath191 with the tangle @xmath227 , so decorated , to get the morphism @xmath228 . now we are ready to define the homomorphism for a tangle @xmath227 . [ tanglemap ] suppose we have a @xmath201-tangle @xmath227 . we define the homomorphism for the tangle , denoted by @xmath229 , to be @xmath230 where @xmath228 is as in definition [ componenttanglemap ] and the sum runs over all colorings @xmath218 . [ functorproposition ] for a tangle @xmath231 in @xmath191 and a tangle @xmath232 in @xmath233 , we have @xmath234 where @xmath235 in @xmath236 means gluing @xmath232 on the top of @xmath231 . here , of course , we assume that the top of @xmath231 and the bottom of @xmath232 agree . this follows from the functoriality of the original tqft . now we can construct tangle endomorphisms for an extended closed @xmath7-manifold with an embedded colored trivalent graph , and choice of infinite cyclic cover . given @xmath138 , we choose a surgery presentation @xmath137 . we put one black dot somewhere on each component of @xmath91 away from @xmath129 . by doing a @xmath2-surgery along @xmath128 , we obtain @xmath237 with link @xmath91 and trivalent graph @xmath114 , where @xmath129 can be completed to @xmath238 for some point @xmath108 on @xmath239 . we cut @xmath240 along @xmath238 . then we obtain a tangle @xmath241 in @xmath233 . here let @xmath243 denote tangle endomorphism associated to @xmath241 . [ blackdots ] if @xmath241 is constructed as above , then the sse class of @xmath243 is independent of the positioning of the black dots . by definition , we can move a black dot on the component of the tangle @xmath241 anywhere without changing the tangle endomorphism @xmath243 . we move the black dot to near bottom or near top and cut the tangle @xmath241 into two tangles @xmath192 and @xmath179 , where @xmath179 is a trivial tangle with the black dot . for an example , see figure [ blackdot ] . then we switch the position of @xmath192 and @xmath179 and move the black dot in resulting tangle to near the other end of that component . then we do the process again . by doing this , we can move it to any arc of the tangle @xmath241 , which belongs to the same component of the link @xmath91 . but for each step , @xmath244 is strong shift equivalent to @xmath245 . therefore , the lemma is true . is the trivial part with the black dot.,title="fig:",width=115,height=115 ] is the trivial part with the black dot.,title="fig:",width=28,height=115 ] is the trivial part with the black dot.,title="fig:",width=115,height=115 ] thus the sse class of the tangle endomorphism @xmath243 constructed as above depends only on a surgery presentation @xmath137 . thus we can denote this class by @xmath246 . [ main ] let @xmath247 and @xmath248 be two surgery presentations for @xmath138 , an extended closed @xmath7-manifold with an embedded colored trivalent graph , and choice of infinite cyclic cover . then @xmath249 thus we may denote this sse class by @xmath250 . this theorem will be proved in section [ pmt ] , after the way has been prepared in sections [ tvm ] and , [ relation ] . in [ introduction ] , we introduced the basic idea of the turaev - viro endomorphism . in this section , we will include the technical details . the discussion in this section and the next section works for @xmath7-manifolds with an embedded @xmath108-admissibly colored trivalent graph . for simplicity , we usually omit mention of the trivalent graph . thus we will write @xmath251 instead of @xmath252 . this is according to the philosophy that we should think of the colored trivalent graph @xmath135 as simply some extra structure on @xmath8 . [ lagrangianchange ] let @xmath253 be an extended cobordism from @xmath254 to itself and @xmath255 be an extended cobordism from @xmath256 to itself . then @xmath257 is strong shift equivalent to @xmath258 . first we notice that @xmath259 then we have @xmath260 here we consider @xmath261 as a cobordism from @xmath256 to @xmath254 . then by the functoriality of @xmath226 , we have the conclusion . [ surfacechange ] suppose we have a closed extended 3-manifold @xmath133 with an infinite cyclic covering . we obtain two extended fundamental domains @xmath262 and @xmath263 by slicing along two extended surfaces @xmath33 and @xmath85 which are dual to @xmath12 . we obtain two morphisms @xmath264 with weight @xmath265 respectively such that the closures of both cobordism having weight @xmath37 . then @xmath266 we just need prove the case where @xmath13 and @xmath67 are disjoint from each other . see ( * ? ? ? * proof of theorem 8.2 ) , @xcite . since @xmath67 is disjoint from @xmath13 , we can choose a copy of @xmath85 inside @xmath267 . we cut along @xmath67 and get two @xmath7-manifolds @xmath268 . we assign to @xmath268 extended @xmath7-manifold structures , denoted by @xmath269 and @xmath270 , such that if we glue @xmath271 to @xmath192 along @xmath67 , we get @xmath267 back . we need to choose appropriate weights @xmath272 for @xmath268 . using definition [ extendedgluing ] , we see that such @xmath272 exists . now we just need prove that if we glue @xmath192 to @xmath179 along @xmath13 , we obtain @xmath273 . actually , it is easy to see that after gluing , we have the right base manifold and lagrangian subspace . what we need to prove is that we get the right weight . this follows from lemma [ closure ] . as a consequence of the two lemmas above , we have the following : [ invariant ] for a tuple @xmath133 and @xmath274 given as in lemma [ surfacechange ] , the strong shift equivalent class of the map @xmath275 is independent of the choice of the extended surface @xmath276 . thus we may denote this sse class by @xmath277 . next , we work towards constructing a fundamental domain for an extended @xmath7-manifold @xmath133 with an infinite cyclic covering . suppose we have a surgery presentation @xmath132 in standard form for @xmath133 , here @xmath278 ( * ? ? ? * lemma(2.2 ) ) . we do @xmath2-surgery along @xmath128 and get a link @xmath91 in @xmath279 . we cut @xmath280 along the 2-sphere containing @xmath129 in this product structure and obtain a tangle @xmath179 in @xmath233 in standard form . here , we say that a tangle is in standard form if it comes from slicing a surgery presentation diagram in standard form . then we drill out tunnels along arcs which meet the bottom and glue them back to the corresponding place on the top . we obtain a cobordism @xmath281 from @xmath282 to itself with a link @xmath283 embedded in it as in figure [ fundamentaldomain ] , where @xmath282 is a genus @xmath284 closed surface . see ( * ? ? ? * figure 3 ) for example . moreover , we identify @xmath282 with a standard surface as pictured in figure [ standardsurface ] . we denote by @xmath285 the lagrangian subspace spanned by the curves labelled by @xmath286 in figure [ standardsurface ] . we assign the lagrangian subspace @xmath285 to each connected component of the boundary of @xmath287 . moreover , we assign the weight @xmath288 to it . thus we obtain an extended cobordism @xmath289 . containing a framed link @xmath290 . if we do extended surgery along @xmath290 , we get a fundamental domain @xmath14 . if , instead , we color @xmath290 by @xmath291 , we obtain another cobordism @xmath292.,width=115,height=115 ] [ weight ] the closure of @xmath293 is @xmath294 , where @xmath104 is a @xmath2-framed unknot . it is easy to see that the closure of @xmath287 is @xmath295 . then we just need to prove that the weight of the closure is @xmath288 . by the gluing formula and definition [ selfgluing ] , we have that the weight on @xmath295 is @xmath296}(\sigma_g\cup\bar{\sigma}_g ) ) . \notag\ ] ] now let @xmath297 then @xmath298}(\sigma_g\cup\bar{\sigma}_g)&=&i_{\sigma_g\cup\bar{\sigma}_g,\sigma_g\times[0,1]}^{-1}(0 ) \notag\\ & = & < ( a_i ,- a'_i),(b_i ,- b'_i)\mid i=1,\cdots , g>. \notag\\ \lambda(\sigma_g\cup\bar{\sigma}_g)&=&\lambda_{\mathcal{a}}\oplus\lambda_{\mathcal{a } } \notag\\ & = & \{(x , y)\mid x\in < a_1,\cdots , a_g > , y\in < a'_1,\cdots , a'_g>\ } \notag\end{aligned}\ ] ] so @xmath299 therefore , @xmath300}(\sigma_g\cup\bar{\sigma}_g)\cap[\lambda(\sigma_g\cup\bar{\sigma}_g)+\lambda_{e}(\sigma_g\cup\bar{\sigma}_g ) ] \notag\\ & = & < ( a_i ,- a'_i)\mid i=1,\cdots , g>. \notag\end{aligned}\ ] ] it is easy to see that the bilinear form defined in @xcite is identically @xmath2 on @xmath301 . so we have @xmath302}(\sigma_g\cup\bar{\sigma}_g))=0 . \notag\ ] ] then we get the conclusion . let @xmath303 be the result of extended surgery along the embedded link @xmath283 in @xmath293 constructed as above starting with a standard surgery presentation diagram for @xmath133 . @xmath303 is a fundamental domain for @xmath133 . the closure of @xmath303 can be obtained by performing extended surgery on the closure of @xmath293 . this uses the commutative property of gluing discussed in @xcite . thus the closure of @xmath14 is diffeomorphic to @xmath8 , and by ( * ? ? ? * lemma 2.2 ) , we see that the closure of @xmath303 has weight @xmath37 . . let @xmath304 be the extended cobordism obtained by coloring the link @xmath283 in @xmath293 by @xmath291 . the sse class @xmath277 is given by @xmath305 the equality @xmath306 follows from the surgery axiom ( * ? ? ? * lemma 11.1 ) for extended surgery . in this section , we will prove the following theorem . [ tvist ] if @xmath133 is an extended @xmath7-manifold with an infinite cyclic covering having a surgery presentation @xmath132 in standard form , then @xmath307 . for simplicity , we indicate the proof in case that @xmath133 does not have a colored trivalent graph . the argument may easily be adapted to the more general case . we obtain a tangle @xmath308 from the surgery presentation @xmath132 , and we place black dots on segments in the top part . we will directly compute two matrices for these two endomorphisms with respect to some bases . * step 1 : compute the entry for the turaev - viro endomorphism . * we will use the basis in @xcite for @xmath309 , where @xmath282 is genus @xmath284 surface . specifically we choose our spine to be a lollipop graph , as in @xcite . we show one example of elements as in figure [ basis1 ] . constructed in @xcite.,width=134,height=115 ] using the method employed in @xcite , we can compute the entries of the matrix , with respect to this basis by computing the quantum invariants of colored links in a connected sum of @xmath310 s . we have @xmath311 where @xmath312 as in figure [ tventry ] and @xmath313 as in figure [ js ] . is consisted of components colored with @xmath291.,title="fig:",width=96,height=115]0.5 in is consisted of components colored with @xmath291.,title="fig:",width=96,height=115 ] is consisted of components colored with @xmath291.,title="fig:",width=96,height=76]0.5 in is consisted of components colored with @xmath291.,title="fig:",width=96,height=76 ] by using fusion and lemma 6 in @xcite and the fact that @xmath314 we have @xmath315 where @xmath316 is the @xmath2-framing unknot colored with @xmath291 . * step 2 : compute the entry for tangle endomorphism . * which do not automatically vanish under @xmath317 have this form.,width=134,height=134 ] by gluing the tangle in @xmath233 to the basis element in figure [ basis4 ] , we can see that @xmath318 for @xmath319 a legal coloring , and is zero otherwise . * step 3 : the two matrices are strong shift equivalent . * by above discussion , it is easy to see that if the matrix for turaev - viro endomorphism is @xmath156 , then a matrix for tangle endomorphism is the block matrix @xmath320 we see that this block matrix is strong shift equivalent to @xmath156 by proposition [ ker ] . [ transforminvariant ] the transformation process in proposition [ standardization ] does not change the strong shift equivalent class of the tangle endomorphism . a small extended kirby-@xmath21 move adds a @xmath321 framed @xmath291 to all the different decorations of @xmath241 which go into the definition of @xmath243 . this would seem to multiply @xmath243 by @xmath322 . but a small extended kirby-@xmath21 move also changes @xmath93 by @xmath323 and thus changes the weight @xmath288 of @xmath324 by @xmath325 . these two effects of the move cancel out and @xmath243 is unchanged . the small kirby-@xmath90 moves preserves all the summands of @xmath243 , by a well known handle slide property of @xmath326 see ( * ? ? ? * lemma 21 ) for instance . two tangles related by a @xmath129 move are obtained by cutting @xmath280 along two different @xmath327 s . suppose if we cut @xmath280 along @xmath328 , we obtain a tangle @xmath308 . if we cut along @xmath329 , we obtain a tangle @xmath330 . by those two cutting , we obtain two homomorphisms @xmath331 and @xmath332 now suppose we cut @xmath280 along @xmath333 and @xmath334 , we get a @xmath335-tangle in @xmath336 , denoted by @xmath231 , and a @xmath201-tangle , denoted by @xmath232 . @xmath231 defines a homomorphism @xmath337 and @xmath232 defines a homomorphism @xmath338 it is easy to see that @xmath339 and @xmath340 therefore , @xmath317 is strong shift equivalent to @xmath341 [ tv ] suppose we have two surgery presentations @xmath247 and @xmath248 for @xmath342 in standard form , then @xmath343 this easily follows from propositions [ tvist ] . by proposition [ standardization ] , and lemma [ transforminvariant ] , we can transform @xmath247 and @xmath248 so that they are standard without changing the sse class of their induced tangle endomorphism . then the result follows from lemma [ tv ] . in this section , we assume , for simplicity , that @xmath108 is odd . similar formulas could be given for @xmath108 even , by the same methods . we let @xmath344 denote the bracket evaluation of a knot diagram of @xmath0 with zero writhe colored @xmath345 at a primitive @xmath31th root of unity @xmath94 . letting @xmath104 denote the unknot , we have that @xmath346 . in particular , @xmath347 . this is one normalization of the colored jones polynomial at a root of unity . [ sym ] using ( * ? ? ? * lemma 6.3 ) , we have that : @xmath348 without losing information , we can restrict our attention to @xmath349 for @xmath350 for other @xmath351 , @xmath352 for some @xmath353 using the above equations . let @xmath354 denote 0-framed surgery along an oriented knot @xmath0 in @xmath5 decorated with a meridian to @xmath0 colored @xmath345 and a longitude little further away from @xmath0 colored @xmath355 and equipped with the weight zero . let @xmath12 be the homomorphism from @xmath356 to @xmath15 which sends a meridian to one . let @xmath357 denote the sse class of the turaev - viro endomorphism @xmath358 . the vector space associated to a 2-sphere with just one colored point which is colored by an odd number is zero . using this fact , and a surgery presentation , one sees that @xmath359 the second author studied @xmath360 @xcite . the idea of adding the longitude with varying colors is due to viro @xcite . the least interesting case , of this next theorem , when @xmath361 already appeared in ( * ? ? ? * corollary 8.3 ) . [ coloredjonesequaltracesum ] for @xmath362 , @xmath363 one has that @xmath2-framed surgery along @xmath0 with the weight zero is the result of extended surgery of @xmath5 with weight zero along a zero - framed copy of @xmath0 . if we add then a zero - framed meridian of @xmath0 to this framed link description , we undo the surgery along @xmath0 and we get back an extended surgery description of @xmath5 , also with weight zero . a longitude to @xmath0 colored @xmath355 and placed a little outside the meridian will go to a longitude of @xmath0 colored @xmath355 in @xmath5 , which is of course isotopic to @xmath0 . but adding a zero - framed meridian to the framed link changes @xmath364 in the same way as cabling by @xmath365 . if we cable the meridian of @xmath0 by @xmath366 instead of by @xmath291 , and calculate @xmath364 , we get @xmath367 by the trace property of tqft @xcite . thus @xmath368 dividing by @xmath113 yields the result . thus the colored jones is determined by the traces of the @xmath369 the next theorem shows that the @xmath370 determine the traces of the @xmath371 . [ traceequalcoloredjonessum ] for @xmath372 , @xmath373 more generally : @xmath374 . by the trace property of tqft , @xmath375 direct calculation of @xmath376 from the definition yields @xmath113 times the bracket evaluation of @xmath0 cabled by @xmath291 together with the meridian colored @xmath377 and the longitude further out colored @xmath378 . these skeins all lie in a regular neighborhood of @xmath0 with framing zero . these skeins can then be expanded as a linear combination of the core of this solid torus with different colors . the operation of encircling an arc colored @xmath379 with loop colored @xmath378 in the skein module of a local disk has the same effect as multiplying the arc by @xmath380 by ( * ? ? ? * lemma 14.2 ) . note the idempotents @xmath381 are only defined for @xmath382 it is well known that the @xmath101 satisfy a recursive formula which can be used to extend the definition of @xmath101 for all @xmath383 . this is given @xcite as follows : @xmath384 , @xmath385 is the zero framed core of a solid standard solid torus , and @xmath386 . in the skein module of a solid torus , we have @xmath387 . using these rules , the expansion can be worked out to be @xmath388 the second equation is worked out in a similar way . notice that , in the summation on the right of the first equation in theorem [ trace ] , @xmath389 for @xmath390 sometimes appears . this can be rewritten using remark [ sym ] as @xmath391 for @xmath392 . we remark that using ( * ? ? ? * corollary 2.8 ) , one can see that the turaev - viro polynomials of @xmath357 will have coefficients in a cyclotomic ring of integers , if @xmath108 is an odd prime or twice an odd prime . in this section , we wish to illustrate with some concrete examples how to calculate the @xmath357 using tangle morphisms in the case @xmath393 ( which is the first interesting case ) . for both examples , we check our computation against an identity from the previous section . the first example is the @xmath103-twist knot with meridian colored @xmath2 or @xmath90 and longitude colored @xmath90 . we then verify directly the equation in theorem [ coloredjonesequaltracesum ] for the case @xmath393 , @xmath394 , and @xmath0 is the @xmath103-twist knot . the second example we study is the knot @xmath395 with the meridian and longitude uncolored . we work out , using tangle morphisms , the traces of the turaev - viro endomorphism . we then verify the equation in theorem [ traceequalcoloredjonessum ] when @xmath393 , @xmath396 , and @xmath397 . we pick an orthogonal basis for the module associated with a 2-sphere with some points , and use this basis to work out the entries on the matrix for the tangle endomorphism coming from a surgery presentation . the bases are represented by colored trees in the 3-ball which meet the boundary in the colored points as in figure [ basis4 ] . here we will refer to these colored trees as basis - trees . each entry is obtained as a certain quotient . the numerator is the evaluation as a colored fat graph in @xmath5 obtained from the tangle closed off with the source basis - tree at the bottom and the target basis - tree at the top . the denominator is the quotient as the evaluation of the double of the target basis element . in both examples , we use a surgery presentation , with one surgery curve with framing @xmath106 . thus the initial weight of @xmath5 , denoted @xmath288 above , should be @xmath148 , so the weight of @xmath5 after the surgery is zero . this puts a factor of @xmath398 in front of the tangle endomorphism . there is also a uniform factor of @xmath113 coming from the single black dot on a strand with two endpoints . we put this total factor of @xmath399 in front . we also have @xmath400 prefactors where @xmath345 is the color of the strand with the black dot , and these factor vary from entry to entry . to simplify our formulas , when @xmath393 , we use @xmath401 to abbreviate @xmath402 , @xmath403 to abbreviate @xmath404 and @xmath405 to denote @xmath406 . a tangle @xmath179 for the @xmath103-twist knot with meridian and longitude is given in figure [ double ] . -twist knot with meridian and longitude . the straight line is from the meridian and the circle is from the longitude . we have also chosen a position for the black dot.,width=115,height=115 ] if we denote by @xmath407 the tangle @xmath179 with meridian colored by @xmath2 and longitude colored by @xmath90 , and let @xmath192 denote a 2-sphere with two uncolored points , then we obtain a map @xmath408 by using the trivalent graph basis in @xcite , @xmath409 where @xmath410 are as in figure [ basis0 ] . where @xmath192 is a 2-sphere with two uncolored points , title="fig:",width=76,height=48]0.5 in where @xmath192 is a 2-sphere with two uncolored points , title="fig:",width=76,height=48 ] with respect to this basis , we have @xmath411 we follow the convention that the columns of the matrix for a linear transformation with respect to a basis are given the images of that basis written in terms of that basis . the characteristic polynomial of this matrix ( i.e. the turaev - viro polynomial ) has coefficients in @xmath412.$ ] if we denote by @xmath232 the tangle @xmath179 with meridian colored by @xmath90 and longitude colored by @xmath90 , and let @xmath192 denote a 2-sphere with two uncolored points and one point colored @xmath90 , then we obtain a map @xmath413 by using the trivalent graph basis in @xcite , @xmath414 where @xmath415 are as in figure [ basis3 ] . where @xmath192 is a 2-sphere with two uncolored points and one point colored 2,title="fig:",width=76,height=48]0.2 in where @xmath192 is a 2-sphere with two uncolored points and one point colored 2,title="fig:",width=76,height=48]0.2 in where @xmath192 is a 2-sphere with two uncolored points and one point colored 2,title="fig:",width=76,height=48 ] with respect to this basis , we have @xmath416 by proposition [ ker ] , we also have @xmath417 this last expression lies in @xmath412 $ ] for all @xmath103 . one has that : @xmath418 moreover , we used recoupling theory as in @xcite to calculate the @xmath90-colored jones polynomial of @xmath103-twist knot directly to obtain : @xmath419 we used mathematica to verify that the two calculations agree for all @xmath103 . in this section , we will compute the turaev - viro endomorphism and quantum invariant of @xmath420 when @xmath94 is a primitive @xmath421th root of unity and verify that the trace of the turaev - viro endomorphism equals to quantum invariant . by @xmath395 , we mean the knot as pictured in @xcite , which is the mirror image of the knot as pictured in @xcite . a tangle @xmath179 for @xmath420 is as in figure [ sixtwo ] . , with a choice for the position for the black dot.,width=96,height=115 ] so we obtain a map @xmath422 where @xmath192 is a 2-sphere with four uncolored points . we use a trivalent graph basis in @xcite for @xmath423 as in figure [ basis ] . , where @xmath192 is a 2-sphere with four uncolored points.,width=288,height=384 ] with respect to this basis , we can obtain a @xmath424 matrix , which is in the strong shift equivalence class of the turaev - viro endomorphism . however , by proposition [ ker ] applied twice in succession , it is enough to consider the minor given by the first five rows and columns . we thus obtain a @xmath425 matrix : @xmath426 the turaev - viro polynomial ( at @xmath393 ) is the characteristic polynomial of the above matrix , namely : @xmath427 we also note that @xmath428 the left hand side of the first equation in theorem [ traceequalcoloredjonessum ] , with @xmath396 , and @xmath429 is by definition , the quantum invariant of @xmath420 . the right hand side is , by direct computation : @xmath430 therefore , we verify a case of the first equation in theorem [ traceequalcoloredjonessum ] . the second author was partially supported by nsf - dms-0905736 . the first author was partially supported by a research assistantship funded by this same grant . we would like to thank the referee for useful suggestions for improving the exposition . t. ohtsuki , _ equivariant quantum invariants of the infinite cyclic covers of knot complements _ , intelligence of low dimensional topology 2006 , 253 - 262 , ser . knots everything 40 , world sci . hackensack , nj , 2007 .

maximum likelihood estimation in statistics leads to the problem of maximizing a product of powers of polynomials . we study the algebraic degree of the critical equations of this optimization problem . this degree is related to the number of bounded regions in the corresponding arrangement of hypersurfaces , and to the euler characteristic of the complexified complement . under suitable hypotheses , the maximum likelihood degree equals the top chern class of a sheaf of logarithmic differential forms . exact formulae in terms of degrees and newton polytopes are given for polynomials with generic coefficients .

math0406533

in algebraic statistics @xcite , a _ model for discrete data _ is a map @xmath0 whose coordinates @xmath1 are polynomial functions in the parameters @xmath2 . the parameter vector @xmath3 ranges over an open subset @xmath4 of @xmath5 such that @xmath6 lies in the positive orthant @xmath7 . the image @xmath8 represents a family of probability distributions on an @xmath9-element state space , provided we make the extra assumption that @xmath10 is the zero polynomial . a given data set is a vector @xmath11 of positive integers . the problem of _ maximum likelihood estimation _ is to find parameters @xmath3 which best explain the data @xmath12 . this leads to the following optimization problem : @xmath13 under suitable assumptions we have an optimal solution @xmath14 to the problem ( [ mle ] ) , which is an algebraic function of the data @xmath12 . our goal is to compute the degree of that algebraic function . we call this number the _ maximum likelihood degree _ of the model @xmath15 . equivalently , the _ ml degree _ is the number of complex solutions to the critical equations of ( [ mle ] ) , for a general data vector @xmath12 . in this paper we prove results of the following form : [ thm1 ] let @xmath1 be polynomials of degrees @xmath16 in @xmath17 unknowns . if the maximum likelihood degree of the model @xmath18 is finite then it is less than or equal to the coefficient of @xmath19 in the generating function @xmath20 equality holds if the coefficients of the polynomials @xmath21 are sufficiently generic . as an example , consider a model given by @xmath22 quadratic polynomials in @xmath23 parameters . the solution to ( [ mle ] ) satisfies the two critical equations @xmath24 if the @xmath21 s are general quadrics then these equations have @xmath25 complex solutions . the formula for the maximum likelihood degree in theorem [ thm1 ] gives @xmath26 for special quadrics @xmath21 , the ml degree can be much lower than @xmath25 . a familiar example is the _ independence model for two binary random variables _ : @xmath27 here the ml degree is only one because the maximum likelihood estimate @xmath14 is a rational function (= algebraic function of degree one ) of the data @xmath12 : @xmath28 this paper is organized as follows . in section 2 we present the algebraic geometry for studying critical points of a rational function @xmath29 on an irreducible projective variety @xmath30 . the critical equations @xmath31 are interpreted as sections of the sheaf @xmath32 of @xmath33-forms with logarithmic singularities along the divisor @xmath34 defined by @xmath35 . in theorem [ chernclassthm ] , we show that if @xmath34 is a global normal crossing divisor then the ml degree equals the degree of the top chern class of @xmath36 . if @xmath30 is projective @xmath17-space then this leads to theorem [ mle ] . in section 3 we study the case when @xmath30 is a smooth toric variety , and we derive a formula for the ml degree when the @xmath21 s are laurent polynomials which are generic relative to their newton polytopes . for instance , example [ rectangles ] shows that the ml degree is @xmath37 if we replace ( [ 2by2table ] ) by @xmath38 section 4 is concerned with the relationship of the ml degree to the bounded regions of the complement of @xmath39 in @xmath5 . the number of these regions is a lower bound to the number of real solutions of the critical equations , and therefore a lower bound to the ml degree . we show that for plane quadrics all three numbers can be equal . however , for other combinations of plane curves the ml degree and the number of bounded regions diverge , and we prove a tight upper bound on the latter in theorem [ thm : viro ] . also , following work of terao @xcite and varchenko @xcite , we show in theorem [ thm - linear ] that the ml degree coincides with the number of bounded regions of the arrangement of hyperplanes @xmath40 when the @xmath21 s are ( not necessarily generic ) linear forms . section 5 revisits the ml degree for toric varieties , replacing the smoothness assumption by a much milder condition . theorem [ toricformula4 ] gives a purely combinatorial formula for the ml degree in terms of the newton polytopes of the polynomials @xmath21 . this section also discusses how resolution of singularities can be used to compute the ml degree for nongeneric polynomials . section 6 deals with topological methods for determining the ml degree . theorem [ euler ] shows that , under certain restrictive hypotheses , it coincides with the euler characteristic of the complex manifold @xmath41 , and theorem [ semiconti ] offers a general version of the semi - continuity principle which underlies the inequality in theorem [ thm1 ] . in section 7 we relate the ml degree to the sheaf of logarithmic vector fields along @xmath34 , which is the sheaf dual to @xmath42 . this paper was motivated by recent appearances of the concept of ml degree in statistics and computational biology . chor , khetan and snir @xcite showed that the ml degree of a phylogenetic model equals @xmath43 , and geiger , meek and sturmfels @xcite proved that an undirected graphical model has ml degree one if and only if it is decomposable . the notion of ml degree also makes sense for certain parametrized models for continuous data : drton and richardson @xcite showed that the ml degree of a gaussian graphical model equals @xmath44 , and bout and richards @xcite studied the ml degree of certain mixture models . the ml degree always provides an upper bound on the number of local maxima of the likelihood function . our ultimate hope is that a better understanding of the ml degree will lead to the development of custom - tailored algorithms for solving the critical equations @xmath45 . there is a need for such new algorithms , given that methods currently used in statistics ( notably the _ em - algorithm _ ) often produce only local maxima in ( [ mle ] ) . in this section we work in the following general set - up of algebraic geometry . let @xmath30 be a complete factorial algebraic variety over the complex numbers @xmath46 . we also assume that @xmath30 is irreducible of dimension @xmath47 . in applications to statistics , the variety @xmath30 will often be a smooth projective toric variety . suppose that @xmath48 is a rational function on @xmath30 . since @xmath30 is factorial , the local rings @xmath49 are unique factorization domains . this means that the function @xmath35 has a global factorization which is unique up to constants : @xmath50 here @xmath51 is a prime section of an invertible sheaf @xmath52 where @xmath53 is the divisor on @xmath30 defined by @xmath51 . in our applications we usually assume that @xmath54 where @xmath9 is the number considered in the introduction . for instance , if @xmath1 are polynomials and @xmath55 then @xmath56 ; namely , @xmath57 are the homogenizations of @xmath58 using @xmath59 , and @xmath60 ( see the proof of theorem [ thm1 ] for details ) . by ( [ globfac ] ) , we can write the divisor of the rational function @xmath35 uniquely as @xmath61 where the @xmath62 s are ( possibly negative ) integers . let @xmath34 be the reduced union of the codimension one subvarieties @xmath63 , or , as a divisor , @xmath64 . we are interested in computing the critical points of the rational function @xmath35 on the open set @xmath65 complementary to the divisor @xmath34 . especially , we wish to know the number of critical points , counted with multiplicities . a critical point is by definition a point @xmath66 where the differential @xmath33-form @xmath67 vanishes . if @xmath68 is a smooth point on @xmath30 , and @xmath69 are local coordinates , then @xmath70 . hence @xmath68 is a critical point of @xmath35 if and only if @xmath71 we next rewrite the critical equations ( [ criticalequations ] ) using the factorization ( [ globfac ] ) . around each point @xmath66 , we may choose a local trivialization for the sheaf @xmath52 and express @xmath51 locally by a regular function . by slight abuse of notation , we denote that regular function also by @xmath51 . for instance , if @xmath55 then this means replacing the homogeneous polynomial @xmath51 by a dehomogenization . since @xmath35 has neither zeros nor poles on the open set @xmath72 , the vanishing of @xmath73 is equivalent to the vanishing of the logarithmic derivative @xmath74 we now recall some classical definitions and results concerning the sheaf of differential @xmath33-forms with logarithmic singularities along @xmath34 . the standard references on this subject are dligne s book @xcite and saito s paper @xcite . we define @xmath75 as a subsheaf of the sheaf @xmath76 of @xmath33-forms with poles at most on @xmath34 and of order one . this sheaf is the image of the natural map @xmath77 which is given by the inclusion @xmath78 and the homomorphisms sending @xmath79 . for experts we note that our definition differs from the one in @xcite when @xmath34 is not normal crossing . saito s sheaf is the double dual of our @xmath80 , which explains why his is always locally free when @xmath30 is a surface ( * ? ? ? * corollary 1.7 ) . ours need not be locally free even for surfaces . however , our definition gives a natural exact sequence . [ exactlem ] if @xmath30 is factorial and complete then we have an exact sequence @xmath81 the local sections of the sheaf @xmath82 are rational @xmath33-forms which can be written as @xmath83 , where @xmath84 is a regular @xmath33-form . ssince the @xmath85 s are distinct prime divisors and @xmath30 is factorial , the local rings @xmath86 are discrete valuation rings with parameter @xmath51 . thus @xmath87 is invertible in this local ring for @xmath88 , and @xmath89 is regular if and only if @xmath51 divides @xmath90 . this implies that the homomorphism which sends @xmath89 to the vector @xmath91 is well defined , and it induces an isomorphism from the quotient @xmath92 onto @xmath93 . assume now that @xmath30 is smooth . then both sheaves @xmath94 and @xmath95 are locally free of rank @xmath96 . hence the intermediate sheaf @xmath97 is torsion free of the same rank . our next result shows that @xmath97 is locally free if and only if the divisors @xmath85 are smooth and intersect transversally . [ prop1 ] let @xmath66 be a smooth point , @xmath98 local coordinates at @xmath68 and @xmath99 the divisors which contain @xmath68 . then the sheaf @xmath97 is locally free at @xmath68 if and only if the @xmath100-matrix @xmath101 has rank @xmath102 at @xmath68 . any local section of @xmath82 can be written in the form @xmath103 this observation gives rise to a local exact sequence @xmath104 the surjective map on the right takes @xmath105 to the sum on the right hand side of ( [ omegasum ] ) . the injective map on the left takes the @xmath102-tuple @xmath106 to @xmath107 the exactness of the sequence ( [ localexseq ] ) follows from the proof of lemma [ exactlem ] . if the section @xmath89 in ( [ omegasum ] ) is identically zero in @xmath108 then @xmath89 is in particular regular , and so @xmath51 divides each @xmath90 . now , since @xmath30 is reduced , a coherent sheaf @xmath109 is locally free of rank @xmath17 if and only if @xmath110 for each point @xmath68 . since tensor product is right exact , it follows that this condition is verified for @xmath97 if and only if the matrix of @xmath111 , evaluated at @xmath68 , has rank precisely @xmath102 . since the functions @xmath112 vanish at @xmath68 , this is exactly the asserted condition that the jacobian marix @xmath113 has rank @xmath102 at @xmath68 . in the above situation where @xmath30 is smooth and @xmath75 is locally free we shall say that the divisor @xmath34 has _ global normal crossings _ ( or gnc ) . [ chernclassthm ] let @xmath30 be smooth and assume that @xmath34 is a gnc divisor . then 1 . the section @xmath114 of @xmath75 does not vanish at any point of @xmath34 , 2 . if the divisor @xmath34 intersects every curve in @xmath30 ( in particular , if @xmath34 is ample ) then @xmath115 vanishes only on a finite subset of @xmath116 , 3 . if the above conclusions hold , then the number of critical points of @xmath35 on @xmath72 , counted with multiplicities , equals the degree of the top chern class @xmath117 we abbreviate @xmath118 . by the proof of proposition [ prop1 ] it follows that if @xmath119 has rank @xmath102 at @xmath68 , then @xmath75 is locally free of rank @xmath17 with generators @xmath120 and some choice of @xmath121 of the @xmath122 . if we write @xmath123 in this basis , the coefficients of @xmath124 are the constants @xmath62 while the coefficients of the @xmath122 are some regular functions . the first assertion follows immediately since the exponents @xmath62 are all nonzero . the second assertion follows from the first : let @xmath125 be the zero set of the section @xmath123 . since @xmath125 does not intersect @xmath34 , it follows that @xmath126 . thirdly , if @xmath109 is a locally free sheaf of rank @xmath17 on a smooth variety @xmath30 of dimension @xmath17 , and @xmath127 is a section of @xmath128 with a zero scheme @xmath125 of dimension @xmath129 , then the length of @xmath130 equals the degree of the top chern class @xmath131 . the _ total chern class _ of a sheaf @xmath109 is the sum @xmath132 . this is a polynomial in @xmath133 whose coefficients are elements in the chow ring @xmath134 . recall that every element in @xmath135 has a well - defined _ degree _ which is the image of its degree @xmath17 part under the degree homomorphism @xmath136 . suppose that @xmath30 is smooth and @xmath34 is a gnc divisor on @xmath30 which intersects every curve . then the number of critical points of @xmath35 , counted with multiplicities , is the degree of the coefficient of @xmath19 in the following polynomial : @xmath137.\ ] ] the total chern class @xmath138 is multiplicative with respect to exact sequences , i.e. , if @xmath139 is an exact sequence of sheaves , then @xmath140 . hence the sequence ( [ lemmasequence ] ) implies the result . in the next section , we apply the formula ( [ totalchernclass ] ) in the case when @xmath30 is a smooth projective toric variety . the chow group @xmath141 has rank one and is generated by the class of any point . this canonically identifies @xmath141 with @xmath142 and so any top chern class can be considered to be a number . [ toricformula ] suppose @xmath30 is a smooth toric variety with boundary divisors @xmath143 and @xmath34 is gnc and meets every curve . the number of critical points of @xmath35 , counted with multiplicity , equals the coefficient of @xmath19 in @xmath144.\ ] ] by virtue of equation we need only compute the total chern class @xmath145 . for this we use the exact sequence in @xcite , @xmath146 where @xmath147 , and the fact that @xmath148 is trivial . we now apply the results of the previous section to models @xmath149 . to illustrate how this works , we first prove theorem [ thm1 ] for generic polynomials . the proof of the statement that the ml degree of generic polynomials is an upper bound on the ml degree of special polynomials ( when this number is finite ) is deferred to theorem [ toricformula3 ] which is a generalization of theorem [ thm1 ] . see also theorem [ semiconti ] where this semi - continuity principle is stated in general . the polynomials @xmath150 are assumed to be generic among all ( nonhomogeneous ) polynomials of degrees @xmath16 in @xmath151 , and @xmath152 are positive integers . we take @xmath30 to be projective space @xmath153 with coordinates @xmath154 . our object of interest is the following rational function on @xmath55 : @xmath155 the global factorization ( [ globfac ] ) of this @xmath156 has @xmath56 prime factors , namely , @xmath157 and @xmath158 with @xmath159 . the chow ring of @xmath55 is @xmath160/ \langle h^{d+1 } \rangle$ ] , where @xmath161 represents the hyperplane class . by our genericity hypothesis , the @xmath162 prime factors of @xmath156 are smooth and global normal crossing . they correspond to the following divisor classes : @xmath163 projective space @xmath153 is a smooth toric variety with @xmath164 torus - invariant divisors @xmath165 , each having the same class @xmath161 . hence the formula in ( [ toricformula2 ] ) specializes to @xmath166 since we work in the chow ring of projective space @xmath153 , the coefficient of @xmath167 is the same as the coefficient of @xmath19 in the generating function in ( [ thm1formula ] ) . we now generalize our results from polynomials of fixed degrees to laurent polynomials with fixed newton polytopes . recall that the _ newton polytope _ of a laurent polynomial @xmath168 is the convex hull of the set of exponent vectors of the monomials appearing in @xmath35 with nonzero coefficient . given a convex polytope @xmath169 with vertices in @xmath170 , by a _ generic laurent polynomial with newton polytope @xmath171 _ we will mean a sufficiently general @xmath172-linear combination of monomials with exponent vectors in @xmath173 . in the next theorem we consider @xmath9 laurent polynomials @xmath174 having respective newton polytopes @xmath175 . because the @xmath21 s are laurent polynomials , i.e. , their monomials may have negative exponents , we only consider those critical points of @xmath176 which lie in the algebraic torus @xmath177 . the number of such critical points ( counted with multiplicity ) will be called the _ toric ml degree _ of the rational function @xmath35 . let @xmath178 denote the _ minkowski sum _ of the given newton polytopes , and let @xmath30 be the projective toric variety defined by @xmath171 . let @xmath179 be the primitive inner normal vectors of the facets of @xmath171 . they span the rays of the fan of @xmath30 . let @xmath180 denote the corresponding torus - invariant divisors on @xmath30 . each of the newton polytopes @xmath181 is the solution set of a system of linear inequalities of the specific form @xmath182 the divisor on @xmath30 defined by the laurent polynomial @xmath21 is linearly equivalent to @xmath183 . the @xmath184 are integers which can be positive or negative . the divisor on @xmath30 defined by @xmath185 is linearly equivalent to @xmath186 we abbreviate the _ support _ of this divisor by @xmath187 a toric variety @xmath30 is smooth if all the cones in its normal fan are unimodular . [ toricformula3 ] if the toric variety @xmath30 is smooth and the toric ml degree of the rational function @xmath35 is finite then it is bounded above by the coefficient of @xmath19 in the following generating function with coefficients in the chow ring of @xmath30 : @xmath188 equality holds if each @xmath21 is generic with respect to its newton polytope @xmath181 . note that theorem [ thm1 ] is the special case of theorem [ toricformula3 ] when @xmath181 is the standard @xmath17-dimensional simplex @xmath189 scaled by a factor of @xmath190 . let us first assume that @xmath21 is a generic laurent polynomial with newton polytope @xmath181 . let @xmath191 $ ] be the homogeneous coordinate ring @xcite of @xmath30 with one variable for each torus - invariant divisor @xmath165 . given a laurent polynomial @xmath192 with newton polytope @xmath181 , the corresponding rational function on @xmath30 is @xmath193 where @xmath85 is as defined above and @xmath51 is homogeneous of degree @xmath85 . therefore the rational function on @xmath30 we are interested in is @xmath194 we next show that the divisor of @xmath156 is gnc . note that @xmath51 is a generic section of a line bundle on @xmath30 that is generated by its sections . this implies , by the bertini - sard theorem and by induction on @xmath9 , that the divisors @xmath195 meet transversally in the dense torus of @xmath30 . for points in the boundary of @xmath30 , we simply restrict to the torus orbit determined by the corresponding facet where the restricted @xmath51 s remain generic sections of the restricted bundles . the reduced divisor of poles and zeros of @xmath156 is @xmath196 + @xmath197 where @xmath198 is defined as in ( [ whatisi ] ) . since @xmath199 is the divisor corresponding to @xmath171 it is ample on @xmath30 by construction . so @xmath199 meets every curve on @xmath30 and therefore so does @xmath34 and we can apply corollary [ toricformula ] . a variable @xmath200 appears as a factor in @xmath156 if and only if @xmath201 , in which case @xmath202 appears in both the numerator and denominator of ( [ toricformula2 ] ) , and we get the expression ( [ toricchern ] ) . consider now arbitrary laurent polynomials @xmath203 in @xmath151 such that @xmath204 has only finitely many critical points in @xmath205 . let @xmath206 be the coefficient of @xmath19 in ( [ toricchern ] ) . let @xmath207 be the space of all @xmath9-tuples of laurent polynomials with the given newton polytopes . consider the critical equations of @xmath208 and clear denominators . the resulting collection of @xmath17 laurent polynomials defines an algebraic subset @xmath209 in the product space @xmath210 . saturate @xmath209 to remove any components along the hypersurfaces @xmath40 and get a new algebraic subset @xmath211 . the map from @xmath211 onto @xmath207 is dominant and generically finite , and the generic fiber of this map consists of @xmath206 points . our given laurent polynomials @xmath203 represent a point @xmath212 in @xmath207 . let @xmath213 be the isolated critical points of @xmath35 . for each @xmath214 , consider any irreducible component @xmath215 of @xmath211 containing the point @xmath216 in @xmath217 . by krull s principal ideal theorem , the component @xmath215 of @xmath211 has codimension @xmath218 and hence it has dimension @xmath219 . as the generic fiber is finite , the dimension of @xmath220 is exactly @xmath221 and the projection to @xmath207 is dominant . since @xmath222 is an isolated solution of the critical equations , the projection map to @xmath207 is open ( * ? ? ? * ( 3.10 ) ) , so the intersection of @xmath215 with an open neighborhood of @xmath216 maps onto an open neighborhood of @xmath212 . hence every generic point @xmath223 near @xmath212 has a preimage @xmath224 near @xmath216 , and these preimages are distinct for @xmath225 . we conclude that @xmath226 . this semicontinuity argument is called the `` specialization principle '' stated in mumford s book ( * ? ? ? * ( 3.26 ) ) and also works when the @xmath222 have multiplicities , as shown in theorem [ semiconti ] below . @xmath227 we illustrate theorem [ toricformula3 ] with two examples which we revisit in section 5 . [ rectangles ] consider @xmath9 generic polynomials @xmath228 where the support of @xmath21 consists of monomials @xmath229 with @xmath230 and @xmath231 , and suppose the @xmath62 s are generic . the newton polytope of @xmath21 is the rectangle @xmath232 the minkowski sum of these rectangles is another rectangle , and @xmath233 . in the numerator of ( [ toricchern ] ) , the contribution of the two torus - invariant divisors @xmath34 and @xmath234 corresponding to the left and the bottom edge of this rectangle survives . the denominator comes from the product of the divisors of @xmath1 : @xmath235 now , the coefficient of the term @xmath236 modulo the chow ring relations @xmath237 gives the toric ml degree @xmath238 [ haseightrays ] let @xmath239 be generic polynomials in @xmath240 and @xmath241 with supports @xmath242 the corresponding newton polytopes @xmath243 are shown in figure [ fig2 ] . the normal fan of the minkowski sum has eight rays and is shown in figure [ fig3 ] . theorem [ toricformula3 ] applies because the toric surface @xmath30 is smooth . we label the eight rays by @xmath244 in counterclockwise order , starting with @xmath245 . the chow ring @xmath246 is the polynomial ring @xmath247 $ ] modulo the ideal @xmath248 the three divisors corresponding to the polygons @xmath249 in figure [ fig2 ] are @xmath250 if all @xmath62 are positive , then the support of the divisor @xmath251 is @xmath252 . it follows that the toric ml degree is the coefficient of @xmath236 in @xmath253 this coefficient is @xmath254 , which means that the toric ml degree is @xmath255 . the _ toric ml degree of the model _ @xmath256 is the toric ml degree defined above for generic @xmath12 . in this case , there is no cancellation among the coefficients in ( [ whatisi ] ) , and @xmath198 is the set of all indices @xmath257 such that for some @xmath181 the supporting hyperplane normal to @xmath258 does not pass through the origin . the toric ml degree of @xmath15 is a numerical invariant of the polytopes @xmath259 . a combinatorial formula for this invariant will be presented in theorem [ toricformula4 ] of section 5 . as in the introduction , we consider @xmath9 polynomials @xmath260 in @xmath17 unknowns @xmath261 . we now assume that all coefficients of the @xmath21 s are real numbers , and we also assume that @xmath152 are positive integers . however , we do not assume that the union of the divisors of the @xmath21 s has global normal crossings . this is the case of interest in statistics . consider the arrangement of hypersurfaces defined by the @xmath21 s and let @xmath262 be the complement of this arrangement . a connected component of @xmath263 is a _ bounded region _ if it is bounded as a subset of @xmath5 . then the following observation holds . [ bounded ] for any polynomial map @xmath264 and any @xmath265 , @xmath266 the function @xmath267 is continuous , and on the boundary of the closure of each bounded region its value is zero . hence it has to have at least one ( real ) critical point in the interior of each region . the second inequality holds trivially , since the ml degree was defined as the number of critical points of @xmath268 in @xmath269 , counted with multiplicities . this observation raises the question whether the inequalities above could be realized as equalities . we next show that this is the case when @xmath150 are quadrics in the plane . here the ml degree is @xmath270 by theorem [ thm1 ] . for each @xmath9 , there are @xmath9 quadrics @xmath150 in @xmath271 such that @xmath272 hence all critical points are real . we will take @xmath9 quadrics that define `` nested '' ellipses with center at the origin , as suggested by figure [ fig1 ] . the proof follows by induction : assume we have @xmath273 bounded regions with @xmath274 ellipses . observe that the @xmath275st ellipse contains @xmath276 bounded regions . then we add a new long and skinny ellipse which replaces the @xmath276 regions with @xmath277 regions . the total count comes out to be @xmath278 . we will see such an equality holding for @xmath9 linear hyperplanes in @xmath5 below . however , even in the plane @xmath271 , the number of critical points and the number of bounded regions of @xmath263 diverge for curves of degree @xmath279 . theorem [ thm1 ] implies that for @xmath9 generic plane curves of degrees @xmath280 the ml degree is @xmath281 the optimal upper bound for the number of bounded regions of @xmath263 is smaller than the ml degree , by the following unpublished result due to oleg viro . [ thm : viro ] ( viro ) let @xmath150 be real plane curves of degrees @xmath280 , and let @xmath282 be the number of odd degree curves among them . then @xmath283 and this bound is optimal . the proof is by induction . for @xmath284 the bound above is harnack s inequality @xcite , and it is optimal . suppose the statement is proved for @xmath274 curves , and @xmath285 defines a curve of degree @xmath286 . we will take this new curve so that it has the maximum number of bounded regions allowed by harnack s inequality , i.e. @xmath287 if @xmath286 is even , and one less than that if @xmath286 is odd . one can achieve @xmath288 by taking a curve with @xmath289 unnested ovals and one more distinguished piece ( that gives an extra bounded region when @xmath286 is even ) such that some line intersects this distinguished piece in exactly @xmath286 points . we can arrange this last curve in such a way that when we superimpose the distinguished piece on the arrangement given by @xmath274 curves , the last curve will intersect the curve given by @xmath21 in @xmath290 points . now if we trace this last distinguished piece , every time we encounter an intersection point an extra bounded region is created , except the last point in case @xmath286 is odd . together with the remaining @xmath289 ovals we get the bound . in order to get any meaningful lower bound on the number of bounded regions of @xmath291 one needs to make some assumptions . without any assumptions the lower bound is zero : for @xmath21 of even degree we take an empty ( real ) curve , and for @xmath21 of odd degree we take the union of an empty curve with a line . if we let all the lines intersect in a single point there will not be any bounded region . if we insist on at least having a gnc configuration , then by the same construction the lower bound we get is the number of bounded regions in a generic arrangement of @xmath282 lines where @xmath282 is the number of odd degrees @xmath190 . this idea leads us to studying the ml degree of a hyperplane arrangement . [ thm - linear ] let @xmath15 be given by @xmath9 linear polynomials @xmath1 with real coefficients . then the ml degree of @xmath15 is equal to the number of the bounded regions of @xmath291 , and all critical points of the optimization problem ( [ mle ] ) are real . this theorem does not assume any hypothesis such as global normal crossing . under the gnc hypothesis , the hyperplanes would be in general position and the number of bounded regions equals @xmath292 , as predicted by theorem [ thm1 ] . theorem [ thm - linear ] is essentially due to varchenko @xcite . we shall give a new proof . in light of proposition [ bounded ] , we need to show that the number of bounded regions of @xmath291 equals the number of complex solutions of the ml equations . let @xmath293 for @xmath294 . the ml equations are @xmath295 consider the map @xmath296 given by @xmath297 . here @xmath298 is the homogenization of @xmath21 . we let @xmath299 be the central hyperplane arrangement in @xmath300 given by the @xmath51 . we assume that the intersection of all the hyperplanes in @xmath299 is just the origin ; otherwise , the linear forms @xmath51 depend on fewer than @xmath17 coordinates , and then we get infinitely many critical points . the zariski closure of @xmath301 in @xmath302 is a @xmath17-dimensional complex variety @xmath303 . the solution set on @xmath303 of the @xmath17 linear equations @xmath304 consists of finitely many points provided @xmath152 are generic . obviously , the solutions to ( [ new - eq ] ) lift to such complex solutions . in other words , the degree of the projective variety @xmath303 is an upper bound on the ml degree of @xmath15 . now we will compute the degree of @xmath303 . this variety is the projective spectrum of the @xmath305-graded algebra @xmath306\,$ ] where @xmath307 . terao ( * ? ? ? * theorem 1.4 ) showed that the hilbert series of @xmath308 is equal to @xmath309 here @xmath310 is the mbius function of the intersection lattice @xmath311 of the arrangement @xmath312 . from this series we shall determine the leading coefficient of its hilbert polynomial . this coefficient has the form @xmath313 , where @xmath314 is the degree of @xmath303 . for large enough @xmath315 , the coefficient of @xmath316 in the hilbert series ( [ terao ] ) equals @xmath317 this is the hilbert polynomial of the graded algebra @xmath308 . its leading term is @xmath318 we conclude that the degree of the projective variety @xmath303 is @xmath319 . by zaslavsky @xcite , this number equals the number of bounded regions of @xmath291 . a family of important statistical models where theorem [ thm - linear ] applies is the _ linear polynomial model _ of @xcite . such a model is given by a polynomial in @xmath315 unknowns @xmath320 with indeterminate coefficients , @xmath321 together with @xmath9 data points @xmath322 . the model is parametrized by @xmath323 the ml degree is the number of bounded regions of this arrangement . we now return to the setting of section 3 , with the aim of relaxing the restrictive smoothness hypothesis in theorem [ toricformula3 ] . our aim is to derive a combinatorial formula for the toric ml degree of any model @xmath324 defined by generic laurent polynomials satisfying a mild hypothesis . the derivation of theorem [ toricformula4 ] involves resolution of singularities in the toric category . in the end of the section we shall comment on using resolution of singularities for bounding the ml degree in general . given a polytope @xmath171 in @xmath5 and a linear functional @xmath325 on @xmath5 , we write @xmath326 for the face of @xmath171 at which @xmath325 attains its minimum . two linear functionals @xmath325 and @xmath327 are _ equivalent _ if @xmath328 . the equivalence classes are the relative interiors of cones of the _ inner normal fan _ @xmath329 . if @xmath123 is a cone in @xmath329 , or @xmath123 is a cone in any fan which refines @xmath329 , then we write @xmath330 for @xmath331 in the relative interior of @xmath123 . if @xmath35 is a polynomial with newton polytope @xmath171 then @xmath332 denotes the _ leading form _ consisting of all terms of @xmath35 which are supported on @xmath333 . as in section 3 , let @xmath1 be laurent polynomials with newton polytopes @xmath334 . consider any fan @xmath335 which is a common refinement of the inner normal fans suppose @xmath337 is a cone in @xmath335 and let @xmath338 be the dimension of @xmath339 . there exists a @xmath338-dimensional linear subspace @xmath340 of @xmath5 and vectors @xmath341 such that @xmath342 lies in @xmath340 for all @xmath343 . the subspace @xmath340 is unique and satisfies @xmath344 . let @xmath345 denote the _ normalized mixed volume _ on the subspace @xmath340 . here `` normalized '' refers to the lattice @xmath346 , as is customary in toric geometry @xcite . for any @xmath338-element subset @xmath347 of @xmath348 we abbreviate @xmath349 and @xmath350 if @xmath351 . if @xmath352 and @xmath353 then we simply write @xmath354 for the mixed volume in ( [ mixedvolume ] ) . if @xmath355 and @xmath356 is full - dimensional then ( [ mixedvolume ] ) equals @xmath33 ; this happens in the last sum of ( [ longmultline ] ) . we are now ready to state our more general toric ml degree formula . as in section 3 , let @xmath30 be the toric variety corresponding to the minkowski sum @xmath357 and @xmath358 the normal fan with rays @xmath359 . we consider the function @xmath360 . each polytope @xmath181 corresponds to a divisor @xmath85 so the divisor of @xmath35 is @xmath361 . let @xmath198 be the support of @xmath34 as in . label the rays of @xmath358 so that @xmath362 are the indices not in @xmath198 . for each subset @xmath363 of @xmath364 let @xmath365 denote the smallest cone of @xmath335 which contains the vectors @xmath366 for @xmath367 . if no such cone exists then @xmath365 is just a formal symbol and the expression ( [ mixedvolume ] ) is declared to be zero for @xmath368 . the mild smoothness hypothesis we need is that every singular cone of @xmath335 contains at least one ray from @xmath198 . equivalently all cones @xmath365 are smooth . [ toricformula4 ] suppose every singular cone of @xmath358 contains some ray in the support of the divisor @xmath34 . then , the toric ml degree of the rational function @xmath35 is bounded above by the following alternating sum of mixed volumes : @xmath369 equality holds if each @xmath21 is generic relative to its newton polytope @xmath181 . in order to apply corollary [ toricformula ] we must resolve the singularities of @xmath30 . for toric varieties this is done in two steps . first we get a simplicial toric variety without adding any new rays to the fan . second we resolve the remaining singular ( but simplicial ) cones by adding new rays . this procedure is described in detail in @xcite . typically the first step involves taking the pulling subdivision at each ray in the fan . however , under the given hypothesis it is enough to perform pulling subdivisions only at the rays in the support of @xmath34 to obtain a simplicial fan @xmath370 . this fine detail will be important below . our hypothesis holds for this intermediate fan as well , and subsequently we take a smooth refinement @xmath371 of @xmath370 by adding new rays in the relative interiors of each of the singular cones . let @xmath372 be the induced map . we will show that we get no new critical points under the resolution . hence the number of critical points can be computed on @xmath373 . we finally claim that the chern class formula expands into the given combinatorial formula . we investigate critical points of the pullback of our rational function : @xmath374 for generic @xmath21 , the same argument as in the proof of theorem [ toricformula3 ] shows that the reduced divisor of poles and zeros @xmath375 of @xmath376 is gnc . what we must show is that all critical points of @xmath376 on @xmath373 are off the exceptional locus hence actually critical points of @xmath156 on @xmath30 . there are two types of new cones in @xmath371 . the first come from the triangulation step . by our construction , any such cone must contain a ray @xmath258 in the support of @xmath34 . this ray corresponding to the strict transform under @xmath377 is in the support of @xmath378 , and its variable appears as a factor in @xmath376 . by part 1 of theorem [ chernclassthm ] , @xmath376 has no critical points along the torus - invariant divisor @xmath379 , hence no critical points on any torus orbit contained in @xmath379 . the second type of new cone comes from the desingularization step . these cones all contain at least one new ray @xmath380 corresponding to an exceptional divisor @xmath381 in @xmath373 . we will show there are no critical points on @xmath381 . equivalently we show that there are no critical points on each torus orbit contained in @xmath381 . given a torus orbit let @xmath382 be the corresponding cone of @xmath383 containing @xmath380 . there is some minimal cone @xmath337 of @xmath335 containing @xmath382 . let @xmath384 be any cone of @xmath383 containing @xmath382 that is maximal with respect to being contained in @xmath337 . since @xmath337 is refined in @xmath383 it must be a singular cone , and so by the hypothesis it has some generating ray in the support of @xmath385 , or equivalently the linear function of this cartier divisor is not identically zero on @xmath337 . the pullback keeps the same linear functional which can not be zero on the subset @xmath384 of @xmath337 . as a consequence , @xmath384 contains a ray @xmath258 in the support of @xmath378 . this means @xmath386 appears as a factor in @xmath376 for some nonzero integer @xmath387 . if @xmath258 is a generator of @xmath382 , then as above there are no critical points on @xmath379 and thus no critical points on the orbit corresponding to @xmath382 . suppose on the other hand @xmath258 is not a generator of @xmath382 . let @xmath388 be the variables corresponding to the generators of @xmath382 in an affine chart of @xmath373 that contains @xmath384 . note that the variable @xmath200 corresponding to @xmath258 is not among these variables . because @xmath337 is the minimal cone of @xmath335 containing @xmath382 , the face @xmath389 is contained in the face @xmath390 , and hence it is contained in the face @xmath391 . so @xmath392 , obtained by setting @xmath388 to zero , does not contain any of the variables corresponding to @xmath384 ; in particular it does nt contain @xmath200 . on the other hand , @xmath393 where @xmath394 is in the ideal generated by @xmath388 . since @xmath200 is not among the @xmath395 we have @xmath396 where @xmath397 means @xmath398 . hence we have @xmath399 we conclude that @xmath376 has no critical points on the torus orbit corresponding to @xmath382 as desired . thus the toric ml degree of @xmath35 on @xmath373 is the same as that on @xmath35 . since @xmath376 has no critical points on the exceptional locus and @xmath34 is ample on @xmath30 , @xmath400 meets any curve off the exceptional locus and therefore the ml degree must be finite . it is computed in the chow ring of @xmath373 as the coefficient of @xmath19 in @xmath401 here @xmath402 are the strict transforms of the @xmath165 not in the support of @xmath385 . the @xmath403 are those exceptional divisors which are not in the support of @xmath404 , and @xmath405 are the proper transforms of the divisor classes of the @xmath51 s . we can now expand our chern class product replacing @xmath406 in the denominator by @xmath407 in the numerator . the intersection product of any collection of prime torus - invariant divisors is the cycle of the cone they span or @xmath129 if there is no such cone . hence the coefficient of @xmath19 is the sum of all terms of the form @xmath408 here @xmath409 and @xmath410 ranges over all dimension @xmath387 cones of @xmath383 spanned by rays not in the support of @xmath411 . this product is exactly the mixed volume @xmath412 . to finish we note that if @xmath410 contains an exceptional divisor @xmath403 , the minimal cone @xmath337 of @xmath335 containing @xmath410 must have dimension strictly larger than @xmath387 . this is because @xmath410 does not have any rays in the support of @xmath411 hence is not maximal in @xmath337 . as a consequence all of the faces @xmath413 have a translate that lies in a subspace of dimension strictly less than @xmath414 and the corresponding mixed volume is @xmath129 . in conclusion , the exceptional divisors do not contribute to the top chern class product and the formula reduces to the stated one . in two variables we recover a particularly simple formula : [ areaarea ] let @xmath415 be generic laurent polynomials in two variables @xmath416 with newton polygons @xmath417 . if the origin lies on none of the lines spanned by edges of their minkowski sum @xmath357 then the toric ml degree equals the area of @xmath171 plus the areas of each of the @xmath181 . this is a special case of theorem [ toricformula4 ] when no facets pass through the origin . therefore the only term is @xmath418 the euclidean area of each @xmath181 is @xmath419 . the euclidean area of the minkowski sum @xmath420 equals @xmath421 . the stated formula is the sum of these expressions . now that we are equipped with the volume formulas in theorem [ toricformula4 ] and corollary [ areaarea ] , let us revisit the two - dimensional examples from section 3 . the newton polygons @xmath422 in example [ rectangles ] are axis - parallel parallelograms . the first term in the formula ( [ rectangleformula ] ) is the area of their minkowski sum @xmath423 , and the second term is the sum of the areas of the @xmath181 , as in corollary [ areaarea ] . the third and fourth term are the two correction terms stemming from the fact that the origin is a vertex of each newton polytope . these terms disappear if we replace one @xmath21 by @xmath424 . the number @xmath255 in example [ haseightrays ] can also be derived using theorem [ toricformula4 ] . the three polygons in figure [ fig2 ] have areas @xmath425 and @xmath426 respectively . the area of their sum is @xmath427 . the two divisors corresponding to @xmath428 and @xmath429 pass through the origin and yield correction terms of @xmath33 and @xmath430 respectively . finally we add back @xmath33 for the vertex at the origin . altogether @xmath431 . .1 cm our discussion so far indicates that we get the sharpest results when @xmath30 is smooth and @xmath34 is gnc . in the toric case the smoothness hypothesis could be largely removed as in theorem [ toricformula4 ] . the gnc condition can also be relaxed for certain other cases as we saw in the previous section . in general , if the pair @xmath432 does not satisfy the smoothness and gnc hypotheses then we must appeal to hironaka s theorem on resolution of singularities ( see e.g. @xcite ) . this furnishes a proper projective morphism @xmath433 such that @xmath373 is smooth and @xmath434 has gnc . we need to compute the divisor @xmath435 of the pullback of the function @xmath35 . if @xmath405 is the proper transform of the divisor @xmath85 and @xmath436 are the exceptional divisors of @xmath377 then @xmath437 where @xmath438 are certain ( possibly negative ) integers . these integers are @xmath439 where @xmath440 is the multiplicity of the full transform of @xmath85 along @xmath441 . the underlying reduced divisor is @xmath442 . the number of critical points is now gotten by applying theorem [ chernclassthm ] to @xmath443 instead of @xmath432 . this procedure can be very complicated in practice . we illustrate it with a simple example . [ 4 curves ] let @xmath23 , @xmath22 , @xmath444 , @xmath445 , @xmath446 , and @xmath447 . the divisor @xmath34 is not a gnc divisor since at the origin all the four curves defined by @xmath448 meet . in order to resolve this singularity we blow up @xmath449 at @xmath450 to obtain @xmath373 which is smooth . we note that @xmath451 is @xmath452 where @xmath161 is the proper transform of the generic hyperplane section , @xmath234 is the exceptional divisor , and @xmath453 is the proper transform of a line through the origin . we have four cases : we consider first the general case where @xmath454 and @xmath455 . in this case @xmath375 consists of the proper transforms of the four original curves , the exceptional divisor , and the pullback of the line at infinity . after cancellations , we just need to compute the coefficient of @xmath236 in @xmath456 where @xmath457 and @xmath458 are the irreducible divisors corresponding respectively to the circle and the ellipse . this coefficient is @xmath459 . in @xmath373 the two curves intersect in three points , and their self - intersection also yields three points . hence the ml degree is nine . in the special case where @xmath460 , we need to compute the coefficient of @xmath236 in @xmath461 , which is @xmath462 . since @xmath463 the ml degree drops down to seven . if @xmath464 then the coefficient of @xmath236 in @xmath465 is @xmath466 . since @xmath467 , the ml degree is five . finally , if both @xmath468 and @xmath469 are zero , then the coefficient of @xmath236 in @xmath470 is @xmath471 : since @xmath472 , the ml degree further drops down to three . the number of bounded regions of the complement of the four curves in @xmath271 is seven . by proposition [ bounded ] , this is a lower bound on the ml degree when all @xmath62 are positive . this example shows that , for specific _ negative _ values of the @xmath62 s , the number of critical points may be smaller than this lower bound . a well - known result in the theory of hyperplane arrangements @xcite states that the number of bounded regions of a real arrangement equals the euler characteristic of the complement of its complexification . the euler characteristic is @xmath473 where @xmath474 is the mbius number of the intersection lattice , and , by theorem [ thm - linear ] , this is precisely the ml degree of the associated linear model . here we extend this relationship between topology and the ml degree to statistical models which are given by nonlinear polynomials @xmath21 . working in the general setting of section 2 , we shall prove the following : [ euler ] let @xmath30 be a smooth complete algebraic variety over @xmath172 of dimension @xmath17 , and let @xmath34 be the reduced divisor associated to @xmath267 . assume that the hypotheses ( a ) , ( b ) and ( c ) below hold . then the ml degree equals @xmath475 where @xmath476 is the topological euler characteristic . invoking hironaka s theorem on resolution of singularities , we fix a blow up @xmath477 such that @xmath478 is smooth , and the rational function @xmath35 pulls back to a proper morphism @xmath479 onto @xmath480 . the three * hypotheses * are as follows : 1 . the inverse image @xmath481 of the divisor @xmath34 can be written as @xmath482 where @xmath483 is the support of the divisor @xmath484 of the pullback @xmath479 of the rational function @xmath35 , while @xmath485 is the horizontal divisor consisting of the sum of all components of @xmath375 which map onto @xmath486 . the restriction of @xmath479 to @xmath487 is a topological fiber bundle over @xmath488 . the number of critical points of @xmath479 on @xmath489 is finite . [ wowo ] hypothesis ( a ) is crucial and depends on the exponents @xmath62 . for instance , consider the rational function @xmath490 on @xmath491 . we get @xmath478 by blowing up the origin twice . the exceptional curve of the first blow - up belongs to the fiber @xmath492 and hence is not supported on @xmath484 . hypothesis ( a ) is not satisfied for this example . if we take instead @xmath493 then hypothesis ( a ) is satisfied because the exceptional curve belongs to the fiber @xmath494 and is hence supported on @xmath484 , hypothesis ( b ) implies that the cohomology ranks of @xmath487 can be computed from that of @xmath495 and the fibers using knneth s formula . the alternating sum of the ranks is zero for the fibers , and we get @xmath496 . in fact , the hypothesis ( b ) could be replaced by the more general condition @xmath496 . any proper map @xmath497 for @xmath30 smooth is a topological fiber bundle if it is a submersion , i.e. @xmath498 for all points in @xmath30 . therefore to check hypothesis ( b ) , we need only find a controlled stratification ( in the sense of thom - mather theory @xcite ) of @xmath485 into locally closed smooth sets such that for all points on each strata @xmath499 , @xmath500 . in particular , this last condition will imply that the critical points of @xmath479 on @xmath489 are the same as the critical points of @xmath35 on @xmath41 . our method follows closely the proof of the lefschetz hyperplane section theorem ( cf . moreover , since the complement is not necessarily compact we shall use borel - moore homology @xcite ( see also @xcite ) . we note that for compact spaces @xmath30 the ordinary homology groups coincide with the borel - moore homology groups . in the borel - moore homology theory we have the following useful exact sequence to be used below : if @xmath30 is locally compact , @xmath156 is closed in @xmath30 , and @xmath501 , we have @xmath502 thus in this situation the borel - moore euler characteristic is additive : @xmath503 finally , if @xmath504 is an even - dimensional orientable manifold then poincar duality holds between borel moore homology and ordinary cohomology , and @xmath505 coincides with the topological euler number @xmath506 . in our situation we get @xmath507 the last equation follows from ( [ additive ] ) . hypothesis ( b ) implies @xmath508 ( see remark [ wowo ] ) , and hence it suffices to show that the ml degree equals @xmath509 in other words , we may now simply erase the tilde and consider the case when @xmath30 is smooth and @xmath35 defines a proper morphism @xmath510 . let @xmath511 denote the set of critical points of @xmath35 on @xmath30 . by hypothesis ( c ) , this set is finite and the ml degree equals its cardinality counting multiplicities : @xmath512 the multiplicity @xmath513 of a critical point @xmath514 of @xmath35 is known as the _ milnor number _ at @xmath514 of the hypersurface @xmath515 . milnor @xcite showed that this algebraic invariant of a singularity has the following topological interpretation . consider a coordinate chart around the point @xmath514 and intersect the fiber @xmath516 with a ball of radius @xmath517 around @xmath514 . for @xmath518 this intersection is the _ milnor fiber_. milnor @xcite showed that the milnor fiber is homotopy equivalent to a bouquet of @xmath513 spheres of dimension @xmath519 . each singular fiber is obtained ( cf . @xcite ) from a smooth fiber by replacing the milnor fiber by a contractible set . the borel - moore exact sequence ( [ bmsequence ] ) implies that the euler number of a singular fiber @xmath376 is obtained from the euler number of a smooth fiber @xmath156 by adding @xmath520 . then the euler number of the union of the singular fibers equals @xmath521 times the euler number of a smooth fiber @xmath156 plus the correction @xmath522 . applying knneth s formula to the fiber bundle defined by @xmath35 on @xmath41 minus the union of the singular fibers , and then applying the additivity formula ( [ additive ] ) , we conclude that @xmath523 as desired . @xmath524 for another illustration consider example [ 4 curves ] with @xmath491 . the generic ml degree was @xmath43 but it decreased by @xmath430 when @xmath525 is a general solution of @xmath526 . this is consistent with theorem [ euler ] , because for such @xmath12 the divisor @xmath34 loses the component at infinity . the difference is a projective line minus @xmath527 points , which has euler number @xmath528 . consider our hypotheses when @xmath529 is the blow - up of @xmath30 at the origin . if @xmath530 then the exceptional curve is part of @xmath531 and theorem [ euler ] is valid . on other hand , if @xmath532 then it maps to @xmath533 under a rational map of degree @xmath534 , so hypothesis ( a ) does not hold . the philosophy of this example is that , even if the divisor @xmath34 is locally biholomorphic to an arrangement of hyperplanes , genericity of the exponents @xmath62 may be necessary for the topological formula of theorem [ euler ] to hold . theorems [ thm1 ] , [ toricformula3 ] and [ toricformula4 ] offer combinatorial formulae for the ml degree and hence ( using theorem [ euler ] ) for the euler number of the complement of an arrangement of generic hypersurfaces @xmath535 in @xmath536 or in a toric manifold . in each of these theorems , the combinatorial number becomes an upper bound for the ml degree when the coefficients of the @xmath21 s are special . this semi - continuity principle will be explained by the following general topological result , in which also the underlying manifold is allowed to vary . [ semiconti ] assume we are given a one - parameter smooth proper family @xmath537 of complex manifolds over the unit disk @xmath538 , and a family of rational functions @xmath539 on @xmath537 , such that 1 . for @xmath540 the divisor @xmath541 defined by @xmath539 has gnc , and 2 . for @xmath542 the divisor @xmath543 defined by @xmath544 has the same homology class as @xmath545 for the natural differentiable trivialization of the family @xmath537 . then the ml degree of @xmath544 is less than or equal to the ml degree of @xmath539 . in order to understand the second hypothesis , let us recall ehresmann s theorem @xcite : any proper submersion @xmath546 of differentiable manifolds is a differentiable fiber bundle , i.e. , if @xmath504 is a sufficiently small open set in @xmath547 , there is a local diffeomorphism between @xmath548 and @xmath549 , for a fiber @xmath156 , and this diffeomorphism is compatible with the two projections to @xmath504 . _ sketch of proof . _ let @xmath550 be the total space of the family @xmath551 . we consider the function @xmath552 , given by @xmath553 . consider the locus @xmath554 given by the vanishing of the vertical differential of @xmath212 : this is the local complete intersection defined by the @xmath17 partials @xmath555 , where @xmath98 are local coordinates on the fibers provided by the implicit function theorem . at each critical point @xmath514 of @xmath544 , the locus @xmath556 has dimension @xmath33 , and thus , in a neighborhood of @xmath514 , the morphism @xmath557 is finite , whence its degree is locally constant . this establishes the desired semi - continuity . in general , it might be difficult to show that a given rational function @xmath544 has a perturbation as above , or we might want to calculate the ml degree with more algebraic precision . the results in the next section may help . in this section we will show that the formula for the ml degree given by the logarithmic chern number ( theorem [ chernclassthm ] ) holds in greater generality . returning to the setting of chapter 2 , we consider logarithmic vector fields along @xmath34 . again , our definition differs slightly from the one given by saito @xcite . recall that the _ tangent sheaf _ @xmath560 is @xmath561 , the dual of the @xmath33-forms on @xmath30 . the inclusion of @xmath95 into @xmath75 , studied in lemma [ exactlem ] , dualizes to an inclusion of the logarithmic vector fields into the tangent sheaf . [ moresheaves ] we have the following exact sequence of sheaves on @xmath30 : @xmath562 if @xmath30 is smooth , then the rightmost homomorphism is onto , and the total chern class of the sheaf of logarithmic vector fields equals @xmath563 dualizing the sequence @xmath564 , we get @xmath565 hence the @xmath566-sequence gotten by dualizing the sequence ( [ lemmasequence ] ) has the form @xmath567 this is the first statement of proposition [ moresheaves ] . if @xmath30 is smooth then the cotangent sheaf @xmath95 is free , and we have @xmath568 . the formula ( [ chernoftheta ] ) follows from the multiplicativity of the total chern class . [ logvecchara ] the two leftmost maps in the exact sequence of proposition [ moresheaves ] characterize the logarithmic vector fields on @xmath30 along @xmath34 as those vector fields @xmath569 which satisfy @xmath570 for all @xmath214 . in other words , for each @xmath571 , the vector field @xmath572 has the property that there exist functions @xmath90 such that @xmath573 . an interesting case will be the one where @xmath30 is smooth and the sheaf @xmath574 has zero - dimensional support and length @xmath514 . here the top chern class of @xmath558 and the top chern class of @xmath75 differ by @xmath575 , and the count in part 3 of theorem [ chernclassthm ] changes accordingly . as in the proof of theorem [ chernclassthm ] , set @xmath576 and let @xmath577 be the subscheme of @xmath30 defined by the vanishing of @xmath123 . the restriction of @xmath578 to the open set @xmath579 is the critical locus of @xmath35 . the reason why the sheaf @xmath558 is important is that it enters directly into the algebraic description of @xmath578 . indeed , the section @xmath123 of @xmath580 corresponds to an exact sequence @xmath581 where @xmath582 is the kernel of @xmath583 . hence the @xmath566-sequence gotten by dualizing the previous sequence has the form @xmath584 this leads to the following more refined formula for the ml degree . [ thetatheorem ] assume that @xmath30 is smooth , @xmath34 meets every curve , the sheaf @xmath558 is locally free , and @xmath125 does not intersect the divisor @xmath34 . then the number of critical points of @xmath35 on @xmath585 equals @xmath586 . the sequence ( [ zsequence ] ) gives the following description of the equations defining the critical points . the ideal sheaf of the subscheme @xmath578 is generated by the functions @xmath587 , where @xmath588 varies among the logarithmic vector fields and @xmath589 is derived from @xmath588 as in remark [ logvecchara ] . this holds because the section @xmath590 factors through the homomorphism @xmath591 , thus the homomorphism dual to the section @xmath123 also factors through the dual homomorphism @xmath592 . let @xmath595 with @xmath596 and @xmath597 . a vector field @xmath598 is logarithmic if and only if @xmath599 and @xmath600 , for some function @xmath601 . observe that @xmath602 . we conclude that the sheaf @xmath558 is locally free of rank @xmath603 . let @xmath608 with @xmath609 and @xmath610 . logarithmic vector fields @xmath588 have the form @xmath611 with @xmath612 divisible by @xmath613 . this implies @xmath614 , @xmath615 , and @xmath616 , for any functions @xmath617 . thus @xmath558 is locally free of rank two . the origin does not belong to the subscheme @xmath125 if and only if @xmath618 . * acknowledgements : * this project grew out of discussions with lior pachter at the december 2003 workshop on _ computational algebraic statistics _ at the american institute for mathematics ( aim ) at palo alto . many thanks to both lior and aim for getting us started , and many thanks to msri berkeley and the organizers of the spring 2004 msri program _ topology of real algebraic varieties _ , which gave the four of us the opportunity to collaborate . we also thank oleg viro for letting us include theorem [ thm : viro ] . amit khetan was supported by an nsf postdoctoral fellowship ( dms-0303292 ) . bernd sturmfels was supported by the hewlett packard visiting research professorship 2003/2004 at msri berkeley and in part by the nsf ( dms-0200729 ) . b. chor , a. khetan and s. snir . maximum likelihood on four taxa phylogenetic trees : analytic solutions , _ the @xmath619th annual conference on research in computational molecular biology recomb 2003 _ , berlin , april 2003 , pp .

the covariance graph ( aka bi - directed graph ) of a probability distribution is the undirected graph where two nodes are adjacent iff their corresponding random variables are marginally dependent in . implies that their corresponding random variables are marginally independent in . this difference in the definition is important in this paper . ] in this paper , we present a graphical criterion for reading dependencies from , under the assumption that satisfies the graphoid properties as well as weak transitivity and composition . we prove that the graphical criterion is sound and complete in certain sense . we argue that our assumptions are not too restrictive . for instance , all the regular gaussian probability distributions satisfy them .

1010.4504

the covariance graph ( aka bi - directed graph ) of a probability distribution @xmath0 is the undirected graph @xmath1 where two nodes are adjacent iff their corresponding random variables are marginally dependent in @xmath0 . covariance graphs were introduced in @xcite to represent independence models . since then , they have received considerable attention . see , for instance , @xcite . the works @xcite are particularly important for the interpretation of covariance graphs in terms of independencies . specifically , these works introduce a graphical criterion for reading independencies from the covariance graph @xmath1 of a probability distribution @xmath0 , under the assumption that @xmath0 satisfies the graphoid properties and composition . in this paper , we show that @xmath1 can also be used to read dependencies holding in @xmath0 . specifically , we present a graphical criterion for reading dependencies from @xmath1 under the assumption that @xmath0 satisfies the graphoid properties , weak transitivity and composition . we also prove that our graphical criterion is sound and complete . here , complete means that it is able to read all the dependencies in @xmath0 that can be derived by applying the graphoid properties , weak transitivity and composition to the dependencies used in the construction of @xmath1 and the independencies obtained from @xmath1 . we also show that there exist important families of probability distributions that satisfy the graphoid properties , weak transitivity and composition . these include , for instance , all the regular gaussian probability distributions . note that this paper would be unnecessary if @xmath0 satifies all and only the independencies that can be read from @xmath1 via the graphical criterion in @xcite , i.e. @xmath0 is faithful to @xmath1 . we will see that one can not safely assume faithfulness in general . therefore , one is only entitled to assume that @xmath0 satifies all ( but not necessarily only ) the independencies that can be read from @xmath1 via the graphical criterion in @xcite , i.e. @xmath0 is markov wrt @xmath1 . this is actually the reason of being of this paper . two previous works that somehow address the problem of reading dependencies off covariance graphs are @xcite . these works propose to determine whether two random variables @xmath2 and @xmath3 are dependent given some other random variables @xmath4 by , first , constructing the covariance graph of the conditional probability distribution given @xmath4 of any set of random variables that includes @xmath2 and @xmath3 and , then , checking if the nodes corresponding to @xmath2 and @xmath3 are adjacent in the covariance graph constructed . therefore , these works construct multiple covariance graphs , one for each conditional probability distribution of interest , from which only the dependencies used in their construction are read . the work presented in this paper is radically different : we only construct the covariance graph of the probability distribution at hand and read from it many more dependencies than those used in its construction . while this is the first work where a sound and complete graphical criterion for reading dependencies off covariance graphs is developed , it is worth mentioning that there already exist sound and complete graphical criteria for reading dependencies off other graphical models . for instance , there exists a sound and complete graphical criterion for reading dependencies off the concentration graph ( aka minimal undirected independence map or markov network ) of a probability distribution that satisfies the graphoid properties @xcite , or the graphoid properties and weak transitivity @xcite . as a matter of fact , the graphical criterion that we present in this paper is dual to the one in @xcite . there also exists a sound and complete graphical criterion for reading dependencies off the bayesian network ( aka minimal directed independence map ) of a probability distribution that satisfies the graphoid properties @xcite , the graphoid properties and weak transitivity @xcite , or the graphoid properties and weak transitivity and composition @xcite . in the last two references , the bayesian networks are restricted to be polytrees . note that @xcite address a related but not more general problem than the one in this paper , since neither concentration graphs nor bayesian networks include covariance graphs . related more general problems than the one studied in this paper have been recently addressed , though . for instance , a method to read dependencies from multivariate regression graphs , which include covariance graphs , is proposed in @xcite . the author also presents necessary and sufficient conditions for the method to be sound . these conditions are the same as the ones considered in this paper , namely the graphoid properties plus weak transitivity and composition . unlike in this paper , no proof of completeness of the method proposed appears in @xcite . another related more general work is @xcite , where the author shows how summary graphs , which include covariance graphs , can help to detect which dependencies remain undistorted and which do not after marginalization and/or conditioning in a probability distribution generated over a so - called parent graph . it should be pointed out that that the probability distribution is generated over a parent graph implies that it satisfies the same conditions as the ones considered in this paper ( * ? ? ? * proposition 3 ) . again , unlike in this paper , the completeness question is not addressed in @xcite . finally , it should be noted that @xcite make use of the graphical criterion presented in @xcite for reading independencies from loopless mixed graphs , which include multivariate regression graphs , summary graphs and parent graphs . this criterion is sound and complete in certain sense , given that the graphoid properties and composition hold ( * ? ? ? * theorem 3 ) . these conditions are , in fact , not only sufficient but necessary too ( * ? ? ? * section 6.3 ) . we think that the work presented in this paper can be of great interest for the artificial intelligence community . graphs are one of the most commonly used metaphors for representing knowledge because they appeal to human intuition @xcite . furthermore , graphs are parsimonious models because they trade off accuracy for simplicity . consider , for instance , representing the independence model induced by a probability distribution as a graph . though this graph is typically less accurate than the probability distribution ( the graph may not represent all the ( in)dependencies and those that are represented are not quantified ) , it also requires less space to be stored and less time to be communicated than the probability distribution , which may be desirable features in some applications . thus , it seems sensible developing tools for reasoning with graphs . our graphical criterion is one such a tool : as the graphical criterion in @xcite makes the discovery of independencies amenable to human reasoning by enabling to read independencies off a covariance graph @xmath1 without numerical calculation , so does our graphical criterion with respect to the discovery of dependencies . there are fields where discovering dependencies is more important than discovering independencies @xcite . it is in these fields where we believe that our graphical criterion has greater potential . in bioinformatics , for instance , the nodes of @xmath1 may represent ( the expression levels of ) some genes under study . bioinformaticians are typically more interested in discovering gene dependencies than independencies , because the former provide contexts in which the expression level of some genes is informative about that of some other genes , which may lead to hypothesize dependencies , functional relations , causal relations , the effects of manipulation experiments , etc . see , for instance , @xcite for an application of covariance graphs to bioinformatics under the name of relevance networks . the rest of the paper is organized as follows . we start by reviewing some concepts in section [ sec : preliminaries ] . we show in section [ sec : wtc ] that assuming the graphoid properties , weak transitivity and composition is not too restrictive . we prove in section [ sec : indep ] that the existing graphical criterion for reading independencies from covariance graphs is complete in certain sense . this result , in addition to being important in its own , is important for reading as many dependencies as possible from covariance graphs . we introduce in section [ sec : dep ] our graphical criterion for reading dependencies from covariance graphs and prove that it is sound and complete in certain sense . finally , we close with some discussion in section [ sec : discussion ] . in this section , we introduce some concepts and results that are used later in this paper . we first recall some results from graphical models . see , for instance , @xcite for further information . let @xmath5 be a finite set of size @xmath6 . the elements of @xmath7 are not distinguished from singletons and the union of the sets @xmath8 is written as the juxtaposition @xmath9 . we assume throughout the paper that the union of sets precedes the set difference when evaluating an expression . unless otherwise stated , all the graphs in this paper are defined over @xmath7 . if a graph @xmath1 contains an undirected ( respectively directed ) edge between two nodes @xmath10 and @xmath11 , then we say that @xmath12 ( respectively @xmath13 ) is in @xmath1 . if @xmath13 is in @xmath1 then @xmath10 is called a parent of @xmath11 in @xmath1 . let @xmath14 denote the set of parents in @xmath1 of the nodes in @xmath15 . a route from a node @xmath10 to a node @xmath16 , denoted @xmath17 , in a graph @xmath1 is a sequence of nodes @xmath18 such that there exists an edge in @xmath1 between @xmath19 and @xmath20 for all @xmath21 . a path is a route @xmath17 in which the nodes @xmath18 are distinct . a route @xmath17 is called undirected if @xmath22 is in @xmath1 for all @xmath21 . a node @xmath10 is an ancestor of a node @xmath16 in @xmath1 if there is a route @xmath17 in @xmath1 such that @xmath22 or @xmath23 is in @xmath1 for all @xmath21 . let @xmath24 denote the set of ancestors in @xmath1 of the nodes in @xmath15 . a node @xmath16 is a descendant of a node @xmath10 in @xmath1 if there is a route @xmath17 in @xmath1 such that @xmath22 or @xmath23 is in @xmath1 for all @xmath21 and @xmath23 is in @xmath1 for some @xmath21 . a chain graph ( cg ) is a graph ( possibly ) containing both undirected and directed edges and such that no node is a descendant of itself . an undirected graph ( ug ) is a cg containing only undirected edges . a directed and acyclic graph ( dag ) is a cg containing only directed edges . a set of nodes of a cg is connected if there exists an undirected route in the cg between every pair of nodes in the set . a connectivity component of a cg is a connected set that is maximal with respect to set inclusion . the moral graph of a cg @xmath1 , denoted @xmath25 , is the ug where two nodes are adjacent iff they are adjacent in @xmath1 or they are both in @xmath26 for some connectivity component @xmath27 of @xmath1 . the subgraph of a cg @xmath1 induced by @xmath15 , denoted @xmath28 , is the graph over @xmath29 where two nodes are connected by a ( un)directed edge if that edge is in @xmath1 . let @xmath30 , @xmath31 and @xmath32 denote three disjoint subsets of @xmath7 . we say that @xmath30 is separated from @xmath31 given @xmath32 in a cg @xmath1 if every path in @xmath33 from a node in @xmath30 to a node in @xmath31 has some node in @xmath32 . we denote such a separation statement by @xmath34 . let @xmath35 denote a vector of random variables and @xmath36 @xmath37 its subvector @xmath38 . we use upper - case letters to denote random variables and the same letters in lower - case to denote their states . unless otherwise stated , all the probability distributions in this paper are defined over @xmath39 . let @xmath30 , @xmath31 , @xmath32 and @xmath40 denote four disjoint subsets of @xmath7 . we represent by @xmath41 that @xmath42 is independent of @xmath43 given @xmath4 in a probability distribution @xmath0 . we represent by @xmath44 that @xmath41 does not hold . a probability distribution @xmath0 is a graphoid if it satisfies the following properties : symmetry @xmath45 , decomposition @xmath46 , weak union @xmath47 , contraction @xmath48 , and intersection @xmath49 . we say that a graphoid @xmath0 is a wtc graphoid if it satisfies the following two additional properties : weak transitivity @xmath50 with @xmath51 , and composition @xmath52 . let @xmath30 , @xmath31 and @xmath32 denote three disjoint subsets of @xmath7 . we denote by that a cg @xmath1 represents that @xmath42 is independent of @xmath43 given @xmath4 . we denote by @xmath53 that @xmath54 does not hold . in this paper , we are interested in the classic lauritzen - wermuth - frydenberg interpretation of cgs as independence models , which is based on the following graphical criterion . [ def : inde ] given a cg @xmath1 , @xmath54 if @xmath34 . however , in this paper we are also interested in the dual interpretation of ugs as independence models that builds on the following graphical criterion . [ def : inde2 ] given an ug @xmath1 , @xmath54 if @xmath55 . the following rephrasing of the graphical criterion in definition [ def : inde2 ] may be easier to recall : @xmath54 if every path in @xmath1 from a node in @xmath30 to a node in @xmath31 has some node outside @xmath56 . when an ug is interpreted according to the graphical criterion in definition [ def : inde ] we call it a concentration graph , and when it is interpreted according to the graphical criterion in definition [ def : inde2 ] we call it a covariance graph . a probability distribution @xmath0 is markov wrt a cg , concentration graph or covariance graph @xmath1 when @xmath41 if @xmath54 for all @xmath30 , @xmath31 and @xmath32 disjoint subsets of @xmath7 . a probability distribution @xmath0 is faithful to a cg , concentration graph or covariance graph @xmath1 when @xmath41 iff @xmath54 for all @xmath30 , @xmath31 and @xmath32 disjoint subsets of @xmath7 . the concentration graph ( aka minimal undirected independence map or markov network ) of a probability distribution @xmath0 is the ug @xmath1 where two nodes @xmath57 and @xmath58 are adjacent iff @xmath59 . the covariance graph ( aka bi - directed graph ) of a probability distribution @xmath0 is the ug @xmath1 where two nodes @xmath57 and @xmath58 are adjacent iff @xmath60 . a wtc graphoid @xmath0 is markov wrt both its covariance graph @xmath1 and its concentration graph @xmath61 . however , neither @xmath53 nor @xmath62 implies @xmath44 , unless @xmath0 is faithful to @xmath1 or @xmath61 . this is actually the reason of being of this paper . this paper is devoted to the study of wtc graphoids . we show in this section that wtc graphoids are worth studying because they include important families of probability distributions . for instance , any regular gaussian probability distribution is a wtc graphoid ( * ? ? ? * sections 2.2.2 , 2.3.5 and 2.3.6 ) . the following theorem introduces another interesting family of wtc graphoids . [ the : fcg ] let @xmath1 be a cg . any probability distribution @xmath0 that is faithful to @xmath1 is a wtc graphoid . let @xmath63 be any regular gaussian probability distribution that is faithful to @xmath1 . such probability distributions exist due to ( * ? ? ? * theorems 1 and 2 ) . since @xmath0 and @xmath63 are faithful to @xmath1 , @xmath41 iff @xmath54 iff @xmath64 for all @xmath30 , @xmath31 and @xmath32 disjoint subsets of @xmath7 . therefore , @xmath0 is a wtc graphoid because @xmath63 is a wtc graphoid . the previous theorem is meaningful only if we prove that , for any cg , there exist probability distributions that are faithful to it . we do so in the following theorem . [ the : fcg2 ] let @xmath1 be a cg . if each random variable in @xmath39 has a finite prescribed sample space with at least two possible states , then there exists a discrete probability distribution with the prescribed sample spaces for the random variables in @xmath39 that is faithful to @xmath1 . on the other hand , if the sample space of each random variable in @xmath39 is @xmath65 , then there exist a regular gaussian probability distribution that is faithful to @xmath1 and a continuous but non - gaussian probability distribution that is faithful to @xmath1 . the first and second statements in the theorem are proven in ( * ? ? ? * theorems 3 and 5 ) and ( * ? ? ? * theorems 1 and 2 ) , respectively . the third statement in the theorem can easily be proven by using copulas @xcite as follows . let @xmath0 denote any regular gaussian probability distribution that is faithful to @xmath1 . derive the gaussian copula for @xmath0 . the copula represents the independence model of @xmath0 stripped from its univariate marginals . therefore , the copula together with a set of arbitrary univariate marginals can be used to generate a multivariate probability distribution whose independence model is the one dictated by the copula and whose univariate marginals are the given ones . the desired result is achieved if the arbitrary marginals are chosen so that they are continuous but non - gaussian . see @xcite for more details . it is worth mentioning that the results in ( * ? ? * theorems 3 and 5 ) and ( * ? ? ? * theorems 1 and 2 ) are actually stronger than the first and second statements in the previous theorem . specifically , the results reported there are that , in certain measure - theoretic sense , almost all the discrete probability distributions and regular gaussian probability distributions that are markov wrt a cg are faithful to it . finally , note that the marginals and conditionals of a regular gaussian probability distribution are regular gaussian probability distributions and , thus , wtc graphoids . in fact , this property can be generalized to all the wtc graphoids . the following theorem , originally reported in ( * ? ? ? * theorem 5 ) , formalizes this result . [ the : chickeringandmeek2002 ] let @xmath0 be a wtc graphoid and let @xmath15 . then , @xmath66 is a wtc graphoid . if @xmath67 has the same ( in)dependencies for all @xmath68 , then @xmath67 for any @xmath68 is a wtc graphoid . it is worth noting that many members of the families of wtc graphoids that we have presented in this section are markov wrt their covariance graphs but not faithful to them . hence , the need to develop a graphical criterion for reading dependencies from the covariance graph of a wtc graphoid . for example , consider any discrete , regular gaussian , or continuous but non - gaussian probability distribution @xmath0 that is faithful to a cg with @xmath69 as induced subgraph . then , the covariance graph @xmath1 of @xmath0 has @xmath70 as induced subgraph and , thus , @xmath0 is not faithful to @xmath1 since @xmath71 but @xmath72 for some @xmath73 . this example is based on @xcite . the interested reader is referred to these works for a characterization of the independence models that can be represented exactly by dags but not by covariance graphs . the graphical criterion in definition [ def : inde2 ] is sound for reading independencies from the covariance graph @xmath1 of a wtc graphoid @xmath0 , that is , it only identifies independencies in @xmath0 ( * ? ? ? * ; * ? ? ? * proposition 2.2 ) . in this section , we show that this graphical criterion is complete in the sense that it identifies all the independencies in @xmath0 that can be identified by studying @xmath1 alone . this completeness result , in addition to being important in its own , is crucial for reading as many dependencies as possible from @xmath1 , as we will see in the next section . in order to prove the referred completeness result , it suffices to prove that there exist wtc graphoids that are faithful to @xmath1 , because @xmath1 is their covariance graph and they only have the independencies that the graphical criterion in definition [ def : inde2 ] identifies from @xmath1 . therefore , we can not derive more independencies from @xmath1 alone than those identified by this graphical criterion , because @xmath0 may be one of the wtc graphoids that are faithful to @xmath1 . the following two theorems prove the desired result . [ the : fcvg ] let @xmath1 be a covariance graph . if each random variable in @xmath39 has a finite prescribed sample space with at least two possible states , then there exists a discrete probability distribution with the prescribed sample spaces for the random variables in @xmath39 that is faithful to @xmath1 . on the other hand , if the sample space of each random variable in @xmath39 is @xmath65 , then there exist a regular gaussian probability distribution that is faithful to @xmath1 and a continuous but non - gaussian probability distribution that is faithful to @xmath1 . we start by proving the first statement in the theorem . first , we create a dag from @xmath1 as follows : replace each edge @xmath74 in @xmath1 with @xmath75 where @xmath76 is a newly created node . call the resulting dag @xmath61 and let @xmath77 denote all the newly created nodes . it is easy to see that @xmath78 iff @xmath54 for all @xmath30 , @xmath31 and @xmath32 disjoint subsets of @xmath7 . let @xmath79 denote a vector of random variables such that each of them has any finite sample space with at least two possible states . let @xmath80 denote any discrete probability distribution that is faithful to @xmath61 . such probability distributions exist by ( * ? ? ? * theorem 7 ) . note that , for any @xmath30 , @xmath31 and @xmath32 disjoint subsets of @xmath7 , @xmath81 iff @xmath82 iff @xmath78 iff @xmath54 . consequently , @xmath83 is faithful to @xmath1 . the second statement in the theorem can be proven in much the same way as the first if @xmath80 denotes now any regular gaussian probability distribution that is faithful to @xmath61 . such probability distributions exist by ( * ? ? ? * theorem 3.2 ) . note that @xmath83 is regular gaussian . finally , the third statement in the theorem can be proven by using copulas as we did in the proof of theorem [ the : fcg2 ] . an alternative proof of the second statement in the theorem above follows from ( * ? ? ? * theorem 7.5 ) . specifically , @xcite introduces a new class of graphical models called ancestral graphs , whose edges can be undirected , directed or bi - directed ( @xmath84 ) . covariance graphs are equivalent to ancestral graphs with only bi - directed edges . * theorem 7.5 ) proves that , for any ancestral graph , there is a regular gaussian probability distribution that is faithful to it . in appendix a , we strengthen the second statement in the theorem above by proving that , in certain measure - theoretic sense , almost all the regular gaussian probability distributions that are markov wrt a covariance graph are faithful to it . although this result is not used in this paper , we consider it to be interesting in its own and , thus , we decide to report on it . [ the : fcvg2 ] let @xmath1 be a covariance graph . any probability distribution @xmath0 that is faithful to @xmath1 is a wtc graphoid . let @xmath63 be any regular gaussian probability distribution that is faithful to @xmath1 . such probability distributions exist due to theorem [ the : fcvg ] . since @xmath0 and @xmath63 are faithful to @xmath1 , @xmath41 iff @xmath54 iff @xmath64 for all @xmath30 , @xmath31 and @xmath32 disjoint subsets of @xmath7 . therefore , @xmath0 is a wtc graphoid because @xmath63 is a wtc graphoid . as explained at the beginning of this section , the previous two theorems imply that the graphical criterion in definition [ def : inde2 ] is complete for reading independencies from the covariance graph @xmath1 of a wtc graphoid @xmath0 , in the sense that it identifies all the independencies in @xmath0 that can be identified by studying @xmath1 alone . an equivalent formulation of this result is that the graphical criterion is complete in the sense that it identifies all the independencies that are shared by all the wtc graphoids whose covariance graph is @xmath1 . finally , it is worth mentioning that the graphical criterion in definition [ def : inde2 ] is not complete in the more stringent sense of being able to identify all the independencies in @xmath0 . actually , no sound graphical criterion for reading independencies from @xmath1 is complete in this latter sense . an example illustrating this follows . [ exa : example ] let @xmath0 and @xmath85 denote two wtc graphoids that are faithful to the cgs in the left and center of table [ tab : example ] , respectively . such wtc graphoids exist by ( * ? ? ? * theorems 1 and 2 ) . note that @xmath86 whereas . let @xmath1 and @xmath61 denote the covariance and concentration graphs of @xmath0 , respectively . likewise , let @xmath87 and @xmath88 denote the covariance and concentration graphs of @xmath85 , respectively . note that @xmath1 , @xmath61 , @xmath87 and @xmath88 are all the complete graph over @xmath89 . now , let us assume that we are dealing with @xmath0 . then , no sound graphical criterion entails @xmath86 from @xmath1 because this independence does not hold in @xmath85 , and it is impossible to know whether we are dealing with @xmath0 or @xmath85 on the sole basis of @xmath1 . in this section , we present the main contribution of this paper : we introduce a graphical criterion for reading dependencies from the covariance graph of a wtc graphoid and prove that it is sound and complete in certain sense . if @xmath1 is the covariance graph of a wtc graphoid @xmath0 then we know , by definition of @xmath1 , that @xmath60 for all the edges @xmath74 in @xmath1 . we call these dependencies the dependence base of @xmath0 . further dependencies in @xmath0 can be derived from the dependence base via the wtc graphoid properties . for this purpose , we rephrase the wtc graphoid properties in their contrapositive form as follows . symmetry @xmath90 . decomposition @xmath91 . weak union @xmath92 . contraction @xmath93 is problematic for deriving new dependencies because it contains a disjunction in the consequent and , thus , we split it into two properties : contraction1 @xmath94 , and contraction2 @xmath95 . likewise , intersection gives rise to intersection1 @xmath96 , and intersection2 @xmath97 . note that intersection1 and intersection2 are equivalent and , thus , we refer to them simply as intersection . similarly , weak transitivity gives rise to weak transitivity1 @xmath98 , and weak transitivity2 @xmath99 . finally , composition @xmath100 gives rise to composition1 @xmath101 , and composition2 @xmath102 . since composition1 and composition2 are equivalent , we refer to them simply as composition . the independence in the antecedent of any of the properties above holds if it can be read off @xmath1 via the graphical criterion in definition [ def : inde2 ] . this is the best solution we can hope for because , as discussed in the previous section , this graphical criterion is sound and complete for wtc graphoids . moreover , this solution does not require more information than what it is available , namely @xmath1 or equivalently the dependence base of @xmath0 . we define the wtc graphoid closure of the dependence base of @xmath0 as the set of dependencies that are in the dependence base of @xmath0 plus those that can be derived from it by applying the nine properties above . let @xmath30 , @xmath31 and @xmath32 denote three disjoint subsets of @xmath7 . we say that @xmath30 is connected to @xmath31 given @xmath32 in an ug @xmath1 if there exist two nodes @xmath103 and @xmath104 such that there exists a * single * path between @xmath57 and @xmath58 in @xmath1 whose nodes are all outside @xmath105 . we denote such a connection statement by @xmath106 . we denote by @xmath107 that an ug @xmath1 represents that @xmath42 is dependent on @xmath43 given @xmath4 . we can now introduce our graphical criterion for reading dependencies from the covariance graph of a wtc graphoid . [ def : de ] given the covariance graph @xmath1 of a wtc graphoid , @xmath107 if @xmath108 . the following rephrasing of the graphical criterion in the previous definition may be easier to recall : @xmath107 if there exist two nodes @xmath103 and @xmath104 such that there exists a * single * path between @xmath57 and @xmath58 in @xmath1 whose nodes are all in @xmath109 . interestingly , the graphical criterion in the previous definition is dual to the following graphical criterion , which we developed in @xcite for reading dependencies from the concentration graph of a wtc graphoid . [ def : de2 ] given the concentration graph @xmath1 of a wtc graphoid , @xmath107 if @xmath106 . we proved in ( * ? ? ? * theorems 5 and 6 , example 3 ) that the graphical criterion in definition [ def : de2 ] is sound and complete in certain sense . we prove in the following two theorems that the graphical criterion in definition [ def : de ] is also sound and complete in certain sense . [ the : sound ] let @xmath1 be the covariance graph of a wtc graphoid @xmath0 . if @xmath107 , then @xmath44 is in the wtc graphoid closure of the dependence base of @xmath0 . let @xmath107 hold due to a path @xmath110 with @xmath103 and @xmath104 . we prove the theorem by induction over the length of @xmath110 . we first prove it for length one . let @xmath111 denote the largest subset of @xmath32 such that there is a path in @xmath1 from @xmath57 to every node in @xmath111 and all the nodes in these paths are in @xmath112 . then , @xmath113 because , otherwise , there would be two paths between @xmath57 and @xmath58 whose nodes are all in @xmath109 , which would contradict that @xmath107 holds due to @xmath110 . thus , @xmath114 . moreover , @xmath60 because @xmath57 and @xmath58 are adjacent in @xmath1 . then , @xmath115 by symmetry and decomposition , which together with @xmath114 imply @xmath116 by symmetry and contraction2 . note that if @xmath117 , then @xmath60 directly implies @xmath116 . in any case , @xmath116 implies @xmath118 by decomposition . now , note that @xmath119 by definition of @xmath111 and thus @xmath120 , which implies @xmath121 by symmetry and weak union , which together with @xmath118 imply @xmath122 by contraction2 . note that if @xmath123 , then @xmath116 directly implies @xmath122 . in any case , @xmath122 implies @xmath44 by symmetry and decomposition . assume as induction hypothesis that the theorem holds when the length of @xmath110 is smaller than @xmath124 . we now prove it for length @xmath124 . let @xmath125 be any node in @xmath110 except @xmath57 and @xmath58 . note that @xmath126 and thus @xmath127 , which implies @xmath128 . moreover , note that @xmath129 holds due to the subpath of @xmath110 between @xmath57 and @xmath125 , which we denote as @xmath130 . to see it , note that @xmath130 is the only path between @xmath57 and @xmath125 in @xmath1 whose nodes are all in @xmath131 , because if there were two such paths then there would be two paths between @xmath57 and @xmath58 in @xmath1 whose nodes are all in @xmath109 , which would contradict that @xmath107 holds due to @xmath110 . likewise , @xmath132 . moreover , @xmath129 and @xmath132 imply respectively @xmath133 and @xmath134 by the induction hypothesis , which together with @xmath128 imply @xmath122 by weak transitivity1 , which implies @xmath44 by symmetry and decomposition . finally , note that the above derivation of @xmath44 only makes use of the dependence base of @xmath0 and the nine properties introduced at the beginning of this section . thus , @xmath44 is in the wtc graphoid closure of the dependence base of @xmath0 . note that we do not make use of the composition property in the proof above . however , we do use the fact that the graphical criterion in definition [ def : inde2 ] is sound . the proof of this fact in ( * ? ? ? * ; * ? ? ? * proposition 2.2 ) does make use of the composition property . [ the : complete ] let @xmath1 be the covariance graph of a wtc graphoid @xmath0 . if @xmath44 is in the wtc graphoid closure of the dependence base of @xmath0 , then @xmath107 . let @xmath61 denote the concentration graph that has the same vertices and edges as @xmath1 . in other words , @xmath1 and @xmath61 are the same ug but with different interpretations . note that @xmath107 iff @xmath135 , which follows from the fact that @xmath107 iff @xmath108 iff @xmath136 iff @xmath135 . clearly , all the dependencies in the dependence base of @xmath0 are identified by the graphical criterion in definition [ def : de ] . therefore , it only remains to prove that this graphical criterion satisfies the nine properties introduced at the beginning of this section . we do so below with the help of @xmath61 . note that the graphical criterion in definition [ def : de2 ] applied to @xmath61 satisfies the nine properties introduced at the beginning of this section ( * ? ? ? * theorem 6 , example 3 ) . * symmetry @xmath137 . + trivial . * decomposition @xmath138 . + @xmath107 implies @xmath135 by definition , which implies @xmath139 by weak union , which implies @xmath140 by definition . * weak union @xmath141 . + @xmath142 implies @xmath143 by definition , which implies @xmath139 by decomposition , which implies @xmath140 by definition . * contraction1 @xmath144 . + @xmath140 and @xmath145 imply respectively and @xmath146 by definition , which imply by contraction2 , which implies @xmath147 by definition . * contraction2 @xmath148 . + @xmath140 and @xmath149 imply respectively @xmath139 and @xmath150 by definition , which imply @xmath143 by contraction1 , which implies @xmath142 by definition . * intersection @xmath151 . + @xmath140 and @xmath145 imply respectively and @xmath146 by definition , which imply by composition , which implies @xmath152 by definition . * weak transitivity1 @xmath153 . + @xmath154 , @xmath155 and @xmath54 imply respectively , @xmath156 and @xmath157 . moreover , and @xmath156 imply respectively and @xmath158 by symmetry and weak union , which together with @xmath157 imply respectively and @xmath159 by symmetry and contraction1 , which together with @xmath157 imply @xmath160 by weak transitivity2 , which implies @xmath161 by definition . * weak transitivity2 @xmath162 . + trivial because the antecedent involves a contradiction . to see it , note that @xmath154 , @xmath155 and @xmath163 imply respectively , @xmath156 and @xmath164 . moreover , @xmath165 and @xmath156 imply respectively and @xmath158 by symmetry and weak union , which together with @xmath164 imply respectively @xmath166 and @xmath159 by symmetry and composition , which together with @xmath164 imply a contradiction as shown in ( * ? ? ? * theorem 6 ) . * composition @xmath167 . + @xmath140 and @xmath54 imply respectively @xmath139 and @xmath157 by definition , which imply @xmath168 by intersection , which implies @xmath147 by definition . while theorem [ the : sound ] was somewhat expected because if there is a single path between @xmath57 and @xmath58 in @xmath1 whose nodes are all in @xmath109 then there is no possibility of path cancellation , the combination of theorems [ the : sound ] and [ the : complete ] is rather exciting : we now have a simple graphical criterion to decide whether a given dependence is or is not in the wtc graphoid closure of the dependence base of @xmath0 , i.e. we do not need to try to find a derivation of it , which is usually a tedious task . we devote the rest of this section to some observations that follow from the previous two theorems . a sensible question to ask is whether the graphical criterion in definition [ def : de ] is complete in the sense of being able to identify all the dependencies shared by all the wtc graphoids whose covariance graph is a given ug . the answer is no . an illustrative example follows . [ exa : example2 ] let @xmath1 denote the ug in table [ tab : example ] . consider any wtc graphoid @xmath0 whose covariance graph is @xmath1 . such wtc graphoids exist by theorems [ the : fcvg ] and [ the : fcvg2 ] . then , @xmath169 or @xmath170 because otherwise @xmath171 by intersection , which is a contradiction because @xmath172 implies @xmath173 by theorem [ the : sound ] . assume @xmath169 . note that @xmath174 implies @xmath175 by theorem [ the : sound ] . then , @xmath169 and @xmath175 together with @xmath176 , which follows from @xmath177 , imply @xmath178 by weak transitivity1 . likewise , @xmath179 when assuming @xmath170 . then , @xmath178 or @xmath179 , which imply @xmath180 by decomposition . however , @xmath181 does not hold . note that the fact that the graphical criterion definition [ def : de ] is not complete in the latter sense implies that it is neither complete in the more stringent sense of being able to identify all the dependencies in the wtc graphoid at hand . actually , no sound graphical criterion for reading dependencies from the covariance graph of a wtc graphoid can be complete in this more stringent sense . to see it , consider again example [ exa : example ] . let us assume that we are dealing with @xmath85 . then , no sound graphical criterion entails @xmath182 from @xmath87 because this dependence does not hold in @xmath0 , and it is impossible to know whether we are dealing with @xmath0 or @xmath85 on the sole basis of @xmath87 . it is worth mentioning that the graphical criteria in definitions [ def : de ] and [ def : de2 ] complement each other , as each of them can read dependencies than the other can not . to see it , consider the wtc graphoid @xmath0 in example [ exa : example ] . then , @xmath183 and thus @xmath60 by theorem [ the : sound ] . however , this dependence can not be derived from @xmath61 because @xmath184 does not hold . on the other hand , @xmath185 and thus @xmath186 by ( * ? ? ? * theorem 5 ) . however , this dependence can not be derived from @xmath1 because @xmath187 does not hold . again , a sensible question to ask is whether the joint use of the graphical criteria in definitions [ def : de ] and [ def : de2 ] is complete in the sense of being able to identify all the dependencies shared by all the wtc graphoids whose covariance and concentration graphs are two given ugs . the answer is no . an illustrative example follows . [ exa : example3 ] let @xmath1 denote the ug in table [ tab : example ] . let @xmath61 denote the complete graph over @xmath188 . consider any wtc graphoid @xmath0 whose covariance and concentration graphs are @xmath1 and @xmath61 , respectively . such wtc graphoids exist . to see it , it suffices to take any wtc graphoid that is faithful to @xmath1 , which exists by theorems [ the : fcvg ] and [ the : fcvg2 ] . recall that we have proven in example [ exa : example2 ] that @xmath189 . however , neither @xmath181 nor @xmath190 holds . note that the fact that the joint use of the graphical criteria in definitions [ def : de ] and [ def : de2 ] is not complete in the latter sense implies that it is neither complete in the more stringent sense of being able to identify all the dependencies in the wtc graphoid at hand . actually , no pair of sound graphical criteria for reading dependencies from the covariance and concentration graphs of a wtc graphoid can be complete in this more stringent sense . to see it , consider again example [ exa : example ] . let us assume that we are dealing with @xmath85 . then , no pair of sound graphical criteria entails @xmath182 from @xmath87 and @xmath88 because this dependence does not hold in @xmath0 , and it is impossible to know whether we are dealing with @xmath0 or @xmath85 on the sole basis of @xmath87 and @xmath88 . the following corollary extends to wtc graphoids a result originally proven in ( * ? ? ? * theorem 3 ) for gaussian probability distributions . the extension is straightforward thanks to the graphical criterion in definition [ def : de ] . let @xmath1 be the covariance graph of a wtc graphoid @xmath0 . if @xmath1 is a forest , then @xmath0 is faithful to @xmath1 . assume to the contrary that @xmath0 is not faithful to @xmath1 . since @xmath1 is the covariance graph of @xmath0 , the previous assumption is equivalent to assume that there exist three disjoint subsets of @xmath7 , here denoted @xmath30 , @xmath31 and @xmath32 , such that @xmath53 but @xmath41 . however , @xmath53 implies that there must exist a path in @xmath1 between some node @xmath103 and some node @xmath104 whose nodes are all in @xmath109 . furthermore , since @xmath1 is a forest , that must be the only such path between @xmath57 and @xmath58 in @xmath1 . however , this implies @xmath107 and thus @xmath44 by theorem [ the : sound ] , which is a contradiction . given the covariance graph @xmath1 of a wtc graphoid @xmath0 , @xmath53 does not imply @xmath44 . this is actually the reason of being of this paper . however , if @xmath1 is a forest , then the previous corollary proves that @xmath53 does imply @xmath44 and , moreover , that this way of reading dependencies from @xmath1 is complete in the strictest sense discussed above . the following corollary extends to wtc graphoids a result originally proven in ( * ? ? ? * lemma 5 ) for gaussian probability distributions . the extension is straightforward thanks to the graphical criteria in definitions [ def : de ] and [ def : de2 ] . let @xmath1 and @xmath61 be , respectively , the covariance and concentration graphs of a wtc graphoid @xmath0 . then , @xmath1 and @xmath61 have the same connected components . moreover , if a connected component in @xmath1 ( respectively @xmath61 ) is a tree then the corresponding connected component in @xmath61 ( respectively @xmath1 ) is the complete graph . first , we prove that @xmath1 and @xmath61 have the same connected components . if two nodes @xmath57 and @xmath58 belong to the same connected component in @xmath1 , then @xmath191 for some @xmath192 and thus @xmath122 by theorem [ the : sound ] . however , if @xmath57 and @xmath58 belong to different connected components in @xmath61 , then @xmath193 and thus @xmath194 , which is a contradiction . on the other hand , if two nodes @xmath57 and @xmath58 belong to the same connected component in @xmath61 , then @xmath195 for some @xmath192 and thus @xmath122 by ( * ? ? ? * theorem 5 ) . however , if @xmath57 and @xmath58 belong to different connected components in @xmath1 , then @xmath196 and thus @xmath194 , which is a contradiction . now , take any connected component @xmath197 in @xmath1 that is a tree . we prove that the corresponding connected component @xmath198 in @xmath61 is the complete graph . if two nodes @xmath57 and @xmath58 belong to @xmath197 , then @xmath199 and thus @xmath59 by theorem [ the : sound ] , which implies that @xmath57 and @xmath58 are adjacent in @xmath198 . finally , take any connected component @xmath198 in @xmath61 that is a tree . we prove that the corresponding connected component @xmath197 in @xmath1 is the complete graph . if two nodes @xmath57 and @xmath58 belong to @xmath198 , then @xmath184 and thus @xmath60 by ( * ? ? ? * theorem 5 ) , which implies that @xmath57 and @xmath58 are adjacent in @xmath197 . note that the opposite of the second statement in the previous corollary is not true . the following example illustrates this . let @xmath1 ( respectively @xmath61 ) be the covariance ( respectively concentration ) graph of a wtc graphoid @xmath0 . assume that @xmath1 ( respectively @xmath61 ) is the complete graph and @xmath0 is faithful to it . such wtc graphoids exist by theorems [ the : fcvg ] and [ the : fcvg2 ] ( respectively theorems [ the : fcg ] and [ the : fcg2 ] ) . then , @xmath0 has no independencies . consequently , @xmath61 ( respectively @xmath1 ) is the complete graph and @xmath0 is faithful to it . in this paper , we have provided new insight into covariance graphs by introducing a graphical criterion for reading dependencies from the covariance graph of a wtc graphoid . we have shown that wtc graphoids are not a too restrictive family of probability distributions by showing that it includes interesting discrete , gaussian , and continuous but non - gaussian probability distributions . we have proven that the new graphical criterion is sound and complete in certain sense . in order to prove these properties , we have had to prove first that the graphical criterion in @xcite for reading independencies from the covariance graph of a wtc graphoid is complete in certain sense . we have done so by proving that there are discrete , gaussian , and continuous but non - gaussian probability distributions that are faithful to any covariance graph . this result is also important because it implies that there exist probability distributions that covariance graphs can represent exactly but cgs can not . therefore , covariance graphs complement cgs . the following example illustrates this . consider the covariance graph @xmath200 . consider any cg @xmath61 that represents the same independencies as @xmath1 . note that @xmath61 must have some edge between @xmath57 and @xmath58 , @xmath58 and @xmath125 , @xmath125 and @xmath201 , and @xmath201 and @xmath57 . however , @xmath57 and @xmath125 can not be adjacent in @xmath61 because @xmath202 . likewise , @xmath58 and @xmath201 can not be adjacent in @xmath61 because @xmath203 . then , @xmath204 , @xmath205 because otherwise either @xmath206 or @xmath207 or @xmath208 , whereas @xmath209 and @xmath210 and @xmath211 . however , such an @xmath61 implies @xmath212 whereas @xmath203 . consequently , no cg can represent the same independencies as @xmath1 . this implies that every probability distribution that is faithful to @xmath1 ( recall that such probability distributions exist by theorem [ the : fcg2 ] ) is not faithful to any cg . it is fair mentioning that we are not the first to note that covariance graphs complement other more popular graphical models . for instance , it follows from @xcite that covariance graphs complement dags . our example above simply extends this observation to cgs . another consequence of the faithfulness result in theorem [ the : fcg2 ] is that it proves wrong the misconception that covariance graphs are densely connected because their edges represent marginal dependencies . specifically , the theorem implies that there are probability distributions that are faithful to any covariance graph , no matter its topology . interestingly , the graphical criterion developed in this paper is dual to the one presented in @xcite for reading dependencies from the concentration graph of a wtc graphoid . this duality resembles the duality existing between the graphical criteria for reading independencies from concentration and covariance graphs @xcite . we have also shown that the new graphical criterion and the one presented in @xcite complement each other , as there may be dependencies that only one of them can identify . finally , we have pointed out some limitations of the graphical criterion introduced in this paper that suggest future lines of research . for instance , it remains an open question whether it is possible to develop a similar graphical criterion that is complete in a stricter sense than the one used in this paper . it also remains an open question whether our faithfulness result in appendix a for regular gaussian probability distributions can be extended to discrete probability distributions with the help of the parameterizations in @xcite . another line of action is the application of our graphical criterion in bioinformatics . in such an application , the covariance graph has to be learnt from gene expression data via , for instance , hypothesis tests . the data available for learning is typically scarce due to the high cost associated with its production . in this scenario , covariance graphs are easier to learn than bayesian networks or concentration graphs , which are the graphical models commonly used in bioinformatics , because the former involves testing for marginal ( in)dependencies whereas the latter involve testing for conditional ( in)dependencies . we do not suggest with this that one should quit using bayesian networks and concentration graphs in bioinformatics . specifically , bayesian networks have an important advantage over covariance graphs , namely that they can provide us with insight into the mechanistic or causal process underlying the data at hand . what we suggest is that when the data at hand is not considered enough to learn a reliable bayesian network , one may be willing to learn a less informative but more reliable model such as a covariance graph , particularly now that we have graphical criteria for reading both dependencies and independencies off it . note that this paper only studies the structure of covariance graphs and , thus , it does not deal with their parameterization and/or parameter estimation . the interested reader is referred to @xcite for gaussian models and @xcite for discrete models . however , it may be worth warning here that finding the maximum likelihood estimates of the parameters of a covariance graph can be hard , depending on the parameterization considered . this is particularly true for discrete models @xcite . the reason is that a marginal independence can imply complicated parameter constraints in some parameterizations , because a marginal dependence is a property of the joint probability distribution rather than of the relevant marginal probability distribution . an early work that showed the latter is @xcite , where the author gives an example of two - way interactions implying three - way interaction . in summary , learning the structure of a covariance graph may be simpler than learning the structure of a concentration graph . however , estimating the parameters of the former may be harder than estimating the parameters of the latter . we strengthen the second statement in theorem [ the : fcvg ] by proving that , in certain measure - theoretic sense , almost all the regular gaussian probability distributions that are markov wrt a covariance graph are faithful to it . although this result is not used in this paper , we consider it to be interesting in its own and , thus , we decide to report on it . we start by recalling some results from matrix theory . see , for instance , @xcite for more information . let @xmath213 denote a square matrix . let @xmath214 with @xmath215 denote its submatrix @xmath216 . the determinant of @xmath217 can recursively be computed , for fixed @xmath218 , as @xmath219 , where @xmath220 denotes the matrix produced by removing the row @xmath221 and column @xmath222 from @xmath217 . note then that @xmath223 is a real polynomial in the entries of @xmath217 . if @xmath224 then the inverse of @xmath217 can be computed as @xmath225 for all @xmath226 . we say that @xmath217 is strictly diagonally dominant if @xmath227 for all @xmath218 , where @xmath228 denotes absolute value . a matrix @xmath217 is hermitian if it is equal to the matrix resulting from , first , transposing @xmath217 and , then , replacing each entry by its complex conjugate . clearly , a real symmetric matrix is hermitian . a real symmetric @xmath229 matrix @xmath217 is positive definite if @xmath230 for all non - zero @xmath231 . we continue by proving some auxiliary results . we assume hereinafter that the sample space of each random variable in @xmath39 is @xmath65 . let @xmath232 denote the set of regular gaussian probability distributions @xmath0 such that @xmath233 for any two nodes @xmath57 and @xmath58 that are not adjacent in @xmath1 . note that these are exactly the regular gaussian probability distributions that are markov wrt @xmath1 ( * ? ? ? * ; * ? ? ? * proposition 2.2 ) . we parameterize each probability distribution @xmath234 with its mean vector @xmath235 and covariance matrix @xmath236 . note that the values of some of these parameters are determined by the values of other parameters or by @xmath1 . specifically , the following constraints apply : * @xmath237 for all @xmath238 because @xmath236 is symmetric . * @xmath239 for all @xmath238 such that @xmath221 and @xmath222 are not adjacent in @xmath1 . to see it , recall that if @xmath221 and @xmath222 are not adjacent in @xmath1 then @xmath240 and , thus , @xmath239 ( * ? ? * corollary 2.3 ) . hereinafter , the parameters whose values are not determined by the constraints above are called non - determined ( nd ) parameters . however , the values the nd parameters can take are further constrained by the fact that these values must correspond to some probability distribution in @xmath232 . in other words , the values the nd parameters can take are constrained by the fact that @xmath236 must be positive definite . that is why the set of nd parameter values satisfying this requirement are hereinafter called the nd parameter space for @xmath232 . we do not work out the inequalities defining the nd parameter space because they are irrelevant for our purpose . the number of nd parameters is what we call the dimension of @xmath1 , and we denote it as @xmath241 . specifically , @xmath242 where @xmath243 and @xmath244 are , respectively , the number of nodes and edges in @xmath1 : * @xmath243 due to @xmath235 . * @xmath243 due to entries in the diagonal of @xmath236 . * @xmath244 due to the entries below the diagonal of @xmath236 that are not identically zero . to see this , recall from the constraint c2 that there is one entry below the diagonal of @xmath236 that is not identically zero for each undirected edge in @xmath1 . [ lem : f1 ] let @xmath1 be a covariance graph of dimension @xmath241 . the nd parameter space for @xmath232 has positive lebesgue measure wrt @xmath245 . since we do not know a closed - form expression of the nd parameter space for @xmath232 , we take an indirect approach to prove the result . recall that , by definition , the nd parameter space for @xmath232 is the set of real values such that , after the extension determined by the constraints c1 and c2 , @xmath236 is positive definite . therefore , the nd parameters @xmath235 can take values independently of the nd parameters in @xmath236 . however , the nd parameters in @xmath236 can not take values independently one of another because , otherwise , @xmath236 may not be positive definite . however , if the entries in the diagonal of @xmath236 take values in @xmath246 and the rest of the nd parameters in @xmath236 take values in @xmath247 $ ] , then the nd parameters in @xmath236 can take values independently one of another . to see it , note that in this case @xmath236 will always be hermitian , strictly diagonally dominant , and with strictly positive diagonal entries , which implies that @xmath236 will always be positive definite ( * ? ? ? * corollary 7.2.3 ) . the subset of the nd parameter space of @xmath232 described in the paragraph above has positive volume in @xmath245 and , thus , it has positive lebesgue measure wrt @xmath245 . then , the nd parameter space of @xmath232 has positive lebesgue measure wrt @xmath245 . [ lem : polynomial ] let @xmath1 be a covariance graph . for every @xmath226 and @xmath248 , there exists a real polynomial @xmath249 in the nd parameters in the parameterization of the probability distributions in @xmath232 such that , for every @xmath234 , @xmath250 iff @xmath249 vanishes for the nd parameter values coding @xmath0 . let @xmath236 denote the covariance matrix of @xmath0 . note that @xmath250 iff @xmath251 ( * ? ? ? * proposition 5.2 ) . recall that @xmath252 with @xmath253 moreover , note that @xmath254 because @xmath236 is positive definite @xcite . then , @xmath250 iff @xmath255 . moreover , as noted in section [ sec : preliminaries ] , @xmath256 is a real polynomial in the entries of @xmath236 . thus , @xmath250 iff a real polynomial @xmath257 in the entries of @xmath236 vanishes . recall that each entry of @xmath236 that is not identically zero corresponds to one of the nd parameters in the parameterization of the probability distributions in @xmath232 . therefore , @xmath257 can be expressed as a real polynomial @xmath249 in the nd parameters . therefore , @xmath250 iff @xmath249 vanishes for the nd parameter values coding @xmath0 . we interpret the polynomial in the previous lemma as a real function on a real euclidean space that includes the nd parameter space for @xmath232 . we say that the polynomial in the previous lemma is non - trivial if not all the values of the nd parameters are solutions to the polynomial . this is equivalent to the requirement that the polynomial is not identically zero . [ lem : f2 ] let @xmath1 be a covariance graph of dimension @xmath241 . the subset of the nd parameter space for @xmath232 that corresponds to the probability distributions in @xmath232 that are not faithful to @xmath1 has zero lebesgue measure wrt @xmath245 . recall that @xmath232 contains exactly the regular gaussian distributions that are markov wrt @xmath1 . therefore , for any probability distribution @xmath234 not to be faithful to @xmath1 , @xmath0 must satisfy some independence that is not entailed by @xmath1 . that is , there must exist three disjoint subsets of @xmath7 , here denoted as @xmath29 , @xmath258 and @xmath259 , such that @xmath260 but @xmath261 . however , if @xmath260 then @xmath262 for some @xmath263 and @xmath264 . furthermore , if @xmath261 then @xmath250 by symmetry and decomposition . by lemma [ lem : polynomial ] , there exists a real polynomial @xmath249 in the nd parameters in the parameterization of the probability distributions in @xmath232 such that , for every @xmath265 , @xmath266 iff @xmath249 vanishes for the nd parameter values coding @xmath63 . furthermore , @xmath249 is non - trivial ( * ? ? ? * theorem 3.1 ) . let @xmath267 denote the set of solutions to the polynomial @xmath249 . then , @xmath267 has zero lebesgue measure wrt @xmath245 because it consists of the solutions to a non - trivial real polynomial in real variables ( the nd parameters ) @xcite . then , @xmath268 has zero lebesgue measure wrt @xmath245 , because the finite union of sets of zero lebesgue measure has zero lebesgue measure too . consequently , the subset of the nd parameter space for @xmath232 that corresponds to the probability distributions in @xmath232 that are not faithful to @xmath1 has zero lebesgue measure wrt @xmath245 because it is contained in @xmath269 . in summary , it follows from lemmas [ lem : f1 ] and [ lem : f2 ] that , in the measure - theoretic described , almost all the elements of the nd parameter space for @xmath232 correspond to probability distributions in @xmath232 that are faithful to @xmath1 . since this correspondence is clearly one - to - one , it follows that almost all the regular gaussian distributions in @xmath232 are faithful to @xmath1 . we think that this result can easily be extended to strictly positive discrete probability distributions with the help of the parameterizations proposed in @xcite . we do not elaborate further on this issue in this paper though . a word of caution is due at this point . it may be tempting to infer from the measure - theoretic results above that every regular gaussian probability distribution @xmath0 one encounters in reality is almost surely faithful to its covariance graph @xmath1 . this may lead one to conclude that our graphical criterion for reading from @xmath1 dependencies holding in @xmath0 is not needed , since @xmath53 almost surely implies @xmath44 . this argument is valid if @xmath0 has been drawn from @xmath232 at random . however , we believe that @xmath0 is more likely to have been carefully engineered ( e.g. by natural evolution in the case of the gene networks mentioned in section [ sec : intro ] ) than to have been drawn at random . consequently , and despite the measure - theoretic results above , we think that one can not safely assume that @xmath0 is almost surely faithful to @xmath1 . hence , the need of the graphical criterion proposed in this paper . we thank the anonymous reviewers and the editor for their thorough review of this manuscript . we are particularly grateful to the second reviewer for suggesting us to use copulas in the proofs of theorems [ the : fcg2 ] and [ the : fcvg ] . this work is funded by the center for industrial information technology ( ceniit ) and a so - called career contract at linkping university . mathias drton and thomas s. richardson . a new algorithm for maximum likelihood estimation in gaussian graphical models for marginal independence . in _ proceedings of the nineteenth conference on uncertainy in artificial intelligence _ , 184 - 191 , 2003 . jose m. pea , roland nilsson , johan bjrkegren and jesper tegnr . identifying the relevant nodes without learning the model . in _ proceedings of the twentysecond conference on uncertainy in artificial intelligence _ , 367 - 374 , 2006 . jose m. pea , roland nilsson , johan bjrkegren and jesper tegnr . an algorithm for reading dependencies from the minimal undirected independence map of a graphoid that satisfies weak transitivity . _ journal of machine learning research _ , 10:1071 - 1094 , 2009 .

by combining the numerical solution of the nonlinear hydrodynamics equations with the solution of the linear inhomogeneous zerilli - moncrief and regge - wheeler equations we investigate the properties of the gravitational radiation emitted during the axisymmetric accretion of matter onto a schwarzschild black hole . the matter models considered include quadrupolar dust shells and thick accretion disks , permitting us to simulate situations which may be encountered at the end stages of stellar gravitational collapse or binary neutron star merger . we focus on the interference pattern appearing in the energy spectra of the emitted gravitational waves and on the amount of excitation of the quasi - normal modes of the accreting black hole . we show that , quite generically in the presence of accretion , the black hole ringdown is not a simple superposition of quasi - normal modes , although the fundamental mode is usually present and often dominates the gravitational - wave signal . we interpret this as due to backscattering of waves off the non - exponentially decaying part of the black hole potential and to the finite spatial extension of the accreting matter . our results suggest that the black hole qnm contributions to the full gravitational wave signal should be extremely small and possibly not detectable in generic astrophysical scenarios involving the accretion of extended distributions of matter .

gr-qc0610131

it is well known within the framework of black - hole perturbation theory @xcite that quasi - normal modes ( qnms ) ( i.e. , exponentially damped harmonic oscillations ) dominate the gravitational - wave response of a non - spherically distorted black hole if the corresponding frequencies are part of the fourier spectrum of the external source that moved the hole away from its equilibrium state . this perturbation is then radiated away in form of gravitational radiation until the black hole returns to its unperturbed , quiescent state . in practice , the qnms excitation is triggered if the frequency of the perturbing agent ( e.g. an external matter source moving close to the black hole ) is sufficiently close to the fundamental frequencies of the black hole , which then acts as an excited oscillator . as a result , the qnms and in particular the fundamental mode ( which is the one at the highest frequency and with the smallest damping time ) represent the main feature of the gravitational waves emission only for a sufficiently compact perturbation ; i.e. , for sources whose characteristic scale is comparable with the width of the peak of the potential . these results are well - known since the early studies of press @xcite ( see also vishveshwara @xcite ) , who considered the scattering of gaussian gravitational - wave packets off a schwarzschild black hole and noticed that the excitation of the qnms of the black hole is more efficient for very narrow packets . approximate relations to compute the efficiency of excitation of the various qnms were introduced by leaver @xcite and by andersson @xcite for gaussian pulses initial data with variable width . since the qnms spectrum is entirely determined by the black hole properties ( i.e. , mass , spin and charge ) , it is expected that the detection of the qnms ringdown would provide a unique opportunity to unveil the physical properties of a black hole . for this reason , the excitation of black hole qnms have been studied in various astrophysical scenarios , such as the gravitational collapse of a ( rotating ) neutron star to a black hole or the collision of two black holes . the presence of the qnms in these situations has been confirmed either through the use of perturbation theory with various degrees of sophistication @xcite , or through fully relativistic numerical simulations @xcite . however , under more general and realistic conditions , such as the excitation of the qnms by accretion of matter , and that may be encountered in gravitational stellar collapse or binary neutron star mergers , the gravitational - wave response of a black hole can be more complex . for instance , the simulations performed by @xcite showed that the gravitational - wave signal is not simply given by the superposition of exponentially - damped sinusoids at the qnms frequencies and that , in some cases , the qnm ringing is only weakly excited and analysis in the frequency domain are needed @xcite . on the other hand , `` backscattering '' effects related to the slowly - decaying features of the scattering potential and interference effects turn out to play a crucial role for the correct interpretation of the results ; a detailed discussion of these effects can be found in ref . @xcite . in the case of a schwarzschild black hole of mass @xmath0 , we recall that the appearance of qnms ringing is related to the peak of the curvature potential ( also referred to as the zerilli or regge - wheeler potential ) , that is located at @xmath1 ; the backscattering , on the other hand , is related to the fact that the curvature potential decays as @xmath2 for @xmath3 . this behavior is responsible for the late time power - law tail @xmath4 of the gravitational - wave signal , where @xmath5 refers to the radiation multipole . therefore , while early - time ringing is the result of a superposition of exponentially damped sinusoids , and is dominated by the fundamental quasi - normal mode , at later times the ringing dies out and the signal is dominated by tail effects . however , in the transition from the `` ringdown '' to the `` tail '' phase , additional oscillations appear that can not be attributed to any of the two regimes and that also seem to depend on the choice of the initial data ( see @xcite ) . as we shall show in this paper for a broad sample of initial data , there could be intermediate regimes where the `` ringdown '' and the `` tail '' terms of the potential produce competing effects , so that the qnms ringing and the backscattering effects can overlap , generating complex waveforms . these effects were first noticed in refs . @xcite , although not discussed in detail there . in a recent paper @xcite , hereafter paper i , a general analysis of the gravitational radiation emitted as a result of anisotropic accretion of matter shells onto nonrotating black holes and neutron stars was presented . that investigation made use of a procedure that combines the solution of the linearized einstein equations for the metric perturbations with fully nonlinear hydrodynamics simulations . although the study of black hole perturbations produced by infalling matter has a long history and rich literature @xcite , the approach outlined in paper i proved to be useful for a number of reasons : _ i ) _ it provided additional information on the black - hole s response to the dynamics of point - like particles in the vicinity of black holes ; _ ii ) _ it helped understanding the basic black - hole s response to extended matter perturbations ; _ iii ) _ it represented an effective way of studying black - hole physics in a linear regime without having to resort to full - scale numerical relativity simulations . one of the main results of paper i was that , in the idealized accretion processes considered , most of the energy is released at frequencies lower than that of the fundamental quasi - normal mode ( qnm ) of the black hole , the spectrum consisting of a complex pattern , mostly produced during the accretion process rather than in the ringdown phase . more precisely , the gravitational - wave emission was found to be dominated by a collection of interference `` fringes '' at frequencies of about a few hundred hz , rather than by a single monochromatic peak at the ( higher ) frequency of the fundamental mode of the black hole . moreover , the width of these fringes was found to decrease rapidly with the initial position of the matter source . these results , which were already observed in other works @xcite in the case of a point - like particle falling onto a schwarzschild black hole , also showed that the appearance of interference fringes in the energy spectra is much larger when the accreting matter is a shell of finite size and that the efficiency in gravitational - wave emission is much reduced , becoming almost two orders of magnitude smaller than in the case of point - like particles . an important feature of the calculations carried out in paper i was the minimization of the initial gravitational - wave content ; this turned out to be crucial to illustrate that the interference pattern was mainly due to the finite radial extension of the accreting source . the aim of the present paper is twofold . firstly , we intend to complete the discussion started in paper i on accreting quadrupolar shells onto a schwarzschild black hole by extending the parameter space of the initial models and by analyzing their impact on the gravitational - wave emission . in particular we study how the energy emitted in gravitational waves , and the corresponding spectra , depend on the compactness of the shells as well as on their initial locations . in doing so we show that , for a finite - size source , the ringing of the black hole is much more complex than a simple superposition of qnms and that the energy spectra ( and in particular the interference fringes ) are dependent on the choice of the initial data . secondly , we improve the astrophysical relevance of our study by analyzing the gravitational radiation produced by thick accretion disks @xcite which accrete onto the black hole on dynamical timescales . we recall , in fact , that quasi - periodic oscillations of thick accretion disks ( or tori ) orbiting around schwarzschild or kerr black holes have been recently addressed as promising sources of gravitational waves @xcite in the khz range . the paper is organized as follows : in sec . [ analytic ] we review the theory of odd- and even - parity nonspherical perturbations of schwarzschild spacetime , writing the inhomogeneous zerilli - moncrief and regge - wheeler equations in a form suitable for time - domain calculations . [ num_app ] briefly describes the numerical approach adopted for the simulations , while sec . [ results ] is devoted to the discussion of the results . finally , sec . [ conclusions ] provides a summary of the most important results and presents our conclusions . unless otherwise specified , we choose geometrized units ( @xmath6 ) , and the black hole mass @xmath0 is the unit of length . the theory of gauge - invariant nonspherical metric perturbations of a schwarzschild spacetime has a long history which has been recently reviewed in refs . we here simply recall that in this approach the spacetime metric @xmath7 is described by the background schwarzschild metric @xmath8 plus a nonspherical perturbation @xmath9 . as a result of the spherical symmetry , @xmath9 can be expanded in three odd - parity and seven even - parity multipoles @xmath10 with the odd multipoles transforming as @xmath11 and the even ones as @xmath12 under a parity transformation @xmath13 . the presence of matter around the black hole is accounted through a `` source '' term in the linearized einstein s equations , represented by a stress energy tensor @xmath14 which can also be expanded in multipoles . from the multipoles of the perturbed metric , it is possible to build odd , @xmath15 , and even - parity , @xmath16 , gauge invariant quantities which are solution of two equations , the regge - wheeler @xcite and zerilli - moncrief @xcite equations , respectively and whose expression in schwarzschild coordinates is given by @xmath17 here the upper indices refer to the odd and even case respectively and @xmath18 $ ] is the `` tortoise '' coordinate ( see refs . @xcite for the most general covariant form of these equations ) . explicit expressions for the scattering potentials @xmath19 and @xmath20 are well known and can be found , for example , in ref . the source terms in schwarzschild coordinates can also be found in @xcite , while the corresponding expressions valid for any coordinate slicing of schwarzschild spacetime are reported in detail in ref . once @xmath21 and @xmath22 are known , the gravitational - wave amplitude can be computed as @xmath23 where @xmath24 are the @xmath25 spin - weighted spherical harmonics @xcite and the emitted power in gravitational waves is simply given by @xmath26 where the overdot refers to a derivative with respect to the schwarzschild coordinate time . specifying suitable initial data for nonspherical metric perturbations taking place in astrophysical events is not trivial . the standard approach , in the case of even - parity perturbations , exploits the fact that @xmath21 can be written in terms of two gauge - invariant multipoles , namely the perturbed conformal factor @xmath27 and the gravitational - wave degree of freedom @xmath28 as introduced , for instance , in eq . ( 8) of paper i and whose relationship with the zerilli - moncrief equation is given by eq . ( 18 ) of the same reference @xcite as a result , the perturbed hamiltonian equation and the equation for the momentum constraint , i.e. , eqs . ( 45 ) and ( 46 ) of @xcite , provide a system of coupled ordinary differential equations for the unknowns ( @xmath27 , @xmath28 , @xmath29 , and @xmath30 ) . the indetermination inherent in the solution of this system can be overcome , for example , by assuming @xmath31 @xcite , where @xmath32 , so that the hamiltonian constraint simply reads @xmath33\partial_{r _ * } k_{\ell m}\nonumber\\ & -\left(1-\dfrac{2m}{r}\right)\left[\dfrac{\lambda}{r^2}+ \beta\dfrac{\lambda+2}{2r^2}\right]k_{\ell m}=-8\pi t^{\lm}_{00 } \ \ , \end{aligned}\ ] ] where @xmath34 and @xmath35 are the multipoles of the energy density @xmath36 of the matter source according to the multipolar decomposition given by eq . ( 13 ) of paper i ( see also ref . the case @xmath37 corresponds to conformally flat initial data , since in this case the perturbed metric in isotropic coordinates is the schwarzschild one modulo a conformal transformation . depending on the problem under investigation , we can either choose initial conditions that are time - symmetric ( i.e. , for which @xmath38 ) or not . in the first case , if the hamiltonian constraint is satisfied , the momentum constraint is automatically satisfied too . in the second case , on the other hand , the momentum constraint must also be solved for @xmath29 if we provide a suitable ansatz for @xmath39 . whether it is possible to use time - symmetric initial data depends essentially on the astrophysical scenario under investigation and thus on the form of the stress - energy tensor of the matter , whose multipoles appear as sources in the hamiltonian and momentum constraint equations . clearly , time - symmetry is satisfied for matter configurations in equilibrium , or initially at rest and subsequently falling radially ; in this case only the numerical solution of the hamiltonian constraint is needed . this is the case , for example , of dust shells initially at rest at a finite radius . because this configuration is intrinsically non time symmetric , it will invevitably produce spurious radiation . a way to minimize the latter is to freeze the sources of the perturbation equations up until the radiation has left the grid . this is the approach we will follow in most of our simulations . the numerical approach adopted in our simulations is the same as the one described in paper i. we only recall here that our hybrid approach implies the solution of both the nonlinear , relativistic hydrodynamics equations on a fixed schwarzschild background , in the two spatial dimensions @xmath40 and @xmath41 ( since we restrict ourselves to axisymmetric flows ) , and of the regge - wheeler and zerilli - moncrief , in the radial direction @xmath42 . such `` test - fluid approximation '' is valid as long as the mass of the accretion flow is much smaller than the mass of the black hole ; in our simulations this mass ratio is @xmath43 the hydrodynamics equations , once cast in conservation form , are solved using godunov - type methods based on riemann solvers ( see e.g. @xcite and references therein ) , while eqs . ( [ r - w ] ) can be written as a first - order hyperbolic system over a one - dimensional ( 1d ) grid expressed in terms of the radial coordinate @xmath42 and solved with some standard finite - difference method such as lax - wendroff ( see paper i for details ) . the computational domain used for the hydrodynamical evolution consists of @xmath44 grid - points , geometrically distributed along @xmath40 [ that is , with grispacing @xmath45 with @xmath46 and uniformly distributed along @xmath41 . when we evolve quadrupolar shells of dust plunging from large distances we set @xmath47 and @xmath48 , while for the evolution of fluid accretion disks we use @xmath49 and @xmath50 to reach the desired truncation error . the hydrodynamical grid extends from a point slightly outside the event horizon up to a finite radius , that depends on the initial data , much smaller than the extraction radius . at the inner boundary we impose inflowing boundary conditions , while at the outer one we fix the conditions of a tenuous stationary spherical atmosphere . the radial grid used for the time evolution of the regge - wheeler and zerilli - moncrief equations , on the other hand , is much more extended in radius than the hydrodynamical one and overlaps with the latter . to avoid spurious boundary effects , the 1d grid typically extends from @xmath51 to @xmath52 and is covered by @xmath53 cells . at the inner and outer boundary of this grid we use standard sommerfeld outgoing conditions . we stress that , due to the long - range action of the potential , these conditions are not completely non - reflecting , which is the reason why we still need large grids to properly capture the late time power - law decay of the waveforms ( see below ) . we recall in passing that exactly non - reflecting boundary condition for the zerilli - moncrief and regge - wheeler equations have been recently proposed in @xcite . while the gravitational waveforms are extracted using both the regge - wheeler and zerilli - moncrief equations , we also use as a comparison the newtonian quadrupole formula to compute the amount of gravitational waves emitted . we use the modified definition of the quadrupole moment proposed in ref . @xcite and which reads @xmath54 where @xmath55 is the conserved rest - mass density , @xmath56 is the relativistic lorentz factor , @xmath57 is the relativistic enthalpy , @xmath58 is the pressure , and @xmath59 . expression represents a slight modification of the newtonian standard quadrupole formula ( sqf@xmath60 ) and it was shown to provide the best agreement with the gauge - invariant waveforms when compared with alternative definitions @xcite . as mentioned in the introduction , for a certain set of initial conditions the ringdown phase of the gravitational - wave signal from a perturbed black hole shows the presence of the first qnms of the black hole ( typically , the fundamental mode and the first overtone ) together with `` backscattering '' or `` tail''effects . while the less damped modes are related to the interaction of the perturbations with the peak of the black hole regge - wheeler potential , the tail - effect can not be associated to qnms and are instead due to the long - range features of the potential and in particular to its slow decay with radius . + in order to single out the contribution of the peak of the curvature potential on the generation of the resulting gravitational - wave spectrum , while suppressing the effect of the tail term , we solve a simplified version of eq . ( [ r - w ] ) in which the regge - wheeler potential is replaced by the so - called pschl teller potential @xcite , which is exponentially decaying for @xmath61 . we therefore evolve in time an equation of the type @xmath62 where @xmath63 is given by @xmath64 } \ ] ] and @xmath65 is the position of the peak of the regge - wheeler potential @xmath66 and @xmath67 is determined through the matching of the second derivatives at @xmath65 ; i.e. , @xmath68 at @xmath69 . in the following , for simplicity we will consider only the @xmath70 odd - parity perturbations . the pschl teller potential is shown in the top left panel of fig . [ fig_2 ] together with @xmath71 . although the exponential decay of the former is evident for large @xmath42 , it is also apparent that it provides a good approximation to the regge - wheeler potential near the peak . the accuracy of the approximation was first studied in ref . @xcite , where it was pointed out that the qnms of the pschl - teller potential can be computed analytically . the frequencies found for the lower modes agree within a few percent with those obtained using the true curvature potential @xmath72 , computed numerically in @xcite . we solve eq . ( [ wave_equation ] ) using a lax - wendroff method on an evenly spaced @xmath42 grid with @xmath73 and with initial data given by a gaussian pulse @xmath74 $ ] where @xmath75 is the initial position of the pulse and @xmath76 a normalization constant . the initial pulse is considered to be purely ingoing ( i.e. , with @xmath77 ) . we compare the signal extracted at @xmath78 for different values of @xmath79 using either the regge - wheeler or the pschl - teller potential . this is summarized in fig . [ fig_2 ] , which shows the waveforms obtained for different values of @xmath80 and @xmath79 and expressed in the retarded time , @xmath81 . note that the waveforms from the pschl - teller have been shifted in time so as to overlap the maxima and the minima of the ringdown phase of the regge - wheeler potential . this is necessary because the scattering of the pulse starts earlier for the regge - wheeler potential than for the pschl - teller one . let us focus first on the top right panel of fig . [ fig_2 ] , which refers to a very narrow pulse ( @xmath82 ) initially located at @xmath83 and shows with a solid and a dashed line the waveforms obtained with the regge - wheeler and pschl - teller potentials , respectively . clearly , the ringing is very similar in the two cases both in the wavelength and in the amplitude , but differences become apparent after @xmath84 , when the tail term of the regge - wheeler potential becomes dominant and the gravitational - wave signal is driven by the backscatter due to the @xmath2 decay and asymptotes the expected late - time decay according to price s law @xcite . since the pschl - teller potential decays exponentially , no tail effects are found and the signal is still given by a superposition of exponentially damped harmonic oscillations . smaller differences between the waveforms are however present also in the early part of the waveform ( cf . , the inset of the top right panel of fig . [ fig_2 ] ) and are , again , due to the fact that the interaction of the perturbation with the regge - wheeler potential starts `` earlier '' ( i.e. , at larger radii ) than in the case of the pschl - teller potential . the two bottom panels of fig . [ fig_2 ] refer instead to a wave - packet that is initially closer to the black hole ( i.e. , @xmath85 ) and show the impact on the waveforms of an increasing width of the wave - packet . most notably , the two panels show that as @xmath79 is progressively increased , the effects due to the slow - decay of the regge - wheeler potential become progressively more pronounced , with the ringdown lasting progressively less and with the backscattering being correspondingly anticipated . for example , when @xmath86 ( left - bottom panel of fig . [ fig_2 ] ) and despite the fundamental mode is still recognizable , other oscillations are present already after @xmath87 , while for @xmath88 ( right - bottom panel ) the global waveform is essentially overwhelmed by backscattering , the qnms are almost absent and the signal is dominated by the `` tail '' . an additional increase in the pulse width would make the qnms disappear completely . clearly , this behavior is not present in the case of a scattering off a pschl - teller potential , which is much less sensitive to the finite - size of the perturbation as long as it is smaller than the scale - height set by the exponential decay . although just a toy - problem , this simple comparison of the black - hole response when modeled with the pschl - teller potential is very useful to clarify that in the case of general initial data with a finite size , the shape of the curvature potential strongly affects the gravitational waveform . in the next sections we will refine this finding by considering more realistic perturbations such as those produced by the accretion of extended distributions of matter . following the analysis presented in paper i , we here consider a number of aspects of gravitational - wave emission resulting from the accretion of quadrupolar shells of dust which were not investigated in detail before . more specifically , we will here discuss : _ i _ ) a more detailed analysis of the black hole ringdown phase , focusing on black hole qnms and on backscattering effects related to the non - exponential decay of the black hole potential ; _ ii _ ) the mechanism responsible for the production of the interference fringes in the energy spectra ; _ iii _ ) the effect of using conformally - flat initial data ; _ iv _ ) a more detailed analysis of the energy released in gravitational waves . as in paper i , the rest - mass density is parameterized as @xmath89\sin^2\vartheta \ , \end{aligned}\ ] ] where @xmath90 is the initial position of the `` center '' of the shell and @xmath79 controls its compactness . the background rest - mass density @xmath91 is chosen to be very small ( i.e. , @xmath92 ) to simulate the vacuum outside the black hole , while the normalization constant @xmath93 is obtained from the condition that the volume integral of eq . ( [ shell ] ) gives @xmath94 , the total mass of the shell , which we choose to be @xmath95 . as we mentioned in sec . [ sbsc : id ] ( and paper i ) , the initial profile for @xmath96 is obtained after solving the hamiltonian constraint for @xmath97 with @xmath37 . furthermore , to minimize the impact of surious radiation and produce initial data that is a `` almost '' time - symmetric , the shell is kept frozen at @xmath90 ( i.e. , the hidrodynamics equations are not evolved ) up until the spurious initial gravitational - wave pulse leaves the numerical grid . we shall show at the end of this section how waveforms and energy spectra change when `` genuine '' time - symmetric and conformally flat initial data are implemented . + we start by considering the plunge from @xmath98 with @xmath99 ( this particular model was discussed also in paper i ) and whose complete waveform is shown in the left panel of fig . as expected , the gravitational - wave signal contains signatures of both the black hole qnms and of the backscattering effects . the first ones are triggered when the maximum of the rest - mass density crosses the peak of the potential at @xmath100 yielding the largest contribution to the waveform at @xmath101 . we have verified the presence of the black hole qnms with complex frequency @xmath102 ( with @xmath103 ) by fitting the waveform from this time onward with a superposition of modes of the form @xmath104\exp[\im(\omega_i)t]$ ] , where the free coefficients @xmath105 and @xmath106 are real . the right panel of fig . [ fig1 ] reports the result of such fit obtained using the first five qnms , and which is essential to properly capture the early - time part of the waveform . ( adding higher modes does not improve the fit , while considering less modes is not enough to reproduce accurately the waveform . ) when analyzed on a logarithmic scale ( see bottom part of the right panel of fig . [ fig1 ] ) , it becomes evident that after @xmath107 the backscattering effects become dominant and the waveform is no longer a simple superposition of exponentially decaying modes . additional information on the black hole response comes from the study of the energy spectra for shells of different initial width @xmath79 and starting from @xmath98 . this is summarized in fig . [ fig3 ] , whose left panel shows the energy spectra of the complete waveforms and the characteristic interference patterns already introduced in paper i. clearly , the interference effects at higher frequencies ( i.e. , for @xmath108 ) as well as the efficiency in the emission of energy via gravitational waves are increased as the compactness of the shell is increased , progressively tending to what is expected for a point like particle ( cf . the top - right panel in fig . [ fig5 ] ) . equally clear is that the energy spectra do not show any peculiar behavior around the qnms frequencies ( i.e. , for @xmath109 ) but also that the behavior of the spectra at low frequencies does not change significantly with the matter compactness . as argued in paper i and in refs . @xcite , interference fringes naturally arise in the energy spectrum generated by the superposition of two waveforms separated by a certain time lag @xmath110 . indeed , given two time series @xmath111 and @xmath112 , the power spectral density resulting from their superposition is given by @xmath113 is the fourier transform of @xmath111 . as a result , such a spectral density will have peaks with a constant spacing given by @xmath114 . if @xmath115 , the minima occur at @xmath116 and the spectrum is said to have @xmath117 frequency modulations . the modulation is always smaller for generic values of @xmath118 . the energy spectra displayed in the left panel of fig . [ fig3 ] can be explained within this general picture . the series of fringes is mainly determined by the interference of two bursts of radiation with different amplitude and separated in time by the scattering off the curvature potential at different times . in the case of a shell of width @xmath99 , the two main pulses responsible for this modulation can be distinguished in the waveform shown in the left panel of fig . these correspond to the peak produced by the initial motion of @xmath96 and appearing at very early times ( i.e. , at @xmath119 ) , and to the one emitted when the bulk of accreting matter crosses the peak of the zerilli potential ( i.e. , around @xmath120 ) . is initially nonzero close to the peak of the zerilli potential , some backscattering is also possible before the bulk of the shell reaches the peak of the potential itself . this explains the small oscillation in the waveform at @xmath119 . ] the time - lag between the two pulses is @xmath121 , which gives a separation between the peaks of the spectrum @xmath122 . this value is in good agreement ( given the difficulty in unambiguously catch the time when the ringdown starts ) with the @xmath123 that can be read off from the solid line in the left panel of fig . [ fig3 ] . the right panel of fig . [ fig3 ] , on the other hand , shows the spectra obtained by performing a fourier transform of the signal in the `` ringdown '' phase only . the two dotted vertical lines indicate the @xmath124 ( fundamental ) and @xmath125 ( first overtone ) qnm frequencies of the black hole . it is interesting to note that with the exception of very wide shells ( i.e. , with @xmath126 ) which are not able to excite cleanly the qnms , the maximum of the spectra always lies between the two lines , suggesting the presence of the two modes in the waveform . another parameter influencing the energy spectra is the initial location of the shells @xmath90 . as mentioned above , the largest part of the gravitational - wave signal is emitted when the `` center '' of the shell reaches the peak of the zerilli potential and this will clearly depend on the initial position of the shell . shells that start further away will have longer infalling times , larger separations between the first and second peaks in the waveforms , and thus smaller separation in the interference fringes of the energy spectra . in order to reproduce this dependence we have performed a number of simulations for shells of fixed initial width ( @xmath99 ) accreting from different initial locations @xmath90 . two representative and extreme cases for @xmath127 and @xmath128 are shown in fig . [ fig5 ] , whose left panels display the waveforms , while the right ones the energy spectra . overall , these plots show a number of interesting and general properties of the excitation of black hole oscillations through accreting matter . firstly , as the infalling time is increased , the ringdown phase is progressively dominated by the non - oscillatory tail , essentially as a result of the `` spreading '' of the shell induced by the tidal field . this behavior is indeed consistent with the analysis carried out in ref . @xcite and which pointed out that the excitation of the black hole qnms through a continuous flux of infalling particles is made difficult by the presence of destructive interferences effects . secondly , the efficiency in gravitational - wave emission decreases for shells falling from larger distances and is significantly smaller than for point - like particles . this can be easily appreciated in the right panels of fig . [ fig5 ] , when comparing energy spectra associated with shells ( solid lines ) with the corresponding @xmath70 energy spectra of a particle plunging radially from the same initial position ( dashed line ) . note that the efficiency is in this case of at least a couple of orders of magnitude larger ( the data have been properly rescaled to aid the comparison ) . thirdly , the energy emission is progressively peaked toward lower frequencies as @xmath90 increases . this behavior can be made more quantitative by defining the _ `` characteristic frequency '' _ of the energy spectrum as the weighted average @xmath129 and by computing how this characteristic frequency changes along a sequence of shells of fixed initial compactness and falling from increasingly larger radii . a summary of this dependence is offered in fig . [ fig6b ] for shells with @xmath130 . by following a phenomenological approach , if one defines @xmath131 and @xmath132 the figure shows that the data can be very well fitted by means of a quadratic law like @xmath133 with @xmath134 , @xmath135 and @xmath136 . the interpretation of this relation is still unclear but certainly deserves further attention . dependence of the characteristic frequency in the energy spectrum on the initial position of the shell.,width=8 ] as we anticipated at the beginning of this section , an important comment is worth making on the influence of the initial amount of gravitational radiation on the waveforms and on the energy spectra . two different approaches are possible in this respect and it is important to bear in mind that they do not yield identical results . the first one consists in selecting a time - symmetric and conformally flat ( or non - conformally flat ) initial profile of the zerilli - moncrief function from the solution of hamiltonian constraint . as we discussed in sec . [ sbsc : id ] , this does not prevent that spurious radiation is produced as the evolution starts . the second approach , which has been the one adopted here , consists , after the solution of the constraint , in the removal of the spurious burst of gravitational radiation by evolving the perturbation equations but _ not _ the perturbing sources . in this way the initial radiation is allowed to leave the computational domain and the evolution can therefore start self - consistently once the initial , nonstationary part of the solution has been removed . figure [ fig7 ] summarizes the impact on the energy spectra and on the waveforms of the two approaches for a shell initially at rest at @xmath137 . plotted with the dashed line in the main panel is the energy spectrum in the case of actual conformally flat initial data involving the solution of hamiltonian constraint only ; on the other hand , the solid line refers to the spectrum one obtains when the initial pulse is allowed to be radiated away . note the much stronger modulation present in the former case and the corresponding larger variation in the initial part of the waveform as shown in the inset . note that we would have found larger differences for smaller @xmath90 and smaller differences for larger @xmath90 . furthermore , we note that if we start with non - conformally flat initial data ( that is , throught the solution of hamiltonian constraint with @xmath138 ) we would obtain larger differences in the early part of the waveforms and thus larger modulations in the energy spectrum . this behavior is qualitatively similar to what discussed in ref . @xcite in the case of particles plunging radially from finite distance conformally flat initial data and spurious bursts of radiation : energy spectra obtained by solving the hamiltonian constraint with @xmath37 only ( dashed line ) and when the initial gravitational - wave pulse is additionally eliminated ( solid line ) . the shell has @xmath130 , it falls from @xmath137 and the signal is extracted at @xmath139 . the inset shows the corresponding initial part of the waveform.,width=8 ] [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] we conclude this section by commenting on the amount of energy released in gravitational waves as a function of both @xmath90 and @xmath79 . this is shown in fig . [ fig8 ] , which displays in the main panel the normalized energy emitted in the lowest multipole as a function of the initial location @xmath90 and for a shell of initial width @xmath130 . note that in the case of a shell this is a monotonically decreasing function of the radial distance , an opposite behavior to that seen for accreting particles where , for conformally flat initial data , the energy has a local minimum and then increases monotonically with @xmath90 @xcite , asymptoting the davis - ruffini - press - price limit @xcite . in addition , the inset in fig . [ fig8 ] displays a comparison between the energy emitted through the whole waveform ( empty circles ) and the one computed considering the ringdown phase only ( filled circles ) . note that while the black hole ringdown contributes for @xmath140 of the total energy in the case of small @xmath90 , this value goes down to @xmath141 for larger initial distances as a result of the progressive loss of compactness experienced by the shell during the infall . _ main panel : _ total energy radiated in gravitational waves ( @xmath142 multipole ) as a function of the initial location @xmath90 of the center of a shell with @xmath130 . _ inset : _ total energy versus @xmath79 for shells falling from @xmath98 ( open circles ) . the filled circles refer to the energy contribution coming from from the black hole ringdown only.,width=8 ] while our model with infalling quadrupolar shells allows us to capture some of the essential features of the gravitational - wave emission from extended matter sources , such as the excitation of the black hole qnm ringdown and the presence of interference effects in the energy spectra , it nevertheless remains a useful toy model . to improve on this and to examine a scenario which is astrophysically more realistic , we will now consider the gravitational - wave emission resulting from the accretion of geometrically thick disks ( or tori ) orbiting around a non - rotating black hole . such systems are believed to form in a variety of different ways , such as during the last stages of gravitational collapse or in the merger of binary neutron stars . in addition , if compact enough and undergoing oscillations , these systems generate gravitational - wave signals within the sensitivity curve of ground - based interferometers @xcite . hereafter , and as in the case of dust shells , we will assume that the mass of the torus @xmath94 is much smaller than that of the black hole . detailed descriptions on how to build equilibrium configurations of barotropic thick disks orbiting black holes are given in refs . @xcite , but we here recall that these objects have traditionally been described as obeying a polytropic equation of state @xmath143 , with @xmath144 . because in pure orbital motion , the fluid four - velocity is given by @xmath145 and it describes a non - keplerian rotation around the central black hole , with a specific angular momentum @xmath146 distribution which is essentially unknown . table [ table4 ] lists the main features of the tori considered in our simulations , such as the values of the specific angular momentum @xmath5 ( assumed constant throughout the disk ) , of the potential gap @xmath147 between the inner edge of the disk and the cusp , and of the mass of the disk @xmath94 . in order to induce a non - trivial dynamics in these otherwise stationary disks , the initial models have to be perturbed in some way , such as by adding a radial velocity @xmath148 , where @xmath149 is a parameter and @xmath150 is the radial velocity of the spherical stationary atmosphere surrounding the torus @xcite . the rest - mass density of the atmosphere is low enough not to influence the dynamics of the tori ( its density is several orders of magnitude smaller than the maximum density at the center of the disk ) . as in the case of infalling dust shells , the initial data for the perturbation fields @xmath151 is obtained either through the solution of the hamiltonian constraint ( for even - parity perturbations ) with conformally flat initial data , or assuming stationarity , i.e. , through eqs . ( [ r - w ] ) with @xmath152 . clearly , in either case this initial data is not consistent with the hydrodynamical sources once they are perturbed . this mismatch inevitably introduces a unphysical initial burst of gravitational radiation that is however easy to distinguish and remove from the analysis . we note that the accretion of matter onto the black hole can occur on different timescales depending on the perturbation and we will consider here the two limiting situations that could be encountered in astrophysical scenarios . in the first case we will consider a perturbation which is large enough to cause a runaway accretion of the disk onto the black hole ; we will refer to this as the `` hypercritical accretion '' scenario ( see sect . [ has ] ) . in the second case , on the other hand , the perturbation is smaller and it will induce a series of quasi - periodic oscillations , each accompanied by an episode of accretion onto the black hole ; we will refer to this as to the `` quasi - period accretion '' scenario ( see sect . [ qpa ] ) . as mentioned above , simulations of hyper - accreting disks can be performed by simply choosing a sufficiently large initial value of the radial velocity perturbation . under this condition , the centrifugal barrier can not counteract the injection of additional kinetic energy and the whole disk is `` pushed '' towards the black hole , rapidly accreting onto it in less than one orbital period . hereafter we will concentrate on the dynamics of model @xmath153 , to which a velocity perturbation with @xmath154 is added . for this model the hydrodynamical grid runs from @xmath155 to @xmath156 and it is covered by @xmath157 grid - points , geometrically spaced in the coordinate @xmath40 , with @xmath158 and @xmath159 . the range of the @xmath42 1d grid is @xmath160 $ ] , to guarantee that the outer boundary does not influence the slope of the late time power - law tail , and is covered by@xmath161 cells . the left panel of fig . [ fig9 ] shows , in terms of the tortoise coordinate , five snapshots of the evolution of the rest - mass density on the equatorial plane . note that as time proceeds , the disk is compressed , the density increases and matter quickly starts falling onto the black hole . this continues until the entire disk has been accreted . a summary of this is shown in the spacetime diagram in the right panel of fig . [ fig9 ] , which presents the motion of the projection on the equatorial plane of an effective `` center - of - mass '' of the disk , defined as @xmath162 where @xmath163 ( for convenience the spacetime diagram is shown using the tortoise radial coordinate @xmath42 ) . correspondingly , the inset in fig . [ fig9 ] depicts the time evolution of the mass accretion rate for model @xmath153 . the vertical dashed line in the right panel of fig . [ fig9 ] is approximately the location of the peak of the zerilli potential , @xmath164 ( @xmath165 ) and can be used to identify @xmath166 as the time at which the center of mass crosses the peak of the potential . furthermore , this panel also shows that the center of mass leaves the hydrodynamical grid ( @xmath167 ) at @xmath168 and from this time onward , the gravitational - wave signal is dominated by the qnms of the black hole . model @xmath153 : gravitational waveform ( extracted at @xmath78 ) . the presence of two ringings is evident . the first one is related to the gravitational - wave content present in the initial profile of the zerilli - moncrief equation . the second ringing ( magnified in the inset ) is determined by the object crossing the peak of the potential . the dashed line , proportional to @xmath169 , shows that the late - time behavior of the waveform is properly captured.,width=8 ] the gauge - invariant gravitational waveform produced by the hypercritical accretion has a dominant @xmath142 character when @xmath170 , and @xmath171 , @xmath172 . ] and is displayed in a logarithmic scale in fig . [ fig10 ] as measured by an observer located at @xmath78 . besides the initial burst of spurious radiation related to the initial data , and which is no longer present by @xmath173 , the waveforms exhibits two main features . the first one is determined by the infalling matter as its center of mass approaches the maximum of the potential at @xmath174 . the second one , instead , starts at @xmath175 and represents the ringing resulting from the perturbation experienced by the black hole through the rapid accretion of the torus ( see the mass accretion rate inset in the right panel of fig . [ fig9 ] ) . this becomes apparent when comparing the features of the waveform in fig . [ fig10 ] with the matter dynamics in fig . the ringdown phase ends with the usual power - law tail ; this is highlighted by the dot - dashed line in fig . [ fig10 ] which is proportional to @xmath169 and shows that for @xmath176 the waveform has only a quadrupolar nature decaying as @xmath177 . model @xmath178 : time evolution of the projection of the center of mass in the equatorial plane . after an initial transient , the dynamics is characterized by constant amplitude quasi - periodic oscillations . the inset shows the mass accretion rate.,width=8 ] in summary , the analysis of the process shows that a very rapid and hypercritical accretion of a compact distribution of matter onto a black hole produces gravitational - wave emission that is qualitatively similar to that of either relatively narrow dust shells plunging from a large distance or wide dust shells accreting from a small distance . furthermore , the gravitational - wave signal shows oscillations that reflect the excitation of the black hole qnms ( especially the fundamental one ) but these oscillations are in general so weak that the non - oscillatory tail determined by the long - range properties of the scattering potential soon dominates the signal . a phenomenology complementary to the one discussed in the previous section for a hypercritical accretion can be studied when the accretion takes place in a quasi - period fashion . indeed , there are at least two different ways of triggering a quasi - periodic accretion and we will discuss them separately . the first possibility has already been discussed in sect . [ tori ] and consists in adding a perturbation in the radial velocity of the torus parameterized in terms of an analytic accretion solution . as a concrete example we will now concentrate on the dynamics of model @xmath178 , describing a disk with a comparatively high value of the specific angular momentum , @xmath179 , and subject to an initial radial velocity perturbation with @xmath180 . figure [ figv_5 ] shows the motion of the center of mass on the equatorial plane , indicating that after an initial transient , a phase of quasi - periodic , ( almost ) constant amplitude oscillations follows , in which the disk periodically approaches the black hole before the centrifugal barrier pushes it back , past its original position . as the torus nears the black hole , part of its matter is spilled through the cusp , resulting in a quasi - periodic accretion of matter onto the black hole ; this is shown by the inset of fig . [ figv_5 ] which reports the evolution of the mass accretion rate ( see ref . @xcite for a more detailed discussion ) . note that the accretion is essentially shut - off as the torus moves away from the black hole . this quasi - periodic dynamics of the torus is clearly imprinted onto the gravitational - wave signal and this is shown in the left panel of fig . [ figv_3 ] , where the @xmath142 gauge - invariant waveform ( extracted at @xmath181 ) is superposed to the signal extracted using the quadrupole formula ( sqf@xmath60 ) . note that the very early time part of the gauge - invariant signal has been removed to avoid influences coming from the burst related to the initial data . the waveforms show a small burst at @xmath182 , that corresponds to the initial accretion of matter , followed by regular oscillations that mirror the motion of the torus in the potential well . some differences between the gauge - invariant and the sqf@xmath60 waveforms are recognizable in the initial stages , with the first one showing highly damped high - frequency oscillations for @xmath183 superposed to the main , quasi - periodic signal . since this high - frequency pattern is absent in the quadrupole waveform , we conclude that it originates from curvature backscattering of gravitational waves emitted in the initial infalling phase . we also note that the differences between the waveforms computed through the gauge - invariant formalism and the quadrupole formula using the quadrupole tensor defined by eq . ( [ sqf1 ] ) become larger if the initial velocity perturbation is increased . this is simply due to the fact that the sqf@xmath60 is valid in the weak - field and slow - motion approximation and thus ceases to be accurate when the dynamics of the oscillations is nonlinear , with the velocities becoming relativistic . on the other hand , the velocity - dependent terms in the source function of the zerilli - moncrief equation are obviously accurate also in a relativistic regime . the energy spectrum for this signal is shown as a solid line in the right panel of fig . [ figv_3 ] and refers to the signal from @xmath184 in order to eliminate the unphysical initial burst . the spectrum shows the presence of a broad - band peak at low frequencies , related to the highly damped oscillation in the early time part of the @xmath142 signal , together with narrow peaks in the harmonic ratio @xmath185 , which instead mirror the non - linear coupling of modes in an oscillating torus ( see refs . @xcite for a detailed discussion of the eigenfrequencies and their astrophysical implications ) . such peaks are present in both the @xmath142 and @xmath186 energy spectra ( dashed line in fig . [ figv_3 ] ) . the latter is generated by a waveform @xmath187 whose amplitude is roughly two orders of magnitude smaller than the @xmath142 one . it is interesting to note that while for the @xmath142 multipole the first frequency of the disk is the dominant one , the second frequency has the largest power for the @xmath186 multipole . the energy spectrum shows no distinctive signature at @xmath188 that could be related with the excitation of the qnms of the black hole ( cf . , right panel of fig . [ fig3 ] ) . this is in contrast with the results of ref . @xcite , which have simulated a similar source of perturbations for the black hole and have indeed found a minute high - frequency contribution in the energy spectra related to the emission from the black hole . it is presently unclear what is the origin of this difference , but it is likely that the hybrid approach proposed in ref . @xcite , and which combines a solution in the time - domain for the hydrodynamics and one in the frequency - domain for the perturbative equations , is better suited to extract the extremely small contributions coming from the black hole . a second possibility of triggering quasi - periodic oscillations in the disk , and hence a quasi - periodic accretion , is by an instantaneous reduction of the specific angular momentum . the main difference with the dynamics discussed before is that this reduction produces a continuous mass accretion , which is however modulated quasi - periodically as the torus approaches or moves away from the black hole . this behavior is shown in fig . [ fig13 ] for model @xmath189 , in which the specific angular momentum is reduced from @xmath179 to @xmath190 . more specifically , the left panel of fig . [ fig13 ] shows the motion of the center of mass on the equatorial plane and the evolution of the mass - accretion rate in the inset . because of the intrinsic reduction of the centrifugal support and the resulting continuous accretion , the oscillations are quasi - periodic in time but with decaying amplitude , so that by @xmath191 they have essentially disappeared and the torus accretes at an almost constant rate , with the center of mass progressively approaching the black hole . this different quasi - periodic dynamics of the torus is also imprinted onto the quadrupolar waveform and this is shown in the right panel of fig . [ fig13 ] , which presents both the gauge - invariant signal as measured by an observer located at @xmath181 ( solid line ) and the one computed with the sqf@xmath192 ( dashed line ) . note that the differences between the signals are very small initially , but they grow with time and by @xmath193 , when the angular momentum loss induces a global radial motion of the disk , the mean value of the gauge - invariant signal progressively drifts away from zero . this different behavior can be understood simply if one bears in mind that during this stage the disk is undergoing an almost steady - accretion phase . as a result , while the disks mass - quadrupole is nonzero and large , its time variations , and hence the corresponding gravitational - wave signal as computed in the sqf@xmath60 , are essentially zero . the perturbative approach , on the other hand , is able to capture the perturbations induced by this steady - state accretion and this is reflected in the gradual and secular increase of the gravitational - wave signal shown in the right panel of fig . [ fig13 ] . of course , this behavior is the same already encountered in the case of accretion of dust shells and it is , once again , generated by curvature backscattering off the `` tail '' of the potential . note that during the final stages of the accretion , i.e. , for @xmath194 , the motion of the center of mass exhibits an exponentially rapid motion toward the black hole which , however , is not reflected neither in the mass - accretion rate , nor in the gravitational - wave signal . the reason for this is that , in practice , the rest - mass contained in the torus in these final stages is extremely small and the rest - mass density is so diluted that it becomes comparable to that of the atmosphere surrounding the torus . as a result , the effective perturbation induced onto the black hole in the final stages of the accretion is vanishingly small . in summary , the analysis of this process shows that the gravitational - wave signal produced by a quasi - periodic accretion can be rather different , depending on whether the amplitude of the oscillations is ( almost ) constant in time and the accretion rate is also periodic , or the oscillations have decreasing amplitude and the accretion rate is constant with a periodic modulation . in the first case , the gravitational - wave signal will reflect faithfully the dynamics of the matter with constant - amplitude waveforms averaging to zero and a spectrum showing peaks at the eigenfrequencies of the oscillating matter . in the second case , on the other hand , the waveforms will have a decreasing amplitude and will not average to zero as a result of the underlying continuous accretion . by performing numerical simulations that combine the solution of the nonlinear hydrodynamics equations with that of the linear inhomogeneous zerilli - moncrief and regge - wheeler equations , we have studied the features of the gravitational - wave signals generated by the accretion of matter onto a schwarzschild black hole . as extended and accreting matter - sources we have considered quadrupolar shells of dust falling radially from a finite distance , as well as geometrically thick disks undergoing either bursts of hypercritical accretion or quasi - periodic oscillations . in both cases we find that the gravitational - wave signal _ is not _ a simple superposition of the black hole qnms and that the latter can not be found in the energy spectra at times . rather , we find that quite generically the signal contains important contributions coming from radiation scattering off the tail of the curvature potential and producing a characteristic pattern of interference fringes in the energy spectra . while the relevance of this contribution differs according to the specific source considered , it is generically present as long as the source of perturbations is extended and the scattering potential does not have an exponential decay with radius . this conclusion , which was already reported in previous studies involving simpler sources , has been here confirmed unambiguously by studying the scattering off a fictitious potential , the pschl - teller potential , which however decays exponentially with radius . these generic properties of the gravitational - wave emission coming from black holes perturbed by extended sources represent important differences with respect to corresponding properties of signals produced by very compact sources , such as point - like particles . of course this conclusion prevents from the derivation of a simple and generic description of the gravitational - wave signal which would be independent of the properties of the perturbing source , but it also opens the exciting perspective of deducing many of the physical properties of the source through a careful analysis of the waveforms produced . overall , the results presented here make us confident that the black hole qnm contributions to the full gravitational - wave signal should be extremely small in generic astrophysical scenarios involving the accretion of extended distributions of matter . on the other hand , it should also be stressed that the time - domain analysis carried out here may not be the most accurate to extract the contributions coming from the perturbed black hole when these are several orders of magnitude smaller than those coming from the source itself or from the backscattering off the potential . in these cases , however , a hybrid approach such as the one proposed in ref . @xcite , combining the solution in the time - domain for the hydrodynamics with one in the frequency - domain for the perturbative equations , may be better suited . it is a pleasure to thank e. berti , h.r . beyer , s. bernuzzi , r. de pietri , i. olabarrieta , a. tartaglia and m. tiglio for useful discussions and comments . part of this work was carried out by an through visits in valencia and at the center of computation and technology ( cct ) at louisiana state university ; the support of the della riccia foundation , of cct and of ilias is gratefully acknowledged . jaf acknowledges financial support from the spanish _ ministerio de educacin y ciencia _ ( grant aya 2004 - 08067-c03 - 01 ) . the computations were performed on the cluster for numerical relativity _ `` albert '' _ at the university of parma . 200 # 1 _ gr - qc/__#1 _ # 1#2#3 phys . * # 1 * , # 2 ( # 3 ) # 1#2#3 phys . rev . d , * # 1 * , # 2 ( # 3 ) # 1#2#3 phys . b , * # 1 * , # 2 ( # 3 ) # 1#2#3 phys . reports , * # 1 * , # 2 ( # 3 ) # 1#2#3 physica , * # 1 * , # 2 ( # 3 ) # 1#2#3 j. comput . , * # 1 * , # 2 ( # 3 ) # 1#2#3 j. math . # 1 * , # 2 ( # 3 ) # 1#2#3 computer phys . , * # 1 * , # 2 ( # 3 ) # 1#2#3 class . quantum grav . , * # 1 * , # 2 ( # 3 ) # 1#2#3 computers math . , * # 1 * , # 2 ( # 3 ) # 1#2#3 math . , * # 1 * , # 2 ( # 3 ) # 1#2#3 astrophys . j. , * # 1 * , # 2 ( # 3 ) # 1#2#3 astrophys . j. lett . , * # 1 * , # 2 ( # 3 ) # 1#2#3 astrophys . j. suppl . , * # 1 * , # 2 ( # 3 ) # 1#2#3 acta astronomica , * # 1 * , # 2 ( # 3 ) # 1#2#3 sov . , * # 1 * , # 2 ( # 3 ) # 1#2#3 siam j. sci . * # 1 * , # 2 ( # 3 ) #1#2#3 astron . astrophys . , * # 1 * , # 2 ( # 3 ) # 1#2#3 mon . not . * # 1 * , # 2 ( # 3 ) # 1#2#3 proc . london , ser . a , * # 1 * , # 2 ( # 3 ) # 1#2#3 i.j.m.p . c * # 1 * , # 2 ( # 3 ) # 1#2#3 phys . lett . a * # 1 * , # 2 ( # 3 ) # 1#2#3 progr . . phys . * # 1 * , # 2 ( # 3 ) s. chandrasekhar , _ the mathematical theory of black holes _ , ( oxford university press , oxford , 1992 ) . kokkotas and b. schmidt , _ quasi - normal modes of stars and black holes _ , living rev . relativity , pub . lrr-1999 - 2 . frolov and i.d . novikov , _ black hole physics _ , kluwer academic publishers , dordrecht 1998 . w.h . press , astrophys . j. letters * 170 * , l105 ( 1971 ) . vishveshwara , nature ( london ) , * 227 * , 936 ( 1970 ) . leaver , phys . d * 34 * , 384 ( 1986 ) ; errata , phys . d * 38 * , 725 ( 1988 ) . n. andersson , phys . d * 51 * , 353 ( 1995 ) , * 52 * , 1808 ( 1995 ) , * 55 * , 468 ( 1997 ) . c.o . lousto and r.h . price , phys . d * 55 * , 2124 ( 1997 ) . cunningham , r.h . price , and v. moncrief , . e. seidel , phys . rev d * 44 * , 950 ( 1991 ) . l. baiotti , i. hawke , l. rezzolla and e. schnetter , phys . rev . lett . * 94 * , 131101 ( 2005 ) . l. baiotti and l. rezzolla , phys . . lett . * 97 * , 141101 ( 2006 ) . f. pretorius , phys . lett . * 95 * , 121101 ( 2005 ) . m. campanelli , c.o . lousto , p. marronetti , and y. zlochower , phys . lett . * 96 * 111101 ( 2006 ) . baker , j. centrella , d.i . choi , m. koppitz , and j. van meter , phys . * 96 * 111102 ( 2006 ) . p. papadopoulos and j.a . font , phys . d * 59 * , 044014 ( 1999 ) . a. nagar , g. diaz , j.a . pons , and j.a . font , phys . d * 69 * , 124028 ( 2004 ) . a. nagar , j.a . font , o. zanotti and r. de pietri , phys . d * 72 * , 024007 ( 2005 ) . v. ferrari , l. gualtieri and l. rezzolla , phys . d * 73 * 124028 ( 2006 ) . shapiro and i. wasserman , astrophys . j. * 260 * , 838 ( 1982 ) . petrich , s.l . shapiro and i. wasserman , astrophys . j. * 58 * , 297 ( 1985 ) . h. sotani and m. saijo , phys.rev . d * 74 * , 024001 ( 2006 ) . k. martel and e. poisson , phys . d * 66 * , 084001 ( 2001 ) . m. abramovicz , m. jaroszyski and m. sikora , astron . astrophys . * 63 * , 221 ( 1978 ) . j.a . font and f. daigne , mon . not . 334 * , 383 ( 2002 ) . o. zanotti , l. rezzolla , and j.a . font , mon . not . . soc . * 341 * , 832 ( 2003 ) . o. zanotti , j.a . font , l. rezzolla , and p. j. montero mon . not . . soc . * 356 * , 1371 ( 2005 ) . l. rezzolla , t. w. maccarone , si . yoshida and o. zanotti , mon . not . . soc . * 344 * , l37 ( 2003 ) . l. rezzolla , si . yoshida and o. zanotti , mon . not . . soc . * 344 * , 978 ( 2003 ) . a. nagar and l. rezzolla , class . * 22 * , r167 ( 2005 ) . k. martel and e. poisson , phys . d * 71 * , 104003 ( 2005 ) . t. regge and j.a . wheeler , phys . rev . * 108 * , 1063 ( 1957 ) . f.j . zerilli , phys . d * 2 * , 2141 ( 1970 ) . v. moncrief , ann . ( n.y . ) * 88 * , 323 ( 1974 ) . o. sarbach and m. tiglio , phys . d * 64 * , 084016 ( 2001 ) . goldberg , j. macfarlane , e.t . newman , f. rohrlich and e.c.g . sudarsahn , j. math . phys . * 8 * , 2155 ( 1967 ) . u.h . gerlach and u.k . sengupta , phys . d * 19 * , 2268 ( 1979 ) ; * 22 * , 1300 ( 1980 ) ; j.m . martn - garca and c. gundlach , phys . rev d * 59 * , 064031 ( 1999 ) ; c. gundlach and j.m . martn - garca , phys . d * 61 * , 084024 ( 2000 ) ; j.m . martn - garca and c. gundlach , phys . d * 64 * , 024012 ( 2001 ) . j.a . font , _ numerical hydrodynamics in general relativity _ , living rev . relativity , pub . lrr-2003 - 4 . lau , j. math . phys . * 46 * , 102503 ( 2005 ) . g. pschl and e. teller , z. phys . * 83 * , 143 ( 1933 ) . beyer , commun . * 204 * , 397 ( 1999 ) . v. ferrari and b. mashhoon , phys . d , * 30 * , 295(1984 ) . leaver , proc . a. , * 402 * , 285 ( 1985 ) . price , phys . d * 5 * , 2419 ( 1972 ) ; * 5 * , 2439 ( 1972 ) . e. berti , v. cardoso and c. m. will , in the proceedings of the vii international conference of the hellenic astronomical society , _ gr - qc/0601077_. a. nagar , t. damour and a. tartaglia , to appear in the special isuue of classical and quantum gravity based around the new frontiers in numerical relativity meeting , albert einstein insitutute , golm , july 17 - 21 , 2006 ; _ gr - qc/0612096_. m. davis , r. ruffini , w.h . press , and r.h . price , phys . * 27 * , 1466 ( 1971 ) ; m. davis , r. ruffini and j. tiomno , phys . d * 5 * , 2932 ( 1972 ) .

we present an mhd simulation of the emergence of a radio minihalo in a galaxy cluster core in a `` secondary '' model , where the source of the synchrotron - emitting electrons is hadronic interactions between cosmic - ray protons with the thermal intracluster gas , an alternative to the `` reacceleration model '' where the cosmic ray electrons are reaccelerated by turbulence induced by core sloshing , which we discussed in an earlier work . we follow the evolution of cosmic - ray electron spectra and their radio emission using passive tracer particles , taking into account the time - dependent injection of electrons from hadronic interactions and their energy losses . we find that secondary electrons in a sloshing cluster core can generate diffuse synchrotron emission with luminosity and extent similar to observed radio minihalos . however , we also find important differences with our previous work . we find that the drop in radio emission at cold fronts is less prominent than that in our reacceleration - based simulations , indicating that in this flavor of the secondary model the emission is more spatially extended than in some observed minihalos . we also explore the effect of rapid changes in the magnetic field on the radio spectrum . while the resulting spectra in some regions are steeper than expected from stationary conditions , the change is marginal , with differences in the synchrotron spectral index of , depending on the frequency band . this is a much narrower range than claimed in the best - observed minihalos and produced in the reacceleration model . our results provide important suggestions to constrain these models with future observations .

1403.6743

a number of relaxed , cool - core clusters are hosts to faint , diffuse synchrotron radio sources called `` radio minihalos '' , with a radius comparable to the size of the cooling region ( @xmath1 100 - 300 kpc ) and a steep spectrum ( @xmath2 ; @xmath3 ) . they arise from cosmic - ray electrons ( hereafter cre ) emitting in the @xmath4 g magnetic fields in the cluster . these are relatively rare sources , with currently only around 10 clusters with confirmed detections . examples include perseus @xcite , a2029 @xcite , ophiuchus @xcite , rxc j1504.1 - 0248 @xcite , and rxj 1347 - 1145 @xcite , along with the newest detections in @xcite . the origin of radio mini - halos in cool core clusters , as well as their possible connection with giant radio halos , is still unclear ( see , e.g. * ? ? ? * for a review ) . though the clusters which host these sources typically also have active galactic nuclei , due to the fact that the radiative timescale of the electrons at the required energies for the observed emission ( @xmath5 years ) is much shorter than the time required for these electrons to diffuse across the cooling region , they alone can not account for the origin of minihalos . most cool - core clusters also possess spiral - shaped cold fronts @xcite , which are believed to be the product of the central cold gas `` sloshing '' in the cluster potential minimum due to interactions with subclusters . a number of simulation works have investigated the importance of this phenomenon ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? most relevant for this work , in ( * ? ? ? * hereafter zml11 ) we showed that such magnetic field amplification occurs near the cold front surfaces due to the local velocity shear , and that the magnetic field within the core region is stronger than it was previous to the onset of sloshing . the relevance of gas sloshing to radio minihalos was first pointed out by @xcite , who discovered a correlation between sloshing cold fronts and minihalo emission in two galaxy clusters . a number of other examples have since been observed ( e.g. * ? ? ? * ; * ? ? ? * ) . in these cases , the minihalo emission is well - confined to the area on the sky delineated by the cold front surfaces , suggesting a causal connection between sloshing motions and radio minihalos . one possible origin for the cre which produce the observed synchrotron radiation of minihalos is reacceleration of a lower energy population of cre due to turbulence @xcite . in a previous work ( * ? ? ? * hereafter z13 ) , we investigated the possibility that radio minihalos are due to turbulent reacceleration in sloshing cluster cores . the sloshing motions produced turbulent velocities of @xmath6 on scales of 10s of kpc within the sloshing region . we showed that the radio emission produced in these simulations was consistent with the properties of observed minihalos where sloshing cold fronts are also found . in particular , the minihalo emission was confined to the core region and possessed a steep spectrum . there is an alternative perspective on the origin diffuse radio emission in clusters of galaxies , known as the `` hadronic '' or `` secondary '' model ( see , e.g. * ? ? ? ? * ; * ? ? ? * ) . cosmic - ray protons ( hereafter crp ) are believed to fill the cluster volume , having been accelerated from the thermal population to relativistic energies by supernovae , agn , and shocks associated with cosmological structure formation ( for a review see * ? ? ? * ) these protons will undergo interactions with the thermal proton population , producing pions which will decay into secondary products , including cosmic - ray electrons and positions : @xmath7 since the crp have very long radiative loss times compared with the age of the cluster , these interactions should continuously provide a fresh population of cre at a range of energies . in this model , it is these cre that produce the observed radio emission . these hadronic interactions should also produce a flux of @xmath8-rays , but so far no confirmed detections of such emission from galaxy clusters have been made . upper limits on the @xmath8-ray flux from several experiments ( see , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) indicate the ratio of the crp energy to the thermal energy of the gas is at most @xmath91 - 2% . still , contrary to the case of nearby giant radio halos , these limits do not put significant tension on a secondary origin of mini - halos @xcite . for example , @xcite proposed that mini - halos are primarily of hadronic origin , while giant radio halos experience a transition from a central hadronic emission to a leptonic emission component in the external regions due to cre reacceleration . in the hadronic model , under relatively quiescent conditions , the electron spectrum will reach a steady - state condition where the energy losses and the gains due to injection balance each other out @xcite . if the spectrum of crp is a power law with spectral index @xmath10 , at high energies , where the radiative ( synchrotron and inverse - compton ) losses dominate , the cre spectrum has the form @xmath11 , which results in a synchrotron emissivity that depends on the properties of the crp and the plasma as @xmath12 where @xmath13 is the number density of thermal particles , @xmath14 is the magnetic field strength , @xmath15 g is the equivalent magnetic field strength of the cmb , @xmath16 is the synchrotron spectral index , and @xmath17 is the energy density of crp . under the steady - state assumption and where radiative losses dominate , @xmath18 , using a canonical value @xmath19 this implies that for @xmath20 the radio emission is roughly independent of the magnetic field strength and for @xmath21 it is suppressed . though this spectral index value is consistent with the range of spectral indices found in minihalo sources , this simplified picture implies that the synchrotron spectrum should have the same slope everywhere throughout the minihalo . so far , the few available spatially - resolved observations of the minihalo spectral slopes indicate that this is likely not the case @xcite , although more detailed analyses are needed . under stationary conditions , and if diffusion and/or transport of crp and cre is not important , this power - law slope should extend to high frequencies , without any steepening ( distinct from the turbulent acceleration model , which predicts steepening at high frequencies due to the balance between reacceleration and losses on the cre ) . however , the conditions of cluster cores with minihalos probably do not lend themselves to a such a steady - state assumption , given the existence of cold fronts which betray the presence of sloshing motions . as previously noted ( zml11 ) , sloshing motions can significantly amplify the magnetic field strength on short timescales . @xcite demonstrated that a rapidly increasing magnetic field strength will result in a steeper spectrum than the steady - state case ( so long as @xmath20 ) . in the present case , magnetic field amplification near cold front surfaces would cause spectral steepening of the radio emission moving outward along the spiral arm of the sloshing cold fronts , away from the cluster center . the same magnetic field amplification has possible implications for the extent of the radio emission as well . @xcite predicted that as a result of the strengthening of the magnetic field , the radio emission would have a steep gradient at the front surface , since inside the front @xmath20 , and outside , @xmath21 . the appeal of this model for the steepening of the radio spectrum and the radial extent of minihalos is that it relies on a process , namely the amplification of the magnetic field by sloshing , that is strongly evidenced by previous theoretical and simulation work , and is independent of the details of the underlying crp spectrum and spatial distribution . for this reason , we believe it deserves close examination . in this work , we use the same cluster setup as in z13 to determine if the amplification of the magnetic field strength by the sloshing motions is sufficient _ by itself _ to produce the two distinct characteristics mentioned above , e.g. , emission confined to the cold front region with a steep spectrum , steeper than that expected from the canonical crp slope under stationary conditions . we follow the evolution of cre spectra associated with passive tracer particles which are advected with the flow of gas , determining the injection into and radiative losses on the spectrum along each trajectory . in this way we can determine how deviations from a steady - state configuration affect the surface brightness of the minihalo emission as well as its spectral shape . this paper is organized as follows . in section 2 we describe our method for evolving the cre spectra and our assumptions regarding their injection by hadronic processes . in section 3 we detail the results of our simulations . in section 4 we discuss the implications of this work , and in section 5 we make our conclusions . thoughout this work we assume a flat @xmath22cdm cosmology with @xmath23 = 0.7 and @xmath24 = 0.3 . for this work , we used the magnetohydrodynamic ( mhd ) simulation of a sloshing cluster core from z13 . we briefly outline the numerical methods and initial setup of this simulation , and refer the reader to zml11 and z13 for further algorithmic details . we performed our mhd simulation of gas sloshing using the ` flash ` code @xcite . ` flash ` solves the equations of ideal mhd using a directionally unsplit staggered mesh algorithm ( usm ; * ? ? ? * ) . for simulating the gravitational potential of the two clusters , we use the `` rigid body '' approach described in @xcite . our initial conditions consist of a massive ( @xmath25 kev ) cool - core galaxy cluster , and a smaller ( @xmath26 ) gasless subcluster , separated at an initial distance @xmath27 = 3 mpc , an initial impact parameter @xmath28 = 500 kpc , and placed on a bound mutual orbit . gas sloshing is initiated in the massive cluster by the gravitational encounter with the subcluster . our simulation employs adaptive mesh refinement ( amr ) . the size of the computational domain is @xmath29 = 2 mpc , with a finest cell size of @xmath30 = 1 kpc ( a discussion of the impact of the finite resolution of our simulation on the amplification of the magnetic field strength is given in the appendix ) . the maximum amr resolution covers a spherical region of @xmath31 kpc , centered on the dm peak of the main cluster , which encompasses all of the phenomena of interest in this study . we set up the initial magnetic field on the grid with a gaussian random field with a kolmogorov spectrum . the magnetic field strength has a radial profile that declines as the square root of the thermal pressure ( e.g. , a constant @xmath32 = 100 ) . as detailed in z13 , we allowed this configuration to relax for several gyr . the resulting magnetic field profile from this relaxation peaks at roughly @xmath33 g at the cluster center , decreases to @xmath95 @xmath34 g at the cooling radius , and drops below @xmath35 ( the redshift - dependent equivalent magnetic field strength of the cmb energy , @xmath36 ) at @xmath37 kpc ( cf . figure 1 of zml11 ) . finally , our simulation includes passive tracer particles flowing with the icm , which carry with them cre spectra which we evolve according to the method in section [ sec : method_spec ] . tracer particles are stored at intervals of 10 myr and from these snapshots the individual particle trajectories are extracted . the entire simulation contains approximately 10 million passive tracer particles initially distributed with their number density proportional to the local gas density . the local conditions of the gas at any time , such as the density of thermal particles and the magnetic field , can be determined by linearly interpolating between their values at the epochs corresponding to the datasets . we model the cre component as time - dependent spectra @xmath38 of the cre momentum @xmath39 along the tracer particle trajectories ( in post - processing ) . in this approximation , the cre are assumed to passively advect with the tracer particles once they are injected . along each tracer particle trajectory , we employ a kinetic equation for @xmath38 ( equivalent to a fokker - planck equation without the diffusive terms ) : @xmath40 = q(p , t)\ ] ] where @xmath41 is the sum of the standard radiative and coulomb losses : @xmath42 \\ \left.\frac{dp}{dt}\right|_{\rm coul } & = & -3.3 \times 10^{-29}n_{\rm th}\left[1+\ln\left(\frac{\gamma / n_{\rm th}}{75}\right)\right],\end{aligned}\ ] ] where the redshift is evolved according to our assumed cosmology . we assign @xmath43 to the epoch @xmath44 = 5 gyr of the simulation , in order to reproduce some of the observed nearby clusters exhibiting cold fronts in their cores . our calculations do not include reacceleration of cre by shocks or turbulence , or changes in energy due to adiabatic compression and expansion of the gas . to solve this kinetic equation we logarithmically sampled the cre spectrum @xmath45 100 times in the range of @xmath46 $ ] , and evolved it on this momentum grid forward in time using a finite difference method . at each timestep , we separate the calculation into two steps , `` advection '' and `` source term '' , using operator splitting . in the `` advection '' step , we compute the change in the spectrum in each momentum cell due to the change in the flux of electrons @xmath47 $ ] from the adjacent cells on the momentum grid . to ensure stability , we applied a minmod slope limiter to this flux at each cell boundary . following the `` advection '' step , we apply the `` source term '' operation @xmath48 by computing the change in the spectrum at each momentum cell using the time - interpolated value of the source term . finally , at the lower momentum boundary we impose outflow boundary conditions , and at the higher momentum boundary we extend the spectrum into the boundary cells with a constant logarithmic slope , a sensible choice since at such high energies in a steady - state the spectrum is dominated by the competition between injection and radiative losses and the spectrum is essentially a pure power law with index @xmath49 . we verify our model for the evolution of the spectrum by confirming that it can reproduce analytic predictions for the evolution of the cre spectrum under simplified conditions ( e.g. , radiative and coulomb losses without injection , and a steady - state balance between injection and losses ; see the left panel of figure [ fig : cre_spec ] . ) the injection rate is determined by the rate of interactions of crp with the thermal population . we assume that the crp momentum ( @xmath50 ) spectrum can be approximated by a single power - law , @xmath51 , where @xmath52 is a normalization constant and @xmath53 , a value within the range ( @xmath54 2.0 - 2.5 ) implied by various mechanisms capable of injecting relativistic crp into the icm ( e.g. , structure formation shocks , injection by radio galaxies , supernova remnants ; * ? ? ? * ; * ? ? ? * ) . for consistency with previous literature , we follow in part the derivation of the electron source function of @xcite . corresponds to the particle momentum but in their derivation it corresponds to the momentum nomalized by @xmath55 . ] the pion production spectrum is @xmath56 where @xmath57 is the inelastic p - p cross section and @xmath58 is the average momentum of a single produced pion , and we have employed the delta - function approximation . we may integrate this source function with respect to the crp momentum distribution to obtain @xmath59 where @xmath60 . using the transformation law for distribution functions and using the mean value of the cre momentum in the relativistic limit , we approximate the cre momentum source function by @xmath61 assuming that the mean momentum of the produced secondary electrons in the laboratory frame is given by @xmath62 . previous works ( see , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have commonly approximated the proton - proton cross section by an energy - averaged constant value , which results in a pure power - law form for the source function , and hence the steady - state electron cre is also a power - law with spectral index @xmath63 , and the synchrotron spectral index is @xmath64 . in this work we also account for the weak energy dependence of the cross section on the proton energy found by @xcite : @xmath65 ^ 2 \label{eqn : sigma_pp } { \rm mb}\ ] ] where @xmath66 and @xmath67 is the threshold energy of production of @xmath68 mesons . our inclusion of a weak energy dependence of @xmath69 results in a steady - state cre spectrum with a slope that is slightly shallower than @xmath49 . from numerical tests assuming a steady - state spectrum of cre we have determined that the observed spectral index for magnetic fields of @xmath70 g will be roughly within the range of @xmath71 , with a typical value of @xmath72 ( see figure [ fig : alpha_histogram ] ) , between 327 and 1420 mhz . ] a very minor deviation from the value of 1.15 expected by treating @xmath73 as constant . the cre in our simulations are produced solely by hadronic interactions between crp and thermal protons . crp are believed to be injected in the icm by a variety of processes , including shocks from supernovae , large - scale structure formation , and active galactic nuclei @xcite . in particular , diffusion ( along the field lines ) is an energy - dependent process that depends on the spectrum of mhd waves and on the physics of the interplay / scatter between crs and these waves . the time necessary for crs to diffuse over distances @xmath29 is @xmath74 , where @xmath75 is the spatial diffusion coefficient . this process is expected to affect our results if crs diffuse on scales of about 100 kpc within a time frame of @xmath9 1 gyr , which would imply a diffusion coefficient @xmath76@xmath77/s that is very large , although it can not be ruled out . sloshing can generate substantial turbulence in the core region . under these conditions , crs are expected to be transported as passive scalars by the turbulent gas - flow @xcite . on scales similar to that of the turbulence or larger , this induces a diffusive process that is energy independent with a diffusion coefficient @xmath78 , where @xmath79 and @xmath80 are the velocities and scales of the largest turbulent eddies . according to numerical simulations of sloshing motions in galaxy clusters , @xmath81 @xmath77/s @xcite , implying typical transport - diffusion scales of @xmath82 kpc on a timescale of 1 gyr for crp . previous works @xcite have studied the generation , advection and energy evolution of crp in the cosmological context . our focus in this work is on the effects produced by subsonic sloshing motions in a idealized , relatively relaxed cluster . therefore , we have chosen to model the evolution of the cre spectrum under the simplifying assumption that the crp energy density is proportional to the thermal energy , @xmath83 , e.g. , we assume passive advection of the crp population with the icm and ignore the propagation effects and energy loss processes mentioned above ( we briefly discuss the limitations of this approach in section [ sec : limitations ] . ) we choose a reference value of @xmath84 , at the upper limit placed on the energy density of cosmic - ray protons by the fermi , hess , and magic non - detections of @xmath8-ray emission from neutral pions , which are produced in the same hadronic interactions and subsequently decay @xcite . to get a sense for how deviations from steady - state conditions affect the resulting electron spectrum of the particles , we will first examine the properties of the tracer particles . the left panel of figure [ fig : cre_spec ] shows a typical example of a tracer particle with a steady - state cre spectrum at the epoch @xmath85 gyr , approximately 2 gyr past the core passage , a moment at which there is strong magnetic field amplification due to sloshing . the magnetic field strength for this particular particle , however , is not changing rapidly at this epoch . the black dashed lines indicate the values of @xmath8 that roughly correspond to the frequencies of 327 mhz and 1420 mhz at @xmath86 g . at low energies , where no injection occurs ( due to the energy threshold for pion production , cf . equation [ eqn : sigma_pp ] ) and coulomb losses dominate , the spectrum is relatively flat , whereas at higher energies it transitions to a power - law shape with @xmath87 , where it is dominated by the tradeoff between injection and radiative losses . for comparison , the steady - state spectrum for the same gas and magnetic field conditions is also plotted in the left panel of figure [ fig : cre_spec ] . the near equivalence of these two spectra indicates that the magnetic field strength and local density of particles are varying smoothly enough that the steady - state assumption is adequate . the center panel of figure [ fig : cre_spec ] shows a typical example of a cre spectrum of a tracer particle with a steep spectrum ( @xmath88 ) at the epoch @xmath85 gyr . the magnetic field of this particle is undergoing significant amplification at this epoch . the black dashed lines indicate the values of @xmath8 that roughly correspond to the frequencies of 327 mhz and 1420 mhz at @xmath89 g . this spectrum , in contrast to the example spectrum that is close to the steady - state case , has a steep transition segment between the flat segment and the high - energy , @xmath72 segment which is determined by the tradeoff between injection and radiative losses . the right panel of figure [ fig : cre_spec ] shows a typical example of a cre spectrum of a tracer particle with a shallow spectrum ( @xmath90 ) at the epoch @xmath85 gyr . the magnetic field strength of this particle is decreasing rapidly at this epoch . the black dashed lines indicate the values of @xmath8 that roughly correspond to the frequencies of 327 mhz and 1420 mhz at @xmath91 g . over most of the range of particle energies , the spectral index is nearly constant , except at very high energies where it steepens due to the @xmath92 dependence of the radiative losses . it is instructive to examine the behavior of the spectral index of individual tracer particles to see how they change with time , and how this is related to the change with time of the magnetic field strength . figure [ fig : alpha_evolution ] shows the change in the spectral index and the magnetic field strength for three different particles each which at some point have steep and shallow spectra , respectively . the particles we have examined in figure [ fig : cre_spec ] are indicated by the green curves . these curves demonstrate that while most of the time the spectral index is fairly constant and close to the steady - state value , that when the magnetic field increases rapidly it steepens , and when there is a rapid decrease it shallows . although these effects are clear and theoretically expected @xcite for particular tracers , we will show later ( within this section and section [ sec : spec_index ] ) that they can not be representative . figure [ fig : alpha_histogram ] shows a binned histogram of the radio spectral index @xmath16 for all of the tracer particles at the epoch @xmath44 = 3.5 gyr . it has a narrow peak at @xmath72 and a small skew toward steeper spectral indices . the spectra of particles gets as flat as @xmath93 and as steep as @xmath94 , but the number of these particles is dwarfed by the number of particles with spectra near the steady - state slope . to get a sense of how the spectral index depends on the magnetic field , figure [ fig : b_vs_alpha ] shows a phase plot of the radio spectral index versus the magnetic field strength for all of the particles in the simulation at the epoch @xmath44 = 3.5 gyr . at low magnetic field strengths , the fitted spectral index is very close to the steady - state expectation for nearly all tracer particles . the lack of steep spectral indices at these magnetic field strengths is readily understood ; in this regime @xmath21 . the fact that we also do not observe many particles to have shallower spectra at these magnetic field strengths is due to the fact that at lower magnetic field strengths the observed synchrotron frequencies are sampling higher cre energies , which will be within the part of the spectrum that is determined essentially completely by the tradeoff between radiative losses and injection . as we go to higher magnetic field strengths , we see a broader distribution of spectral indices . some of these spectra are somewhat flatter than the steady - state expectation , which correspond to particles where the magnetic field has been dropping rapidly ( figure [ fig : alpha_evolution ] ) . we find that the number of these particles increases up to @xmath95 and then drops off , due to the increasing importance of the synchrotron over the inverse - compton losses . similarly , beginning around @xmath95 , the population of tracer particles with very steep spectra increases , corresponding to particles which are experiencing rapid increases in magnetic field strength . nevertheless , the number of tracer particles with nearly steady - state spectra are dominant . figure [ fig:327_maps ] shows maps of the x - ray emission of the thermal gas with contours of radio surface brightness at the frequency of 327 mhz overlaid for several different epochs of the simulation where the crp energy follows the thermal energy density . the resulting radio emission is diffuse and fills the cool - core region of the cluster , similar to observed minihalos . it is fairly extended , and largely reflects the shape of the x - ray emission , which is expected in this case since @xmath96 , which is the same dependence on density as the x - ray emission . in particular , the spiral shape of the sloshing gas is readily seen in the radio contours . figure [ fig : profiles ] shows profiles of the radio surface brightness at the frequencies of 153 , 327 , and 1420 mhz corresponding to the black profiles in figure [ fig:327_maps ] at the epoch @xmath44 = 3.5 gyr . additionally , we have plotted the x - ray surface brightness along these profiles , and rescaled the overall normalization of both so that their radial dependence may be closely compared . along the southern profiles , there are drops in radio emission across the cold front surface of an order of magnitude over @xmath920 kpc . along the ne profile , there is a similar drop , but it is over a larger radial range , @xmath940 kpc . along the nw profile , the radio emission drops off with radius at roughly the same rate as the x - ray emission . this plot also demonstrates that the overall shape of the radio emission is roughly independent of frequency along these profiles . this is expected , since once non - stationary effects play a minor role , the spectrum is a power law and the properties of the minihalo simply scale with frequency . therefore , the main driver of the behavior of the minihalo profile is the strength of the magnetic field along this profile . along most directions , @xmath97 below the front and @xmath98 above , with the exception of the nw profile , where the magnetic field does not decrease substantially across the front surface ( cf . figure [ fig : scaledb ] and section [ sec : radial_extent ] ) . some observed minihalos ( e.g. , rxj 1720.1 + 26 , ms 1455.0 + 2232 * ? ? ? * ; * ? ? ? * ) typically have well - defined edges or appear to be confined to the core region , whereas others are more amorphous and/or extended ( e.g. , a2029 , ophiuchus * ? ? ? * ; * ? ? ? this potentially tells us something about the origin and /or transport of the relativistic particles . unfortunately , in this work we have only simulated one cluster configuration , so we do not have any statistics for comparison , but we can at least provide some indications of what may be expected for future study . in our reacceleration simulation of z13 , the radio emission was confined completely within the cold fronts , whereas the radio emission in our hadronic model is more extensive , particularly in the northern direction . to illustrate this , figure [ fig : compare_profiles ] shows profiles of the radio emission along the northwest and southeast profiles from the hadronic simulation and the reacceleration simulation of z13 . also plotted for comparison are 617 mhz radial profiles of the radio emission of rxj 1720.1 + 26 @xcite , a galaxy cluster with sloshing cold fronts and a minihalo qualitatively similar in shape to that from our simulations . the radii of the profiles from rxj 1720.1 + 26 have been rescaled such that the radii of the x - ray cold fronts are the same as in the simulations . both the hadronic and reacceleration models result in drops in radio emission at the cold front profiles , but the drops in the reacceleration model are much steeper , and appear very similar to the drops in radio emission in rxj 1720.1 + 26 . there is a strong increase in the reacceleration profile at @xmath99 130 kpc due to a patch of strong turbulence in this region , but even without this feature the drop in radio emission would be very pronounced . these results indicate that different physics potentially produces distinguishable predictions for the radial profiles of minihalos that can serve as a novel test to discriminate between different models for the origin of the emitting cre . measurements of the spectral index provide crucial information on the acceleration and transport physics of the cr . in the previous section , we have shown that in the hadronic model that the vast majority of our tracer particles have essentially stationary electron spectra , and that consequently the synchrotron spectrum only depends on the input spectrum of crp . this is different from reacceleration models , where turbulence plays a role in shaping the electron / synchrotron spectum and non - stationary conditions apply ( by definition ) . consequently the study of spectra is a powerful test to discriminate between the predictions of these two models , though in practice having multiple observations ranging over more than 1 decade in frequency will be necessary to start distinguishing between different models . figure [ fig : compare_spectra ] shows the total radio spectrum in w / hz within a radius of 300 kpc from the cluster potential minimum at several different epochs . spectra from z13 at similar epochs for the reacceleration model are also plotted for comparison . our hadronic spectra are essentially power - law in shape over a large range of radio frequency , with the spectral index near the steady - state value @xmath72 . in contrast , the reacceleration spectra gently steepen at high frequency , with different slopes depending on the epoch . to create maps of the spectral index @xmath16 , we reblock our intensity images by a factor of 8 and fit the spectrum in each reblocked pixel within a frequency range to a power - law , assuming 10% errors on the radio intensity in each band ( typical errors for the radio intensity in minihalo observations are within the @xmath95 - 10% range * ? ? ? * ) . figure [ fig : spec_index_maps ] shows maps of the spectral index fitted between 327 and 1420 mhz at several epochs of the simulation . we find a fairly uniform spectral index throughout the sloshing region , close to the steady - state expectation of @xmath72 , with the slopes in a narrow range of @xmath100 . near the cluster center , the spectrum becomes slightly flatter than the steady - state expectation ( @xmath101 ) , consistent with the increasing importance of coulomb losses in the higher density region . near the cold front edges , where the magnetic field is increasing rapidly due to shear amplification , we find that the spectral index increases to a maximum of @xmath102 . this range of observed @xmath16 , roughly within @xmath103 is well within the observational uncertainties of the spectral index , e.g. figure 6 of @xcite , and figure 8 of @xcite , indicating that when filtered by observational responses , this map might be difficult to distinguish from a spatially uniform distribution of spectral indices . though we found that the spectral index for individual tracer particles in these highly magnetized regions could be as steep as @xmath88 ( see section [ sec : tracer_particles ] ) , there are simply too few of these steep - spectrum particles to affect the spectrum of the emission projected along the line of sight . these results are inconsistent with the argument of @xcite for minihalos , who argued that rapid increases in magnetic field strength could explain steep - spectrum radio emission . lofar and lwalwa / index.html ] will observe radio sources at much lower frequencies , down to @xmath940 mhz . at these frequencies , the radiative loss timescales for the corresponding electrons are much longer and the effects of a change in the magnetic field would be more long - lasting . figure [ fig : spec_index_maps_low ] shows maps of the spectral index fitted between 60 and 153 mhz at the same epochs of the simulation . here , there is again steepening of the radio spectrum at the cold front surface , but this too is limited , up to @xmath104 , only slightly stronger than @xmath16 measured in the higher - frequency band . at early epochs ( @xmath105 gyr ) , we also see some marginal steepening of the radio spectrum just outside the core region ( @xmath106 kpc ) , where the magnetic field has been recently strongly amplified due to the recent passage of the subcluster and resulting compression of the plasma . in some observed minihalos , the radio emission is strongly confined to the core region , or the envelope of the cold fronts , if they are clearly observed . in our simulations , we find that the radial profile of the minihalo emission typically decreases at a more rapid rate approaching the cold fronts , but this is not always the case ( in particular the nw profile from figure [ fig : profiles ] ) . a drop in radio emission across a cold front in the secondary model primarily results from two effects : the first is the density jump across the cold front , which is typically a factor of @xmath91.5 - 2 for most cold fronts . the contribution of the density jump to the radio drop will be modest , since for the case where the crp energy density is proportional to the thermal density , @xmath107 , and the thermal pressure is roughly continuous across cold front surfaces ( except in a very thin layer at the cold front surfaces where the enhanced magnetic pressure due to shear amplification may become significant ) . the main contribution to the drop in radio emission will arise from the difference in the magnetic field across the front @xcite . below the front , the field will be amplified to @xmath108 , and the radio emission is roughly independent of the magnetic field strength . when the fronts reach a large radius , above the front the unamplified field will likely be weak due to the decline of magnetic field strength with radius , with @xmath109 , and the radio emission will drop off rapidly as the square of the field . figure [ fig : scaledb ] shows the mass - weighted projected magnetic field at the epoch @xmath85 gyr scaled by @xmath35 . throughout the volume of the sloshing region , the magnetic field has been amplified , and is stronger than @xmath35 . for the most part , just outside of the cold fronts the field drops to @xmath110 . however , the region to the northwest is filled with strong magnetic field outside of the cold fronts , due to strong velocity shears in that region . along projected radii in that direction , there will be no drop in emission as one crosses the cold front surface . clearly , in the hadronic model the radial extent of the minihalo emission depends critically on the magnetic field structure . this , of course , depends to a certain extent on our initial choice for the magnetic field strength . in some clusters , the magnetic field strength may be even higher than in our setup ( given by @xmath111 , see section [ sec : mhd ] ) due to turbulent - driven amplification caused by agn , minor mergers , and accretion . more simulations with different magnetic field configurations would be needed in order to determine the conditions under which minihalos would have steep drops in emission at cold front surfaces in the hadronic model . the significance of the drop in emission due to a sharp decrease in the magnetic field strength across the front is also affected by the strong redshift dependence of the cmb energy density , with @xmath112 . according to @xcite , rapid magnetic field amplification can cause significant departures from steady - state conditions , which produces synchrotron spectral steepening under the hypothesis of a secondary origin of the emitting particles . if strong , these effects may have important consequences for interpreting the origin of minihalos and giant halos . in our simulations we do not see significant effects of this nature . we find that the total spectrum of the minihalo is fairly consistent with a simple power - law in agreement with the expectation based on steady - state conditions . though we do see some steepening of the synchrotron spectrum near the cold front surfaces that we can confirm is indeed due to magnetic field amplification in these regions , we find that the effect is barely observable , even at very low observing frequencies where it should be maximized . section 4.3.3 of @xcite determines the change in spectral index @xmath16 for a sudden increase in the cooling rate over the injection rate , parameterized by @xmath113 from figure 16 of @xcite , we may determine the approximate increase in the magnetic field strength associated with a given spectral steepening , which will serve as an independent check on our model for the steepening of the cre spectra . it is instructive to examine the necessary increase in the cooling rate to produce the steepest spectra of any of the particles in the simulation ( @xmath94 ) and the steepest spectra we measure in our spectral index maps ( @xmath114 ) . for our injection index of @xmath115 , a sudden increase of @xmath116 is required to steepen the synchrotron spectrum to @xmath94 , whereas for steepening to @xmath114 a more modest increase of @xmath117 is required . what qualifies as a `` sudden '' increase ? formally , since in steady - state conditions the characteristic timescale is @xcite : @xmath118\nu_{1.4}^{-\frac{1}{2}}(1+z)^{-\frac{7}{2}}~{\rm gyr},\ ] ] any sudden increase in the cooling rate must occur on a timescale shorter than this in order to significantly affect the spectrum . for @xmath119 , @xmath120 gyr for electrons emitting at frequencies between 327 and 1400 mhz . motivated by this , we examine the trajectories of a few tracer particles with extreme ( @xmath94 ) spectral indices , to see if these conditions prevail . figure [ fig : cooling_evolution ] shows the evolution in @xmath121 for three tracer particles with spectral indices @xmath122 . @xmath121 for each particle is normalized to unity at time @xmath123 , and all times are given with respect to the epoch @xmath124 at which the spectral index is measured , marked by the black dashed line . each of these particles indeed experiences a sudden increase in @xmath121 , followed by a rapid decrease . the dot - dashed lines in the figure indicate the necessary @xmath121 to steepen the spectrum to the given spectral index from a steady - state , estimated from figure 16 in @xcite . all of the curves show an increase in @xmath121 that is at least this much , and this occurs over roughly a cooling time ( @xmath125 gyr ) , with the most significant increase occurring over a small fraction of that time . we therefore find that the evolution of the cre energies for the steepest - spectrum tracer particles in our simulation is consistent with the results of @xcite . the fact that we find so few tracer particles with very steep spectra indicates that these conditions ( required to produce such steepening ) do not prevail throughout most of the cluster core . though there is significant magnetic field amplification ( see figure [ fig : scaledb ] ) , it does not usually occur fast enough to steepen the spectrum significantly . in this work , we have adopted a simple model for hadronically originating cre , by relaxing the typical steady - state assumption and allowing for the time evolution of the cre spectrum due to changes in the local gas and crp density and the magnetic field strength , as well as the evolution of the cmb energy density with redshift . relaxing the steady - state assumption is meaningful in our simulation , due to our high spatial and time resolution which are able to resolve adequately the amplification of the magnetic field . as a significant simplification , we also assumed that at any given epoch the energy density of crp was proportional to the energy density of the thermal gas , and that the input spectrum of the crp is a constant power - law spectrum ( excepting the cutoff at the low - energy threshold ) . this approach is therefore limited , as it relies on a simplified treatment of the spatial distribution and energetics of the crp . first , we are not able to model in a self - consistent fashion the transport and diffusion of crp and its connection with the gas dynamics in the sloshing region . turbulence might transport crp on a scale that could be of the order of the cluster core , potentially inducing a spatial distribution of crp that is broader than that of the thermal plasma . momentum - diffusion of crs that is mediated by the scattering of particles with mhd waves in the plasma is a different process that is more difficult to model due to the uncertainties in the physics of ( small - scale ) turbulence in these environments . in addition , @xcite and @xcite proposed that super - alfvnic cr streaming is possible in the icm . in particular , @xcite calculated the suppression of the streaming instability in a turbulent flow showing that under these conditions the self - generated waves do not limit the particle drift velocity to the alfvn speed . however , this does not automatically imply that streaming is efficient , because the background turbulence ( necessary to suppress the instability ) still provides a source of scattering that may make the transport of crs diffusive and potentially inefficient ( see * ? ? ? * for a discussion ) . the combination of these effects may significantly affect the expected properties of minihalos with respect to our simplified treatment . turbulent transport and streaming would be the most important effects , and they typically produce a broader spatial distribution of crp with respect to our simplified model . however , if these effects are important , they would produce synchrotron distribution even broader than those obtained by our model , making the difference between reacceleration and hadronic models even more relevant . in particular , the radial drops in hadronic models could become even smoother , increasing the tension with observations . another possibility that we can not rule out , which would produce the opposite effect , is that the crp which are ultimately responsible for the radio emission are generated in the cluster center ( e.g. , from agn activity ) , and then diffuse outward , but are confined to the region bounded by the cold fronts due to the fact that the field direction is largely tangential at the front surfaces ( zml11 ) and thus would prevent diffusion across them . to investigate these and other effects , future studies of sloshing and minihalos based on simulation methods that self - consistently follow the transport of crp will be needed . second , given the simplified assumptions in our model with respect to the crp , the spatial and spectral distribution of the radio emission that is produced in our model is determined almost entirely by the spatial distribution and time evolution of the magnetic field . it is therefore crucial that we adequately resolve this evolution and are not missing any significant sources of further amplification . the magnetic field in our cluster core will be amplified by shearing motions localized at the cold fronts , as well as turbulence generated by sloshing . @xcite identified shear flows at cold front surfaces as the most relevant mechanism for the amplification of the magnetic field , which was confirmed in ideal mhd simulations by zml11 . however , turbulence will also play a role . in our inviscid simulation , turbulent dissipation is entirely numerical , and determined by the spatial resolution . therefore , though we resolve gas motions down to scales of @xmath9kpc , the scale at which dissipation of the turbulent cascade begins to set in is slightly larger , around @xmath910 kpc ( see z13 and references therein ) . in any case , any further field amplification provided by turbulent motions on scales that are not resolved by our simulations is likely to be a small fraction of what we are able to accomplish with our high - resolution simulation . the turbulence is injected via the sloshing motions on scales of @xmath9100 kpc , and cascades down to the previously mentioned scales where it is dissipated by our finite resolution . the energy flux through the turbulent cascade , @xmath126 ( assuming a kolmogorov spectrum ) is constant with respect to the scale @xmath127 . given this fact , the kinetic energy available for conversion into magnetic energy at these unresolved scales will be small compared with the energy available at the scales we do resolve . therefore , we do not expect that there is any significant room for further amplification by turbulence by resolving these smaller scales . even an increase in the magnetic field energy by a few tens of percent would not significantly change our conclusions . additionally , our previous work ( z13 ) showed that the kinetic energy in turbulence generated by sloshing on scales less than @xmath930 kpc is estimated to be at most a few percent of the thermal energy , and even if all of this energy were to be transferred to the magnetic field it would still not be as large as the contribution from the larger - scale sloshing motions ( see figure 24 , z13 ) . in the appendix , we use a resolution study to demonstrate that we have converged with respect to the overall properties of the magnetic field within the core region . however , it is the case that our specific initial value for the overall ratio of the thermal to the magnetic pressure , @xmath128 , is a free parameter in our simulations . zml11 investigated a range of initial values for @xmath128 ( over an order of magnitude from @xmath9100 - 6400 ) , and demonstrated that the average magnetic field strength within the core region was similar at late times , regardless of the initial field strength . we may predict that for initial @xmath128 higher than our value of 100 ( a weaker initial field ) , that the contrast between the field strength inside and outside the cold front would be larger , and there would be a sharper decrease in the radio emission at the cold front surface . on the other hand , for a lower initial @xmath128 ( a stronger initial field ) , the contrast between the magnetic field strength inside and outside the front would be smaller , and the drop in radio emission would be less significant . third , we assumed a population of crp with a single power - law spectrum . this implies an injection spectrum of cre that is power - law in form , and that any steepening of the radio spectrum at higher frequencies comes from strong departures from the steady - state condition . instead , steepening of the cre spectrum may reflect possible steepening of the crp spectrum . although the momentum distribution of crs is expected to be a power law in a very broad range of astrophysical situations , this is certainly a simplification . in our conditions , momentum - dependent spatial transport of crs could change the spectrum of crp . for example , @xcite discussed the possibility of a steepening of the crp spectral distribution resulting from an efficient momentum - dependent diffusion . this is also not included in our simulations . however , in non steady - state conditions , this process would eventually result in a spectral flattening of the synchrotron emission with distance , because higher energy crp would not be diffusively trapped and propagate faster faster ( this effect is unavoidable in the case that we measure variations of the radio spectrum with distance ) . the observational consequence would be that minihalos would be typically broader if observed at higher frequencies . this differs from expectations based on other models , allowing to discriminate between different origin scenarios for the cre . for example , reacceleration models eventually predict a spectral steepening with radius or more complex spectral variations ( e.g. * ? ? ? * and references therein ) . another limitation is that we do not include the effect of turbulent reacceleration on the spectrum of both crp and cre , a process that is believed to be important , for example , in the case of giant radio halos . in this work , we have employed a mhd simulation of gas sloshing in a galaxy cluster core to model the formation of a radio minihalo from hadronically - originating cosmic - ray electrons . we used passive tracer particles to follow the evolution of cre spectra under the assumption that these electrons are continuously injected by hadronic processes and they undergo radiative ( synchrotron and inverse - compton ) and coulomb losses via interaction with the thermal gas and magnetic fields . our model represents a simple extension beyond that of a steady - state approximation between injection and loss processes in that fast changes in the magnetic field strength are allowed to change the shape of the electron spectrum . the radio emission that is produced in our simulation is similar to that of observed minihalos in terms of being spatially diffuse and also being of similar luminosity . it also follows the overall spiral pattern of the cold fronts , expected due to the amplification of the magnetic field within these structures . in this respect , the model presented here and the results of our earlier work ( z13 ) using a reacceleration model give similar results . going beyond these considerations , the spatial extent and spectral steepening of radio minihalos provide avenues of distinguishing between models with different origin scenarios for the cosmic - ray electrons . therefore , in this work our aim was to determine the effects of a strong , fast amplification of the magnetic field due to sloshing motions on these two properties of the radio emission . this model is attractive because it does not require any further assumptions about the input spectrum and spatial distribution of the cosmic - ray protons . under these assumptions , we find that the radio emission associated with the minihalo is not completely confined within the core region or the area bounded by the cold fronts . along some radial directions that cross cold front surfaces , the radio emission drops steeply at these surfaces , due to the difference in the magnetic field strength across the interface . however , along other directions , the magnetic field strength changes less sharply at the fronts and as a result the radio emission decreases much more smoothly with radius , in concert with the mild decline in x - ray emission . we also find that some steepening of the synchrotron spectrum does occur in some regions due to rapid magnetic field amplification caused by sloshing . for some test particles , this effect is significant , with radio spectra as steep as @xmath129 ( compared to the steady - state , and prevalent , value of @xmath130 ) . however , the emission from these particles makes up a very small fraction of the total emission from the cre . as a result , we find that the total spectrum of the minihalo is essentially a power law , consistent with that expected from assuming steady - state conditions . the effects of rapid magnetic field amplification are more visible when we consider synchrotron spectral index maps . in these , we see some steepening in the region of the cold fronts where magnetic field amplification is stronger . however these effects induce only marginal variations of @xmath131 0.1 - 0.15 between 327 - 1420 mhz and @xmath131 0.2 - 0.25 between 60 - 153 mhz as one goes outward along the cold fronts , where magnetic field amplification is strongest . this differs from our previous results based on the same underlying numerical simulation under the hypothesis that the emitting electrons are reacccelerated by turbulence , where the spectral index is predicted to steepen more than this at high frequencies . based on our numerical studies from this work and z13 , we believe that the issues highlighted above regarding the spatial extent and the spectral slope of the minihalo emission are important to consider when discriminating between different origin models of the emitting cre . for example , our expectations based on our hadronic model are in tension with observations of a few of the best - observed minihalos , which show the radio emission confined to roughly the size of the core region in all directions @xcite . similarly , spectral studies of radio minihalos ( such as in * ? ? ? * ; * ? ? ? * ) indicate that significant spectral steepening is likely to be common in these sources . though the few observations currently available do not permit us to make firm conclusions , addressing these differences between the two models should be a focus of further studies . however , the parameter space explored in this work was necessarily limited . the most obvious extension of this work would need to incorporate the dynamics of the crp self - consistently to determine to what extent it plays a role in shaping the radio emission from minihalos . secondly , though our models resolve the structure of the magnetic field adequately , different initial conditions for the magnetic field strength and spatial configuration could be explored . finally , a fuller picture of the cause of such emission may arise from the combination of the effects of hadronic injection of cre and their subsequent turbulent reacceleration , a merger of the two currently competing models , which has already been investigated for giant radio halos from a theoretical perspective @xcite . jaz thanks uri keshet for useful discussions . analysis of the simulation data was carried out using the amr analysis and visualization toolset yt @xcite , which is available for download at http://yt-project.org . jaz is supported under the nasa postdoctoral program . gb acknowledges partial support by grant prin - inaf-2009 . the software used in this work was in part developed by the doe - supported asc / alliances center for astrophysical thermonuclear flashes at the university of chicago . ackermann , m. , ajello , m. , allafort , a. , et al . 2010 , , 717 , l71 ascasibar , y. , & markevitch , m. 2006 , , 650 , 102 blasi , p. , & colafrancesco , s. 1999 , astroparticle physics , 12 , 169 bonafede , a. , feretti , l. , murgia , m. , et al . 2010 , , 513 , a30 brunetti , g. , setti , g. , feretti , l. , & giovannini , g. 2001 , , 320 , 365 brunetti , g. , & blasi , p. 2005 , , 363 , 1173 brunetti , g. , & lazarian , a. 2007 , , 378 , 245 brunetti , g. , & lazarian , a. 2011 , , 410 , 127 brunetti , g. , & jones , t. w. 2014 , arxiv:1401.7519 burns , j. o. , sulkanen , m. e. , gisler , g. r. , & perley , r. a. 1992 , , 388 , l49 cassano , r. , gitti , m. , & brunetti , g. 2008 , , 486 , l31 dolag , k. , & enlin , t. a. 2000 , , 362 , 151 , a. , antypas , k. , ganapathy , m. k. , reid , l. b. , riley , k. m. , sheeler , d. , siegel , a. , weide , k. extensible component based architecture for flash , a massively parallel , multiphysics simulation code . parallel computing 35 ( 10 - 11 ) , 512522 . enlin , t. , pfrommer , c. , miniati , f. , & subramanian , k. 2011 , , 527 , aa99 the _ fermi_-lat collaboration : ackermann , m. , et al . 2013 , arxiv:1308.5654 fujita , y. , & ohira , y. 2012 , , 746 , 53 ghizzardi , s. , rossetti , m. , & molendi , s. 2010 , , 516 , a32 giacintucci , s. , markevitch , m. , brunetti , g. , cassano , r. , & venturi , t. 2011 , , 525 , l10 giacintucci , s. , markevitch , m. , venturi , t. , et al . 2014 , , 781 , 9 giacintucci , s. , markevitch , m. , brunetti , g. , et al . 2014 , , 795 , 73 gitti , m. , brunetti , g. , & setti , g. 2002 , , 386 , 456 gitti , m. , ferrari , c. , domainko , w. , feretti , l. , & schindler , s. 2007 , , 470 , l25 gitti , m. 2013 , , 436 , l84 govoni , f. , murgia , m. , markevitch , m. , et al . 2009 , , 499 , 371 hlavacek - larrondo , j. , allen , s. w. , taylor , g. b. , et al . 2013 , , 777 , 163 kelner , s. r. , aharonian , f. a. , & bugayov , v. v. 2006 , , 74 , 034018 keshet , u. , markevitch , m. , birnboim , y. , & loeb , a. 2010 , , 719 , l74 keshet , u. , & loeb , a. 2010 , , 722 , 737 keshet , u. 2010 , arxiv:1011.0729 lee , d. , & deane , a. e. 2009 , journal of computational physics , 228 , 952 magic collaboration , aleksi , j. , alvarez , e. a. , et al . 2011 , arxiv:1111.5544 markevitch , m. , & vikhlinin , a. 2007 , , 443 , 1 mazzotta , p. , & giacintucci , s. 2008 , , 675 , l9 miniati , f. 2001 , computer physics communications , 141 , 17 miniati , f. , ryu , d. , kang , h. , & jones , t. w. 2001 , , 559 , 59 murgia , m. , eckert , d. , govoni , f. , et al . 2010 , , 514 , a76 pfrommer , c. , & enlin , t. a. 2004 , , 413 , 17 pfrommer , c. , springel , v. , jubelgas , m. , & ensslin , t. a. 2007 , cosmic frontiers , 379 , 221 pfrommer , c. , enlin , t. a. , & springel , v. 2008 , , 385 , 1211 prokhorov , d. a. , & churazov , e. m. 2013 , arxiv:1309.0197 roediger , e. , brggen , m. , simionescu , a. , et al . 2011 , , 413 , 2057 roediger , e. , & zuhone , j. a. 2012 , , 419 , 1338 sarazin , c. l. 1999 , , 520 , 529 sijbring , d. 1993 , ph.d . thesis , groningen smith , r. k. , brickhouse , n. s. , liedahl , d. a. , & raymond , j. c. 2001 , , 556 , l91 turk , m. j. , smith , b. d. , oishi , j. s. , skory , s. , skillman , s. w. , abel , t. , & norman , m. l. 2011 , , 192 , 9 vazza , f. , brggen , m. , gheller , c. , & brunetti , g. 2012 , , 421 , 3375 vazza , f. , roediger , e. , & brggen , m. 2012 , , 544 , a103 vazza , f. , gheller , c. , & brggen , m. 2014 , , 439 , 2662 wiener , j. , oh , s. , & guo , f. 2013 , arxiv:1303.4746 zandanel , f. , pfrommer , c. , & prada , f. 2014 , , 438 , 124 zuhone , j. a. , markevitch , m. , & johnson , r. e. 2010 , , 717 , 908 ( zmj10 ) zuhone , j. a. , markevitch , m. , & lee , d. 2011 , , 743 , 16 ( zml11 ) zuhone , j. a. , markevitch , m. , brunetti , g. , & giacintucci , s. 2013 , , 762 , 78 ( z13 ) zuhone , j. a. , kunz , m. w. , markevitch , m. , stone , j. m. , & biffi , v. 2014 , arxiv:1406.4031 the finite resolution of our simulations limits the amplification of the magnetic field strength that can occur , either due to turbulence or bulk motions like sloshing ( which is the predominant source of such amplification in our simulation ) . higher resolution simulations will result in stronger magnetic fields in regions where vorticity or shear is significant . to examine the effect of resolution on our simulations , we compare our default simulation with a finest cell width of @xmath132 kpc to an otherwise identical simulation that has been evolved with a finest cell width of @xmath133 kpc . figure [ fig : am06_res_test ] shows slices of the magnetic field strength through the center of the domain for both of these simulations at the epoch @xmath44 = 3.5 gyr . the simulation with higher resolution exhibits more small - scale structure in the magnetic field , along with localized regions where the magnetic field is stronger than the corresponding regions in its lower - resolution counterpart . however , the average field over larger scales is the same in both simulations . figure [ fig : am06_res_test_proj ] shows projections of the magnetic field strength from the two simulations ( weighted by the density squared , see equation [ eqn : ss_j ] ) . to approximate the resolution one would expect from a typical observation of a minihalo , we have smoothed each image with a gaussian with fwhm = 10 kpc . the two maps of the magnetic field strength are very similar , with subtle differences , but the field is not significantly stronger in the higher - resolution simulation . the similarities in field amplification between the two simulations can also be seen in figure [ fig : am06_histogram ] , which shows a binned histogram of the fraction @xmath134 of the total volume @xmath135 ( centered on the cluster potential minimum ) with a given magnetic field strength . the curves for the two simulations are essentially identical up to @xmath136 , and the increase in field strength for the higher - resolution simulation does not extend much farther than this , @xmath137 . the differences between the two simulations also occur within a very small volume , a fraction @xmath138 of the inner @xmath139 volume . finally , we also examine the time dependence of the magnetic field amplification within the cluster core for our two different resolution cases . the left panel of figure [ fig : dbdt ] shows the evolution of the magnetic energy over time within three spherical volumes centered on the cluster potential minimum , defined by the radii @xmath140 100 , 200 , and 300 kpc . in each of these volumes , we see an increase in the magnetic energy due to field amplification by the sloshing motions , followed by a decrease at later times due to the cold fronts expanding beyond the radius of each sphere ( where the bulk of the increase in the magnetic field is occurring ) . both simulations exhibit essentially the same time dependence with respect to the evolution of the magnetic field energy within these volumes . the right panel of figure [ fig : dbdt ] shows the relative difference in the magnetic energy versus time between the two simulations . in general , the higher - resolution case exhibits a greater increase in the magnetic energy over the lower - resolution case , but the difference is very small , up to @xmath915 - 20% more energy at later times . this again indicates that we have adequate resolution in our default simulation to resolve the the amplification of the magnetic field by sloshing . from this investigation , we conclude that for the purposes of this work the spatial resolution of our mhd simulation is adequate , since the magnetic field strength that is achieved throughout the volume of the cold fronts and on resolvable spatial scales is essentially the same between the two different simulations .

most autonomous robotic agents use logic inference to keep themselves to safe and permitted behaviour . given a set of rules , it is important that the robot is able to establish the consistency between its rules , its perception - based beliefs , its planned actions and their consequences . this paper investigates how a robotic agent can use model checking to examine the consistency of its rules , beliefs and actions . a rule set is modelled by a boolean evolution system with synchronous semantics , which can be translated into a labelled transition system ( lts ) . it is proven that stability and consistency can be formulated as computation tree logic ( ctl ) and linear temporal logic ( ltl ) properties . two new algorithms are presented to perform realtime consistency and stability checks respectively . their implementation provides us a computational tool , which can form the basis of efficient consistency checks on - board robots .

1611.03322

a robotic system s decision making is well known to be in need of some hard decision making at times . a most popular example is asimov s laws @xcite , which demonstrate the difficulties to apply logic by robots in practice . a shortened version of these laws is `` 1 . a robot may not allow a human being to come to harm . 2 . a robot must obey the orders given to it by human beings except if the order causes harm to humans . 3 . a robot must protect its own existence as long as such protection does not cause harm to humans . '' assuming these , what would happen to the robot s decision making if a human commands a robot to kill someone , but at the same time threatens to kill himself if the robot does not obey ? in this example the human introduces a contradiction into the logic of the robot . to avoid this the robot may have a complex rule base to provide it with legal and ethical principles and can be equipped by a meta law which says that `` the robot should not allow itself to be dictated by communicated conditions which make its logic contradictory '' . in this example one could say that in legal terms the suicide will remain the sole `` responsibility '' of the threatening person who commands the robot . the problem is not only the imperfection of asimov s robotic laws or that an agent programmer can make mistakes . logical consistency checks are also needed when the robot s perception - based beliefs are wrong . the agent can be programmed to re - examine whether its beliefs may need to be changed as were mistakenly believed to be true or false . this is not unlike enabling the agent to think like poirot , miss marple or sherlock holmes when they are reassessing their initial beliefs or impressions . but there are simpler cases : a robot may decide that the book it sees on the table can not be tom s as that one is in his home . in this paper we address the problem of how a robot can quickly and efficiently resolve inconsistencies in order to make the right decisions . the ability of making fast decisions about logical consistency , and the robot s ability to detect when inconsistency occurs , is an important problem for the future of robotics . it is also of particular importance for logic - based robot control systems , e.g. , @xcite . a typical logic - based robotic system usually contains a belief set , which provides the basis of reasoning for a robot s behaviour @xcite . an inconsistent belief set could lead to a wrong plan causing an unexpected result , e.g. , an unmanned vehicle can hit an obstacle , instead of avoiding it , if it mistakenly believes that any route of avoidance could cause more damage , due to , for instance , mis - perception of the environment . its mis perception could perhaps be corrected if it had been able to combine environmental prior knowledge with current sensing . in a rapidly changing environment bayesian methods can be used to identify and track movements of objects and establish functional relationships , e.g. , @xcite . when faced with balanced probabilities for two hypothetical and competing relationships in the robot s environment , it may need to make a decision based on the application of logic using prior knowledge . discovery of logical inconsistency in geometrical and physical relationships in an environmental model should prompt a robotic agent to revise its perception model of the world . for instance belief - desire - intention ( bdi ) agents should carry out consistency checks in their reasoning cycle in languages such as @xmath0 , @xmath1 and @xmath2 @xcite . in these systems the agent programmer should program logical consistency checks and handling of inconsistencies at design stage of the software . to topic of fast consistency checking by robots has also implications for legal certification of robots . as we humans formulate social and legal behaviour rules in terms of logical implications , the process is likely to be similar for robots and the problem of consistent decisions by robots is an important generic capability . future legal frameworks for certification of robots need to take into account verifiable decision making by robots . consistency checks on a set of logic rules in propositional logic is a textbook problem and has been extended to various types of logic systems in terms of validity , consistency and satisfiability . for instance @xcite provides an authoritative account of the history of logical consistency checking in a propositional logic . relevant methods and algorithms have long been investigated for database systems and rule - based expert systems , e.g. , @xcite , but none has been specifically designed for robotics . query language 4ql @xcite and boolean networks ( bn ) @xcite are very similar to our modelling formalism _ boolean evolution systems_. the former allows a variable to have four values : @xmath3 , @xmath4 , @xmath5 and @xmath6 . the algorithm that computes the unique well - supported model in @xcite can be adapted to check consistency , but it can only deal with one initial evaluation of variables at a time . bn was developed for modelling gene regulatory networks in biology . in bn , a boolean variable can only take either @xmath3 or @xmath4 , while in our formalism , a variable can be initialised as @xmath5 . research on bdi reasoning cycles focuses on runtime detection and resolution of conflicting goals , such as @xcite . no work has been conducted on complex reasoning process , which will be required by autonomous and intelligent robots . for realtime robotic systems it is important to increase solver efficiency to be able to deal with large search spaces with complex reasoning process for both offline and online application . in this respect , the use of binary decision diagram ( bdd ) is very effective by compressing search space through generating a unique and succinct representation of a boolean formula . bdd has been widely adopted for model checking @xcite , and applied successfully to verification of large systems . in this paper we adopt the bdd based symbolic model checking approach @xcite to robotics . to our best knowledge , nothing has been reported on its application on consistency and stability checking of decisions by robots . in this paper we propose a fast method for discovery of inconsistency in a set of logic rules and statements on relationships in a current world model , past actions , planned actions and behaviour rules of a robotic agent . we do not address the problem of how to resolve logical inconsistency , mainly because we hold the view that , to eliminate inconsistencies , a robot can efficiently improve its world model by non - logic based techniques . such techniques can include gathering more perception data , active vision , using alternative action plans or analyzing and deriving spatial temporal models using probabilities . if a single new perception predicate or predicate derived by logic rules of the robot contradicts its otherwise consistent world model , then the robot may apply a set of logic rules to derive a correction of its belief in terms of the predicate . what to derive and analyse for consistency is however a broad topic and lies outside of the scope of this paper . here we focus on fast discovery of inconsistencies which is fundamental for safe operations of autonomous robots . with time it should be a key technical part in the process of legal certification of future autonomous robots . our contribution builds on and develops our past efficient state space generation and parallel computation @xcite methods further . we have previously developed various state space reduction techniques for symbolic model checking via bdds , such as symmetry reduction @xcite and abstraction @xcite . the preliminary results of our techniques have been published in @xcite . in this paper we elucidate the setting for which our techniques are designed and demonstrate their way of using it in robotics . we also extend the techniques to deal with a different semantics and develop a new technique to extract counterexamples efficiently when the system is inconsistent or unstable . the counterexamples are useful for system developers to correct robotic reasoning systems ; they can provide guidance on how to improve the reasoning process of robots . we study the efficiency of the agent s ability to examine the consistency of its beliefs and logic rules and , if inconsistency occurs , generate counterexamples to the rules which can then be used by the robot to resolve inconsistency . our technique can be used both by robot programmers at software design stage and by robots when reasoning . in the former case , system developers can check the logical consistency of reasoning cycles in agent programs at design stage . for each inconsistent check , a counterexample can be produced to help developers understand the source of inconsistency and correct the program . in the latter case , consistency checks are carried out by the robots themselves in realtime and counterexamples are examined to improve reasoning , e.g. , bringing in more sensor data to eliminate ambiguity or bring about alternative decisions about future actions . in section [ sec : system ] we introduce the problem in a robotic framework and its characteristics . in section [ boolevul ] boolean evolution systems are formally represented . in section [ sec : modelling ] , we translate boolean evolution systems into _ transition systems _ which are now widely used in the control systems literature @xcite , which provides the basis of verification . note that in this paper we abstract robotic behaviour to propositional logic to be able to cope with computational complexity of consistency checking . section [ sec : mc ] contains our results on stability of boolean evolution systems in terms of ctl and ltl formulae . an important result states that stability checking can be reduced to a reachability problem which only asks for one fixpoint computation . similarly , consistency checking can be also converted into simple fixpoint computation . section [ sec : case ] presents a case study in a home robotics scenario , which demonstrates the use of uncertain sensory and communication information and a set of rules to satisfy . in section [ sec : exp ] , performance comparison between ctl formulae based solutions and the reachability based algorithms is highlighted and implemented in the symbolic model checker mcmas @xcite . we discuss stability checking under an alternative semantics of evolution in section [ sec : interleaving ] . we conclude the paper in section [ sec : concl ] . our predicates - based knowledge representation of a robot , which is derived from sensing events , remembering the past as well as from prediction of a future environment , is schematically depicted in fig . [ robpreds ] . for new sensory predicates we assume that the robot is able to identify which are uncertain in a probabilistic sense . the following specific problems are to be addressed : 1 . assuming inconsistency occurs , identify which uncertain new sensing predicates in @xmath7 can be made certain within rules @xmath8 based on physical models . 2 . the agent considers a set @xmath9 of actions as its options . for each action @xmath10 in @xmath11 it simulates a physical model over a time horizon and abstracts a set of events @xmath12 for its future consequences . it checks if @xmath13 and its behaviour rules @xmath14 are consistent based on 1 ) and 2 ) . the set @xmath15 of feasible actions @xmath10 in @xmath11 , which are consistent with @xmath14 , are used by the robot to make a final choice of an action using non - logic based evaluations ( for instance using planning ) . in fig . [ robpreds ] the diamonds indicate the procedural locations of logical consistency checks , based on predicates and sets of rules ( logical implications ) . [ h!t ! ] it can however happen that some of the probabilistic sensing of events remain unresolved based on physical models and associated rules : let @xmath16 denote the set of undecided perceptions . the robotic agent needs to check for each of its possible actions what would happen if various combinations of its uncertain perceptions in @xmath17 were true or false . in safety critical situations a robot can not take any action , which could lead to it breaking its rules in some combination of truth values in @xmath17 . checking this can require complex consistency checking to be done while the robot interacts with its environment , hence the efficient methods proposed in this paper are key to timely decisions by a robot . this paper is not committed to any particular type of software architecture . we assume that propositional logic using a predicate system , which can admit arguments but is equivalent to propositional logic ( for decidability properties ) , is used in the robotic software . we also assume that the robot perceives and creates predicates about environmental events and about its actions periodically within a _ reasoning cycle _ performed at an approximately fixed rate per second . at a given reasoning cycle of the robotic agent , indexed with time @xmath18 , the agent holds a set of predicates @xmath19 in its memory , possibly some of these with negation signs . this means that the predicates in @xmath20 split into two disjoint sets as @xmath21 consisting of ones assigned value @xmath3 while the rest the boolean value @xmath4 . such an assignment of boolean values in @xmath20 is called a valuation of the boolean variables in @xmath20 and denoted by @xmath22 . the agent also has a set of rules at time @xmath18 denoted by @xmath23 . the rule set @xmath24 may contain more variables than @xmath20 . those not in @xmath20 are unknown to the agent and its logic reasoning is then interested in the problem of satisfiability of all the logic rules by suitable assignments to the unknown variables . in the following we will drop the time index @xmath18 as we will be interested in the consistency of logic rules at any time , in view of some boolean evaluations . the terms `` variable '' and `` predicate '' will be used interchangeably . our primary problem is that the robotic agent has limited time for logical derivations , when quick response is required , and it needs to assess the following : * are its current evaluations and its rule base consistent in the sense that unknown variables can take on values to satisfy all the rules ? * having answered the previous question negatively , can it modify some of its own boolean evaluations so that its set of predicates becomes consistent with its set of rules ? testing consistency of a set of evaluations can be achieved by checking satisfiability of the conjunction of the evaluations and the rule set , and obtaining consistent values for unknown variables can be done by starting to apply the rules until the boolean evaluation becomes stable , i.e. the logical value of no variable changes any more . however , it can be inefficient to use this method as the number of evaluations may increase exponentially with the number of variables . by analogy to previous definitions @xcite of agentspeak - like architectures for belief - desrie - intention type of robotic agents , we define our reasoning system by a tuple : @xmath25 where : * @xmath26 is the set of all predicates . * @xmath27 is the total atomic belief set . the current belief base at time @xmath18 is defined as @xmath28 . at time @xmath18 beliefs that are added , deleted or modified are considered _ events _ and are included in the set @xmath29 , which is called the _ event set_. events can be either _ internal _ or _ external _ depending on whether they are generated from an internal action , in which case are referred to as `` mental notes '' , or an external input , in which case are called `` percepts '' . * @xmath30 is a set of implication rules . * @xmath31 is the set of executable plans or _ plans library_. current applicable plans at time @xmath18 are part of the subset @xmath32 , this set is also named the _ desire set_. a set @xmath33 of intentions is also defined , which contains plans that the agent is committed to execute . * @xmath34 is a set of all available actions . actions can be either _ internal _ , when they modify the belief base or generate internal events , or _ , when they are linked to external functions that operate in the environment . agentspeak - like languages , including lisa ( limited instruction set architecture ) @xcite , can be fully defined and implemented by listing the following characteristics : * _ initial beliefs_. + the initial beliefs and goals @xmath35 are a set of literals that are automatically copied into the _ belief base _ @xmath36 ( that is the set of current beliefs ) when the agent mind is first run . * _ initial actions_. + the initial actions @xmath37 are a set of actions that are executed when the agent mind is first run . the actions are generally goals that activate specific plans . * _ logic rules_. + a set of logic based implication rules @xmath38 describes _ theoretical _ reasoning about physics and about behaviour rules to redefine the robot s current knowledge about the world and influence its decision on what action to take . * _ executable plans_. + a set of _ executable plans _ or _ plan library _ @xmath39 . each plan @xmath40 is described in the form : @xmath41 where @xmath42 is a _ triggering predicate _ obtained by consistency in @xmath43 and possible valuation for the best choice of @xmath44 from @xmath45 . next the @xmath42 allows the plan to be retrieved from the plan library whenever it becomes true ; @xmath46 is called the _ context _ , which allows the agent to check the state of the world , described by the current belief set @xmath36 , before applying a particular plan ; the @xmath47 form a list of actions to be executed . the above list of steps are cyclically repeated to run the reasoning process of a robotic agent . a binary - decision - diagram ( bdd ) @xcite is a succinct representation of a set of boolean evaluations and , motivated by this , we examine the possibility of applying symbolic model checking via bdds to verify consistency and stability . this way , we avoid the combinatorial explosion of evaluations . we will show that bdd based model checking is very efficient for this task to be carried out in realtime , while the agent needs to give quick responses to its environment . as agent perception processes are often prone to errors in a physical world due to sensor issues or to unfavourable environmental conditions , this is an important problem of robotic systems . we present a formal definition of the consistency checking problems in the next section . a boolean evolution system @xmath48 is composed of a set of boolean variables @xmath49 and a set of evolution rules @xmath50 defined over @xmath51 . a rule @xmath52 is of the form @xmath53 , where @xmath54 is the guard , i.e. , a boolean formula over @xmath51 , and @xmath55 is an assignment that assigns @xmath3 ( `` 1 '' ) or @xmath4 ( `` 0 '' ) to a boolean variable @xmath56 . for simplicity , we write a rule of the form @xmath57 as @xmath58 , and write @xmath59 as @xmath60 . we also group rules with the same guard into one . for example , two rules @xmath61 and @xmath62 can be written as @xmath63 . in practice , the set @xmath51 is usually partitioned into 2 subsets : @xmath64 and @xmath65 , where variables in the former are initialized to either @xmath3 or @xmath4 , and variables in the latter initialized to @xmath5 . accordingly , the guard of a rule can be evaluated to @xmath3 , @xmath4 and @xmath5 . the last case can occur when the guard contains a variable in @xmath65 . to model a predicates - based knowledge representation and reasoning system in fig . [ robpreds ] by a bes , we translate each predicate in @xmath36 , action in @xmath11 and future event in @xmath12 into a boolean variable and each reasoning rule in @xmath66 into a boolean formula . in particular , the uncertain sensing predicates in @xmath67 and future events in @xmath12 are placed in @xmath65 , and those in @xmath68 and actions in @xmath11 are placed in @xmath64 . let @xmath69 be a valuation of the boolean variables , and @xmath70 the value of variable @xmath71 in @xmath69 . we say that a rule @xmath72 is _ enabled _ if its guard @xmath54 is evaluated to @xmath3 on @xmath69 . the new valuation , after applying the evolution rules to @xmath69 , is defined by _ synchronous evolution semantics _ as follows . let @xmath73 be the set of rules that are enabled . the new valuation @xmath74 is the result of simultaneously applying all rules in @xmath75 to @xmath69 . that is , every value of @xmath71 in @xmath74 is defined as follows . @xmath76 the evolution from @xmath69 to @xmath74 is written as @xmath77 . we assume that for each valuation , there exists a non - empty set of enabled rules . a boolean evolution system is _ stable _ if from any valuation and applying the rules recursively , it eventually reaches a valuation @xmath69 where no other valuation can be obtained , i.e. , @xmath78 . we say that @xmath69 is a _ stable _ valuation , written as @xmath79 . whether stability happens is decidable by the agent : it requires that two consecutive steps in the evolution have identical valuations . [ def : incon ] three problems might occur during evolution of a bes : 1 . two enabled rules try to update the same boolean variable with opposite values at some time ; 2 . a variable in @xmath64 is updated to the opposite value of its initial value at some time . a variable in @xmath65 is updated to the opposite value at some time after its value has been determined are initially set to @xmath5 , which can be overwritten using the evolution rules . ] . if any of these problem happens , we say that the system is _ inconsistent_. otherwise , the system is _ consistent_. these problems should be identified when robotic software is programmed . for instance belief - desire - intention rational agent implementations apply the logic rules in each reasoning cycle in @xmath0 , @xmath1 and @xmath2 @xcite . within one reasoning cycle , where the input to the variables in @xmath64 is kept constant . this justifies the second and third problems in definition [ def : incon ] . * example 1 . * @xmath80 this example demonstrates the inconsistency under synchronous semantics . for the initial valuation @xmath81 , both the first and second rules are enabled , which makes @xmath82 and @xmath83 . in the next evolution iteration , the second rule sets @xmath84 to @xmath3 , while the third one sets @xmath84 to @xmath4 . [ example1-cex ] illustrates the evaluation in these evolution iterations . the following result can be used to provide a simple algorithm to solve problem ( 1 ) of the agent . [ theorem : simple ] let @xmath69 be a boolean evaluations of variables in the rule set @xmath85 . then the following hold . if the boolean evolution system is not stable then @xmath69 and @xmath85 are inconsistent which the agent can detect from the same evaluation reoccurring during the boolean evolution . if the evolution is not stable , then during the transition steps between a recurrence of the same evaluation , some evaluations must be different as otherwise the evolution would be stable with the evaluation that occurred at two consecutive identical evaluations . as an evaluation reoccurs , this means that some variable values in that evaluation are changed between the two identical occurrences of valuations . let @xmath86 be such a variable . the logic rules applied , which led to the recurrence of the evaluation , have in fact forced @xmath86 at least once to change to its opposite value and later change back . this means that the rule set @xmath85 for the initial evaluation is inconsistent with the rules , i.e. @xmath85 is not satisfiable by any evaluation which is consistent with the initial evaluation @xmath69 . theorem [ theorem : simple ] shows that stability is guaranteed in consistent systems . for certain systems , however , the inconsistency conditions in definition [ def : incon ] are considered unnecessarily strict in that the initial value of _ known _ variables may not be obtained directly from the environment . hence , these values can sometimes be incorrect . to alleviate this problem , the second and third inconsistency condition in definition [ def : incon ] can be relaxed . using this principle , we say that the second and the third conditions are _ solvable _ if the system eventually reaches a stable state by ignoring these two conditions . this principle makes consistency and stability checking not straightforward any more : some rules can correct the evaluations of some predicates . * example 2 . * @xmath87 this example shows a consistent and stable system , where @xmath88 and @xmath89 . we use a sequence of ` @xmath90 ' , ` @xmath91 ' and ` @xmath92 ' to represent states . for example , the initial state ` @xmath93 ' represents @xmath94 . * for valuation @xmath95 , the evolution is @xmath96 . * for valuation @xmath97 , we have @xmath98 . in this section , we describe how to enable model checking to deal with boolean evolution systems . first , we introduce _ transition systems _ , which are a mathematical formalism that forms the basis of model checking . second , we present an example of encoding a boolean evolution system under the semantics of transition systems using an input language of a model checker . model checking is usually performed on transition systems . here we present the definition of transition systems and the translation of a boolean evolution system into a transition system . a transition system @xmath99 is a tuple @xmath100 , @xmath101 such that * @xmath102 is a finite set of states ; * @xmath103 is a set of initial states ; * @xmath104 is the transition relation ; * @xmath105 is a set of atomic propositions ; * @xmath106 is a labelling function mapping states to the set of atomic propositions @xmath105 . we denote the set of atomic propositions valid in state @xmath107 by @xmath108 . let @xmath109 be a set of states . the function @xmath110 computes the successor states of @xmath111 under @xmath112 . formally , @xmath113 given a boolean evolution system @xmath114 with @xmath115 unknown variables , i.e. , @xmath116 and @xmath117 known variables , i.e. , @xmath118 , let @xmath119 @xmath120 @xmath121 , where @xmath122 is an atomic proposition representing that a variable @xmath123 is @xmath3 , @xmath124 representing that @xmath125 is @xmath4 , and @xmath126 representing that an unknown variable @xmath127 has value @xmath5 . a transition system ( ts ) @xmath99 can be generated from @xmath128 as follows . 1 . @xmath102 is composed of all @xmath129 valuation of @xmath51 . @xmath130 is composed of @xmath131 valuations , where variables in @xmath64 can take either @xmath3 or @xmath4 , and variables in @xmath65 take @xmath5 . 3 . a transition @xmath132 iff @xmath133 . in the presence of inconsistent update of some boolean variables , the successor valuation is chosen randomly , which results in multiple successor states . for example , consider a valuation @xmath107 from where a boolean variable @xmath86 can be updated to @xmath3 by rule @xmath134 and @xmath4 by rule @xmath135 . in the transition system , @xmath107 has two successor states , i.e , valuation : one state contains @xmath97 and the other contains @xmath95 . if there are @xmath136 boolean variables that are updated inconsistently in @xmath107 , then @xmath107 has @xmath137 successor states . @xmath138 is defined such that for each variable @xmath139 , @xmath140 iff @xmath125 is evaluated as @xmath3 , @xmath141 iff @xmath125 is evaluated as @xmath4 , and for each variable @xmath127 , @xmath142 iff @xmath143 is evaluated to @xmath3 . note that all possible input values of variables in @xmath65 are captured by @xmath130 , i.e. , each possible valuation of @xmath65 is encoded into an initial state in @xmath130 . the set of states and the transition relation in a transition system can be illustrated as a direct graph , where a transition @xmath144 is represented by an arrow from @xmath145 to @xmath146 . [ example1-lts ] shows the directed graph for example 1 in section [ sec : system ] . a boolean evolution system can be written as a program in the input language of a symbolic model checker , such as nusmv @xcite . the program is then parsed by the model checker to build a transition system . in this section , we show how to model a boolean evolution system by an ispl ( interpreted system programming language ) @xcite program , inspired by the interpreted system semantics @xcite , and the corresponding transition system can be generated by the model checker mcmas @xcite . we use example 1 to illustrate how to construct an ispl program from the boolean evolution system @xmath147 . an ispl program contains a set of agents , a set of atomic propositions , an expression representing the initial states and a set of logic formulas representing the specification of the system . the structure of the program is as follows : .... agent 1 ... end agent ... agent n ... end agent evaluation ... end evaluation initstates ... end initstates formulae ... end formulae .... where atomic propositions are defined in the section `` evaluation '' and the initial states defined in `` initstates '' . each agent is composed of a set of _ program variables _ , a set of _ actions _ that the agent can execute , a _ protocol _ and an _ evolution function_. each agent has a set of local states that are encoded by its program variables : each valuation of the variables is a local state . its protocol defines a set of enabled actions for each local state , and its evolution function specifies the transition relation among its local states . the structure of an agent @xmath148 is below : .... agent m vars : ... end vars actions = { ... } ; protocol : ... end protocol evolution : ... end evolution end agent .... to encode a bes into an ispl program , we only need one agent , and this agent has only one action , which is enabled in every local state . in the rest of this section , we give details of the construction of the ispl program . the definition of actions and protocol is omitted as they do not affect the translation of the bes . 1 . as explained before , we do not directly list all states in the state space @xmath102 of the corresponding transition system . instead , we define program variables to match variables in @xmath128 . each variable in @xmath64 is translated into a boolean variable in ispl and each variable in @xmath65 into an enumerated variable with three values @xmath149 , @xmath150 and @xmath151 . the corresponding ispl code for example 1 is as follows . + .... vars : a : boolean ; b : { true , false , unknown } ; c : { true , false , unknown } ; end vars .... each evolution rule is translated into a guarded transition `` @xmath84 ` if ` @xmath54 '' in ispl , where guard @xmath54 is a boolean expression over variables , and @xmath84 is a set of assignments . indeed , the semantics of a guarded transition matches exactly that of an evolution rule . the rules in example 1 are translated into the ispl code below . + .... evolution : b = false if a = true ; c = true if a = true ; b = false if c = false ; end evolution .... 3 . as each variable in @xmath65 in @xmath128 is initialized to _ unknown _ , we need to specify this in the initial state section @xmath152 in an ispl program . the following code is generated for example 1 . + .... initstates m.b = unknown and m.c = unknown ; end initstates .... + note that ` m ` is the name of the agent , which encapsulates the variables and transitions , and ` m.x ` refers to the variable ` x ` in ` m ` . an atomic proposition in ispl is of the form `` @xmath153 ` if ` @xmath54 '' , where @xmath153 is the name of the atomic proposition , and @xmath54 is a boolean expression that defines the set of states @xmath153 holds . that is , @xmath153 holds in any state whose corresponding valuation satisfies @xmath54 . the ispl code for example 1 is below . + .... evaluation a_true if m.a = true ; a_false if m.a = false ; b_true if m.b = true ; b_false if m.b = false ; b_unknown if m.b = unknown ; c_true if m.c = true ; c_false if m.c = false ; c_unknown if m.c = unknown ; end evaluation .... the above construction steps suggests that a compiler can be produced without difficulties to automatically generated ispl code from a given boolean evolution system . although we have shown the possibility of coding a boolean evolution system in ispl , we would like to emphasize that compilers for other symbolic model checkers can also be constructed when necessary . for example , the semantics of the input language of nusmv is similar to that of ispl in this setting as we do not use the capability of specifying actions and protocols in ispl . computation tree logic ( ctl ) @xcite and linear time temporal logic ( ltl ) @xcite are the most popular logics adopted in verification of transition systems to specify properties that a system under investigation may possess . ctl is a branching time logic , which considers all possibilities of future behaviour , while ltl only deals with one possible future behaviour at a time . in this section , we use ctl to formulate stability and inconsistency checks due to the efficient implementation of ctl model checking . but we also discuss the application of ltl when possible . ltl can be specified by the following grammar @xcite : @xmath154 ctl on the other hand is given by the extended grammar @xcite : @xmath155 both ctl and ltl are defined over paths in a transition system . given a transition system @xmath156 , a path @xmath157 is a ( finite or infinite ) sequence of states such that for each pair of adjacent states , there exists a transition in the system , i.e. , @xmath158 for all @xmath159 and @xmath160 for all @xmath161 . we denote the @xmath162-th state in the path @xmath163 , i.e. , @xmath164 , by @xmath165 . the satisfaction of ctl and ltl in @xmath99 is defined as follows . [ ctl_sat ] given a transition system @xmath166 and a state @xmath167 , the satisfaction for a ctl formula @xmath168 at state @xmath107 in @xmath169 , denoted by @xmath170 , is recursively defined as follows . * @xmath171 iff @xmath172 ; * @xmath173 iff it is not the case that @xmath174 ; * @xmath175 iff @xmath176 and @xmath177 ; * @xmath178 iff there exists a path @xmath163 starting at @xmath107 such that @xmath179 . * @xmath180 iff there exists a path @xmath163 starting at @xmath107 such that @xmath181 for all @xmath182 ; * @xmath183 iff there exists a path @xmath163 starting at @xmath107 such that for some @xmath182 , @xmath184 ; * @xmath185 iff there exists a path @xmath163 starting at @xmath107 such that for some @xmath182 , @xmath186 and @xmath187 for all @xmath188 ; * @xmath189 iff for all paths @xmath163 starting at @xmath107 , we have @xmath179 . * @xmath190 iff for all paths @xmath163 starting at @xmath107 , we have @xmath181 for all @xmath182 ; * @xmath191 iff for all paths @xmath163 starting at @xmath107 , there exists @xmath182 such that @xmath184 ; * @xmath192 iff for all paths @xmath163 starting at @xmath107 , there exists @xmath182 such that @xmath186 and @xmath187 for all @xmath188 ; [ ltl_sat ] given a transition system @xmath166 and a state @xmath167 , the satisfaction for a ltl formula @xmath168 at state @xmath107 in @xmath169 , denoted @xmath170 , is recursively defined as follows . * @xmath171 iff @xmath172 ; * @xmath173 iff it is not the case that @xmath174 ; * @xmath175 iff @xmath176 and @xmath177 ; * @xmath193 iff for all paths @xmath163 starting at @xmath107 , we have @xmath179 . * @xmath194 iff for all paths @xmath163 starting at @xmath107 , we have @xmath181 for all @xmath182 ; * @xmath195 iff for all paths @xmath163 starting at @xmath107 , there exists @xmath182 such that @xmath184 ; * @xmath196 iff for all paths @xmath163 starting at @xmath107 , there exists @xmath182 such that @xmath186 and @xmath187 for all @xmath188 ; when we verify whether a ctl / ltl formula @xmath168 holds on a model , we check if this formula is satisfied by all initial states , denoted by @xmath197 . in particular , when we say that an ltl @xmath168 holds in the model , every path from every initial state has to satisfy @xmath168 . more details of ctl and ltl , as well as the difference between them , can be found in @xcite . [ thm : incons ] the first category of inconsistency can be checked by the following ctl formula @xmath198 if a system is inconsistent due to the first case , then there must exist a state that has two successor states such that a variable is evaluated to _ true _ in one successor state , and to _ false _ in the other . the ctl formula @xmath199 captures this scenario for variable @xmath125 . the negation @xmath200 excludes the occurrence of inconsistency caused by @xmath125 . operator @xmath201 guarantees that inconsistency does not occur in any states . note that it is not necessary to consider a case like @xmath202 for an _ unknown _ variable because it can not be assigned to @xmath5 during evolution . if the above formulae are evaluated to true , then the boolean evolution system is consistent . note that this category of inconsistency can not be checked by an ltl formula because ltl can only specify linear properties . however , a small modification would make ltl work again on checking consistency . lemma [ thm : incons ] searches for occurrences of inconsistency by checking if two opposite values of a variable can be reached from one state . the following theorem focuses on looking for such a state to perform consistency checks . [ thm : incons1 ] checking the first category of inconsistency can be transformed into a reachability problem as follows . 1 . for each pair of rules @xmath203 and @xmath204 , check if @xmath205 and @xmath206 assign opposite values to the same boolean variable . if the answer is yes and @xmath207 , then we add a new atomic proposition @xmath208 that holds in states satisfying @xmath209 . 2 . let @xmath210 be the set of all newly added propositions in the previous step . the first category of inconsistency can be identified by the ctl formula @xmath211 or the ltl formula @xmath212 the system is consistent if formula [ eqn : incons1_ctl ] or [ eqn : incons1_ltl ] is true . if a state satisfies a proposition @xmath213 , which is constructed as @xmath214 , then both guards @xmath215 and @xmath216 are satisfied in the state . thus , the corresponding rules @xmath217 and @xmath218 are enabled in the state . as these two rules set opposite values to a variable , inconsistency occurs in this state if it is reachable from an initial state . @xmath219 captures all states where inconsistency could happen , and @xmath220 and @xmath221 examine if any of these states can be reached from an initial state . note that @xmath222 and @xmath223 . although the second and the third cases are not needed in the relaxed inconsistency conditions , we still present the temporal logic properties for checking them . the second category of inconsistency can be checked by the following ctl formula @xmath224 or ltl formula @xmath225 if this case occurs , then there must exist a state @xmath107 that has a successor state @xmath226 such that a variable is evaluated to @xmath3 in @xmath107 and @xmath4 in @xmath226 , or @xmath4 in @xmath107 and @xmath3 in @xmath226 . the ctl formulas @xmath227 and @xmath228 ( @xmath229 ) capture this scenario for variable @xmath125 . the ltl formulas @xmath230 and @xmath231 have the same effect . the negation @xmath200 excludes the occurrence of inconsistency caused by @xmath125 . operator @xmath201 ( or @xmath232 ) guarantees that inconsistency does not occur in any states . the third category of inconsistency can be checked in the same way over the _ unknown _ variables . [ thm : stability ] the stability problem can be checked by the following ctl formula @xmath233 or ltl formula @xmath234 if the above ltl or ctl formula is evaluated to true , then the boolean evolution system is stable . in a stable system , every path leads to a stable state , where no _ unknown _ variable will change its value any more . therefore , one of three cases @xmath235 , @xmath236 or @xmath237 for _ unknown _ variable @xmath125 holds in the stable state . the last case means that the _ unknown _ variable remains @xmath5 during the evolution . the _ known _ variables can not take value _ unknown_. thus , we do not need to consider them being @xmath5 in the ltl formula . the operator @xmath221 specifies that this stable state will be reached eventually . the ctl formula can be reasoned in a similar way . although theorem [ thm : incons1 ] provides a simpler ctl / ltl formula than lemma [ thm : incons ] , in practice , it can be improved further . in order to check if the formula @xmath238 is satisfied in a transition system @xmath156 by symbolic model checking , we need to compute the set of states satisfying the formula @xmath238 using a fixed - point computation . let @xmath239 represent the set of states satisfying @xmath168 . the fixed - point computation begins with a set @xmath240 and computes a sequence of sets such that @xmath241 until @xmath242 . the detailed algorithm is presented in algorithm [ algo : ef ] . algorithm [ algo : ef ] could be time - consuming for a large system . fortunately , we can utilise a characteristic of model checking to avoid the problem of checking @xmath220 . a model checker generates only reachable states , which can be reached from initial states , and perform model checking algorithms on the reachable states . to identify inconsistency , showing the existence of a reachable state with two conflict successor states is sufficient . as model checkers only work on reachable states , the existence of a bad state can be converted into non - emptiness of the set of states satisfying @xmath210 defined in theorem [ thm : incons1 ] , returned by a model checker . therefore , the fixed - point computation for @xmath220 can be avoided . indeed , checking existence of bad states can be integrated into the process of generation of reachable state space . once a bad state is found , the process can be aborted to give fast feedback to the programmer . for a large system , the ctl formula specified in lemma [ thm : stability ] involves a large number of conjunction clauses @xmath243 or @xmath244 and each @xmath201 requires a computational expensive fixed - point computation , as @xmath245 . therefore , model checking this formula could be time consuming . the following theorem tells us that stability checking can be reduced to a reachability problem , which only asks for one fixed - point computation . [ thm : reachability ] stability in a consistent system @xmath246 can be checked in the following three steps . 1 . find the set @xmath55 of states that only have self - loop transitions , i.e. , @xmath247 ; 2 . find the set @xmath248 of states that can reach states in @xmath55 ; 3 . check if @xmath249 . if the answer is yes , then the system is stable if it is consistent . from the definition of stability , we know that a stable valuation corresponds to a state that only has self - loops , and vice versa . in a consistent system , a state can not enter a non - stable loop if it can reach a stable state . otherwise , there exists a state that has two successor states , which contradicts the assumption that the system is consistent . step 3 checks if there exists an initial state that can not reach a stable state . the existence of such a state means that the system contains a non - stable loop . for instance the cudd library @xcite can be used to manipulate bdds in mcmas . the first step can be implemented using the function `` cudd_xeqy '' in cudd , which constructs a bdd for the function @xmath250 for two sets of bdd variables @xmath153 and @xmath251 . when applied to the transition relation in the system , this function simply enforces that the successor state of a state @xmath107 is @xmath107 itself , i.e. , a self - loop . the second step can be achieved by the classic model checking algorithm for @xmath220 . the third step is done by checking if @xmath252 is a zero bdd , which means that the result from the set subtraction @xmath252 is empty . therefore , this algorithm runs more efficiently than model checking the long formula in lemma [ thm : stability ] . in practice , this stability check can be combined with consistency checks . during the generation of the reachable state space , we check if the system is consistent using theorem [ thm : incons1 ] . if the generation is not aborted due to the occurrence of inconsistent states , then a stability check is executed . a common question asked after a formula is model checked is whether a witness execution or counterexample can be generated to facilitate deep understanding of why the formula holds or does not hold in the system . in our situation , it is natural to ask the model checker to return all evolution traces that lead to inconsistency or instability . we will show how to compute traces in mcmas for inconsistency first and for instability afterwards . it is usually good in practice to generate the shortest traces for counterexamples / witness executions in order to decrease the difficulty of understanding them . to achieve this for our setting , we utilize the approach of construction of state space in mcmas . starting from the set of initial states @xmath130 , mcmas can compute the state space in the following manner @xcite . in this algorithm , @xmath102 is the set of reachable states and @xmath253 , initialised as @xmath130 , is the set of states that their successor states need to be computed , which is done by the function @xmath254 . in each iteration , we compute the successors @xmath255 of @xmath253 , and remove from @xmath255 the states that have been processed before by @xmath256 . this iteration continues until no new states can be added to @xmath102 , i.e. , @xmath257 . we modify the state space generation algorithm to store every intermediate @xmath253 : in each iteration @xmath162 , we change @xmath253 to @xmath258 . the modified algorithm is shown in algorithm [ algo : modified ] . [ algo : modified ] [ thm : cex ] a shortest trace leading to an inconsistent state by enabling the rules @xmath259 and @xmath260 , can be achieved in the following steps . 1 . starting from @xmath261 , we test each @xmath258 to search for the smallest index @xmath136 such that @xmath262 . 2 . we pick up an arbitrary state @xmath263 from @xmath264 and compute its predecessor @xmath265 in @xmath266 by using the reversed transition relation @xmath267 such that @xmath268 . if @xmath263 has multiple predecessors , then we pick up an arbitrary one to be @xmath265 . in the same way , we compute a predecessor of @xmath265 in @xmath269 . this process continues until we find a state @xmath270 in @xmath271 , which is @xmath130 . to find the shortest counterexamples for unstable loops , we need to identify all such loops first , and for each loop , we test each @xmath258 from @xmath261 if it contains a state in the loop , i.e. , if @xmath272 , where @xmath273 is the set of states in the loop . next we apply the second step in theorem [ thm : cex ] to generate the shortest trace . now we focus on how to find all unstable loops efficiently . [ lem : scc ] given a consistent system , none of the unstable loops interfere with each other . if the conjunction of two loops is not empty , then there exists a state such that it has two outgoing transitions , one in each loop . hence , this state leads to the occurrence of inconsistency . due to lemma [ lem : scc ] , finding unstable loops is equivalent to finding non - trivial strongly connected components ( sccs ) when the system is consistent . there are several scc identification algorithms in the literature working on bdd representation of state spaces @xcite . the more efficient one was reported in @xcite . but before we apply these algorithms , we could remove states that can not reach any unstable loops from the state space in order to speed up the computation . those states are identified as @xmath248 in the second step of stability checking in theorem [ thm : stability ] . in this section we illustrate the use of our consistency and stability checking techniques on an example scenario which could occur to a household robot . the robot with arms observes an object rolling across a table . it needs to decide whether to stop it or allow it to drop off from the table . the object can be a glass or a light effect . it would be unwise to stop the object if it is the latter case . the robot may resort to more accurate sensors to decide how to react . the model is formalized in a boolean evolution system based on the perception structure in fig . [ rescyc ] . 1 . feasible sensing possibilities ( @xmath36 ) are : * @xmath274 : object o rolls across table * @xmath275 : senses that object o is rolling across table * @xmath276 : sensed o but there is no real object o * @xmath277 : derived that light effect moving across table , there was no real object o sensed + in @xmath36 , the uncertain sensing events ( @xmath278 ) are @xmath275 and @xmath277 . action possibilities ( @xmath11 ) are : * @xmath279 : stop rolling object by arm * @xmath280 : remain idle 3 . future events predicted ( @xmath12 ) are : * @xmath281 : object o falls * @xmath282 : object o breaks * @xmath283 : object o is useless * @xmath284 : handling of object o * @xmath285 : the robot has made the correct observation * @xmath286 : the robot chooses the correct action 4 . naive physics rules ( @xmath8 ) are : * @xmath287 : the object will fall if not stopped * @xmath288 : if the object falls it will break * @xmath289 : if object is stopped it will not fall * @xmath290 : if object will not fall then it will not break 5 . general rules - values and moral consequences rules ( @xmath14 ) are : * @xmath291 * @xmath292 * @xmath293 * @xmath294 * @xmath295 * @xmath296 the robot starts with a simple but fast reasoning cycle by considering each action individually using observation only . the criteria for choosing the correct action is to guarantee the following goals . * @xmath297 * @xmath298 * @xmath299 when only one of events @xmath274 and @xmath276 is true , the robot can make its decision easily . however , it is difficult to do so when both events are true . the reasoning process is as follows . 1 . _ evaluation of action choice 1 _ : goals + @xmath280 + this choice results that @xmath286 becomes false . evaluation of action choice 2 _ : goals + @xmath279 + this results inconsistency in the reasoning process as shown in fig . [ fig - example ] , which demonstrates the evolution of the value of @xmath300 . to resolve the inconsistency , the robot needs to acquire information from more sensors , which would instantiate the two sensing events @xmath275 and @xmath277 in @xmath278 with two extra physical rules . * @xmath301 * @xmath302 if these two sensing events do not become true simultaneously , then the robot can make the correct decision . our consistency and stability checking techniques of this kind can be used in both offline and online modes . in the online mode , counterexamples are used to assist the system to acquire more information , i.e. , fixing the uncertain sensing events , or adjusting the possible actions that can be take , in order to solve inconsistency or instability problems in a consistency resolution cycle . our case study demonstrates an application of the online mode . [ rescyc ] illustrates the consistency resolution cycle that can be implemented in agent programs . in the offline mode , users can apply the techniques to their rule - based reasoning system to check consistency and stability . if a problem is found , users can revise their system using generated counterexamples as a guidance . for example , if we add the following extra general rules in the case study , * @xmath303 * @xmath304 the robot could be trapped in a circular reasoning process when @xmath274 is true land @xmath276 is false because @xmath305 and @xmath306 this circular reasoning process can be captured by our stability check . we have integrated the algorithms in theorem [ thm : incons1 ] and [ thm : reachability ] into the model checker mcmas @xcite . the implementation slightly deviated from theorem [ thm : incons1 ] in order to maximize the utility of the internal encoding of the transition relation in mcmas . instead of finding all pair of conflicting rules , we built a single bdd for each variable @xmath307 . we first collect all rules @xmath308 that set @xmath307 to @xmath3 , and all rules @xmath309 that set @xmath307 to @xmath4 . second , we generate a bdd @xmath310 representing @xmath311 if the bdd is not empty , then there exists a pair of conflicting rules that might be enabled simultaneously . searching for a bad state is done by testing the conjunction of @xmath310 and @xmath102 , the set of reachable states . to demonstrate their performance , we applied them to the following example , where @xmath312 and @xmath313 . in the experiment , we fixed @xmath314 to 32 , and generated a series of models by replicating the group of variables @xmath315 . in the largest model , this group has ten copies , i.e. , @xmath136 ranges from @xmath91 to @xmath316 , which means the total number of variables is @xmath317 . each variable in @xmath64 requires one bdd variable to encode , as one bdd variable can represent two values 0 and 1 , perfectly matching boolean values @xmath4 and @xmath3 . each variable in @xmath65 needs two bdd variables because it has three values . therefore , the total number of bdd variables in the largest model is 383 . * example . * + @xmath318 @xmath319 the experimental results are listed in table [ table - exp ] . for each model , we present the number of variables and corresponding bdd variables in parentheses , as well as the number of reachable states . the time ( in second ) spent on checking consistency and stability via the ctl formulae [ eqn : incons ] and [ eqn : ctl ] are shown in the two columns in the middle , and the time for the direct algorithms in theorem [ thm : incons1 ] and [ thm : reachability ] are given in the last two columns . the results clearly demonstrates the huge advantage of using our stability checking algorithm . the performance of our consistency checking algorithm is also excellent , given the fact that the ctl formula [ eqn : incons ] is quite efficient already . note that the time spent on building bdd @xmath310 for each variable is shown to be below 1ms in the penultimate column of the table . .[table - exp ] experimental results . [ cols="^ , > , > , > , > , > " , ] although synchronous semantics have been applied broadly in practice , a differently semantics , _ interleaving semantics _ , still finds its usefulness in case of limited processing power . interleaving means that only one enabled rule , which is usually chosen randomly , is processed at a time . the new valuation @xmath74 under interleaving semantics is the result of applying a rule @xmath320 in @xmath75 to @xmath69 . the rule @xmath320 is chosen non - deterministically . that is , every new value of @xmath71 in @xmath74 is defined as follows . @xmath321 under the relaxed inconsistency conditions , a system is guaranteed to be consistent , if at any given time , only one rule is processed . therefore , the first inconsistent condition is not satisfied any more . however , the interleaving semantics possesses different characteristics during stability checking . a stable system under the synchronous semantics may become unstable . let us re - examine example 2 using the interleaving semantics . we can construct a path that visits unstable states as follows . for valuation @xmath97 , we have @xmath322 . the infinite loop @xmath323 makes the system unstable . however , the infinite loop is quite special in that the rule @xmath324 is enabled infinitely often in state @xmath325 , which is the beginning of the unstable loop . in practice , such infinite loops rarely happen because of randomness of choice . once this rule is executed , the unstable loop is interrupted , and the system becomes stable . this observation leads to the introduction of fairness into stability checking . fairness @xcite has been studied and applied to many temporal logic - based verification , including both ctl and ltl . various types of fairness constraints have been brought up . among them , the most popular ones are _ unconditional _ , _ strong _ and _ weak _ fairness . in this section , _ strong _ fairness is sufficient to exclude the above unrealistic / unfair paths . strong fairness under interleaving semantics requires that , in every infinite path , an evolution rule has to be executed infinitely often if it is enabled infinitely often . for transition systems , strong fairness is composed of a set of fairness constraints , written as @xmath326 where each constraints @xmath327 specifies that if @xmath328 occurs infinitely often , then @xmath329 has to occurs infinitely often as well . strong fairness rules out unrealistic evolution paths , where some enabled rules are consistently ignored . therefore , it allows more systems evaluated as stable . for example 2 , we only need one fairness constraint : @xmath330 this example suggests that the generation of a fairness constraint from a rule can be straightforward , which can be achieved by following the syntactic form of the rule . however , strong fairness still can not prevent some stable system under synchronous semantics from being unstable . the following example demonstrates an unstable system under strong fairness . in this example , @xmath88 and @xmath331 * example 3 . * @xmath332 for the initial valuation @xmath97 , we have @xmath333 . the unstable loop @xmath334 can not be broken because the only rule that can break it , i.e. , @xmath335 is disabled in state @xmath336 . enforcing strong fairness in the verification of ctl formulas can be transformed into finding strongly connected components ( sccs ) , which in turn can be implemented using graphic algorithms @xcite . as we do not consider inconsistency for the interleaving semantics , the stability can be checked by ltl model checkers , such as spin @xcite and nusmv . the verification of an ltl formula @xmath337 under strong fairness can be achieved directly by checking a combined ltl formula @xmath338 where @xmath339 is of the form of formula ( [ eqn : fair ] ) . for stability checking the @xmath337 is taken as in ( [ eqn : ltl ] ) . note that the algorithm in theorem [ thm : reachability ] does not work here because multiple successor states do not mean inconsistency any more . spin uses explicit model checking techniquesfor the verification . it requires that every initial valuation has to be enumerated explicitly , which is very inefficient . nusmv adopts the method in @xcite to check ltl formulae symbolically via bdds , which can be more efficient for our purpose . now the question is how we identify rules that need to be guaranteed for execution by strong fairness . human guidance on the selection of rules would be ideal . when it is not available , we need to find a solution to allow automatic selection . a simple method is to put all rules under the protection of fairness . this solution does not request the modification of an existing model checker . however , it only works for a small rule set . a large number of rules would render @xmath339 a large ltl formula containing equal number of constraints as the number of rules . an alternative solution utilises a sequence of verification to search for a fair unstable loop . starting with no fairness constraints , we check formula ( [ eqn : ltl ] ) solely on the converted transition system . if the result is @xmath4 , which means the system may be unstable , then we ask the model checker to generate a counterexample , i.e. an unstable loop . we examine each state in the loop to look for enabled rules that are never executed in the loop . if no such rule is found , then the system is unstable under strong fairness . otherwise , we put the unfairly treated rules into @xmath339 and re - start the verification . this process is carried out iteratively until the system is proven to be stable or unstable under strong fairness . although the idea of this solution is not complex , its implementation requires to build an extension of a model checker , e.g. , nusmv , which is not a trivial task . further , its performance would be degraded when the number of iterations increases . this paper has solved the problem of efficiency for logical consistency checks of robots by adopting symbolic model checking based on binary decision diagrams . in addition to specifying stability and consistency as ctl and ltl formulas , also efficient symbolic algorithms have been applied to speed up consistency and stability checking . timely decision making by robots can be vital in many safety critical applications . the most basic task they need to check is the consistency of their assumptions and their rules . speed of computation hence affects quality and safety of robots . as a first step towards application of our approach , we have embedded it within the framework lisa @xcite for reasoning of robots . further direct use of the techniques is in rule - based reasoning systems before they are deployed on robots . the counter - examples , which can generated by the techniques presented , can demonstrate the reasons for possible violation , which can help software developers revising their designs . sometimes it can be time - consuming to modify the design of a complex system , possibly the violation is tolerable or is very rare during run - time . in these cases counter - examples can be used as a guide to correct the reasoning of a robot while in action . future work in this research area can target the implementation of our approach in practical programming @xcite , and to aid finding solutions to making inconsistent / unstable systems consistent and stable . based on the results an iterative design process can be defined to enable a programmer to control an agent s decision making . our plans are also to integrate consistency checks in lisa @xcite into the control code of unmanned aerial vehicles ( uavs ) and unmanned ground vehicles ( ugvs ) for practical application . this work was supported by the epsrc project ep / j011894/2 . s. maghsoudi , i. watson , epistemic logic and planning , in : proceedings of knowledge - based intelligent information and engineering systems ( kes04 ) , vol . 3214 of lecture notes in computer science , springer , 2004 , pp . 3645 . m. shanahan , m. witkowski , high - level robot control through logic , in : proceedings of intelligent agents vii agent theories architectures and languages , vol . 1986 of lecture notes in computer science , springer , 2001 , pp . 104121 . m. singh , d.r.parhi , s.bhowmik , s.k.kashyap , intelligent controller for mobile robot : fuzzy logic approach , in : proceedings of international association for computer methods and advances in geomechanics ( iacmag08 ) , 2008 , pp . 17551762 . c. r. torres , j. m. abe , g. lambert - torres , j. i. d. s. filho , h. g. martins , autonomous mobile robot emmy iii , in : proceedings of new advances in intelligent decision technologies , vol . 199 of studies in computational intelligence , springer , 2009 , pp . 317327 . l. mihaylova , a. carmi , f. septier , a. gning , s. k. pang , s. j. godsill , overview of bayesian sequential monte carlo methods for group and extended object tracking , digital signal processing 25 ( 2014 ) 116 . v. morreale , s. bonura , g. francaviglia , m. puccio , f. centineo , g. cammarata , m. cossentino , s. gaglio , practionist : a framework for developing bdi agent systems , in : proceedings of the 7th woa 2006 workshop , from objects to agents , vol . 204 of ceur workshop proceedings , ceur-ws.org , 2006 , pp . 187194 . a. cimatti , e. m. clarke , e. giunchiglia , f. giunchiglia , m. pistore , m. roveri , r. sebastiani , a. tacchella , nusmv 2 : an opensource tool for symbolic model checking , in : proceedings of cav 2002 , vol . 2404 of lecture notes in computer science , springer , 2002 , pp . 359364 .

evolution of stellar bars in disk galaxies is accompanied by dynamical instabilities and secular changes . following the vertical buckling instability , the bars are known to weaken dramatically and develop a pronounced boxy / peanut shape when observed edge - on . using high - resolution-body simulations of stellar disks embedded in live axisymmetric dark matter halos , we have investigated the long - term changes in the bar morphology , specifically the evolution of the bar size , its vertical structure and exchange of angular momentum . we find that following the initial buckling , the bar resumes its growth from deep inside the corotation radius and follows the ultra - harmonic resonance thereafter . we also find that this secular bar growth triggers a spectacular _ secondary _ vertical buckling instability which leads to the appearance of characteristic boxy / peanut / x - shaped bulges . the secular bar growth is crucial for the recurrent buckling to develop . furthermore , the secondary buckling is milder , persists over a substantial period of time , gyr , and can have observational counterparts . overall , the stellar bars show recurrent behavior in their properties and evolve by increasing their linear and vertical extents , both dynamically and secularly . we also demonstrate explicitly that the prolonged growth of the bar is mediated by continuous angular momentum transfer from the disk to the surrounding halo , and that this angular momentum redistribution is resonant in nature a large number of lower resonances trap disk and halo particles and this trapping is robust , in a broad agreement with the earlier results in the literature .

astro-ph0507219

observations and numerical modeling of galactic stellar bars have been frequently accompanied with basic controversies about their origin and evolution . while modern understanding of bar growth in a live and responsive environment is rooted in the angular momentum redistribution between the inner and outer disks , bulges and dark matter halos ( athanassoula 2003 ) , the efficiency of this process is hardly known and its details are still to be investigated . recent efforts include but are not limited to the issues related to the bar lifetime cycles , gas - star interactions , bar amplitudes and sizes , and bar slowdown ( e.g. , bournaud & combes 2002 ; valenzuela & klypin 2003 ; shen & sellwood 2004 ; weinberg 1985 ; hernquist & weinberg 1992 ; debattista & sellwood 1998 , 2000 ) . the bigger issue of course is how the observational and theoretical aspects of bar evolution fit within the emerging understanding of cosmological galaxy evolution ( e.g. , jogee et al . 2004 ; elmegreen et al . 2004 ) , specifically the bar evolution in triaxial halos ( el - zant & shlosman 2002 ; berentzen , shlosman & jogee 2005 ) . on one hand , early self - consistent models of numerical stellar bars have relied heavily on the ostriker & peebles ( 1973 ) result which emphasized the maclauren sequence parameter @xmath2 the ratio of bulk kinetic - to - gravitational energy as the threshold of bar instability . on the other hand , they revealed robustness of the bars once formed , the bars persisted ( athanassoula 1984 and refs . therein ) . the subsequent increase in the particle number above @xmath3 , switching from 2-d to 3-d models with responsive spheroidal components , and introduction of nonlinear physics tools in the orbital analysis have shown a much more complex bar evolution and morphology than was anticipated originally ( e.g. , review by athanassoula 2002a ) . this refers especially to the numerical confirmation that live halos can indeed drive the bar instability rather than damp it ( athanassoula & misiriotis 2002 ; athanassoula 2003 ) . lastly , it is still unclear to what extent and how closely the numerical bars correspond to their observed counterparts . this issue exacerbates galaxy studies because overall , both theoretically and observationally , the bars appear to be among the most important drivers of galactic evolution across a wide range of spatial scales . in this paper , we have revisited some aspects of a self - consistent evolution of stellar bars originating in _ live _ stellar disks embedded in _ live _ dark matter halos by focusing on dynamical and secular changes disk rotation , and to a secular evolution when they develop on a much longer timescale of @xmath4 rotations . ] in these systems . the evolution of _ numerical _ collisionless bars has been characterized so far in the literature by three distinct phases the initial growth , the rapid vertical buckling and the prolonged quasi - steady regime , i.e. , when the bars preserve their basic parameters ( e.g. , sellwood & wilkinson 1993 ) . however , some indication that bars can grow even in the last phase has been noticed already in low - resolution 3-d models with live halos ( e.g. , sellwood 1980 ) . this ability of the bars to grow over extended period of time due to the momentum exchange with the outer disk and especially with the halo has been confirmed recently and analyzed in self - consistent 3-d simulations ( athanassoula 2003 ) . here we attempt to quantify this secular growth in terms of the bar size and its ellipticity , of the angular momentum exchange , and of the ratio of vertical - to - radial dispersion velocities . moreover , we look into the corollaries of such recurrent growth and find that it leads to additional and substantial 2-d and 3-d structural changes in the bar . we , therefore , discuss the observational consequences of this evolution . early in their growth stage , numerical stellar bars experience a dynamical instability the vertical buckling . the bars thicken profoundly , become more centrally - concentrated and acquire a characteristic peanut / boxy shape when seen edge - on ( combes et al . 1990 ; pfenniger & friedli 1991 ; raha et al . 1991 ; berentzen et al . 1998 ; patsis , skokos & athanassoula 2002a ) , while nearly dissolving the outer half of the bar , beyond the vertical inner lindblad resonance ( martinez - valpuesta & shlosman 2004 ) . this happens in live models with both axisymmetric and mildly triaxial halos ( berentzen , shlosman & jogee 2005 ) . these boxy / peanut shapes are similar to bulge shapes observed in edge - on galaxies ( e.g. , jarvis 1986 ; shaw 1987 ; bureau & freeman 1999 ; merrifield & kuijken 1999 ) , which can be found in nearly half of all edge - on disk galaxies ( ltticke , dettmar & pohlen 2000 ) . although observed in numerical simulations a long time ago ( combes & sanders 1981 ) , the origin of boxy / peanut bulge shapes still has two alternative explanations the well - known firehose instability ( e.g. , toomre 1966 ; raha et al . 1991 ; merritt & sellwood 1994 ) and the resonance heating ( e.g. , combes et al . 1990 ; pfenniger & friedli 1991 ; patsis et al . these two views can be reconciled if buckling is responsible for shortening the secular timescale of particle diffusion out of the disk plane and for accelerating the buildup of boxy / peanut bulges which proceeds on a much shorter dynamical timescale instead ( martinez - valpuesta & shlosman 2004 ) . however , is the buckling really necessary for a buildup of these boxy / peanut shaped bulges ? after all , even _ imposing _ vertical symmetry did not eliminate this effect , albeit the buildup proceeded on a much longer timescale ( friedli & pfenniger 1990 ) . where are the observational counterparts of these asymmetric buckled bars ? due to a particular importance of the buckling instability for the evolution of numerical bars and its plausible connection to the buildup of the pronounced 3-d structure there , we have analyzed the bar behavior during and following this instability . specifically , we find that the bars in a live environment are capable of recurrent growth , that the buckling instability is a recurrent event and that the buildup of the 3-d shape is not necessarily a dynamic phenomenon . much of the analysis of bar evolution and the accompaning instabilities is implemented here by means of nonlinear orbit analysis , because for such strong departures from axial symmetry the epicyclic approximation can not be relied upon being wrong quantitatively it frequently leads to qualitative errors . we shall try to avoid the specific jargon associated with this technique where possible . the angular momentum redistribution in the model is quantified using orbital spectral analysis . in section 2 we provide the details of numerical modeling and analysis . section 3 describes the overall results of secular bar evolution . the 3-d bar orbital structure and the inferred vertical structure in the bar are analyzed in section 4 , and the resonant interaction between the disk and the halo in section 5 . discussion and conclusions are given in sections 6 and 7 . to simulate the stellar disk embedded in a live dark matter halo , we have used version ftm-4.4 of the @xmath0-body code ( heller & shlosman 1994 ; heller 1995 ) with @xmath5 . the gravitational forces are computed using dehnen s ( 2002 ) falcon force solver , a tree code with mutual cell - cell interactions and complexity o(n ) . it conserves momentum exactly and is about 10 times faster than an optimally coded barnes & hut ( 1986 ) tree code . the initial density distribution is derived from the fall & efstathiou ( 1980 ) disk - halo analytical model . the system is not in exact virial equilibrium , and therefore must be relaxed iteratively . the halo - to - disk mass ratio within 10 kpc is fixed to unity . the halo has a flat density core of 2 kpc to avoid excessive stochastic behavior associated with the central cusps ( el - zant & shlosman 2002 ) . the disk is exponential and its radial and vertical scalelengths are taken as 2.85 kpc and 0.5 kpc , respectively . the disk and halo cut off radii are 25 kpc and 30 kpc and the initial circular velocity curves for disk and halo components and their sum are given in fig . 1 . the gravitational softening used is 160 pc and toomre s parameter @xmath6 . the adopted units are those of @xmath7 , mass @xmath8 and distance @xmath9 kpc . this leads to the time unit of @xmath10 yrs and a velocity unit of 208 @xmath11 . the energy and angular momentum in the system are conserved to within approximately 1% and 0.05% accuracy , respectively . above @xmath12 our results appear to be reasonably independent of @xmath0 . the model evolution presented here has @xmath13 , with @xmath14 particles in the disk . to1.9 in ' '' '' the self - consistent evolution of stellar bars can be only understood by studying their 3-d structure ( e.g. , pfenniger & friedli 1991 ; skokos , patsis & athanassoula 2002a , b ) . for such in - depth investigation we use the updated algorithm described in heller & shlosman ( 1996 ) , which is based on a comprehensive search for periodic orbits in arbitrary gravitational potentials , and display them in characteristic diagrams ( section 4 ) . these orbits close in the bar frame and provide the backbone for any meaningful analysis of the bar properties . periodic orbits characterize the overall orbital structure of the bar phase space , because each of them traps a region of phase space around it . these trapped orbits have shapes similar to the shapes of the parent periodic orbit . the algorithm is run using the package parallel virtual machine ( pvm ) , which distributes the search for orbits among different processors and computes the orbits using adaptive step size , with a relative accuracy of @xmath15 . we also calculate the stability of these periodic orbits to estimate which orbits can be populated . we track both the 2-d orbital families and the 3-d families bifurcating from planar ( i.e. , equatorial ) orbits at the ( vertically ) unstable gaps . the shapes of these 3-d orbital families will contribute to the shape of the simulated bar , when populated , including the evolving shape of the growing boxy / peanut bulge . to quantify the disk halo interaction and the angular momentum redsitribution in the system , we have developed a package based on the orbital analysis algorithm and which uses the fast fourier transform ( fft ) to find the main orbital frequencies the angular frequency @xmath16 , and the radial and vertical epicyclic frequencies @xmath17 and @xmath18 in the disk and the halo . the particle distribution with the frequency ratio is then determined to find the population of resonant orbits . lastly the change in the angular momentum is computed for each of these particles ( see section 5 ) . we first describe the overall bar evolution during the simulation period of @xmath19 gyr and analyze some of its more important behaviors . the bar develops in otherwise axisymmetric model during the first two - three disk rotations . initially it has an axial ratio ( flatness ) @xmath20 , the same thickness as the disk , but this ratio increases with time by a factor of @xmath21 . the bar starts to brake against the outer disk and the halo as seen in fig . 2b , reaching maximum strength at @xmath22 gyr . at around @xmath23 gyr the bar experiences a vertical buckling instability which affects its 2-d and 3-d appearance . the outer part of the bar nearly dissolves , while overall the bar is weakened dramatically , as shown by the @xmath24 amplitudes , @xmath25 , in fig . 2a . immediately following this buckling , the bar resumes its growth which saturates again at @xmath26 gyr . at this time @xmath25 in the outer part weakens again . the growth is resumed after @xmath27 gyr . a close inspection of the edge - on bar frames during time intervals of @xmath28 gyr and @xmath29 gyr and the analysis described in the next sections , reveal sufficient similarities between these two events both represent the vertical buckling instability in the bar , i.e. , the breaking of symmetry of the bar ( fig . 3 ) . such a recurrent buckling of bars has never been reported in the literature . to3.3 in ' '' '' the stellar bar evolution presented here is characterized by substantial changes in the bar size and strength , and by changes in its 3-d shape ( fig . 3 and animation sequence 1 ) . determination of the live bar size in numerical simulations is not trivial . for example , athanassoula & misiriotis ( 2002 ) and oneil & dubinski ( 2003 ) used a variety of methods and found that some of them give erroneous and unreliable results . the position of @xmath30 maximum does not provide any meaningful estimate for the bar size because the contribution of higher harmonics , such as @xmath31 and 8 , is neglected . to quantify the bar size changes we have used two alternative methods an ellipse fitting to the isodensity curves and the characteristic orbital diagrams . the former method has been previously used to detect and to characterize ` observational ' bars ( e.g. , knapen et al . 2000 ; laine et al . 2002 ; hunt & malkan 2004 ) . its main deficiency when applied to numerical bars is an excessive noise for low - to - moderate @xmath0 and the resulting distribution of the bar ellipticities , @xmath32 , being flat for young unbuckled bars ( martinez - valpuesta & shlosman 2004 ) . however , at later times , with the growth of the central mass concentration , this method becomes more reliable . we find that a consistently reliable estimate of the bar size , @xmath33 , is the radius where @xmath32 declines @xmath34 from its maximal value . the alternative and a _ new _ method used by us here relates the bar extent to the size of the maximal stable orbit of the main orbital family supporting the bar ( more in section 4 ) . both methods produce consistent results with each other . to6.8 in ' '' '' the bar size evolution is shown in fig . 4a , for both methods . it grows initially to 11 kpc , then buckles and shortens to 6 kpc . it then grows again to 13 kpc , where the growth stagnates due to the secondary buckling , for about 3 gyr . after this the bar resumes its growth to about 16 kpc . the size evolution in the @xmath35 plane is accompanied by the vertical thickening of the bar . it does not stop after the first buckling instability , but continues gradually due to the vertical resonance scattering , amplified by the recurrent buckling instability of the bar . the accompanied boxy / peanut shaped bulge also grows with time ( fig . to4.6 in ' '' '' with the bar length , @xmath33 , and its corotation radius , @xmath36 , we can quantify some of the dynamical characteristics of an evolving bar . the ratio @xmath37 is shown in fig . this ratio determines the shape of the offset dust lanes in barred galaxies which delineate shocks in the gas flow ( athanassoula 1992 ) . the observed shapes constrain the ratio @xmath37 to @xmath38 . the modeled ratio typically falls within the required limits except during the first buckling when it is higher , @xmath39 , as noticed already by martinez - valpuesta & shlosman ( 2004 ) . we shall discuss this issue in more detail in section 6 . the buckling is a 3-d phenomena and is most visible in the vertical @xmath40-plane ( fig . 3 ) . to quantify the bar asymmetry , we have calculated the vertical @xmath41 mode amplitude @xmath42 in the @xmath40 plane and follow the maximal distortion of the bar during its buckling periods . two maxima are apparent at @xmath43 gyr and @xmath44 gyr in fig . note , that the vertical asymmetry given by this figure is building up slower and is weaker for the second buckling , i.e. , from @xmath45 gyr to @xmath46 gyr the amplitude is growing to @xmath47 , compared to @xmath48 at @xmath49 gyr . 3 shows this evolution in a more graphical way while the first buckling affects mostly the central few kpc , especially the bar s midplane , the second buckling is most prominent in the outer bar range of 510 kpc and affects the midplane much less visibly . to3.3 in ' '' '' the recurrent buckling can be detected in a number of ways , e.g. , from @xmath25 in the @xmath35-plane ( fig . 2a ) , as mentioned in section 3.1 and above , and from @xmath42 in the @xmath40 plane , which quantifies the breaking of vertical symmetry in the bar ( fig . 5 ) . each of the coefficients emphasizes a different property of this instability . we are basically looking at the same phenomena at different times the first buckling extends over @xmath50 gyr and the second one over @xmath1 gyr . the changes in the orbital structure of the bar during the bucklings will be analyzed in section 4 . we have commented in section 1 that the buckling instability is a collective breaking of a vertical symmetry in the bar . toomre ( 1966 ) has shown that the coupling between the vertical and radial degrees of motion is the prime driver of this instability and the main outcome of it is the equalizing of the velocity dispersion in the @xmath35 plane with the vertical velocity dispersion .- axis , is an adiabatic invariant ] this evolution is characterized by changes in vertical - to - radial velocity dispersion ratio @xmath51 . toomre estimated the critical value for this ratio to lie at @xmath52 for a non - rotating plane - parallel slab , raha et al . ( 1991 ) at 0.06 - 0.3 for a 3-d stellar disk , and sotnikova & rodionov ( 2005 ) at @xmath53 for the central regions embedded in the hot halos . sellwood ( 1996 ) have shown that toomre s limit is violated for many of his stellar models , some remaining unstable up to 0.4 . we plot this ratio for two areas of the bar at two different times . for the first buckling ( fig . 6 ) , the velocity dispersions are calculated in the central kpc , where the maximum effect is expected ( e.g. , fig . 3 ) . during this buckling , we observe first an increase in @xmath54 ( the initial growth of the bar ) , and after @xmath50 gyr a decrease in @xmath54 with a corresponding increase in @xmath55 . when @xmath51 drops below @xmath56 the bar buckles and weakens . the buckling ends when @xmath51 increases to unity . the bar thickens and grows and the vertical ilr moves gradually out . we expect and observe the maximal vertical asymmetry in the outer part of the bar during the second buckling , and therefore calculate @xmath51 at around 7 kpc . again , the gradual increase in @xmath55 and decrease in @xmath54 drive the ratio up from 0.4 to about unity , similar to the first buckling . for both bucklings we observe a very similar evolution in terms of the velocity dispersions and their ratios . to3.4 in ' '' '' so far we have shown that the bar buckles twice during the simulations : first time abruptly , fast and in the central part , and the second time slower , less pronounced and in a different part of the bar . in both bucklings we observe the loss of symmetry in the vertical plane , the drop in @xmath25 and the equilizing of the vertical and radial velocity dispersions . after each buckling the bar becomes more symmetric , with the difference that after the first buckling the asymmetry starts to build anew , and after the second buckling the asymmetry is completely washed out . to summarize , three main factors characterize the end of the buckling instability : @xmath25 starts to decrease , the asymmetry in the @xmath57-plane decreases ( given by @xmath42 ) , and @xmath58 . applying these conditions to our model , the end of the first buckling appears at @xmath59 gyr and the end of the second buckling at @xmath60 gyr . since the buckling leads to a sudden vertical thickening of the bar and is therefore characterized by particle injection above the disk plane , we can monitor this instability by following the particle distribution . when viewed along the bar minor axis , during the first buckling the bar bends and develops a boxy / peanut shaped bulge , while during the second buckling the bar acquires an asymmetric shape which leads to the appearance of an x - shaped bulge ( fig . these shapes have a direct relationship to the population of orbits trapped by the main family of periodic orbits in the bar ( pfenniger & friedli 1991 ; section 4 below ) . stellar bars are 3-d objects which exhibit dynamical and secular evolutionary trends both in their morphology and internal structure . in this section , we study the evolution of the orbital structure in the bar by searching for the _ main _ 2-d and 3-d families of orbits at various snapshots , with a particular emphasis on the recurrent buckling periods . a comprehensive search for the orbits is beyond the scope of this work . we analyze the main parameters of detected orbits and calculate their stability . this section uses a specific terminology developed for nonlinear orbital dynamics ( e.g. , binney & tremaine 1987 ; sellwood & wilkinson 1993 ) . reader unfamiliar with this terminology can skip it and go directly to section 5 . we shall discuss the results of the orbital analysis in the context of the bar evolution in section 6 . the initial vertical axial ratio of the bar which develops in our models is @xmath61 ( nearly a 2-d object ) , and grows dynamically and secularly to @xmath62 over the hubble time . we first analyze the bar s orbital structure in the equatorial plane , then add the 3-d effects accounting for the finite thickness and shape of the bar . in doing so , we limit our analysis to the periodic orbits within @xmath36 , which are largely responsible for the bar shape , and to specific times of bar evolution : @xmath63 gyr ( just prior to first buckling ) , 2.8 gyr ( after the first buckling ) , 7.1 gyr ( during the second buckling ) and 11.8 gyr ( after the second buckling ) . the orbits are searched in the potential symmetrized horizontally with respect to the four quadrants . 5 confirms that the vertical symmetry is violated during the buckling periods . the first buckling is fast and there is little meaning , therefore , to calculate the orbits without some frame averaging the stars see only a time - averaged potential . hence , we resort to a vertically - symmetrised potential at 2.8 gyr . on the other hand , the second buckling is much more gradual and we calculate the orbits in the actual ` raw ' potential at @xmath64 gyr to capture the persisting asymmetry . the third snapshot with the bar being symmetric again is treated similarly to the first snapshot for simplicity . the extent of the orbital families and their stability are displayed by means of characteristic diagrams . the @xmath65 , @xmath66 , or @xmath67 intercept values with the @xmath68 plane are plotted with respect to their jacobi integral , @xmath69 ( e.g. , binney & tremaine 1987 ) . the jacobi ( energy ) integral of motion is conserved along any given orbit in the rotating bar frame . the orbits form curves or families in the characteristic diagrams . the actual trajectories will not coincide exactly with these periodic orbits but may be ` trapped ' in their vicinity . the properties of periodic orbits and their temporal changes , therefore , reflect the bar structural evolution . to4.2 in ' '' '' the most important single - periodic orbits in the bar midplane are those aligned parallel or normal to the bar major axis , the so - called @xmath70 and @xmath71 orbits ( e.g. , contopoulos & papayannopoulos 1980 ; binney & tremaine 1987 ) . the @xmath70 constitutes the main family of orbits which support the bar figure . while this family exists always within @xmath36 , the @xmath71 appear only if the planar inner lindblad resonance(s ) ( ilrs ) are present . typically , numerical bars form with a pattern speed sufficiently high to avoid the ilrs , at least in the beginning , and hence the @xmath70 family completely dominates the bar midplane , short of the corotation region . the vertical shape of the bar is determined by the projection of the populated 3-d orbits onto the corresponding planes . we search for the vertically - unstable gaps in the @xmath70 family . it is at these gaps where the 3-d families bifurcate through @xmath66 and @xmath72 bifurcations ( e.g. , pfenniger 1984 ; pfenniger & friedli 1991 ; skokos et al . 2002a , b ; patsis et al . these gaps coincide with the vertical ilr and other vertical resonances . when an orbital family goes from being stable to unstable , in the mentioned gaps , we get the 3-d prograde families , i.e. , orbits with initial conditions @xmath73 , with @xmath74 . when it goes from being unstable to stable , we get the retrograde families , @xmath75 , with @xmath76 , where @xmath77 , @xmath65 and @xmath66 are oriented along bar s major , minor and vertical axes . while the vertical ilr is associated with the boxy / peanut bulge shapes , which develop as a result of the resonance heating of midplane orbits , patsis et al . ( 2002b ) have found that this effect is not limited to a particular resonance and not to the barred galaxies _ per se _ , but operates equally well in nearly axisymmetric and/or ovally distorted disks . furthermore , the buckling itself is not a necessary condition for these shapes to form , but rather accelerates the process from secular to a dynamical timescale . the main families of 3-d orbits which are responsible for the appearance of the boxy / peanut shaped bulges are ban ( prograde to the bar rotation ) and aban ( retrograde ) families found by pfenniger & friedli ( 1991 ) . they are called @xmath78 and @xmath79 , respectively , and appear as the 3-d generalizations of @xmath70 and @xmath71 families in the nomenclature of skokos et al . ( 2002a ) . in general terms , the ban orbits can be described as @xmath80 " ( @xmath81 ) and @xmath82 " ( @xmath83 ) , and aban as @xmath84 , " when projected onto the @xmath40-plane . both families are @xmath85 ( i.e. , two radial oscillations for two vertical oscillations for one turn ) in the notation of sellwood & wilkinson ( 1993 ) . their projections onto the @xmath35 plane have the @xmath70 orbit shapes . we trace the planar @xmath70 and @xmath71 and the 3-d ban / aban families from the bar initial growth period . for simplicity , we have divided the simulation into a number of characteristic time intervals and discuss them separately . during this period the bar develops and buckles ( figs . it is geometrically thin and is not centrally concentrated , so the vertical ilr is not present . neither present are the planar ilrs and consequently the @xmath71 orbits . the stable 3-d orbits lie close to the midplane . as we stated above , there is little meaning to calculate the orbits in the rapidly varying potential without the proper time - averaging . because the midplane of the bar is bent in the @xmath40 plane ( fig . 3 ) , so are the @xmath70 orbits . to4.3 in ' '' '' the bar has increased its central mass concentration and acquired its boxy / peanut shaped bulge ( see fig . 3 ) . as we stated above , there is little meaning to calculate the orbits in the vertically unsymmetrized potential at the time of the first buckling because of the rapidly varying potential . moreover , the bar midplane is bent in @xmath40 at this time , and so are the @xmath70 orbits ( fig . 3 ) . hence , we have vertically symmetrized the potential at this time frame ( fig . 7b , left frame ) . in the midplane , the @xmath70 orbits dominate and no @xmath71 family exists ( fig . most of the @xmath70 orbits are stable , except for a small gap , where the vertical ilr is located . at @xmath86 gyr the gap is at @xmath87 , where the ban / aban orbits bifurcate ( fig . the ban / aban are traced ( fig . 7b , c ) their origin in @xmath88 plane moves gradually out with time toward higher jacobi energy . by the end of the first buckling it stabilizes and subsequently creeps toward higher @xmath69 as the bar brakes against the halo . ban orbits appear stable from 300 pc above the midplane , with a broad unstable gap at @xmath89 kpc and a narrow stability island ( fig . initially , the stable part of these families is limited to a very small height above the bar midplane , compared to the bar vertical thickness . gradually the stable extent of the ban family is increasing both in @xmath66 and in @xmath77 ( fig . the aban family is unstable when ban is stable and vice versa ( fig . 7c ) , as first noted by pfenniger & friedli ( 1991 ) . the bar is growing both in size and amplitude and exhibits a pronounced boxy profile when viewed along its minor axis . it appears symmetric with respect to the bar midplane up to @xmath45 gyr ( fig . 3 ) , although @xmath42 shows that some small residual asymmetry remains between the bucklings ( fig . after @xmath45 gyr , the vertical symmetry of the bar is broken for an extended period of time @xmath1 gyr ( figs . 3 , 5 ) . the maximal distortion happens at @xmath90 gyr . no @xmath71 orbits exist during this time period and the unstable gap in @xmath70 moved to somewhat higher energies , @xmath91 , due to the bar braking ( fig . additional unstable gap has appeared at @xmath92 , where the 3:2:1 orbit family bifurcates . note that the analysis shown in the @xmath64 gyr frame of fig . 7 is performed in the actual gravitational potential of the @xmath0-body simulation , i.e. , without symmetrization with respect to the @xmath88 plane . therefore , we should not expect to find all the families existing in the symmetrized potentials . but we should be able at least to identify the reason for the vertical asymmetry of the particle distribution in the projection of the 3d - orbits onto the @xmath40 plane from the orbital shapes ( fig . we find the ban family and also the @xmath93 family in the nomenclature of patsis et al . ( 2002a ) . the ban family appears stable everywhere above the midplane , except around @xmath94 kpc ( fig . 7b ) , while aban is unstable everywhere and so is not shown here . for @xmath95 , the ban family is detached from the midplane and starts at @xmath96 kpc . it forms a kind of loop in the characteristic diagram . we do not find any stable orbit below @xmath97 kpc . the @xmath93 family is stable only for a low range in energies and only below the midplane , which makes it important for contributing to the vertical asymmetry in the bar ( fig . the bar slowly regains its vertical symmetry , and becomes symmetric after @xmath60 gyr . as a result of this instability , the bar has developed a pronounced x - shape which differs from the ` usual ' peanut shape because of the extended concave region between the spikes , i.e. , peanuts ( fig . orbits which appear at the same @xmath66 in the characteristic diagram but at different time frames differ in their projections . after the second buckling the orbits are longer and extend higher , i.e. , the convexity and concavity have significantly changed . the vertical ilr has moved to @xmath98 by 11.8 gyr . the ban orbits are stable while aban are unstable . overall , the boxy / peanut bulge has increased in size from @xmath99 kpc ( after the first buckling ) to @xmath100 kpc ( after the second buckling ) . the @xmath71 orbits and consequently two ilrs in the @xmath35-plane appear only after the second buckling , @xmath101 gyr . initially occupying a small range in energies , their extent grows with time . the @xmath70 family is mostly stable up to the ( mostly ) unstable ` shoulder ' in the characteristic diagram ( fig . 7a , right frame ) , which has continued to move out to higher jacobi energies , along with the last stable orbit supporting the bar . we also use the characteristic diagrams to get an independent estimate for the bar physical size . unlike the case of analytical bars where this issue is resolved trivially , to estimate the length of the _ live _ numerical and ` observed ' bars can be more difficult . for example , if the bar potential or bar nonaxisymmetric force is used for this purpose , one can get an erroneous result that the bar extends beyond its corotation radius . if one relies on the density distribution , one can overestimate the size as well because the bar can drive a pair of open spirals . instead , we use the calculated orbital structure of the bar and rely on the properties of @xmath70 family which is generally stable except for narrow gaps ( in jacobi energy ) and in the region around the corotation . specifically , close to the corotation , the @xmath70 curve bends upward ( e.g. , fig . 7a ; binney & tremaine 1987 ) . we use the @xmath77-axis extent of the last stable @xmath70 orbit which lies on this upward branch . it is the slow drift of unstable gaps and of the elbow of the @xmath70-curve towards higher @xmath69 which leads to the trapping of orbits and the secular increase in the bar size . a large number of models here and in berentzen et al . ( 2005 ) have been used for the comparison with the isodensities fit to the bar ( section 3 ) . the bar amplitude @xmath25 shows that the outer half of the bar practically dissolves in the first buckling and weakens substantially in the recurrent buckling ( fig . 2 ) . the ellipse fit to the bar reflects this by a dramatic decrease in the bar length from @xmath102 kpc to @xmath103 kpc in radius . unfortunately , it is more tricky to estimate the bar size from the orbital analysis at this time the underlying potential is time - dependent while the analysis is performed in the frozen potential . nevertheless , both methods agree before and after the the first buckling and capture the change in the growth rate of the bar length after the second buckling . to summarize , the bar size decreases during the first buckling and levels - off during the second one . the simple explanation for the success of the orbital analysis method is that it is the most self - consistent method known to us and relies on the orbit trapping by the stable periodic family whose existence basically defines the bar . we note that the middle frame of fig . 7a does show a large unstable gap just below the elbow of @xmath70 curve . however , we note once more that the potential in this frame has not been vertically symmetrized and that the weakening of the bar during this time of buckling is in fact a direct consequence of this gap . in summary , the characteristic diagrams and their respective orbital families reflect both the dynamical and secular evolution of the bar shown in both fig . 3 and the animation sequence . the ustable gaps , where the vertical families ( ban / aban ) bifurcate , move toward higher jacobi energies while the orbits become more extended along the bar axes . the orbital families evolve not just quantitatively but also qualitatively this shows up in the changing shape of the bar . although we do not measure the population of the different families , the bar shapes appear to be governed mainly by the ban family , which evolves by changing the concavity of its orbits , thus giving the bar the pronounced peanut / boxy / x shapes when viewed along the minor axis . the change in the orbital structure of the bar supports our observation , based on independent arguments , that there is a recurrent buckling . at the secondary buckling time , the vertical families differ above and below the plane . the former are stable orbits extending more along the @xmath77-axis and also giving it more of a boxy shape . the latter ones are stable orbits which give the bar more of a peanut shape . the appearance of the @xmath71 family confirms that the ilrs form only after the second buckling of the bar . the disk region which gives rise to a bar is known to lose its angular momentum ( @xmath104 ) . in principle , a number of components in a galaxy can acquire this momentum , namely , the outer disk , the bulge , and the halo . while the former two components are able to store @xmath104 , the responsive ( i.e. , live ) halo can serve as a particularly large angular momentum sink due to its large mass and low @xmath104 . athanassoula ( 2002b , 2003 ) has shown that the disk - halo interaction is mediated by lower resonances . in order to understand more fully the complex behavior exhibited by the model , such as the initial bar growth , its recurrent bucklings , and subsequent secular growth we find it instructive to determine the balance of the angular momentum and its evolution . we pay particular attention to the resonant interactions between the model components . to5.0 in ' '' '' to5.0 in ' '' '' we first determine the overall @xmath104 balance in the different regions of the disk and the halo . 9 shows the total angular momentum evolution for three different regions in the disk and halo : inner ( @xmath105 kpc ) , intermediate ( @xmath106 kpc ) and outer ( @xmath107 kpc or 15 - 50 kpc ) regions . the inner and intermediate regions are those hosting the bar for various times during its evolution . the time of the bar formation is characterized by a substantial loss of @xmath104 in the inner disk . this angular momentum is redistributed to the intermediate and especially to the outer disk ( fig . 8) . during the first buckling , there is some indication that the inner disk gains some angular momentum , and we return to this issue in section 6.2 . the subsequent growth of the bar is accompanied by a slow @xmath104 growth in the outer disk and the halo . interestingly , after the second buckling @xmath104 stops growing in the outer disk , while the outer halo picks - up @xmath104 at an increasing rate . there is a general loss of angular momentum from the disk and there is a general gain in the halo . the loss of @xmath104 in the disk correlates with the slowdown of the bar , which is known to anti - correlate with the growth of the bar ( e.g. , athanassoula 2003 ) . next , we calculate the contribution of the resonances between disk and halo particles with the bar s @xmath108 to the angular momentum exchange between the disk and the halo . we generally follow the procedure described in athanassoula ( 2002b ) . the principal frequences , @xmath16 and @xmath17 of azimuthal motion and of radial oscillations of the 3-d orbits , have been determined by means of spectral analysis ( binney & spergel 1982 ) . we freeze the gravitational potential in the model at times @xmath86 gyr and 5.2 gyr , but allow the bar to tumble with its original @xmath108 , and integrate the orbits of @xmath109 randomly - picked particles in the disk and the same amount in the halo for an additional 9.4 gyr , about 20 bar tumblings . results are plotted in fig . 10 as a function of the frequency ratio @xmath110 , where @xmath111 particles are binned in the intervals of @xmath112 ( top frames ) . the difference @xmath113 in angular momentum between the two times is given as a function of @xmath114 at @xmath115 gyr ( bottom frames ) . the major resonances are located at @xmath116 ( ilr and olr ) , @xmath117 etc . , with the positive @xmath114 indicating the resonances inside the corotation radius and @xmath118 outside the corotation . the corotation resonance corresponds to @xmath119 . the resonant and near - resonant particle distributions @xmath111 at the time @xmath86 gyr are clearly non - uniform and are permeated by numerous resonances . the trapping of orbits by the resonances is also evident . the dominant resonance in the disk appears to be the ilr which traps the largest number of particles and therefore facilitates the loss of the angular momentum from the disk ( lower panels of fig . note , that we have cut @xmath111 for this resonance to keep the same scales for disk and halo diagrams it extends to @xmath120 . the @xmath70 within the cr and later on @xmath71 ( if populated ) orbits , as well as ban / aban orbits are trapped within this resonance . the broad asymmetric peak ( fig . 10a , left panel ) in the range of @xmath121 is made out of near - resonant orbits in the ( rotating ) boxy / peanut bulge , many of them near resonant ban orbits . those have been injected during the first buckling . they appear on both sides of the cr because @xmath104 along these orbits oscillates wildly , @xmath122 , and it is sometimes difficult to disentangle the prograde from retrograde ones among them . for comparison , we repeated this procedure at @xmath123 gyr and confirmed that more than half of the particles in the above broad peaks have been trapped by the nearby ilr and olr by that time . the resonances for @xmath124 , with cr and olr as the next strongest ones , absorb the angular momentum and their @xmath113 are all positive . the halo is dominated by the corotation resonance which absorbs @xmath104 from the disk . overall , the disk resonances emit and the halo resonances absorb @xmath104 . this behavior was demonstrated by athanassoula ( 2002b , 2003 ) , where similar plots to fig . 10 were given and there is a very good agreement between our results . the only difference we find is that the halo ilr is actually losing @xmath104 in our model , albeit a small amount , unlike in athanassoula s models . the modeled halo extends well beyond the disk and its outer part appears to be actively storing @xmath104 , up to @xmath125 . this effect is confirmed by the overall @xmath104 evolution in the halo ( fig . we find that the trapping of orbits by the resonances is robust almost all the particles trapped at @xmath86 gyr remain trapped at 5.2 gyr . the ability of the barred disk to transfer its angular momentum to the halo while @xmath104 saturates in the outer disk , explains why the bar is able to grow during this time interval as exhibited by the @xmath25 amplitude and the size of the bar ( e.g. , figs . it is this growth of the bar which is ultimately responsible for the recurrent buckling phenomenon analyzed here . we have modeled the evolution of collisionless ( stellar ) bars embedded in responsive ( live ) axisymmetric dark matter halos using high - resolution @xmath0-body simulations . we find that ( 1 ) bars experience secular growth over the simulation ( @xmath126 hubble ) time , except ( 2 ) during well defined time periods when the bars encounter spontaneous breaks of vertical symmetry , so - called buckling ( or firehose ) instability ( e.g. , fig . 3 and the animation sequence 1 ) . we detect such a recurrent buckling during which the bar , and especially its outer half , weakens substantially , but the growth is resumed subsequently . two different techniques have been used to measure the bar size , namely , the nonlinear orbital analysis and the isodensity ellipse fitting . the bar strength has been measured using the amplitude of @xmath24 mode , and the bar vertical asymmetry using the vertical @xmath41 mode . moreover , we have analyzed the bar evolution and its buckling periods by means of the nonlinear orbit analysis and the ratio of vertical - to - planar velocity dispersions . finally , we have examined the bar development in terms of the angular momentum redistribution between various components in the disk - halo system mediated by the resonant interactions using orbital spectral analysis . low resolution @xmath0-body simulations exhibit fast ( e.g. , during one disk rotation ) stellar bar growth , followed by a vertical buckling and secular weakening . high - resolution simulations with @xmath127 show a more complex evolution . they allow for modeling the disks embedded in live halos and account for the resonant interactions between the halo and the disk , adding a new and crucial element to the simulations in the form of angular momentum transfer between these components . the main difference with the previous low - resolution models and that presented in this work is the ability of the stellar bar to strengthen again after its original weakening following the buckling a process which leads to the bar growth and consequently to its recurrent buckling . the bar growth has been associated with the existence of ` sinks ' of angular momentum located elsewhere . athanassoula ( 2002b , 2003 ) has shown that a galactic halo can play such a role and absorb large quantities of angular momentum from the disk / bar region . this effect appears to completely invert the original suggestion by ostriker & peebles ( 1973 ) about the stabilizing function of dark matter halos in disk galaxies against the bar formation instability . angular momentum redistribution , at least in principle , can have contributions both from the non - resonant and resonant interactions between the bar and orbits in the halo . to capture the latter requires large @xmath0 to stabilize the population of resonant particles ( e.g. , weinberg & katz 2002 ) . it is therefore important to demonstrate that resonant particles are indeed responsible for the angular momentum transfer in the first place . we start with the relative contributions to the @xmath104 transfer between the disk and the halo and between the inner ( i.e. , bar unstable @xmath128 kpc ) and outer disks ( fig . although the efficiency of this redistribution is of course model dependent , we nevertheless quote the numbers , assuming that to a certain degree , they are representative . we find that @xmath129 of the angular momentum in the disk is lost to the halo during the evolution . the bar unstable region , @xmath130 kpc has lost about 71% of its original @xmath104 , of which about 38% went to the outer disk and the rest was absorbed by the halo . the most intensive flow of @xmath104 to the outer disk happens in the early stage which ends with the first buckling . the halo particles have much larger dispersion velocities than the disk and until the bar fully develops , their resonant interaction with the @xmath24 asymmetry in the disk is virtually non - existent . after the first buckling , it is the halo which absorbs most of the angular momentum lost by the bar region and from fig . 9 ( lower frames ) it is clear that the lion share of this exchange is resonant and mediated by the cr and the olr in the halo and by the ilr and the cr in the disk , with some contribution from the numerous minor resonances . this results in the increase of the bar size and strength , and leads to the gradual decrease in the ratio of the velocity dispersion , @xmath51 , in a close analogy with the disk evolution preceeding the first buckling . the secondary buckling of the bar , therefore , can be directly traced to its robust growth in our model . the issue of a bar dissolution vs. growth has become controversial recently . while it is beyond the scope of this work , indirectly we do touch it . first , we confirm the results of martinez - valpuesta & shlosman ( 2004 ) that vertical bucklings do not destroy the bar , unlike suggested by raha et al . ( 1991 ; see also sellwood & wilkinson 1993 ) . next , in various axisymmetric models , here and in berentzen et al . ( 2005 ) , a robust growth in the bar is observed after the first buckling , supplemented by the angular momentum exchange between the disk and the halo . athanassoula ( 2003 ) has analyzed the effect of initial conditions on this @xmath104 redistribution , namely , of the initial disk - to - halo mass ratio and of the toomre s q - parameter the angular momentum exchange varied from @xmath131 to @xmath132 with a clear correlation between the amount of the exchanged momentum and the size of the bar . in comparison , our models have a halo - to - disk mass ratio of unity within the central 10 kpc , @xmath133 , and a halo core of @xmath134 kpc . they lie between the models mh1 , m@xmath1353 - 4 and mq4 of athanassoula . interestingly , valenzuela & klypin ( 2003 ) argued that it is the mass and force resolutions of numerical models which dictate the efficiency of @xmath104 transfer to the halo and hence play an important role in the evolution of the bar . insufficient resolution results in an excessive growth of numerical bars and an excessive decrease in the bar pattern speed . however , it seems rather that the physical conditions mentioned above take priority their high resolution models have extended and hot halos . high dispersion velocities in the halo affect the particle trapping by the resonances and so are expected to lower substantially the @xmath104 transfer . if one neglects the dissipative component in a modeled galaxy , the bar pattern speed decreases secularly , except during brief time intervals of internal instabilities in the bar itself . at least in numerical models of stellar bars , their overall slowdown is accompanied by an increase in the bar length , so the bar roughly extends to its corotation radius which increases with time ( athanassoula 1992 ) . alternative theoretical models of bars terminating at the ilr exist ( lynden - bell 1979 ) , but difficulties remain in actually reproducing them in numerical simulations ( but see e.g. , polyachenko & polyacnenko 1994 for the so - called slow bars ) . a sole exception consists of a system of nested bars , where the inner ( nuclear ) bars form within the ilr ( e.g. , shlosman , frank & begelman 1989 ; friedli & martinet 1993 ; englmaier & shlosman 2004 ) . however , a dissipative component is required to be present for self - consistency in this case . how exactly the bar traps the disk orbits in order to increase its length is not known at present . we have demosntrated that while the size of the bar , @xmath33 , drops during the first buckling ( fig . 2a ) , the ratio of @xmath37 ( fig . 1b ) does _ not _ increase dramatically above the range of @xmath38 determined by athanassoula ( 1992 ) to fit the observed shapes of the offset dust lanes in barred galaxies . amazingly , the reason for this is that @xmath36 drops as well due to the sudden increase in the bar pattern speed in the same time interval . this is not unexpected due to a clear trend between @xmath108 and @xmath25 ( athanassoula 2003 ) and a long known fact that the bar lengthens with its slowdown in numerical simulations . what is new here is that we find that @xmath108 increases sharply during the first buckling and this increase apparently correlates with the sudden decrease in the bar length . hence the relation between between @xmath108 , @xmath25 and @xmath33 appears to be more fundamental than anticipated and holds for decelerating and accelerating bars altogether . we associate the continued bar growth with the ability of the halo to absorb angular momentum from the disk region lying within the corotation radius a region which itself expands with time . this process leads to a stronger bar and can be seen as a counterbalance to a number of other processes which have been discussed in the literature within the framework of bar dissolution . recent observational results from galaxy evolution from morphology and seds ( gems , rix et al . 2004 ) survey have shown that the bar size and axial ratio distributions at intermediate redshifts of @xmath136 are compatible with those in the local universe ( jogee et al . 2004 ; elmegreen et al . 2004 ; sheth et al . 2003 ) . our models which exhibit a slow secular bar growth over the hubble time are in general agreement with these results . they do lack the gaseous component which can dramatically shorten the bar life cycle ( e.g. , bournaud & combes 2002 ) . however , it is difficult to understand how the short - lived ( e.g. , less than 1.5 gyr [ bournaud & combes 2002 ; combes 2005 ] ) bars can form at the same rate and with the same sizes and axial ratios they are being destroyed by the gas . yet , taken at the face value , in order to agree with the gems results , the bars must preserve their size and strength distributions nearly unchanged over the last 8 gyr . clearly , the precise role of gas in the bar evolution must still be determined . next we focus on some aspects of the bar 3-d shape evolution in the model , specifically on the axial ratio in the bar midplane and on the bar symmetry in the vertical plane . the midplane axial ratio , or alternatively the bar ellipticity @xmath137 , is measured from the ellipse fitting ( fig . 4c ) and agrees with the evolution of the bar amplitude @xmath25 . at the same time , the variations in @xmath137 appear less dramatic than in @xmath25 , although recurrent bucklings are clearly visible as a decrease or saturation . the ratio @xmath37 has also dynamical implications for the bar . for example , it determines the shape of the offset dust lanes in barred galaxies , which in turn delineate shocks in the gas flow ( athanassoula 1992 ) . the strength of the underlying shocks determines the gas inflow towards the central kpc . a bar which is considerably weakened will slow down the radial gas inflow . the observed shapes constrain the ratio @xmath37 to @xmath38 . the modeled ratio ( fig . 4b ) typically falls within the required limits except during the first buckling when it is higher , @xmath39 . to3.5 in ' '' '' after the second buckling , the bar shows the so - called _ ansae _ ( handles ) on both its ends . we observe them as characteristic density enhancements in the face - on ( fig . 11 ) or edge - on ( fig . 3 ) disks . athanassoula ( 2001 ) related the appearance of _ ansae _ to initial conditions in the models ( e.g. , the halo - to - disk mass ratio ) . it remains unknown why they appear at this particular evolutionary stage , @xmath138 gyr . the _ ansae _ can be seen in some early - type barred galaxies , e.g. , ngc 4262 , ngc 2859 and ngc 2950 ( sandage 1961 ) , ngc 4151 ( mundell & shone 1999 ) , eso 509 - 98 ( buta et al . 1998 ) , etc . vertically , the bar evolves from a geometrically thin configuration , similar to the disk hosting it . the vertical bar buckling , when viewed along the bar s minor axis , shows a rapidly evolving bending which relaxes to a boxy shaped bulge , i.e. , bulge with flat or mildly convex isodensities ( fig . subsequently , the bulge acquires a peanut shape . during the second buckling , for @xmath1 gyr , the vertical asymmetry persists with one - sided boxy and peanut symmetries which derive from the asymmetric ban family ( fig . with the asymmetry washed out , the boxy / peanut bulge / bar has a pronounced x - shape : two pointed spikes with a large concave region in between ( fig . the x - shapes have been seen before in numerical simulations ( e.g. , athanassoula & misiriotis 2002 ) and have been observed as well ( e.g. , ngc 4845 , ngc 1381 , ic 4767 ( whitmore & bell 1988 ) , ic 3370 ( jarvis 1987 ) and am 1025 - 401 ( arp & madore 1987 ; bureau & freeman 1999 ; patsis et al . mihos et al . ( 1995 ) proposed a merging scenario for the formation of the x - shaped bulges , e.g. , a minor merger for hickson 87a galaxy . this merger triggers the bar instability in the disk , followed by the buckling and the x - shaped bulge , when viewed from a specific aspect angle . in principle , there may be more than one cause for the bar buckling and for the formation of peanut / boxy / x - shaped bulges . but when taken together , a high observed frequency of these bulges ( ltticke et al . 2000 ) and the need for only mild asymmetry in the disk for their appearance ( patsis et al . 2002b ) may hint to their intrinsic origin . the vertical asymmetry of the modeled bar has been detected during its recurrent bucklings , for about 1 gyr and @xmath1 gyr , respectively ( figs . 3 , 5 ) . its characteristic shape , in principle , is detectable by observations of edge - on galaxies ( e.g. , fig . 3 and the animation seq . 1 ) , especially during the second buckling due to its prolonged period . observationally , the bucklings differ when viewed edge - on along the bar minor axis . the main difference is the location of the maximal asymmetry it is close to the rotation axis during the first buckling and around the middle region of the bar during the second buckling . ltticke et al . ( 2000 ) present first statistics of edge - on galaxies with the boxy / peanut bulges and show some isophote fits ( their fig . 2 ) . they distinguish between peanut - shaped and boxy - shaped bulges . a total of 27% of the sample of 734 galaxies with boxy / peanut bulges have bulges which are either close to boxy or , due to low resolution , could not be distinguished between boxy and peanuts . some of these bulges show a mild vertical asymmetry similar to that found during the second buckling , e.g. , ngc 4289 a good example of a possible observed secondary buckling . of course , to corroborate the numerical simulations one needs a large , statistically significant sample with high resolution . the main difficulty lies in the unambigious determination of the inclination angle of a galactic disk it must be edge - on within @xmath139 . from a theoretical point of view , it is unclear how wide - spread is the recurrent buckling in stellar bars how sensitive it is to disk and especially halo parameters , such as mass distribution and dispersion velocities . while the condition for a second buckling seem to be directly related to the ability of the bar to strengthen after the initial weakening , this process can depend on a number of additional parameters . for example , athanassoula ( 2003 ) has found that the angular momentum transfer between the disk and the halo depends on the mass distribution in the halo and weakens substantially for hotter halos . berentzen et al . ( 2005 ) have detected a recurrent buckling in their ls2 model with 2 kpc flat core ( live ) halo with a logarithmic potential . at the time of the second buckling , the bar in this model appears somewhat stronger than in the first buckling . moreover , the secondary buckling can be apparently seen in fig . 4 of oneil & dubinski ( 2003 ) and also in fig . 11 of valenzuela & klypin ( 2003 ) , the latter based on the evolution of the bar pattern speed , strength and the angular momentum rate change . secondary buckling is also present in some models of l. athanassoula ( private communication 2005 ) . in all the cases they went unnoticed . lastly , we comment on the addition of the dissipative component to the stellar disk ( berentzen et al . the effect of the clumpy gas component is to weaken the buckling instability , but without quantifying the degree of clumpiness within the central kpc , it is not clear whether the isothermal equation of state used has led in fact to over - damping of this instability . to summarize this work , we have studied the long - term stellar bar evolution in a high - resolution self - consistent model of a disk and a responsive halo . we find that a developing bar goes through the vertical buckling instability which weakens it and dissolves its outer half . subsequently , the bar experiences a renewed growth which leads to a recurrent buckling . this evolution is driven by the resonant interaction between the barred disk and the surrounding halo we quantify this effect by means of the spectral analysis of individual orbits in the disk and the halo and show that the halo particles are trapped by numerous lower resonances with the bar and that this trapping is robust . during these periods of recurrent instability , and especially during the slower second buckling , the bar remains vertically asymmetric for a prolonged @xmath1 gyr time interval which in principle can be detected observationally . however , two issues can potentially complicate this detection . first , it is not clear how widespread are the conditions favorable for the recurrent bar growth , although we have detected it in a number of models with different initial conditions . second , while it was shown that a clumpy gaseous component with an isothermal equation of state in the disk will weaken this instability , the effect of a realistic ism was never estimated . a statistically significant sample of ( nearly ) edge - on galaxies is required to test the prediction of a prolonged vertical asymmetry . we also find that the secular bar growth and the triggered buckling instabilities lead to pronounced changes in the bulge shape it grows both radially and vertically , acquiring a peanut , a boxy and finally the x - shaped appearance . while the bar size approximately follows the 4:1 ( ultra - harmonic ) resonance in the disk , the boxy / peanut bulge size appears to be guided by the vertical ilr . concurrently , the bar is going through a structural evolution new families of 3-d periodic orbits appear ( or become more pronounced ) after the bucklings . finally , we find that the bar strength correlates with its pattern speed , in both strengthening and weakening bars . while it was known already that a stellar bar becomes stronger as it slows down , we detect the reverse trend as well bars that weaken during the buckling speed up their tumbling . moreover , the bar size appears to be sensitive to these changes slowing down bars become longer , while speeding up bars shorten . we gratefully acknowledge insightfull discussions with lia athanassoula , ingo berentzen , amr el - zant , johan knapen and alar toomre . this work has been partially supported by grants from nasa / ltsa5 - 13063 , nasa / atp nag 5 - 10823 , hst ar-09546 and 10284 ( i.s . ) , and by nsf ast-0206251 ( c.h . & i.s . ) . i.m . acknowledges support from pparc . simulations and orbital analysis have been performed on a dedicated linux cluster and we thank brian doyle for technical support . athanassoula , e. 1992 , mnras , 259 , 345 athanassoula , e. 2001 , new horizons of computational science , ( eds . ) t.ebisuzaki & j.makino ( dordrecht : kluwer acad . publ . ) , 312 ( astro - ph/0112076 ) athanassoula , e. 2002a , disks of galaxies : kinematics , dynamics & interactions , ( eds . ) e. athanassoula , a. bosma , and r. mujica ( asp conf . 275 ) , 141 binney , j. , tremaine , s. 1987 , galactic dynamics , princeton univ . press bournaud , f. , combes , f. 2002 , a&a , 392 , 83 bureau , m. , freeman , k.c . 1999 , apj , 118 , 126 buta , r. et al . , 1998 , aj , 116 , 1142 combes , f. sanders , r.h . 1981 , a&a , 96 , 164 combes , f. , debbasch , f. , friedli , d. , pfenniger , d. 1990 , a&a , 233 , 82

in this paper we describe a dynamic data structure that answers one - dimensional stabbing - max queries in optimal time . our data structure uses linear space and supports insertions and deletions in and amortized time respectively . we also describe a space data structure that answers-dimensional stabbing - max queries in time . insertions and deletions are supported in and amortized time respectively .

1109.3890

in the stabbing - max problem , a set of rectangles is stored in a data structure , and each rectangle @xmath8 is assigned a priority @xmath9 . for a query point @xmath10 , we must find the highest priority rectangle @xmath8 that contains ( or is stabbed by ) @xmath10 . in this paper we describe a dynamic data structure that answers stabbing - max queries on a set of one - dimensional rectangles ( intervals ) in optimal @xmath2 time . we also show how this result can be extended to @xmath11 dimensions . * previous work . * the stabbing - max problem has important applications in networking and geographic information systems . solutions to some special cases of the stabbing - max problem play a crucial role in classification and routing of internet packets ; we refer to e.g. , @xcite for a small selection of the previous work and to @xcite for more extensive surveys . below we describe the previous works on the general case of the stabbing - max problem . the semi - dynamic data structure of yang and widom @xcite maintains a set of one - dimensional intervals in linear space ; stabbing - max queries and insertions are supported in @xmath1 time . agarwal et al . @xcite showed that stabbing - max queries on a set of one - dimensional intervals can be answered in @xmath12 time ; the data structure of agarwal et al . @xcite uses linear space and supports updates in @xmath1 time . the linear space data structure of kaplan et al . @xcite supports one - dimensional queries and insertions in @xmath1 time , but deletions take @xmath13 time . in @xcite , the authors also consider the stabbing - max problem for a nested set of intervals : for any two intervals @xmath14 , either @xmath15 or @xmath16 or @xmath17 . their data structure for a nested set of one - dimensional intervals uses @xmath18 space and supports both queries and updates in @xmath1 time . thorup @xcite described a linear space data structure that supports very fast queries , but needs @xmath19 time to perform updates . his data structure supports stabbing - max queries in @xmath20 time and updates in @xmath21 time for any parameter @xmath22 . however the problem of constructing a data structure with poly - logarithmic update time is not addressed in @xcite . agarwal et al . @xcite described a data structure that uses linear space and supports both queries and updates in @xmath1 time for an arbitrary set of one - dimensional intervals . the results presented in @xcite and @xcite are valid in the pointer machine model @xcite . the one - dimensional data structures can be extended to @xmath11 dimensions , so that space usage , query time , and update time increase by a factor of @xmath23 . thus the best previously known data structure @xcite for @xmath4-dimensional stabbing - max problem uses @xmath24 space , answers queries in @xmath25 time , and supports updates in @xmath26 time . kaplan et al . @xcite showed that @xmath4-dimensional stabbing - max queries can be answered in @xmath1 time for any constant @xmath4 in the special case of nested rectangles . * our result . * the one - dimensional data structure described in @xcite achieves optimal query time in the pointer machine model . in this paper we show that we can achieve sublogarithmic query time without increasing the update time in the word ram model of computation . our data structure supports deletions and insertions in @xmath0 and @xmath1 amortized time respectively . as follows from the lower bound of @xcite , any fully - dynamic data structure with poly - logarithmic update time needs @xmath27 time to answer a stabbing - max query that is stabbed by a query point @xmath10 ? ] . thus our data structure achieves optimal query time and space usage . our result can be also extended to @xmath11 dimensions . we describe a data structure that uses @xmath3 space and answers stabbing - max queries in @xmath28 time ; insertions and deletions are supported in @xmath29 and @xmath28 amortized time respectively . moreover , our construction can be modified to support stabbing - sum . ] queries : we can count the number of one - dimensional intervals stabbed by a query point @xmath10 in @xmath2 time . the stabbing - sum data structure also supports insertions and deletions in @xmath1 and @xmath0 amortized time . * overview . * we start by describing a simple one - dimensional stabbing - max data structure in section [ sec : stabover ] . this data structure achieves the desired query and update times but needs @xmath30 space . all intervals are stored in nodes of the base tree of height @xmath2 ; the base tree is organized as a variant of the segment tree data structure . intervals in a node @xmath31 are stored in a variant of the van emde boas ( veb ) data structure @xcite . we can answer a stabbing - max query by traversing a leaf - to - root path in the base tree ; the procedure spends @xmath32 time in each node on the path . in section [ sec : compact ] , we show how all secondary data structures in the nodes of the base tree can be stored in @xmath33 bits of space . the main idea of our method is the compact representation of intervals stored in each node . similar compact representations were also used in data structures for range reporting queries @xcite and some other problems @xcite . however the previous methods are too slow for our goal : we need @xmath34 time to obtain the representation of an element @xmath35 in a node @xmath31 if the representation of @xmath35 in a child of @xmath31 is known . therefore it would take @xmath1 time to traverse a leaf - to - root path in the base tree . in this paper we present a new , improved compact storage scheme . using our representation , we can traverse a path in the base tree and spend @xmath32 time in each node . we believe that our method is of independent interest and can be also applied to other problems . the results for multi - dimensional stabbing - max queries and stabbing - sum queries are described in theorems [ theor : multidim ] and [ theor : sum1d ] ; an extensive description of these results is provided in appendices c and d. * base tree . * in this section we describe a data structure that answers stabbing - max queries in optimal time . endpoints of all intervals from the set @xmath36 are stored in the leaves of the base tree @xmath37 . every leaf of @xmath37 contains @xmath38 endpoints . every internal node , except of the root , has @xmath39 children ; the root has @xmath40 children . throughout this paper @xmath41 denotes an arbitrarily small positive constant . the range @xmath42 of a node @xmath31 is an interval bounded by the minimal and maximal values stored in the leaf descendants of @xmath31 . for a leaf node @xmath43 , the set @xmath44 contains all intervals @xmath8 , such that at least one endpoint of @xmath8 belongs to @xmath43 . for an internal node @xmath31 , the set @xmath45 contains all intervals @xmath8 , such that @xmath46 for at least one child @xmath47 of @xmath31 but @xmath48 . see fig . [ fig : basetree ] for an example . thus each interval is stored in @xmath0 sets @xmath45 . for every pair @xmath49 , @xmath50 denotes the set of all intervals @xmath51 such that @xmath52 for a child @xmath53 of @xmath31 if and only if @xmath54 . for simplicity , we will sometimes not distinguish between intervals and their priorities . internal nodes of the base tree . intervals @xmath55 and @xmath56 are stored in the set @xmath57 , but are not stored in the set @xmath45 . interval @xmath58 is stored in @xmath45 , but @xmath58 is not stored in @xmath57 . interval @xmath59 is stored in both @xmath45 and @xmath57 . moreover , @xmath55 , @xmath59 , and @xmath56 belong to the sets @xmath60 , @xmath61 , and @xmath61 respectively . intervals @xmath59 and @xmath58 belong to @xmath62 and @xmath63 respectively . ] = 1000 * secondary data structures . * for each internal node @xmath31 , we store a data structure @xmath64 described in the following proposition . [ lemma : search1d ] suppose that priorities of all intervals in @xmath45 are integers in the interval @xmath65 $ ] for @xmath66 . there exists a data structure @xmath64 that uses @xmath67 words of space and supports the following queries : for any @xmath68 , the predecessor of @xmath10 in @xmath69 can be found in @xmath34 time and the maximum element in @xmath69 can be found in @xmath32 time . the data structure supports insertions in @xmath34 time . if a pointer to an interval @xmath51 is given , @xmath8 can be deleted in @xmath32 amortized time . it suffices to store all intervals from @xmath69 in a veb data structure @xmath70 . each @xmath70 uses @xmath71 words of space and answers queries in @xmath72 time @xcite . it is a folklore observation that we can modify the veb data structure so that maximum queries are supported in constant time . we also store a data structure @xmath73 that contains the interval @xmath74 with maximal priority among all intervals in @xmath50 for each @xmath49 . since each internal node has @xmath39 children , @xmath73 contains @xmath75 elements . for any query index @xmath76 , @xmath73 reports the largest element among all @xmath77 , @xmath54 . in other words for any child @xmath53 of @xmath31 , @xmath73 can find the interval with the highest priority that covers the range of the @xmath76-th child of @xmath31 . using standard techniques , we can implement @xmath73 so that queries and updates are supported in @xmath32 time . for completeness , we describe the data structure @xmath73 in appendix a. * queries and updates . * let @xmath78 denote the search path for the query point @xmath10 in the base tree @xmath37 . the path @xmath78 consists of the nodes @xmath79 where @xmath80 is a leaf node and @xmath81 is the root node . let @xmath82 be the interval with the highest priority among all intervals that are stored in @xmath83 and are stabbed by @xmath10 . the interval @xmath84 can be found by examining all @xmath40 intervals stored in @xmath85 . suppose that we reached a node @xmath86 and the interval @xmath87 is already known . if @xmath10 stabs an interval @xmath8 stored in @xmath88 , then @xmath89 . therefore @xmath10 stabs an interval @xmath90 if and only if @xmath8 is stored in some set @xmath91 such that @xmath92 and @xmath93 is the @xmath76-th child of @xmath86 . using the data structure @xmath94 , we can find in constant time the maximal interval @xmath95 , such that @xmath96 and @xmath92 . then we just set @xmath97 and proceed in the next node @xmath98 . the total query time is @xmath2 . when we insert an interval @xmath8 , we identify @xmath2 nodes @xmath86 such that @xmath8 is to be inserted into @xmath88 . for every such @xmath86 , we proceed as follows . we identify @xmath99 and @xmath100 such that @xmath8 belongs to @xmath91 . using @xmath101 , we find the position of @xmath8 in @xmath91 , and insert @xmath8 into @xmath91 . if @xmath8 is the maximal interval in @xmath91 , we delete the old interval @xmath102 from the data structure @xmath94 , set @xmath103 , and insert the new @xmath102 into @xmath94 . when an interval @xmath8 is deleted , we also identify nodes @xmath86 , such that @xmath90 . for each @xmath86 , we find the indices @xmath99 and @xmath100 , such that @xmath104 . using the procedure that will be described in the next section , we can find the position of @xmath8 in @xmath91 . then , @xmath8 is deleted from the data structure @xmath101 . if @xmath105 , we remove @xmath102 from @xmath94 , find the maximum priority interval in @xmath91 , and insert it into @xmath94 . we will show in section [ sec : compact ] that positions of the deleted interval @xmath8 in @xmath91 for all nodes @xmath86 can be found in @xmath0 time . since all other operations take @xmath32 time per node , the total time necessary for a deletion of an interval is @xmath2 . unfortunately , the space usage of the data structure described in this section is very high : every veb data structure @xmath70 needs @xmath71 space , where @xmath106 is the highest possible interval priority . even if @xmath66 , all data structures @xmath64 use @xmath107 space . in the next section we show how all data structures @xmath64 , @xmath108 , can be stored in @xmath33 bits without increasing the query and update times . the key idea of our compact representation is to store only interval identifiers in every node @xmath31 of @xmath37 . our storage scheme enables us to spend @xmath34 bits for each identifier stored in a node @xmath31 . using the position of an interval @xmath8 in a node @xmath31 , we can obtain the position of @xmath8 in the parent @xmath109 of @xmath31 . we can also compare priorities of two intervals stored in the same node by comparing their positions . these properties of our storage scheme enable us to traverse the search path for a point @xmath10 and answer the query as described in section [ sec : stabover ] . similar representations were also used in space - efficient data structures for orthogonal range reporting @xcite and orthogonal point location and line - segment intersection problems @xcite . storage schemes of @xcite also use @xmath34 bits for each interval stored in a node of the base tree . the main drawback of those methods is that we need @xmath34 time to navigate between a node and its parent . therefore , @xmath110 time is necessary to traverse a leaf - to - root path and we need @xmath111 time to answer a query . in this section we describe a new method that enables us to navigate between nodes of the base tree and update the lists of identifiers in @xmath32 time per node . the main idea of our improvement is to maintain identifiers for a set @xmath112 in every @xmath108 . when an interval is inserted in @xmath45 , we also add its identifier to @xmath113 . but when an interval is deleted from @xmath45 , its identifier is not removed from @xmath113 . when the number of deleted interval identifiers in all @xmath113 exceeds the number of intervals in @xmath45 , we re - build the base tree and all secondary structures ( global re - build ) . * compact lists . * we start by defining a set @xmath114 . if @xmath31 is a leaf node , then @xmath115 . if @xmath31 is an internal node , then @xmath116 for all children @xmath47 of @xmath31 . an interval @xmath8 belongs to @xmath117 if at least one endpoint of @xmath8 is stored in a leaf descendant of @xmath31 . hence , @xmath118 where the union is taken over all nodes @xmath119 that are situated on the same level of the base tree @xmath37 . since the height of @xmath37 is @xmath0 , the total number of intervals stored in all @xmath117 , @xmath108 , is @xmath120 . let @xmath113 be the set that contains all intervals from @xmath117 that were inserted into @xmath117 since the last global re - build . we will organize global re - builds in such way that at most one half of elements in all @xmath113 correspond to deleted intervals . therefore the total number of intervals in @xmath121 is @xmath122 . we will show below how we can store the represntations of sets @xmath113 in compact form , so that an element of @xmath113 uses @xmath123 bits in average . since @xmath124 , we can also use the same method to store all @xmath45 and @xmath64 in @xmath33 bits . sets @xmath117 and @xmath113 are not stored explicitly in the data structure . instead , we store a list @xmath125 that contains a compact representation for identifiers of intervals in @xmath113 . @xmath125 is organized as follows . the set @xmath113 is sorted by interval priorities and divided into blocks . if @xmath126 , then each block of @xmath113 contains at least @xmath127 and at most @xmath128 elements . otherwise all @xmath129 belong to the same block . each block @xmath130 is assigned an integer block label @xmath131 according to the method of @xcite . labels of blocks are monotone with respect to order of blocks in @xmath125 : the block @xmath132 precedes @xmath133 in @xmath125 if and only if @xmath134 . besides that , all labels assigned to blocks of @xmath125 are bounded by a linear function of @xmath135 : for any block @xmath130 in @xmath125 , @xmath136 . when a new block is inserted into a list , we may have to change the labels of @xmath12 other blocks . for every block @xmath130 , we store its block label as well as the pointers to the next and the previous blocks in the list @xmath125 . for each interval @xmath137 in a block @xmath130 of @xmath113 , the list @xmath125 contains the _ identifier _ of @xmath137 in @xmath125 . the identifier is simply the list of indices of children @xmath47 of @xmath31 , such that @xmath138 . as follows from the description of the base tree and the sets @xmath113 , we store at most two child indices for every interval @xmath137 in a block ; hence , each identifier uses @xmath34 bits . to simplify the description , we sometimes will not distinguish between an interval @xmath8 and its identifier in a list @xmath125 . we say that the position of an interval @xmath8 in a list @xmath125 is known if the block @xmath130 that contains @xmath8 and the position of @xmath8 in @xmath130 are known . if we know positions of two intervals @xmath139 and @xmath140 in @xmath125 , we can compare their priorities in @xmath32 time . suppose that @xmath139 and @xmath140 belong to blocks @xmath132 and @xmath133 respectively . then @xmath141 if @xmath142 , and @xmath143 if @xmath134 . if @xmath144 , we can compare priorities of @xmath139 and @xmath140 by comparing their positions in the block @xmath145 . the rest of this section has the following structure . first , we describe auxiliary data structures that enable us to search in a block and navigate between nodes of the base tree . each block @xmath130 contains a poly - logarithmic number of elements and every identifier in @xmath130 uses @xmath123 bits . we can use this fact and implement block data structures , so that queries and updates are supported in @xmath32 time . if the position of some @xmath137 in a list @xmath125 is known , we can find in constant time the positions of @xmath137 in the parent of @xmath31 . next , we show how data structures @xmath64 and @xmath73 , defined in section [ sec : stabover ] , are modified . finally , we describe the search and update procedures . * block data structures . * we store a data structure @xmath146 that supports rank and select queries in a block @xmath130 of @xmath125 : a query @xmath147 returns the number of intervals that also belong to @xmath148 among the first @xmath149 elements of @xmath130 . a query @xmath150 returns the smallest @xmath151 , such that @xmath152 ; in other words , @xmath150 returns the position of the @xmath149-th interval in @xmath130 that also belongs to @xmath148 . we can answer @xmath153 and @xmath154 queries in a block in @xmath32 time . we can also count the number of elements in a block of @xmath125 that are stored in the @xmath76-th child of @xmath31 , and determine for the @xmath100-th element of a block in which children of @xmath31 it is stored . implementation of @xmath146 will be described in appendix a. for each block @xmath155 and for any child @xmath53 of @xmath31 , we store a pointer to the largest block @xmath156 before @xmath157 that contains an element from @xmath158 . these pointers are maintained with help of a data structure @xmath159 for each child @xmath53 . we implement @xmath159 as an incremental split - find data structure @xcite . insertion of a new block label into @xmath159 takes @xmath32 amortized time ; we can also find the block @xmath156 for any block @xmath155 in @xmath32 worst - case time . using the data structure @xmath160 for a block @xmath157 , we can identify for any element @xmath35 in a block @xmath157 the largest @xmath161 , @xmath162 , such that @xmath163 belongs to @xmath158 . thus we can identify for any @xmath164 the largest element @xmath165 , such that @xmath161 and @xmath166 , in @xmath32 time . * navigation in nodes of the base tree . * finally , we store pointers to selected elements in each list @xmath125 . pointers enable us to navigate between nodes of the base tree : if the position of some @xmath164 is known , we can find the position of @xmath35 in @xmath167 for the parent @xmath109 of @xmath31 and the position of @xmath35 in @xmath168 for any child @xmath47 of @xmath31 such that @xmath168 contains @xmath35 . we associate a _ stamp _ @xmath169 with each element stored in a block @xmath130 ; every stamp is a positive integer bounded by @xmath170 . a pointer to an element @xmath35 in a block @xmath130 consists of the block label @xmath131 and the stamp of @xmath35 in @xmath130 . when an element is inserted into a block , stamps of other elements in this block do not change . therefore , when a new interval is inserted into a block @xmath130 we do not have to update all pointers that point into @xmath130 . furthermore , we store a data structure @xmath171 for each block @xmath130 . using @xmath171 , we can find the position of an element @xmath35 in @xmath130 if its stamp @xmath169 in @xmath130 is known . if an element @xmath35 must be inserted into @xmath130 after an element @xmath172 , then we can assign a stamp @xmath173 to @xmath35 and insert it into @xmath171 in @xmath32 time . implementation of @xmath171 is very similar to the implementation of @xmath146 and will be described in the full version . if @xmath35 is the first element in the block @xmath130 of @xmath125 that belongs to @xmath168 , then we store the pointer from the copy of @xmath35 in @xmath125 to the copy of @xmath35 in @xmath168 . if an element @xmath164 is also stored in @xmath168 and @xmath35 is the first element in a block @xmath174 of @xmath168 , then there is a pointer from the copy of @xmath35 in @xmath125 to the copy of @xmath35 in @xmath168 . such pointers will be called child pointers . for any pointer from @xmath164 to @xmath175 , we store a pointer from @xmath175 to @xmath164 . such pointers will be called parent pointers . see fig . [ fig : pointers ] for an example . we can store each pointer in @xmath1 bits . the total number of pointers in @xmath125 equals to the number of blocks in @xmath125 and @xmath168 for all children @xmath47 of @xmath31 . hence , all pointers and all block data structures use @xmath33 bits . if we know the position of some interval @xmath8 in @xmath176 , we can find the position of @xmath8 in the parent @xmath109 of @xmath119 as follows . suppose that an interval @xmath8 is stored in the block @xmath130 of @xmath176 and @xmath119 is the @xmath76-th child of @xmath109 . we find the last interval @xmath177 in @xmath130 stored before @xmath8 , such that there is a parent pointer from @xmath177 . let @xmath178 denote the number of elements between @xmath177 and @xmath8 in @xmath130 . let @xmath174 be the block in @xmath167 that contains @xmath177 . using @xmath179 , we find the position @xmath180 of @xmath177 in the block @xmath174 of @xmath167 . then the position of @xmath8 in @xmath174 can be found by answering the query @xmath181 to a data structure @xmath182 . using a symmetric procedure , we can find the position of @xmath8 in @xmath176 if its position in @xmath167 is known . * root and leaves of the base tree . * elements of @xmath183 are explicitly stored in the root node @xmath81 . that is , we store a table in the root node @xmath81 that enables us to find for any interval @xmath8 the block @xmath130 that contains the identifier of @xmath8 in @xmath184 and the stamp of @xmath8 in @xmath130 . conversely , if the position of @xmath8 in a block @xmath130 of @xmath184 is known , we can find its stamp in @xmath130 in @xmath32 time . if the block label and the stamp of an interval @xmath8 are known , we can identify @xmath8 in @xmath32 time . in the same way , we explicitly store the elements of @xmath185 for each leaf node @xmath43 . moreover , we store all elements of @xmath183 in a data structure @xmath186 , so that the predecessor and the successor of any value @xmath187 can be found in @xmath0 time . * data structures @xmath73 and @xmath64 . * each set @xmath45 and data structure @xmath64 are also stored in compact form . @xmath64 consists of structures @xmath70 for every pair @xmath68 . if a block @xmath130 contains a label of an interval @xmath188 , then we store the label of @xmath130 in the veb data structure @xmath70 . the data structure @xmath189 contains data about intervals in @xmath190 . for every @xmath191 , we store indices @xmath192 if @xmath188 ; if an element @xmath193 does not belong to @xmath36 ( i.e. , @xmath8 was already deleted ) , then we set @xmath194 . for a query index @xmath76 , @xmath189 returns in @xmath32 time the highest priority interval @xmath193 , such that @xmath188 and @xmath92 . a data structure @xmath189 uses @xmath195 bits of space and supports updates in @xmath32 time . implementation of @xmath189 is very similar to the implementation of @xmath146 and will be described in the full version of this paper . we store a data structure @xmath73 in every node @xmath108 , such that @xmath125 consists of at least two blocks . for each pair @xmath68 , the data structure @xmath73 stores the label of the block that contains the highest priority interval in @xmath69 . for a query @xmath76 , @xmath73 returns the label of the block that contains the highest priority interval in @xmath196 . such queries are supported in @xmath32 time ; a more detailed description of the data structure @xmath73 will be given in appendix a. * search procedure . * we can easily modify the search procedure of section [ sec : stabover ] for the case when only lists @xmath125 are stored in each node . let @xmath86 be a node on the path @xmath197 , where @xmath78 is the search path for the query point @xmath10 . let @xmath82 denote the interval that has the highest priority among all intervals that belong to @xmath198 and are stabbed by @xmath10 . we can examine all intervals in @xmath85 and find @xmath84 in @xmath40 time . the position of @xmath84 in @xmath199 can also be found in @xmath32 time . during the @xmath149-th step , @xmath200 , we find the position of @xmath82 in @xmath201 . an interval @xmath8 in @xmath88 is stabbed by @xmath10 if and only if @xmath104 , @xmath92 , and @xmath93 is the @xmath76-th child of @xmath86 . using @xmath94 , we can find the block label of the maximal interval @xmath95 among all intervals stored in @xmath91 for @xmath92 . let @xmath202 be the block that contains @xmath95 . using the data structure @xmath203 , we can find the position of @xmath95 in @xmath202 . by definition , @xmath97 . although we have no access to @xmath95 and @xmath87 , we can compare their priorities by comparing positions of @xmath95 and @xmath87 in @xmath201 . since the position of @xmath87 in @xmath204 is already known , we can find its position in @xmath201 in @xmath32 time . we can also determine , whether @xmath95 precedes or follows @xmath87 in @xmath201 in @xmath32 time . since a query to @xmath94 also takes @xmath32 time , our search procedure spends constant time in each node @xmath86 for @xmath200 . when we know the position of @xmath205 in @xmath184 , we can find the interval @xmath205 in @xmath32 time . the interval @xmath205 is the highest priority interval in @xmath36 that is stabbed by @xmath10 . hence , a query can be answered in @xmath2 time . we describe how our data structure can be updated in section [ sec : rebal ] . our result is summed up in the following theorem . [ theor:1d ] there exists a linear space data structure that answers orthogonal stabbing - max queries in @xmath2 time . this data structure supports insertions and deletions in @xmath1 and @xmath2 amortized time respectively . suppose that an interval @xmath8 is inserted into @xmath36 . the insertion procedure consists of two parts . first , we insert @xmath8 into lists @xmath201 and update @xmath201 for all relevant nodes @xmath86 of @xmath37 . then , we insert @xmath8 into data structures @xmath101 and @xmath94 . using the data structure @xmath186 , we can identify the segment @xmath206 that precedes @xmath8 in @xmath183 . then , we find the position of @xmath206 in @xmath184 . we also find the leaves @xmath207 and @xmath208 of @xmath37 , in which the left and the right endpoints of @xmath8 must be stored . the path @xmath209 is traversed starting at the root node . let @xmath210 be the segment that precedes @xmath8 in @xmath201 and let @xmath211 be the block that contains @xmath210 . in every node @xmath86 , we insert the identifier for @xmath8 after @xmath210 . we also update data structures @xmath212 and @xmath213 for the block @xmath211 that contains @xmath8 . if the number of intervals in @xmath211 equals @xmath214 , we split the block @xmath211 into two blocks @xmath215 and @xmath216 of equal size . we assign a new label to @xmath216 and update the labels for some blocks of @xmath201 in @xmath217 time . we also may have to update @xmath40 split - find data structures @xmath218 . besides that , @xmath32 child pointers in the parent of @xmath86 and @xmath40 parent pointers in the children of @xmath86 may also be updated . we split the block after @xmath219 insertions ; hence , an amortized cost of splitting a block is @xmath32 . when @xmath8 is inserted into @xmath201 , we find the largest element @xmath220 , such that @xmath221 is also stored in @xmath204 . then , we find the position of @xmath221 in @xmath204 and proceed in the node @xmath93 . let @xmath222 be the lowest common ancestor of @xmath207 and @xmath208 , and let @xmath223 be the path from @xmath222 to @xmath208 . we also insert @xmath8 into @xmath224 for all @xmath225 ; nodes @xmath225 are processed in the same way as nodes @xmath226 during the second stage , we update data structures @xmath101 and @xmath94 for @xmath227 . for any such @xmath86 we already know the block @xmath228 that contains @xmath8 and the position of @xmath8 in @xmath130 . hence , a data structure @xmath189 can be updated in @xmath32 time . if @xmath8 is the only interval in @xmath130 that belongs to @xmath91 for some @xmath99 , @xmath100 , we insert @xmath131 into @xmath229 . if @xmath131 is the greatest label in @xmath229 , we also update @xmath94 . an insertion into a data structure @xmath229 , @xmath230 , takes @xmath34 time . updates of all other structures in @xmath86 take @xmath32 time . hence , an insertion takes @xmath1 time in total . when an interval is deleted , we identify its position in every relevant node @xmath86 . this can be done in the same way as during the first stage of the insertion procedure . then , we update the data structures @xmath101 , @xmath94 , and @xmath231 if necessary . since a block label can be removed from @xmath229 in @xmath32 time , the total time necessary to delete an interval is @xmath0 . * re - balancing of the tree nodes . * it remains to show how the tree can be rebalanced after updates , so that the height of @xmath37 remains @xmath0 . we implement the base tree @xmath37 as a weight - balanced b - tree @xcite with branching parameter @xmath232 and leaf parameter @xmath232 for @xmath233 . we assume that the constant @xmath234 . when a segment @xmath8 is deleted , we simply mark it as deleted . after @xmath235 deletions , we re - build the base tree and all data structures stored in the nodes . this can be done in @xmath122 time . now we show how insertions can be handled . we denote by the weight of @xmath31 the total number of elements in the leaf descendants of @xmath31 . the weight @xmath236 of @xmath31 also equals to the number of segment identifiers in @xmath125 . our weight - balanced b - tree is organized in such way that @xmath237 , where @xmath238 and @xmath239 is the level of a node @xmath31 . a node is split after @xmath240 insertions ; when a node @xmath31 is split into @xmath241 and @xmath242 , the ranges of the other nodes in the base tree do not change ; see @xcite for a detailed description of node splitting . we will show that all relevant data structures can be re - built in @xmath243 time . hence , the amortized cost of splitting a node is @xmath32 . when a segment is inserted , it is inserted into @xmath2 nodes of the base tree . hence , the total amortized cost of all splitting operations caused by inserting a segment @xmath8 into our data structure is @xmath2 . it remains to show how secondary data structures are updated when a node @xmath31 is split . let @xmath109 be the parent of @xmath31 . the total number of segments in @xmath244 and @xmath245 does not exceed @xmath246 . hence , we can construct data structures @xmath247 , @xmath248 and lists @xmath249 , @xmath250 with all block data structures in @xmath251 time . we can find the positions of all elements @xmath252 in @xmath167 , update their identifiers in @xmath167 , and update all auxiliary data structures in @xmath243 time . [ cols="^,^,^ " , ] some of the intervals stored in @xmath45 can be moved from @xmath45 to @xmath57 : if an interval @xmath253 does not cover @xmath42 but covers @xmath254 or @xmath255 , then @xmath8 must be stored in @xmath57 after splitting . see fig . [ fig : split ] for an example . the total number of elements in @xmath167 is @xmath256 ; hence , the total number of blocks in @xmath167 is @xmath257 . identifiers of all segments in @xmath45 are already stored in @xmath167 . we can update all block data structures in @xmath32 time per segment . every update of the data structure @xmath258 takes @xmath123 time . but the total number of insertions into @xmath258 does not exceed the number of blocks in @xmath258 . therefore the total time needed to update @xmath258 is @xmath259 . suppose that @xmath31 was the @xmath260-th child of @xmath109 . some segments stored in @xmath261 for @xmath262 can be moved to @xmath263 , and some segments in @xmath264 for @xmath265 can be moved to @xmath266 . since the total number of blocks in @xmath167 is @xmath267 , all updates of data structures @xmath268 take @xmath269 time . at most @xmath236 segments were stored in @xmath261 and @xmath264 before the split of @xmath31 . hence , we can update all block data structures in @xmath243 time . finally , we must change the identifiers of segments stored in @xmath167 and data structures @xmath189 . recall that an identifier indicates in which children of @xmath109 a segment @xmath8 is stored . suppose that a segment @xmath8 had an identifier @xmath265 in @xmath167 . then its identifier will be changed to @xmath270 after the split of @xmath31 . the total number of segments with incremented identifiers does not exceed @xmath271 . however , each identifier is stored in @xmath34 bits . we can use this fact , and increment the values of @xmath272 identifiers in @xmath32 time using the following look - up table @xmath273 . there are @xmath274 different sequences of @xmath272 identifiers . for every such sequence @xmath55 and any @xmath275 , @xmath276[j]=\alpha'$ ] . here @xmath277 is the sequence that we obtain if every @xmath278 in the sequence @xmath55 is replaced with @xmath279 . thus the list @xmath167 can be updated in @xmath280 time . using the same approach , we can also update data structures @xmath146 and @xmath189 for each block @xmath130 in @xmath281 time . hence , the total time necessary to update all data structures in @xmath109 because some identifiers must be incremented is @xmath243 . since all data structures can be updated in @xmath243 time when a node @xmath31 is split , the total amortized cost of an insertion is @xmath1 . our data structure can be extended to the case of @xmath4-dimensional stabbing - max queries for @xmath11 . in this section we prove the following theorem . [ theor : multidim ] there exists a data structure that uses @xmath282 space and answers @xmath4-dimensional stabbing - max queries in @xmath28 time . insertions and deletions are supported in @xmath283 and @xmath284 amortized time respectively . let @xmath285-stabbing problem denote the @xmath286-dimensional stabbing - max problem for the case when the first @xmath4 coordinates can assume arbitrary values and the last @xmath287 coordinates are bounded by @xmath288 , for @xmath289 . first , we show that the data structure of theorem [ theor:1d ] can be modified to support @xmath290-stabbing queries for any constant @xmath287 . then , we will show how arbitrary @xmath4-dimensional queries can be answered . throughout this section @xmath291 denotes the projection of a rectangle @xmath8 on the @xmath149-th coordinate . we denote by @xmath292 and @xmath293 the @xmath149-th coordinates of endpoints in a rectangle @xmath8 , so that @xmath294 $ ] . * a data structure for the @xmath290-stabbing problem . * the solution of the @xmath290-stabbing problem is very similar to our solution of the one - dimensional problem . we construct the base tree @xmath37 on the @xmath187-coordinates of all rectangles . @xmath37 is defined as in section [ sec : stabover ] , but we assume that @xmath295 . sets @xmath45 , @xmath69 , and @xmath113 are defined as in sections [ sec : stabover ] , [ sec : compact ] . we define @xmath296(u)$ ] as the set of rectangles @xmath8 in @xmath69 such that @xmath297 , @xmath298 , @xmath299 , @xmath300 , @xmath298 , @xmath301 . a rectangle identifier for a rectangle @xmath8 stored in a list @xmath125 consists of one or two tuples with @xmath302 components . the first component of each tuple is an index @xmath151 , such that @xmath8 belongs to the @xmath151-th child of @xmath31 ; remaining @xmath303 components are the last @xmath287 coordinates of the rectangle endpoints . the data structure @xmath73 contains the maximum priority rectangle in @xmath304(u)$ ] for each @xmath305 . @xmath73 can find for any @xmath306 the highest priority rectangle in all @xmath304(u)$ ] , such that @xmath307 , and @xmath308 for @xmath309 . we can implement @xmath73 as for the one - dimensional data structure . data structure @xmath64 consists of veb structures @xmath310(u)$ ] . if a block @xmath130 of @xmath125 contains a rectangle from @xmath304(u)$ ] , then @xmath131 is stored in @xmath310(u)$ ] . we can answer a query @xmath311 by traversing the search path @xmath312 for @xmath313 in the base tree @xmath37 . as in theorem [ theor:1d ] , we start at the leaf node @xmath80 and find the maximal rectangle stored in @xmath314 that is stabbed by the query point @xmath10 . the search procedure in nodes @xmath315 is organized in the same way as the search procedure in section [ sec : compact ] : the rectangle @xmath82 is defined as in section [ sec : compact ] . if @xmath93 is the @xmath76-th child of @xmath86 , we answer the query @xmath316 using the data structure @xmath94 . then we visit the block @xmath202 returned by @xmath94 and find the last rectangle @xmath95 in @xmath202 with an identifier @xmath317 such that @xmath318 for @xmath309 . finally we compare @xmath95 with the rectangle @xmath87 by comparing their positions in @xmath201 . thus we can find the position of @xmath82 in @xmath201 . when we reach the root @xmath81 , @xmath205 is the highest priority segment that is stabbed by @xmath10 . updates of the list @xmath125 and all auxiliary data structures are implemented as in section [ sec : rebal ] . [ lemma:1 g ] there exists a data structure that answers @xmath290-stabbing queries in @xmath0 time and uses @xmath18 space . insertions and deletions are supported in @xmath1 and @xmath2 amortized time respectively . = 1500 * a data structure for the @xmath285-stabbing problem . * the result for @xmath290-stabbing can be extended to @xmath285-stabbing queries using the following lemma . [ lemma : dg ] suppose that there is a @xmath319 space data structure @xmath320 that answers @xmath321-stabbing queries in @xmath322 time ; @xmath320 supports insertions and deletions in @xmath323 and @xmath324 amortized time respectively . + then there exists a @xmath325 space data structure @xmath326 that answers @xmath285-stabbing queries in @xmath327 time ; @xmath326 supports insertions and deletions in amortized time @xmath328 and @xmath329 respectively . the main idea is to construct the base tree @xmath37 on the @xmath4-th coordinates of rectangles and store a data structure for @xmath321-stabbing queries in each tree node . the tree is organized as in section [ sec : stabover ] and the first part of this section . let @xmath330 denote the @xmath4-th coordinate of a point . leaves contain @xmath4-th coordinates of rectangle endpoints . a rectangle @xmath8 is stored in a set @xmath45 for a leaf @xmath31 if @xmath331 and @xmath332 . a rectangle @xmath8 is stored in a set @xmath45 for an internal node @xmath31 if @xmath333 and @xmath334 for at least one child @xmath47 of @xmath31 . @xmath50 is the set of all intervals @xmath51 such that @xmath335 for a child @xmath53 of @xmath31 if and only if @xmath54 . we store a data structure @xmath336 for @xmath321-stabbing queries in each internal node @xmath31 . for a rectangle @xmath188 , @xmath336 contains a rectangle @xmath177 such that @xmath337 for @xmath338 and @xmath339 $ ] . in other words , we replace the @xmath4-th coordinates of @xmath8 s endpoints with @xmath99 and @xmath100 . a query @xmath10 is answered by traversing the search path for @xmath10 . for a leaf node @xmath80 , we examine all rectangles in @xmath85 and find the highest priority rectangle in @xmath85 . in an internal node @xmath86 , @xmath200 , we answer the query @xmath340 using the data structure @xmath341 . the query @xmath342 is obtained from @xmath10 by replacing the @xmath4-th coordinate with an index @xmath76 , such that @xmath93 is the @xmath76-th child of @xmath86 . when a node @xmath86 is visited , our procedure finds the highest priority rectangle @xmath82 that is stored in @xmath88 and is stabbed by @xmath10 . all rectangles that are stabbed by @xmath10 are stored in some @xmath88 . hence , the highest priority rectangle stabbed by @xmath10 is the maximum rectangle among all @xmath82 . since our procedure spends @xmath343 time in each node , the total query time is @xmath28 . when a rectangle @xmath8 is inserted or deleted , we update @xmath0 data structures @xmath336 in which @xmath8 is stored . hence , deletions and insertions are supported in @xmath28 and @xmath344 time respectively . we can re - balance the base tree using a procedure that is similar to the procedure described in section [ sec : rebal ] . * proof of theorem [ theor : multidim ] . * theorem [ theor : multidim ] follows easily from lemmas [ lemma:1 g ] and [ lemma : dg ] . by lemma [ lemma:1 g ] , there exists a linear space data structure that answers @xmath345-stabbing queries in @xmath0 time . we obtain the main result for @xmath4-dimensional stabbing - max queries by applying the result of lemma [ lemma : dg ] to the data structure for @xmath345-stabbing queries . we can also modify our data structure so that it supports stabbing - sum queries : count the number of intervals stabbed by a query point @xmath10 . [ theor : sum1d ] there exists a linear space data structure that answers orthogonal stabbing - sum queries in @xmath2 time . this data structure supports insertions and deletions in @xmath1 and @xmath2 amortized time respectively . the only difference with the proof of theorem [ theor:1d ] is that we store data structures for counting intervals instead of @xmath73 . we maintain a data structure @xmath346 in each internal node @xmath31 . for a query index @xmath76 , @xmath346 reports in @xmath32 time the total number of intervals in @xmath196 . in every leaf node @xmath43 , we maintain a data structure @xmath347 that supports stabbing sum queries on @xmath44 and updates in @xmath34 time . as above , let @xmath348 denote the search path for a query point @xmath10 . in every node @xmath349 , @xmath200 , we count the number @xmath350 of intervals in @xmath196 using @xmath351 ; the index @xmath76 is chosen so that @xmath93 is the @xmath76-th child of @xmath86 . we also compute the number @xmath352 of intervals that belong to @xmath85 and are stabbed by @xmath10 using @xmath353 . an interval @xmath8 is stabbed by @xmath10 either if @xmath8 is stored in @xmath88 , @xmath200 , and @xmath89 or if @xmath8 is stored in @xmath85 and @xmath8 is stabbed by @xmath10 . hence , @xmath10 stabs @xmath354 intervals . each @xmath350 , @xmath200 , can be found in @xmath32 time and @xmath352 can be found in @xmath123 time . hence , a query can be answered in @xmath0 time . now we turn to the description of the data structure @xmath346 . following the idea of @xcite , we store information about the recent updates in a word @xmath130 ; the array @xmath355 reflects the state of the structure before recent updates . @xmath355 is a static array that is re - built after @xmath356 updates of @xmath346 . when we construct @xmath355 , we set @xmath357=\sum_{l\leq f\leq r } |s_{lr}(u)|$ ] . the word @xmath130 contains one integer value @xmath358 for each pair @xmath68 ( @xmath358 can also be negative , but the absolute value of each @xmath358 is bounded by @xmath40 ) . when a segment is inserted into ( deleted from ) @xmath69 , we increment ( decrement ) the value of @xmath358 by @xmath359 . we can find @xmath360 in @xmath32 time using a look - up table . after @xmath361 updates we rebuild the array @xmath355 and set all @xmath362 . the amortized cost of rebuilding @xmath355 is @xmath32 . the total number of segments in @xmath196 equals to @xmath357+\sum_{l\leq f\leq r } m(l , r)$ ] . hence , a query to @xmath346 is answered in @xmath32 time . we implement @xmath355 in such way that each entry @xmath363 $ ] uses @xmath364 bits . thus each data structure @xmath346 uses @xmath365 bits and all @xmath346 , @xmath366 , use @xmath33 bits in total . it remains to describe the data structure @xmath367 . let @xmath368 be the set that contains endpoints of all intervals in @xmath45 . since @xmath31 is a leaf node , @xmath45 contains @xmath369 elements . hence , it takes @xmath123 time to find the rank @xmath370 of @xmath10 in @xmath368 . for each @xmath371 , @xmath372 $ ] equals to the number of intervals that are stabbed by a point @xmath10 with rank @xmath373 . the array @xmath374 enables us to count intervals stabbed by @xmath10 in @xmath32 time if the rank of @xmath10 in @xmath368 is known . since @xmath375 and @xmath372=o(\log^{{\varepsilon } } n)$ ] , the array @xmath374 uses @xmath376 bits of memory . when a set @xmath45 is updated , we can update @xmath374 in @xmath32 time using a look - up table . thus @xmath367 uses linear space and supports both queries and updates in @xmath34 time . 99 p. k. agarwal , l. arge , j. yang , and k. yi , _ i / o efficient structures for orthogonal range max and stabbing max queries _ , proc . esa 2003 , 7 - 18 . p. k. agarwal , l. arge , k. yi , _ an optimal dynamic interval stabbing - max data structure ? soda 2005 , 803 - 812 . s. alstrup , t. husfeldt , t rauhe , _ marked ancestor problems _ , proc . focs 1998 , 534 - 544 . l. arge , j. s. vitter , _ optimal external memory interval management _ , siam j. comput . 32(6 ) , 1488 - 1508 ( 2003 ) . g. e. blelloch , _ space - efficient dynamic orthogonal point location , segment intersection , and range reporting _ soda 2008 , 894 - 903 . b. chazelle , _ a functional approach to data structures and its use in multidimensional searching _ , siam j. on computing , 17 , 427 - 462 ( 1988 ) . h. edelsbrunner , _ a new approach to rectangle intersections _ , part i , int . j. computer mathematics , 13 , 209 - 219 ( 1983 ) . p. van emde boas , r. kaas , e. zijlstra , _ design and implementation of an efficient priority queue _ , mathematical systems theory 10 , 99 - 127 ( 1977 ) . p. gupta and n. mckeown . _ dynamic algorithms with worst - case performance for packet classification _ , proc . ifip networking 2000 , 528- 539 . p. gupta and n. mckeown . _ algorithms for packet classification _ , ieee network , 15(2 ) , 24 - 32 ( 2001 ) . a. feldmann , s. muthukrishnan , _ tradeoffs for packet classification _ , proc . infocom 2000 , 1193 - 1202 . m. l. fredman , d. e. willard , _ trans - dichotomous algorithms for minimum spanning trees and shortest paths _ , j. comput . 48(3 ) , 533 - 551 ( 1994 ) . h. imai and t. asano , _ dynamic orthogonal segment intersection search _ , journal of algorithms 8(1 ) , 1 - 18 ( 1987 ) . a. itai , a. g. konheim , m. rodeh , _ a sparse table implementation of priority queues _ , 8th icalp 1981,417 - 431 . h. kaplan , e. molad , r. e. tarjan , _ dynamic rectangular intersection with priorities _ , proc . stoc 2003 , 639 - 648 . y. nekrich , _ orthogonal range searching in linear and almost - linear space _ , 42(4 ) , 342 - 351 ( 2009 ) . m. ptracu , e. d. demaine , _ tight bounds for the partial - sums problem _ , proc . 15th soda 2004 , 20 - 29 . s. sahni and k. kim , _ @xmath1 dynamic packet routing _ , ieee symposium on computers and communications 2002 , 443 - 448 . s. sahni , k. kim , and h. lu . _ data structures for one - dimensional packet classification using most specific rule matching _ , proc . ispan 2002 , 3 - 14 . r. e. tarjan , _ a class of algorithms which require nonlinear time to maintain disjoint sets _ , j. comput . 18(2 ) , 110 - 127 ( 1979 ) . m. thorup , _ undirected single - source shortest paths with positive integer weights in linear time _ , j. acm 46(3 ) , 362 - 394 ( 1999 ) . m. thorup , _ space efficient dynamic stabbing with fast queries _ stoc 2003 , 649 - 658 . d. e. willard , _ a density control algorithm for doing insertions and deletions in a sequentially ordered file in good worst - case time _ , information and computation 97 , 150 - 204 ( 1992 ) . j. yang , j. widom , _ incremental computation and maintenance of temporal aggregates _ ieee intl . conf . on data engineering 2001 , 51 - 60 . using bitwise operations and table look - ups , we can implement the data structure @xmath73 and the block data structures so that queries are supported in constant time . * data structure @xmath73 . * let @xmath377 be the set of all interval priorities stored in @xmath73 . recall that @xmath377 contains an element @xmath77 for each set @xmath50 , where @xmath77 is the highest priority interval ( or its block label ) stored in @xmath50 . for simplicity , we assume that @xmath378 if @xmath379 . let the rank of @xmath187 in @xmath377 be defined as @xmath380 . let @xmath381 be the set in which every element of @xmath377 is replaced with its rank in @xmath377 . that is , for each @xmath77 stored in @xmath377 we store @xmath382 in the set @xmath381 . each of @xmath383 elements in @xmath381 is bounded by @xmath383 . hence , we can store @xmath381 in one bit sequence @xmath384 ; @xmath384 consists of @xmath385 bits and fits into a machine word . we can store a table @xmath386 with an entry for each possible value of @xmath384 and for each @xmath76 . the entry @xmath387[f]$ ] contains the pair @xmath388 , such that @xmath389 is the highest value in the set @xmath390 . when @xmath73 is updated , the value of some @xmath77 is changed . we store all elements of @xmath377 in an atomic heap @xcite that supports predecessor queries and updates in @xmath32 time . hence , the new rank of @xmath77 in @xmath377 can be found in @xmath32 time . if the rank of @xmath77 has changed , the ranks of all other elements in @xmath377 also change . fortunately , @xmath396 can assume only @xmath383 values . there are @xmath383 elements @xmath396 in a set @xmath381 , and there are @xmath397 different sets @xmath381 . besides that , each @xmath394 fits into one word . hence , we can update the set @xmath381 using a look - up in a table @xmath386 . suppose that the rank of the @xmath76-th element in a set @xmath381 is changed to @xmath100 . then the new set @xmath381 is stored in an entry @xmath398[f][r]$ ] of @xmath386 ; here @xmath384 is the bit sequence that corresponds to the old set @xmath381 . thus an update of @xmath73 can be implemented in @xmath32 time . * data structure @xmath146 . * the data structure @xmath146 for a block @xmath401 is implemented as a b - tree @xmath402 . every leaf of @xmath402 contains @xmath403 identifiers and each internal node has degree @xmath403 for @xmath404 . recall that each identifier consists of at most two indices @xmath405 , @xmath406 such that the corresponding interval is stored in the @xmath405-th and @xmath406-th children of the node @xmath31 in the base tree . all identifiers stored in a leaf node can be packed into @xmath407 bits . in every internal node @xmath408 of @xmath402 , we store the bit sequence @xmath409 . for every child @xmath410 and for every possible value of @xmath151 , @xmath409 contains the total number of indices @xmath151 in the @xmath149-th child @xmath410 of @xmath408 ; @xmath409 consists of @xmath411 bits . using a look - up table , we can count for any @xmath409 and for any @xmath149 the total number of elements stored in the children @xmath412 of @xmath408 . for any @xmath409 and @xmath149 , we can also count the total number of indices @xmath151 in the children @xmath412 of @xmath408 . for any @xmath413 and any @xmath409 , @xmath151 , we can find the largest @xmath149 such that the total number of indices @xmath151 in @xmath414 does not exceed @xmath260 . look - up tables that support such queries use @xmath400 space and can be initialized in @xmath400 time . we need only one instance of each table for all blocks @xmath130 and all @xmath146 . all queries and updates of a data structure @xmath146 can be processed by traversing a path in @xmath402 . using the above described look - up - tables , we spend only constant time in each node of @xmath402 ; hence , queries and updates are supported in @xmath32 time . details will be given in the full version of this paper . data structures @xmath189 and @xmath171 can be implemented in a similar way .

we present a hubble diagram of type ii supernovae using corrected magnitudes derived only from photometry , with no input of spectral information . we use a data set from the carnegie supernovae project i ( csp ) for which optical and near - infrared light - curves were obtained . the apparent magnitude is corrected by two observables , one corresponding to the slope of the plateau in the band and the second a colour term . we obtain a dispersion of 0.44 mag using a combination of the colour and the band and we are able to reduce the dispersion to 0.39 mag using our golden sample . a comparison of our photometric colour method ( pcm ) with the standardised candle method ( scm ) is also performed . the dispersion obtained for the scm ( which uses both photometric and spectroscopic information ) is 0.29 mag which compares with 0.43 mag from the pcm , for the same sn sample . the construction of a photometric hubble diagram is of high importance in the coming era of large photometric wide - field surveys , which will increase the detection rate of supernovae by orders of magnitude . such numbers will prohibit spectroscopic follow - up in the vast majority of cases , and hence methods must be deployed which can proceed using solely photometric data .

1511.05145

a fundamental probe in modern astronomy to understand the universe , its history and evolution , is the measurement of distances . stellar parallax and the spectroscopic parallax allow us to reach @xmath3 100 - 1000 pc , respectively but farther afield other methods are needed . a traditional technique for measuring distances consists in applying the inverse square law for astrophysical sources with known absolute magnitudes , aka , as standard candles . one of the first such objects used in astronomy were cepheid stars . a cepheid star s period is directly related to its intrinsic luminosity @xcite and allows one to probe the universe to 15 mpc . to attain larger distances brighter objects are required . type ia supernovae ( sne ia ) , have an absolute @xmath4-band magnitude about -19.5 -19.2 mag ( depending on the assumptions of h@xmath5 @xcite ) which can be precisely calibrated using photometric and/or spectroscopic information from the sn itself , and be used as excellent distance indicators . indeed , there are two parameters correlated to the luminosity . the first one is the decline rate : sne ia with fast decline rates are fainter and have narrower light - curve peaks @xcite and the second one , the colour @xcite : redder sne ia are fainter . the standardisation of sne ia to a level @xmath3 0.150.2 mag @xcite , led to the measurement of the expansion history of the universe and showed that , contrary to expectations , the universe is undergoing an accelerated expansion @xcite . within this new paradigm , one of the greatest challenges is the search for the mechanism that causes the acceleration , an endeavour that will require exquisitely precise measurements of the cosmological parameters that characterise the current cosmological concordance model , i.e. , @xmath6@xmath7 model . several techniques that offer the promise to provide such constraints have been put forward over recents years : refined versions of the sne ia method ( @xcite ) , cosmic microwave background radiation measurements ( cosmic microwave background explorer , @xcite ; wilkinson microwave anisotropy probe , @xcite ; and more recently the planck mission , @xcite ) , and baryon acoustic oscillation measurements ( @xcite ) . all of the above techniques have their own merits , but also their own systematic uncertainties that could become dominant with the increasingly higher level of precision required . thus , it is important to develop as many methods as possible , since the truth will likely emerge from the combination of different independent approaches . + while sne ia have been used as the primary diagnostic in constraining cosmological parameters , type iip supernovae ( sne iip ) have also been established to be useful independent distance indicators . sne iip are 12 mag less luminous than the sne ia however , their intrinsic rate is higher than sne ia rate @xcite , and additionally the rate peaks at higher redshifts than sne ia @xcite , which motivates their use in the cosmic distance scale ( see @xcite ) . also the fact that in principle they are the result of the same physical mechanism , and their progenitors are better understood than those of sne ia , further encourages investigations in this direction . sne iip are thought to be core - collapse supernovae ( ccsne ) , i.e. , the final explosion of stars with zero - age main - sequence mass @xmath8 8 @xmath9 @xcite . ccsne have diverse classes , with a large range of observed luminosities , light - curve shapes , and spectroscopic features . ccsne are classified in two groups according to the absence ( sne ib / c : @xcite ) or presence ( sne ii ) of lines ( @xcite and references therein ) . additional of the sne iip and sne iil which are discussed later , sne ii are composed by sne iib which evolve spectroscopically from sne iip at early time to deficient few weeks to a month past maximum @xcite and sne iin which have narrow emission lines ( @xcite ) . historically , sne ii were separated in two groups : sne iip ( 70% of ccsne ; @xcite ) , which are characterised by long duration plateau phases ( @xmath10 100 days ) of constant luminosity , and sne iil which have linearly declining light - curve morphologies @xcite . however , as discussed in detail in @xcite , it is not clear how well this terminology describes the diversity of sne ii . there are few sne ii which show flat light - curves , and in addition there are very few ( if any ) sne which decline linearly before falling onto the radioactive tail . therefore , henceforth we simply refer to all sne with distinct decline rates collectively as sne ii , and later further discuss sne in terms of their `` @xmath11 '' plateau decline rates @xcite . @xcite also suggested that the sne ii family forms a continuous class , while @xcite and @xcite have argued for two separate populations . + the most noticeable difference between sne ii occurs during the plateau phase . the optically thick phase is physically well - understood and is due to a change in opacity and density in the outermost layers of the sn . at the beginning the hydrogen present in the outermost layers of the progenitor star is ionised by the shock wave , which implies an increase of the opacity and the density which prevent the radiation from the inner parts to escape . after a few weeks , the star has cooled to the temperature allowing the recombination of ionised hydrogen ( higher than 5000 k due to the large optical depth ) . the ejecta expand and the photosphere recedes in mass space , releasing the energy stored in the corresponding layers . the plateau morphology requires a recession of the photosphere in mass that corresponds to a fixed radius in space so that luminosity appears constant . as @xcite show , this delicate balance is rarely observed and there is significant diversity observed in the @xmath0-band light - curve . to reproduce the plateau morphology , hydrodynamical models have used red supergiant progenitors with extensive h envelopes @xcite . direct detections of the progenitor of sne iip have confirmed these models ( @xcite ) . it has also been suggested that sn iil progenitors may be more massive in the zero age main sequence than sne iip @xcite and with smaller hydrogen envelopes @xcite . + to date several methods have been developed to standardise sne ii . the first method called the `` expanding photosphere method '' ( epm ) was developed by @xcite and allows one to obtain the intrinsic luminosity assuming that sne ii radiate as dilute blackbodies , and that the sn freely expands with spherical symmetry . the epm was implemented for the first time on a large number of objects by @xcite and followed by many studies @xcite . one of the biggest issues with this method is the epm only works if one corrects for the blackbody assumptions which requires corrections factors computed from model atmospheres ( @xcite and see @xcite for the resolution of the epm - based distance problem to sn 1999em ) . also to avoid the problem in the estimation of the dilution factor , @xcite proposed a distance correcting factor that takes into account the departure of the sn atmosphere from a perfect blackbody , the `` spectral - fitting expanding atmosphere method '' ( seam , updated in @xcite ) . this method consists of fitting the observed spectrum using an accurate synthetic spectrum of sne ii , and then since the spectral energy distribution is completely known from the calculated synthetic spectra , one may calculate the absolute magnitude in any band . + a simpler method , also based on photometric and spectroscopic parameters , the `` standardised candle method '' ( scm ) was first introduced by @xcite . they found that the luminosity and the expansion velocity are correlated when the sn is in its plateau phase ( 50 days post explosion ) . this relation is physically well understood : for a more luminous sn , the hydrogen recombination front will be at a larger radius thereby the velocity of the photosphere will be greater @xcite for a given post - explosion time . thanks to this method the scatter in the hubble diagram ( hereafter hubble diagram ) drops from 0.8 mag to 0.29 mag in the @xmath12-band . @xcite improved this method by adding an extinction correction based on the @xmath13 colour at day 50 after maximum . this new method is very powerful and many other studies @xcite have confirmed the possibility to use sne ii as standard candles finding a scatter between 10 and 18% in distance . recently @xcite suggested that using near - infrared ( nir ) filters , the scm , the dispersion can drop to a level of 0.10.15 mag ( using 12 sne iip ) . indeed , in the nir the host - galaxy extinction is less important , thus there may be less scatter in magnitude . note also the work done by @xcite where the authors used the photospheric magnitude method ( pmm ) which correspond to a generalisation of the scm for various epochs throughout the photospheric phase and found a dispersion of 0.12 mag using 13 sne . this is an intrinsic dispersion and is not the rms . + the main purpose of this work is to derive a method to obtain purely photometric distances , i.e , standardise sne ii only using light - curves and colour - curve parameters , unlike other methods cited above which require spectroscopic parameters . this is a big issue , and purely photometric methods will be an asset for the next generation of surveys such as the large synoptic survey telescope ( lsst ; @xcite ) . these surveys will discover such a large number of sne that spectroscopic follow - up will be impossible for all but only for small number of events . this will prevent the use of current methods to standardise sne ii and calculate distances . therefore deriving distances with photometric data alone is important and useful for the near future but also allows us to reach higher distance due to the fact that getting even one spectrum for a sn ii at @xmath14 is very challenging . + the paper is organised as follows . in section 2 a description of the data set is given and in section 3 we explain how the data are corrected for milky way ( mw ) extinction and how the k - correction is applied . in section 4 we describe the photometric colour method ( pcm ) using optical and nir filters and we derive a photometric hubble diagram . in section 5 we present a comparative hubble diagram using the scm . in section 6 we compare our method with the scm and we conclude with a summary in section 7 . the _ _ carnegie supernova project _ _ ( csp , @xcite ) provided all the photometric and spectroscopic data for this project . the goal of the csp was to establish a high - cadence data set of optical and nir light - curves in a well - defined and well - understood photometric system and obtain optical spectra for these same sne . between 2004 and 2009 , the csp observed many low redshift sne ii ( @xmath15 with @xmath16 ) , 56 had both optical and nir light - curves with good temporal coverage ; one of the largest nir data samples . two sn 1987a - like events were removed ( sn 2006v and sn 2006au see @xcite ) living the sample listed in table [ parameters ] with photometric parameters measured by @xcite . note that we do not include sne iib or sne iin . + cccccccc sn & avg & v@xmath17 & v@xmath18 & explosion date & @xmath19 & @xmath11 & optd + & ( mag)&(km s@xmath20)&(km s@xmath20)&(mjd)&(mag 100d@xmath20)&(mag 100d@xmath20)&(days ) + 2004ej & 0.189 & 2723(6 ) & 3045(23 ) & 53224.90(5 ) & @xmath21 & 1.07(0.04 ) & 96.14 + 2004er & 0.070 & 4411(33 ) & 4186(37 ) & 53271.80(4 ) & 1.28(0.03 ) & 0.40(0.03 ) & 120.15 + 2004fc & 0.069 & 1831(5 ) & 1560(20 ) & 53293.50(10 ) & @xmath21 & 0.82(0.02 ) & 106.06 + 2004fx & 0.282 & 2673(3 ) & 2679(3 ) & 53303.50(4 ) & @xmath21 & 0.09(0.03 ) & 68.40 + 2005j & 0.075 & 4183(1 ) & 4530(24 ) & 53382.78(7 ) & 2.11(0.07 ) & 0.96(0.02 ) & 94.03 + 2005z & 0.076 & 5766(10 ) & 6088(25 ) & 53396.74(8 ) & @xmath21 & 1.83(0.01 ) & 78.84 + 2005an & 0.262 & 3206(31 ) & 3541(39 ) & 53428.76(4 ) & 3.34(0.06 ) & 1.89(0.05 ) & 77.71 + 2005dk & 0.134 & 4708(25 ) & 4618(26 ) & 53599.52(6 ) & 2.26(0.09 ) & 1.18(0.07 ) & 84.22 + 2005dn & 0.140 & 2829(17 ) & 2693(20 ) & 53601.56(6 ) & @xmath21 & 1.53(0.02 ) & 79.76 + 2005dw & 0.062 & 5269(10 ) & 4974(23 ) & 53603.64(9 ) & @xmath21 & 1.27(0.04 ) & 92.59 + 2005dx & 0.066 & 8012(31 ) & 7924(31 ) & 53615.89(7 ) & @xmath21 & 1.30(0.05 ) & 85.59 + 2005dz & 0.223 & 5696(8 ) & 5327(27 ) & 53619.50(4 ) & 1.31(0.08 ) & 0.43(0.04 ) & 81.86 + 2005es & 0.228 & 11287(49 ) & 10917(55 ) & 53638.70(10 ) & @xmath21 & 1.31(0.05 ) & @xmath21 + 2005gk & 0.154 & 8773(10 ) & 8588(30 ) & @xmath21 & @xmath21 & 1.25(0.07 ) & @xmath21 + 2005hd & 0.173 & 8323(10 ) & 8246(30 ) & @xmath21 & @xmath21 & 1.83(0.13 ) & @xmath21 + 2005lw & 0.135 & 7710(29 ) & 8079(39 ) & 53716.80(10 ) & @xmath21 & 2.05(0.04 ) & 107.23 + 2006y & 0.354 & 10074(10 ) & 10220(30 ) & 53766.50(4 ) & 8.15(0.76 ) & 1.99(0.12 ) & 47.49 + 2006ai & 0.347 & 4571(10 ) & 4637(30 ) & 53781.80(5 ) & 4.97(0.17 ) & 2.07(0.04 ) & 63.26 + 2006bc & 0.562 & 1363(10 ) & 1476(13 ) & 53815.50(4 ) & 1.47(0.18 ) & -0.58(0.04 ) & @xmath21 + 2006be & 0.080 & 2145(9 ) & 2243(11 ) & 53805.81(6 ) & 1.26(0.08 ) & 0.67(0.02 ) & 72.89 + 2006bl & 0.144 & 9708(49 ) & 9837(50 ) & 53823.81(6 ) & @xmath21 & 2.61(0.02 ) & @xmath21 + 2006ee & 0.167 & 4620(19 ) & 4343(27 ) & 53961.88(4 ) & @xmath21 & 0.27(0.02 ) & 85.17 + 2006it & 0.273 & 4650(9 ) & 4353(23 ) & 54006.52(3 ) & @xmath21 & 1.19(0.13 ) & @xmath21 + 2006ms & 0.095 & 4543(18 ) & 4401(21 ) & 54034.00(13 ) & 2.07(0.30 ) & 0.11(0.48 ) & @xmath21 + 2006qr & 0.126 & 4350(5 ) & 4642(21 ) & 54062.80(7 ) & @xmath21 & 1.46(0.02 ) & 96.85 + 2007p & 0.111 & 12224(25 ) & 12570(35 ) & 54118.71(3 ) & @xmath21 & 2.36(0.04 ) & 84.33 + 2007u & 0.145 & 7791(9 ) & 7795(9 ) & 54134.61(6 ) & 2.94(0.02 ) & 1.18(0.01 ) & @xmath21 + 2007w & 0.141 & 2902(2 ) & 3215(22 ) & 54136.80(7 ) & @xmath21 & 0.12(0.04 ) & 77.29 + 2007x & 0.186 & 2837(6 ) & 3055(16 ) & 54143.85(5 ) & 2.43(0.06 ) & 1.37(0.03 ) & 97.71 + 2007aa & 0.072 & 1465(4 ) & 1826(26 ) & 54135.79(5 ) & @xmath21 & -0.05(0.02 ) & 67.26 + 2007ab & 0.730 & 7056(13 ) & 7091(13 ) & 54123.86(6 ) & @xmath21 & 3.30(0.08 ) & 71.30 + 2007av & 0.099 & 1394(3 ) & 1742(24 ) & 54175.76(5 ) & @xmath21 & 0.97(0.02 ) & @xmath21 + 2007hm & 0.172 & 7540(15 ) & 7241(26 ) & 54335.64(6 ) & @xmath21 & 1.45(0.04 ) & @xmath21 + 2007il & 0.129 & 6454(10 ) & 6146(24 ) & 54349.77(4 ) & @xmath21 & 0.31(0.02 ) & 103.43 + 2007oc & 0.061 & 1450(5 ) & 1184(19 ) & 54382.51(3 ) & @xmath21 & 1.83(0.01 ) & 77.61 + 2007od & 0.100 & 1734(3 ) & 1377(25 ) & 54402.59(5 ) & 2.37(0.05 ) & 1.55(0.01 ) & @xmath21 + 2007sq & 0.567 & 4579(4 ) & 4874(21 ) & 54421.82(3 ) & @xmath21 & 1.51(0.05 ) & 88.34 + 2008f & 0.135 & 5506(21 ) & 5305(25 ) & 54470.58(6 ) & @xmath21 & 0.45(0.10 ) & @xmath21 + 2008k & 0.107 & 7997(10 ) & 8351(27 ) & 54477.71(4 ) & @xmath21 & 2.72(0.02 ) & 87.1 + 2008 m & 0.124 & 2267(4 ) & 2361(8 ) & 54471.71(9 ) & @xmath21 & 1.14(0.02 ) & 75.34 + 2008w & 0.267 & 5757(45 ) & 6041(49 ) & 54485.78(6 ) & @xmath21 & 1.11(0.04 ) & 83.86 + 2008ag & 0.229 & 4439(6 ) & 4428(6 ) & 54479.85(6 ) & @xmath21 & 0.16(0.01 ) & 102.95 + 2008aw & 0.111 & 3110(4 ) & 3438(23 ) & 54517.79(10 ) & 3.27(0.06 ) & 2.25(0.03 ) & 75.83 + 2008bh & 0.060 & 4345(8 ) & 4639(22 ) & 54543.54(5 ) & 3.00(0.27 ) & 1.20(0.04 ) & @xmath21 + 2008bk & 0.054 & 230(4 ) & -50(20 ) & 54542.89(6 ) & @xmath21 & 0.11(0.02 ) & 104.83 + 2008bu & 1.149 & 6630(9 ) & 6683(10 ) & 54566.78(5 ) & @xmath21 & 2.77(0.14 ) & 44.75 + 2008ga & 1.865 & 4639(3 ) & 4584(5 ) & 54711.85(4 ) & @xmath21 & 1.17(0.08 ) & 72.79 + 2008gi & 0.181 & 7328(34 ) & 7103(37 ) & 54742.72(9 ) & @xmath21 & 3.13(0.08 ) & @xmath21 + 2008gr & 0.039 & 6831(41 ) & 6549(46 ) & 54766.55(4 ) & @xmath21 & 2.01(0.01 ) & @xmath21 + 2008hg & 0.050 & 5684(10 ) & 5449(19 ) & 54779.75(5 ) & @xmath21 & -0.44(0.01 ) & @xmath21 + 2009n & 0.057 & 1036(2 ) & 1386(25 ) & 54846.79(5 ) & @xmath21 & 0.34(0.01 ) & 89.50 + 2009ao & 0.106 & 3339(5 ) & 3665(23 ) & 54890.67(4 ) & @xmath21 & -0.01(0.12 ) & 41.71 + 2009bu & 0.070 & 3494(9 ) & 3372(13 ) & 54907.91(6 ) & 0.98(0.16 ) & 0.18(0.04 ) & @xmath21 + 2009bz & 0.110 & 3231(7 ) & 3393(13 ) & 54915.83(4 ) & @xmath21 & 0.50(0.02 ) & @xmath21 + [ parameters ] all the photometric observations were taken at the las campanas observatory ( lco ) with the henrietta swope 1-m and the irne du pont 2.5-m telescopes using optical ( @xmath22 , @xmath23 , @xmath2 , @xmath24 , @xmath4 , and @xmath0 ) , and nir filters ( @xmath25 , @xmath26 , and @xmath27 , see @xcite ) . all optical images were reduced in a standard way including bias subtractions , flat - field corrections , application of a linearity correction and an exposure time correction for a shutter time delay . the nir images were reduced through the following steps : dark subtraction , flat - field division , sky subtraction , geometric alignment and combination of the dithered frames . due to the fact that sn measurements can be affected by the underlying light of their host galaxies , we took care in correctly removing the underlying host - galaxy light . the templates used for final subtractions were always taken months / years after each sn faded and under seeing conditions better than those of the science frames . because the templates for some sne were not taken with the same telescope , they were geometrically transformed to each individual science frame . these were then convolved to match the point - spread functions , and finally scaled in flux . the template images were then subtracted from a circular region around the sn position on each science frame ( see @xcite ) . + observed magnitudes for each sn was derived relative to local sequence stars and calibrated from observations of standard stars in the @xcite ( @xmath28 ) , @xcite ( @xmath29 ) , and @xcite ( @xmath30 ) systems . the photometry of the local sequence stars are on average based on at least three photometric nights . magnitudes are expressed in the natural photometric system of the swope+csp bands . final errors for each sn are the result of the instrumental magnitude uncertainty and the error on the zero point . the full photometric catalog will be published in an upcoming paper ( note that the @xmath0-band photometry has been already published in @xcite ) . + the majority of our spectra were obtained with the 2.5 m irne du pont telescope using the wfccd- and boller and chiven spectrographs ( the last is now decommissioned ) at lco . additional spectra were obtained with the 6.5 m magellan clay and baade telescopes with ldss-2 , ldss-3 , mage ( see @xcite for details ) and imacs together with the ctio 1.5 m telescope and the ritchey - chrtien cassegrain spectrograph , and the new technology telescope ( ntt ) at la silla observatory using the emmi and efosc instruments . the majority of the spectra are the combination of three exposures to facilitate cosmics - ray rejection . information about the grism used , the exposure time , the observation strategy can be found in @xcite . all spectra were reduced in a standard way as described in @xcite and @xcite . briefly , the reduction was done with iraf using the standard routines ( bias subtraction , flat - field correction , 1-d extraction , wavelength and flux calibration ) . the full spectroscopic sample will be published in an upcoming paper and the reader can refer to @xcite and @xcite for a thorough analysis of this sample . in order to proceed with our aim of creating a hubble diagram based on photometric measurements using the pcm , in this section we show how to correct apparent magnitudes for mw extinction ( avg ) and how to apply the k - correction , without the use of observed sn spectra but only with model spectra . in the @xmath0 band the determination of avg can be applied using the extinction maps of @xcite . to convert avg to extinction values in other bands we need to adopt : an extinction law and the effective wavelength for each filter . + sn ii spectra evolve with time from a blue continuum at early times to a redder continuum with many absorption / emission features at later epochs . this implies that the effective wavelength of a broad - band filter also changes with time ( see the formula given @xcite a.21 ) . to calculate effective wavelengths at different epochs we adopt a sequence of theoretical spectral models from @xcite consisting of a sn progenitor with a main - sequence mass of 15 @xmath9 , solar metallicity z=0.02 , zero rotation and a mixing - length parameter of 3 . the choice of this model is based on the fact that it provided a good match to a prototypical sn ii such as sn 1999em . for each photometric epoch , we choose the closest theoretical spectrum in epoch since the explosion , the extinction law from @xcite and in time @xmath31 to obtain the mw extinction in the other filters . + having corrected the observed magnitudes for galactic extinction , we need to apply also a correction attributable to the expansion of the universe called the k - correction ( kc ) . a photon received in one broad photometric band pass in the observed referential has not necessarily been emitted ( rest frame referential ) in the same filter , that is why this correction is needed . for each epoch of each filter we use the same procedure to estimate the kc . here we describe our method step by step for one epoch and a given filter x. 1 . we choose in our model spectral library ( @xcite , model m15mlt3 ) the theoretical spectrum ( rest frame ) closest to the photometric epoch since explosion time ( corrected for time dilatation ) , with a rest - frame spectral energy distribution ( sed ) , @xmath32 . because our library covers a limited range of epochs from 12.2 to 133 days relative to explosion , observations outside these limits are ignored . 2 . we bring the rest - frame theoretical spectrum to the observer s frame using @xmath33 the correction , where @xmath34 is the heliocentric redshift of the sn , @xmath35 where @xmath36 is the wavelength in the observer s frame . we match the theoretical spectrum to the observed photometric magnitudes of the sn @xcite . for this we calculate synthetic magnitudes ( from the model in the observer s frame,@xmath37 ) and compare them to the observed magnitudes corrected for mw extinction . we use all the filters available at this epoch . then we obtain a warping function @xmath38 ( quadratic , cubic , depending on the number of filters used ) and do a constant extrapolation for the wavelengths outside of the range of filters used . with our warping function we correct our model spectrum and obtain @xmath39 . we compute the magnitude in the observer s frame : + @xmath40 + with c the light velocity in @xmath41 s@xmath20 , @xmath42 the planck constant in ergs s , @xmath36 is wavelength , @xmath43 the transmission function of filter @xmath44 and @xmath45 is the zero point of filter @xmath44 ( see @xcite ) . we bring back the warping spectrum to the rest frame @xmath46 and we obtain and calculate the magnitude : + @xmath47 5 . finally we obtain the kc for this epoch as the difference between the observed and the rest frame magnitude , @xmath48 . 6 . to estimate the associated errors , we follow the same procedure but instead of using the observed magnitudes for the warping , we use the upper limit , i.e. , observed magnitudes plus associated uncertainties . as a complementary work on the kc and to validate our method we compare the kc values found using the @xcite model to those computed from our database of observed spectra . in both cases we use exactly the same procedure . first the observed spectrum is corrected in flux using the observed photometry ( corrected for avg ) in order to match the observed magnitudes . the photometry is interpolated to the spectral epoch . in figure [ model_obs ] we show a comparison between the kc obtained with the theoretical models and using our library of observed spectra at different redshifts . as we can see that the kc values calculated with both methods are very consistent . this exercise validates the choice of using the @xcite models to calculate the kc . there are two advantages to use the theoretical models . first we can obtain the kc for nir filters ( @xmath25 , @xmath26 , @xmath27 ) for which we do not have observed spectra , and secondly this method does not require observed spectra which are expensive to obtain in terms of telescope time , and virtually impossible to get at higher redshifts . band . the black dotted line represents x = y . each square represents one observed spectrum of our database . the colour bar on the right side represents the different redshifts.,width=377 ] in this section we present our pcm with which we derive the corrected magnitudes necessary for constructing the hubble diagram solely with photometric data . since we want to examine hubble diagrams from photometry obtained at different epochs , we start by linearly interpolating colours on a daily basis from colours observed at epochs around the epoch of interest . the same procedure is used to interpolate magnitudes . + to correct and standardise the apparent magnitude we use two photometric parameters : @xmath11 which is the slope of the plateau measured in the @xmath0-band @xcite , and a colour term at a specific epoch . the colour term is mainly used to take into account the dispersion caused by the host - galaxy extinction . the magnitude is standardised using a weighted least - squares routine by minimising the equation below : @xmath49 where @xmath50 is the speed of light , @xmath51 the redshift , @xmath52 the observed magnitudes with different filters , and corrected for avg and kc , while @xmath53 , @xmath54 and @xmath55 are free fitting parameters . the errors on these parameters are derived assuming a reduced chi square equal to one . in order to obtain the errors on the standardised magnitudes , an error propagation is performed in an iterative manner . note that @xmath54 is related to host - galaxy @xmath56 if we assume that the colour - magnitude relation is due to extrinsic factors ( the intrinsic colour is degenerate with the @xmath55 ) . we obtain : @xmath57 where @xmath58 is the host - galaxy extinction in the @xmath59 filter and @xmath60 the colour excess . assuming a @xcite law , there is one to one relationship between @xmath56 and @xmath54 . first we obtain the theoretical @xmath61 for different @xmath56 values using the @xcite coefficients ( a and b ) : @xmath62 then we derive @xmath56 from the value of @xmath63 determined from the least - squares fit ( eq 1 ) . we will discuss the resulting @xmath56 values in section 6.5 . + we select only sne located in the hubble flow , i.e. , with @xmath64 @xmath8 3000 km s@xmath20 in order to minimize the effect of peculiar galaxy motions . our available sample is composed of the entire sample in the hubble flow but 3 sne . we eliminate two sne due to the fact that the warping function can not be computed , thus the k - correction ( sn 2004ej and sn 2008k ) . we also take out the outlier sn 2007x and found for this object particular characteristics like clear signs of interaction with the circumstellar medium ( flat h alpha p - cygni profile , see gutirrez et al . in prep . ) . + sne ii are supposedly characterised by similar physical conditions ( e.g. temperature ) when they arrive towards the end of the plateau @xcite that is why we use the end of the optically thick phase measured in the @xmath0 band ( as defined by @xcite ) as the time origin in order to bring all sne to the same time scale . when the end of the plateau is not available we choose 80 days post explosion , which is the average for our sample . + given that sne ii show a significant dispersion in the plateau duration driven by different evolution speeds , we decide to take a fraction of the plateau duration and not an absolute time , to ensure that we compare sne ii at the same evolutionary phase . thus , in the following analysis , we adopt optd*x% as the time variable where optd is the optically thick phase duration and x is percentage ranging between 1 - 100% . + in figure [ rms_time ] we present the variation with evolutionary phase of the dispersion in the hubble diagram using the filters available and the @xmath1 colour . the lowest root mean square ( rms ) values in the optical is found for the @xmath2 band , and at nir wavelength using the @xmath65 band . note that the coverage in the @xmath25 band is better than in the @xmath26 band hence , hereafter we use the @xmath25 band . for these two bands we can obtain the median rms over all the epochs ( from 0.2*optd to 1.0*optd ) and the standard deviation . we find for @xmath2 band 0.47 @xmath66 0.04 mag and for @xmath25 band 0.48 @xmath66 0.04 mag . in figure [ rms_time_couleur ] we do as above but this time we change colours . fixing the @xmath2 band and using different colours we show the variation of the rms . this figure shows that the colour that minimises the rms is @xmath1 ( @xmath67 yields a lower dispersion but the time coverage is significantly less ) . we find a median rms over all the epochs of 0.47 @xmath66 0.05 . for this reason we decide to combine the @xmath2 band and the @xmath1 colour for the hubble diagram . note also that the best epoch for the @xmath2 band is close to the middle of the plateau , 55% of the time from the explosion to the end of the plateau , whereas in the @xmath25 band is later in phase post explosion , around 65% . in general the best epoch to standardise the magnitude is between 6070% of the optd for nir filters and for optical filters between 5060% of optd . physically these epochs correspond in both cases more or less to the middle of the plateau . note that we tried other time origin such as the epoch of maximum magnitude instead of the end of the plateau but changing the reference does not lower the rms . + in figure [ hubble diagram_hubble_flow ] we present a hubble diagram based entirely on photometric data using @xmath11 and colour term for two filters , @xmath2 band and @xmath25 band . in the @xmath2 band the rms is 0.44 mag ( with 38 sne ) which allows us to measure distances with an accuracy of @xmath3 20% . we find the same precision using the @xmath68-band with a rms of 0.43 mag ( 30 sne ) . note that the colour term is more important for the optical filter than for the nir filter . indeed , for the @xmath2 band the rms decreases from 0.50 to 0.44 mag when the colour term is added whereas for the nir filter the improvement is only of 0.004 mag . using all available epochs we find a mean improvement of 0.025 @xmath66 0.011 in @xmath2 band and 0.014 @xmath66 0.013 in @xmath25 band . this shows that the improvement is significant in optical but less in nir . the drop using the optical filter is not surprising because this term is probably at least partly related to host - galaxy extinction which is more prevalent in optical wavelengths than in the nir , so adding a colour term for nir filters does not significantly influence the dispersion . note that if we use the weighted root mean square ( wrms ) as defined by @xcite we find 0.40 mag and 0.36 mag for the @xmath2 band and @xmath25 band , respectively , after @xmath11 and colour corrections . + in the literature the majority of the studies used sne iip for their sample . to check if we can include all the sne ii ( fast- and slow - decliners ) we did some analysis of the sne and investigate if any of the higher residuals arise from intrinsic sn properties . the overall conclusion is that at least to first order , we did not find any correlation between sne ii intrinsic differences ( @xmath11 , optd , ... ) and the hubble residuals . this suggests that sne within the full range of @xmath11 values ( i.e. , all sne ii ) should be include in hubble diagram . + following the work of @xcite for sne ia , we investigated the combined hubble diagram using all the filters available ( by averaging the distance moduli derived in each filter ) but the dispersion obtained is not much better . we found the same correlation between the distance - modulus residuals in one band versus those in another band as found by @xcite , suggesting that the inclusion of multiple bands does not improve the distance estimate . + if we include sne in the hubble flow ( @xmath69 3000 km s@xmath20 ) and very nearby sne ( @xmath70 3000 km s@xmath20 ) for the @xmath2 band the dispersion increases from 0.44 mag to 0.48 mag ( 46 sne ) whereas in the @xmath25 band the rms increase from 0.43 mag to 0.45 ( 41 sne ) . + we also try to use two different epochs , one for the magnitude and the other for the colour but again , this does not improve the rms . finally , we try also to use the total decline rate ( between maximum to the end of the plateau ) instead of the plateau slope . using the total decline rate does not lower the rms ( dispersion around 0.47 mag for 45 sne in the @xmath2 band ) but could be useful for high redshift sne . . in the x - axis we present the time as ( explosion time+optd*x% ) . the black squares present the @xmath4 band , dark blue circle the @xmath23 band , blue cross the @xmath0 band , dark green diamonds the @xmath2 band , green hexagons for the @xmath24 band , yellow pentagons for the @xmath25 band , red plus symbol for the @xmath26 band . the @xmath27 band is not presented because the sampling is as good as it is in the other bands.,width=359 ] band and different colours . in the x - axis we present the time as the optd*x% . the black stars present @xmath71 colour , dark blue squares are for @xmath72 , blue circle @xmath73 , cyan cross @xmath74 , green diamonds @xmath75 , yellow pentagons for @xmath67 , and red hexagons for @xmath1.,width=359 ] + + + the scm as employed by various authors gives a hubble diagram dispersion of 0.250.30 mag @xcite . here we present the hubble diagram using the scm for our sample . to apply the scm , we need to measure the velocity of the sn ejecta . one of the best features is @xmath76 because other iron lines such as @xmath77 can be blended by other elements . expansion velocities are measured through the minimum flux of the absorption component of p - cygni line profile after correcting the spectra for the heliocentric redshifts of the host - galaxies . errors were obtained by measuring many times the minimum of the absorption changing the trace of the continuum . the range of velocities is 18008000 km s@xmath20 for all the sne . because we need the velocities for different epochs in order to find the best epoch ( as done for the pcm ) , i.e. , with less dispersion , we do an interpolation / extrapolation using a power law @xcite of the form : @xmath78 where @xmath79 and @xmath80 are two free parameters obtained by least - squares minimisation for each individual sn and @xmath81 the epoch since explosion . in order to obtain the velocity error , we perform a monte carlo simulation , varying randomly each velocity measurement according to the observed velocity uncertainties over more than 2000 simulations . from this , for each epoch ( from 1 to 120 days after explosion ) we choose the velocity as the average value and the incertainty to the standard deviation of the simulations . the median value of @xmath80 is @xmath820.55@xmath66 0.25 . this value is comparable with the value found by other authors ( @xmath820.5 for @xcite and @xmath820.464 by @xcite and @xmath820.546 by @xcite ) . note that , as found by @xcite , the iron velocity for the fast - decliners ( sne iil ) also follow a power law but with more scatter . indeed for the slow - decliners ( @xmath83 ) we find a median value , @xmath84 whereas for the fast - decliners ( @xmath85 ) we obtain @xmath86 . more details will be published in an upcoming paper ( gutirrez et al . ) . to standardise the apparent magnitude , we perform a least - squares minimisation on : @xmath87 + where @xmath50 , @xmath51 , @xmath88 are defined in section 4.1 and @xmath53 , @xmath54 , and @xmath55 are free fitting parameters . the errors on the magnitude are obtained in the same way as for the pcm but the epoch is different . for the scm , the photospheric expansion velocity is very dependent on the explosion date that is why after trying different epochs and references , we found that the best reference is the explosion time as used in @xcite , @xcite , @xcite . the same epoch for the magnitude , the colour and the iron velocity is employed . just like for the pcm , we use the same colour , @xmath1 , and the same filters ( @xmath2 , @xmath25 band ) . for some sne we are not able to measure an iron velocity due to the lack of spectra ( only one epoch ) and our sample is thus composed of 26 sne . in figure [ hubble diagram_scm ] we present the hubble diagram and the residual for two different filters . the dispersion is 0.29 mag ( or 0.300.28 mag in wrms for the @xmath25 band and @xmath2 band respectively ) for 24 sne ( some sne do not have colour at this epoch ) . these values are some what better than previous studies @xcite where the authors found dispersions around 0.30 mag with 30 sne ( more details in section 6.3 ) . note the major differences between our study and theirs is that they included very nearby sne ( @xmath70 3000 km s@xmath20 ) , only slow - declining sne ii ( sne ii with low @xmath11 , historically referred to as sne iip ) , did not calculate a power - law for each sn as we do , and used a different epoch . note also the work done by @xcite where they applyed the scm to nir filters ( @xmath26-band and @xmath89 colour ) using nearby sne ( 92% of their sample with @xmath70 3000 km s@xmath20 ) , finding a dispersion of 0.39 mag with 12 sne ( see section 6.3 ) . to finish we tried a combination of the pcm and scm , i.e. , adding a @xmath11 term to the scm but this does not improve the dispersion . ) and the epoch chosen with respect to explosion date in days . both plots present the hubble diagram using the sne in the hubble flow . on the top we present the hubble diagram using the @xmath2 band and the colour @xmath1 . on the bottom is the same but we use a nir filter , @xmath25 band . on the bottom of each plot we present the residual . in the residuals plot , the dashed line correspond to the rms.,title="fig:",width=340 ] + ) and the epoch chosen with respect to explosion date in days . both plots present the hubble diagram using the sne in the hubble flow . on the top we present the hubble diagram using the @xmath2 band and the colour @xmath1 . on the bottom is the same but we use a nir filter , @xmath25 band . on the bottom of each plot we present the residual . in the residuals plot , the dashed line correspond to the rms.,title="fig:",width=340 ] above we demonstrate that using two terms , @xmath11 and a colour , we are able to obtain a dispersion of 0.43 mag ( optical bands ) . in this section we try to reduce the rms by using well - observed sne and we compare the pcm to the scm . we also discuss comparisons between the scm using the csp sample with other studies . because the value of the rms is the crucial parameter to estimate the robustness of the method , we also discuss statistical errors . finally , we briefly present the values of @xmath56 derived from the colour term both from pcm and scm . a significant fraction of values from @xcite do not correspond to the slope of the plateau but sometimes to a combination of @xmath19 ( initial decline ) and @xmath11 . indeed , for some sne , it was impossible to distinguish two slopes and the best fit was only one slope . for this reason we decide to define a new sample composed only by 12 sne with values of @xmath19 and @xmath11 and with @xmath64 @xmath8 3000 km s@xmath20 . from this sample and using the @xmath2/@xmath1 combination we obtain a dispersion of 0.39 mag with 12 sne , which compares to 0.48 mag from the entire sample . from the @xmath25 band the dispersion drops considerably from 0.44 to 0.18 mag with only 8 sne . however this low value should be taken with caution due to possible statistical effects which are discussed later ( see section 6.4 ) . in figure [ hubble diagram_compa_r ] and in figure [ hubble diagram_compa_y ] we compare the hubble diagram obtained using the scm and the pcm . for both methods we use the same sne ( hubble flow sample ) , and the same set of magnitude - colour . the dispersion using the @xmath2 band and the @xmath25 band is 0.43 mag for the pcm whereas for the scm is 0.29 . + in general the scm is more precise than the pcm but the dispersion found with the pcm is consistent with the results found by the theoretical studies done by @xcite ( distances accurate to @xmath3 20% ) but the authors used other photometric correlations ( plateau duration ) . unfortunately , as suggested by @xcite using this parameter the prediction is not seen in the observations . we tried to use the optd values as an input instead of the @xmath11 and we did not see any improvement on the dispersion . note also the recent work of @xcite , in which the authors found a correlation between the iron velocity and the @xmath12-band total decline rate . although in this paper we do not use the total decline rate but another quantity related to the plateau slope , our work confirms the possibility of using photometric parameters instead of spectroscopic . + in this section we compare our scm with other studies . first we use only optical filters to compare with @xcite and @xcite . both studies used the @xmath13 colour and also the @xmath12 band . note that @xcite also used the @xmath4 and @xmath0 band but here we consider only the @xmath12 band for consistency . @xcite found a dispersion of 0.38 mag using 40 slow - decliners . in our sample instead of using the @xmath12 band we used the sloan filter , @xmath24 band and @xmath1 colour . using our entire sample , i.e. , sne ( 37 sne in total for all the redshift range ) we derive a dispersion similar to @xcite of 0.32 mag ( epoch : 35 days after explosion ) . we can also compare the parameter @xmath53 derived from the fit . again we obtain a consistent value , @xmath90 whereas @xcite found @xmath91 . the other parameters are not directly comparable due to the fact that the authors assumed an intrinsic colour which is not the case in the current work . using a hubble constant ( @xmath92 ) equal to 70 km s@xmath20 mpc@xmath20 we can translate our zp to an absolute magnitude ( @xmath93 ) m@xmath94 = @xmath8217.12 @xmath66 0.10 mag that it is lower than the results obtained by @xcite ( m@xmath95 = @xmath8217.43 @xmath66 0.10 mag ) . this difference is probably due to the fact that the corrected magnitude has not been corrected for the intrinsic colour in our work . + using 30 slow - declining sne in the hubble flow and very nearby sne ( z between 0.00016 and 0.05140 ) , @xmath13 colour and the @xmath12 band , @xcite derived a dispersion of 0.32 mag which is the same that we obtained . however the parameters derived by @xcite are different . indeed using the same equation ( 5 ) , and the entire sample they obtained @xmath96 , @xmath97 and @xmath98 instead of @xmath90 , @xmath99 and @xmath100 for us . from their zp ( @xmath92=70 km s@xmath20 mpc@xmath20 ) we derive m@xmath95 = -18.00 @xmath66 0.07 mag ( m@xmath94 = -17.12 @xmath66 0.15 mag for us ) . when the authors restrict the sample to objects in the hubble flow , they end up with 20 sne and a dispersion of 0.30 mag . if we do the same cut , we find a dispersion of 0.29 for 24 sne . we obtain consistent dispersion for both samples using similar filters . note that reducing our sample to slow - decliners alone ( @xmath83 , the classical sne iip in other studies ) in the hubble flow does not improve the dispersion . as mentioned in section 5.3 , the difference in dispersion between @xcite and our study can be due , among other things , to the difference in epoch used , or that we calculate a power - law for each sn for the velocity . + with respect to the nir filters @xcite suggested that it may be possible to reduce the scatter in the hubble diagram to 0.10.15 mag and this should then be confirmed with a larger sample and more sne in the hubble flow . the authors used 12 slow - decliners but only one sn in the hubble flow . using the @xmath26 band and the colour @xmath89 they found a dispersion of 0.39 mag against 0.50 mag using the @xmath12 band . from this drop in the nir , the authors suggested that using this filter and more sne in the hubble flow could reduce the scatter from 0.25 - 0.3 mag ( optical studies ) to 0.10.15 mag . with the same filters used by @xcite , and using the hubble flow sample , we find a dispersion of 0.28 mag with 24 sne . this dispersion is 0.1 mag higher than that predicted by @xcite ( 0.10.15 mag ) . to derive the fit parameters , the authors assumed an intrinsic colour @xmath101 @xmath102 1 mag . they obtained @xmath103 and an absolute magnitude m@[email protected] @xmath66 0.25 mag ( @xmath92=70 km s@xmath20 mpc@xmath20 ) . if we use only the sne with @xmath64 @xmath8 3000 km s@xmath20 ( 24 sne ) , we find @xmath105 and zp = @xmath822.44 @xmath66 0.18 which corresponds to m@[email protected] @xmath66 0.18 mag assuming h@xmath5 = 70 km s@xmath20 mpc@xmath20 . if we include all sne at any redshift , the sample goes up to 34 sne and the dispersion is 0.31 mag . from all sne we derive @xmath106 and zp = @xmath822.44 @xmath66 0.20 which corresponds to m@xmath104=-18.21 @xmath66 0.20 . to conclude , the hubble diagram derived from the csp sample using the scm is consistent and some what better with those found in the literature . + more recently , @xcite proposed another method to derive a hubble diagram from sne ii . the pmm corresponds to the generalisation of the scm , i.e. , the distances are obtained using the scm at different epochs and then averaged . using the @xmath13 colour , and the filter @xmath0 , the authors found an intrinsic scatter of 0.19 mag . given that the intrinsic dispersion used by @xcite is a different metric than that used by us ( the rms dispersion ) we computed the latter from their data , obtaining 0.24 mag for 24 sne in the hubble flow . using the @xmath0 band and the @xmath1 colour and doing an average over several epochs we found a dispersion of 0.28 mag which is similar to the value found from the scm and comparable with the value derived by @xcite . from the @xmath25 band and @xmath1 colour we find an identical dispersion of 0.29 mag . in analysing the hubble diagram , the figure of merit is the rms and the holy grail is to obtain very low dispersion in the hubble diagram ( i.e. low distance errors ) . in our work we show that in the @xmath25 band we can achieve a rms around 0.430.48 mag using the hubble flow sample ( 30 sne ) and the entire sample ( 41 sne ) , whereas using the golden sample ( 8 sne ) we obtain a dispersion of 0.18 mag . it is important to know if this decrease in rms is due to the fact that we used well - studied sne within the golden sample or if it is due to the low number of sne . for this purpose we do a test using the monte carlo bootstrapping method . + from our hubble flow sample , we remove randomly one sn and compute the dispersion . we do that for 30000 simulations and the final rms corresponds to the median , and the errors to the standard deviation . then after removing one sn , we remove randomly two sne and again estimate the rms and the dispersion over 30000 simulations . we repeat this process until we have only 4 sne , i.e. , we remove from one sn to ( size available sample - 4 sne ) . for each simulation we compute a new model , i.e. , new fit parameters ( @xmath53 , @xmath61 , and @xmath55 ) . + from this test we conclude that when the number of sne is lower than 1012 sne the rms is very uncertain because the parameters ( i.e. , @xmath53 , @xmath61 , and @xmath55 ) start diverging ( see appendix ) . this implies that the rms is driven by the reduced number of objects so it is difficult to conclude if the model for the golden sample is better because the rms is smaller or because it is due to a statistical effect . as stated in section 4.1 , the @xmath54 colour term is related to the total - to - selective extinction ratio if the colour - magnitude relation is due to extrinsic factors ( dust ) . in the literature , for the mw , @xmath56 is known to vary from one line of sight to another , from values as low as 2.1 @xcite to values as large as 5.6 - 5.8 @xcite . in general for the mw , a value of 3.1 is used which corresponds to an average of the galactic extinction curve for diffuse interstellar medium ( ism ) . using the minimisation of the hubble diagram with a colour term , in the past decade the sne ia community has derived lower @xmath56 for host - galaxy dust than for the mw . indeed they found @xmath56 between 1.52.5 @xcite . this trend was also seen more recently using sne ii @xcite . this could be due to unmodeled effects such as a dispersion in the intrinsic colours ( e.g. @xcite ) . + we follow previous work in using the minimisation of the hubble diagram to obtain constraints on @xmath56 for host - galaxy dust . using the pcm , the hubble flow sample and the @xmath2 band , we find @xmath107 close to 0.98 . using a @xcite law we can transform this value in the total - to - selective extinction ratio , and we obtain , @xmath108 . following the same procedure but using the scm , we also derive low @xmath56 values , but consistent with those derived using the pcm . + at first sight , our analysis would suggest a significantly different nature of dust in our galaxy and other spiral galaxies , as previously seen in the analysis of sne ia and sne ii . however , we caution the reader that the low @xmath56 values could reflect instead intrinsic magnitude - colour for sne ii not properly modelled . to derive the @xmath56 ( or pseudo @xmath56 ) values we assume that all the sne ii have the same intrinsic colours and same intrinsic colour - luminosity relation , however theoretical models with different masses , metallicity , show different intrinsic colours @xcite . disentangling both effects would require to know the intrinsic colours of our sn sample . indeed , with intrinsic colour - luminosity corrections the @xmath54 colour term could change and thus we will be able to derive an accurate @xmath56 . in a forthcoming paper we will address this issue through different dereddening techniques ( de jaeger , in prep . ) that we are currently investigating . using 38 sne ii in the hubble flow we develop a technique based solely on photometric data ( pcm ) to build a hubble diagram based on sne ii . in summary : 1 . using pcm we find a dispersion of 0.44 mag using the @xmath2 band and 0.43 mag with the @xmath25 band , thus using nir filters the improvement is not so significant for the pcm . 2 . the @xmath11 plays a useful role , allowing us to reduce the dispersion from 0.58 mag to 0.50 mag for @xmath2 band . 3 . the colour term does not have so much influence on the nir filters because it is related to the host - galaxy extinction . we find very low ( @xmath61 ) values ( the colour - magnitude coefficient ) . if @xmath61 is purely extrinsic , it implies very low @xmath56 values . the hubble diagram derived from the csp sample using the scm yields to a dispersion of 0.29 mag , some what better than those found in the literature and emphasising the potential of scm in cosmology . it is interesting also to obtain more data and sne for which the initial decline rate and the plateau are clearly visible to try to reduce this dispersion . the pcm is very promising , and more efforts must be done in this direction , i.e. , trying to use only photometric parameters . in the coming era of large photometric wide field surveys like lsst , having spectroscopy for every sne will be impossible hence the pcm which is the first purely photometric method could be very useful . the referee is thanked for their through reading of the manuscript , which helped clarify and improve the paper . support for t. d. , s. g. , l. g. , m. h. , c. g. , f. o. , h. k. , is provided by the ministry of economy , development , and tourism s millennium science initiative through grant ic120009 , awarded to the millennium institute of astrophysics , mas . s. g. , l. g. , h. k. and f.o . also acknowledge support by conicyt through fondecyt grants 3130680 , 3140566 , 3140563 and 3140326 , respectively . the work of the csp has been supported by the national science foundation under grants ast0306969 , ast0607438 , and ast1008343 . m. d. s. , c. c. and e. h. gratefully acknowledge generous support provided by the danish agency for science and technology and innovation realized through a sapere aude level 2 grant . the authors thank f. salgado for his work done with the csp . 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 and of data provided by the central bureau for astronomical telegrams . figure [ bootstrapping_scm ] ( left ) presents the evolution of the rms versus the number of sne for both methods ( pcm and scm ) using the @xmath25 band . for both methods , after a constant median value the rms decreases when the number of sne is lower than 1012 sne because the model starts diverging . indeed if we look at the figure [ bootstrapping_scm ] on the right where the evolution of the fit parameters versus the number of sne for one single epoch ( optd*0.55 ) and the @xmath25 band are presented , we see that for the pcm , @xmath53 , @xmath61 and @xmath55 change significantly when the number of sne is around 12 . the values start diverging for a number of sne smaller than 12 , so this implies that the rms is driven by the reduced number of objects and therefore it will be difficult to conclude between the fact that @xmath61 and @xmath55 are better because we have a better rms or because it is due to a statistical effect . note that the figure does not present directly the value of the fit parameters but a fraction of the value , i.e. , the value divided by the first value plus an offset corresponding to the first value . + band and the @xmath1 colour . the black squares represent the evolution for the pcm whereas the black circles are used for the scm . _ right figure : _ we present the evolution of our fit parameters ( @xmath53 , @xmath61 , and @xmath55 ) versus the number of sne . the black colour represents the @xmath61 , the red is for @xmath55 , and the blue for @xmath53 . the circles are used for the scm and the squares for the pcm.,title="fig:",width=340 ] band and the @xmath1 colour . the black squares represent the evolution for the pcm whereas the black circles are used for the scm . _ right figure : _ we present the evolution of our fit parameters ( @xmath53 , @xmath61 , and @xmath55 ) versus the number of sne . the black colour represents the @xmath61 , the red is for @xmath55 , and the blue for @xmath53 . the circles are used for the scm and the squares for the pcm.,title="fig:",width=340 ]

radial velocities from the 2.1 m telescope at mcdonald observatory supplemented with radial velocities from the coud feed telescope at kpno provide new precise orbits for the double - lined spectroscopic binaries rr lyn ( a3/a8/a6 ) , 12 boo ( f8iv ) , and hr 6169 ( a2v ) . we derive orbital dimensions ( and ) and minimum masses ( and ) with accuracies of 0.06 to 0.9% . the three systems , which have magnitudes of 5.54 , 4.83 , and 6.42 , respectively , are all sufficiently bright that they are easily within the grasp of modern optical interferometers and so afford the prospect , when our spectroscopic observations are complemented by interferometric observations , of fully - determined orbits , precise masses , and distances . in the case of rr lyn , which is also a detached eclipsing binary with a well - determined orbital inclination ( ; khaliullin et al . 2001 ) , we are able to determine the semimajor axis of the relative orbit ,r , primary and secondary radii ofr andr , respectively ; and primary and secondary masses ofm andm , respectively . comparison of our new systemic velocity determination , km s , with the earlier one of , km s , shows no evidence of any change in the systemic velocity in the 40 years separating the two measurements , a null result that neither confirms nor contradicts the presence of the low - mass third component proposed by . our spectroscopic orbit of 12 boo is more precise that that of but confirms their results about this system . our analysis of hr 6169 has produced a major improvement in its orbital elements . the minimum masses of the primary and secondary are 2.20 0.01 and 1.64 0.02 m , respectively . although all three systems have eccentric orbits , the six components of the systems are either pseudosynchronously rotating or very nearly so .

astro-ph0601716

the current generation of ground - based interferometers , such as the palomar testbed interferometer ( pti ) @xcite , the naval prototype optical interferometer ( npoi)@xcite , the infrared optical telescope array ( iota3)@xcite , and the center for high angular resolution in astronomy ( chara ) array @xcite , is advancing stellar astronomy in a number of ways . @xcite , for example , reviewed the state of optical and infrared interferometry . one direction of progress is the increasing number of spectroscopic binaries that are being resolved as visual binaries . this allows the determination of their three - dimensional orbits and the derivation of accurate masses for the component stars and distances to the systems , distances that in many cases are more accurate than those from the _ hipparcos _ satellite . in recognition of this development we have started a program to determine substantially improved spectroscopic orbits for bright , field spectroscopic binaries . the program has two benefits : the provision of new radial velocities and spectroscopic orbits of a quality that matches or exceeds the prospective interferometric observations and , for some binaries , the detection of the secondary spectrum and measurement of secondary radial velocities for the first time . we now briefly consider these two points in turn . while some interferometric studies , such as that of 12 boo @xcite , include complementary new radial velocities , the usual practise is to take the radial velocities for the binary concerned from the literature . the precision of such velocities often falls short of that needed to match the interferometric observations . for example , in their recent determination of the three - dimensional orbit of the bright spectroscopic binary @xmath17 psc , @xcite had to complement their interferometric measurements with radial velocities observed in 1944 and 1945 @xcite . their resulting best - fit solution for the three - dimensional orbit has rms velocity residuals of 4.8 and 3.6 km s@xmath13 for the primary and secondary , respectively . orbits with large velocity residuals are not exceptional because of the generally lower resolution and low signal - to - noise ratio of spectra obtained in the first roughly three - quarters of the twentieth century , for example , of the first 100 systems in the _ eighth catalogue of the orbital elements of spectroscopic binaries _ @xcite , 63 have orbits that were published in 1980 or earlier and 24 have orbits that were published in 1950 or earlier , long before the advent of radial velocity spectrometers and charge - coupled device detectors , which can produce spectra with very high signal - to - noise ratios . similar proportions must apply for all 1469 systems in the catalogue . while these proportions will have improved as a result of the substantial number of new spectroscopic binary orbits that have been published since 1989 @xcite , most such orbits are for newly discovered binaries . with respect to the detection of the secondary spectrum , we note that without secondary radial velocities and a determination of the secondary s spectroscopic orbit , the linear separation between the binary components is unknown and the determination of the three - dimensional orbit is incomplete . increasing the pool of double - lined spectroscopic binaries ( sb2s ) thus increases the number of spectroscopic binaries available for fruitful interferometric observation . in addition , binary systems with components of significantly different masses provide the greatest constraints on evolutionary models . considering that the majority of spectroscopic binaries are single - lined spectroscopic binaries ( sb1s ) , there is ample opportunity here . @xcite , for example , found that two - thirds of the spectroscopic binaries in their _ ninth catalogue _ ( there is no reason to think the catalogue s incompleteness affects this statistic much . ) our program uses new , high - resolution , red - wavelength spectra obtained with the 2.1 m telescope at mcdonald observatory of the university of texas and the coud feed telescope at kitt peak national observatory ( kpno ) . @xcite provided a preliminary description of our program and an initial list of observed stars , which has now been expanded to over 40 systems . these come primarily from a sample of 130 candidate systems obtained by searching the _ eighth catalogue _ for sb2 systems that could profit from renewed spectroscopic observation and sb1 systems with large enough mass functions to suggest that high signal - to - noise ratio spectra might transform them into sb2 systems ( e.g. , * ? ? ? the stars are north of @xmath1840 in declination and generally brighter than @xmath19 mag . others have also seen the need for improved radial velocities for spectroscopic binaries . for example , @xcite has successfully applied the iodine absorption - cell method for determining very precise radial velocities to the measurement of radial velocities of _ both _ components in sb2s . hitherto , this technique , which uses an iodine absorption cell to impose a reference spectrum on the stellar spectrum and is notable for its use in the discovery of extrasolar planets , has been restricted to the radial velocities of single stars or stars with companions of insignificant relative brightness . his pioneering investigation , which was carried out on the keck i telescope with the hires spectrograph , was limited to five objects including a radial - velocity standard and two sb2s . among the latter was 64 psc ( hd 4676 ) , a well - known , bright spectroscopic binary ( f8v , @xmath20 days ) with a three - dimensional orbit determined by @xcite , using their own interferometric observations made with pti and radial velocities from @xcite . konacki s combined fit of his new radial velocities and the @xcite interferometric data leads to better - determined orbital and physical parameters for 64 psc . in particular , the rms velocity residual of 24 m s@xmath13 , determined from his new fit , is a striking improvement compared to the rms residual of 810 m s@xmath13 ( average @xmath21 m s@xmath13 ) given by @xcite for their radial - velocity solution . our new velocities are not as precise as the iodine - cell velocities from the keck i telescope we will see that for 12 boo , which has a similar spectral type to 64 psc , the rms residual for our velocities is 110 m s@xmath13 but their quality is still a good match with that of current interferometric observations . there is thus ample opportunity for radial velocities from small and medium - sized telescopes , measured by traditional methods , to make a worthwhile contribution . here we report new orbit determinations for three bright spectroscopic binaries , rr lyn , 12 boo , and hr 6169 . analysis of our new velocities provides significant improvements in the orbital elements of the systems . indeed , in the case of hr 6169 some of the revisions are substantial . table 1 gives basic data for the systems ; all three are known sb2s , have orbital periods near 10 days , and eccentric orbits . we now briefly look at each system individually . rr lyn has long been known as both an eclipsing and spectroscopic binary . its orbit has a period of 9.95 days and a modest eccentricity , @xmath22 . the star s peculiar abundances are indicated in its metallic - lined spectral classification , which reflects primarily the spectral type of the primary . @xcite classified it as a3/a8/a6 , based on its ca ii k line , its hydrogen balmer lines , and its metallic lines , respectively , while @xcite called the hydrogen lines a7 and the metallic lines f0 . @xcite compared the two components with evolutionary tracks and also used their photoelectric photometry on the @xmath23 system @xcite to place the components of rr lyn in a @xmath24 diagram . from those results they estimated individual spectral types of a6 iv for the primary and f0 v for the secondary . thus , the primary has already begun to leave the main sequence , while the secondary is still ensconced within it . the metallic - line status of the primary is not in doubt @xcite , while that of the secondary remains uncertain , although the abundances of @xcite indicate possible metallicism . @xcite published the first spectroscopic orbit . more recent , but now quite old orbits were published by @xcite and @xcite . kondo s orbit has the advantages that it sampled the primary and secondary radial velocity curves more fully than popper s and that it has a single systemic velocity , while popper determined different systemic velocities for the primary and secondary . photoelectric light curves of rr lyn have been reported by @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . the last of these studies used precise photometry in the @xmath25 and @xmath26 bandpasses @xcite , which are based on but somewhat different from the johnson @xmath27 and @xmath26 photometric bandpasses . @xcite showed that the primary and secondary eclipse depths in @xmath4 are @xmath28 and @xmath29 , respectively , the fractional luminosities , also in @xmath4 , are @xmath30 , @xmath31 , and the orbital inclination is @xmath32 . both @xcite and @xcite needed the addition of third light to solve the light curves , while @xcite , who also did a photometric analysis , found no requirement for third light . this last result is supported by @xcite , who demonstrated that their observations in all filters are fitted by the same geometry and that their model matches the observations without any need to invoke third light . in a later paper , however , @xcite presented new evidence for a third body , based _ not _ on an analysis of the light curve , but on the times of the eclipse minima . the presence of the third star , which @xcite suggested is a very low - mass , low - luminosity object that contributes an insignificant amount of light to the system , is inferred from quasi - periodic oscillations in the times of primary and secondary minima . they proposed an extemely eccentric orbit ( @xmath33 ) with a rather long orbital period of @xmath34 years for it . the 64-year timespan of the published photoelectric minima amounts to only one - and - a - half orbital periods , however , so the case for the third body can not yet be regarded as conclusive . the sb2 system 12 boo is the only one in our trio that already has a determination of its three - dimensional orbit @xcite . the two components have an orbital period of 9.60 days and are of very similar mass , having a secondary to primary mass ratio of 0.97 . the system s combined spectral type is f8 iv @xcite or f8 v @xcite . @xcite found that the primary is leaving the main sequence , and with @xmath35 k it is now marginally _ cooler _ than the secondary ( @xmath36 k ) . @xcite was the first to determine its spectroscopic orbit . newer orbital elements have been published by @xcite and @xcite . @xcite presented the newest set of radial velocities as well as interferometric observations . details for these more recent spectroscopic orbits , including the present investigation , are summarised in table 2 . prior to its publication , we were unaware of the new investigation of 12 boo by @xcite . thus , in spite of the element of ( unintentional ) redundancy here , our own study of 12 boo , based on radial velocities acquired from 2002 to 2005 , is still of interest because our velocities turn out to be significantly more precise , allowing us to determine improved spectroscopic orbital elements . compared to rr lyn and 12 boo , the previous work on hr 6169 has been quite modest . this binary has a combined spectral type of a2 v @xcite or a1 iv @xcite . the first and , to date , only spectroscopic orbit is that of @xcite , which was based on radial velocities from blue photographic spectrograms obtained at the cassegrain focus of the 1.8 m telescope at the dominion astrophysical observatory ( dao ) . this sb2 system has an orbit with a period of 10.56 days and a moderate eccentricity , @xmath37 . our observations were collected from 2002 april through 2005 october at the mcdonald observatory and kpno , with the majority of spectra of all three program stars obtained at mcdonald . the mcdonald observations were acquired with the 2.1 m telescope , the sandiford cassegrain echelle spectrograph @xcite , and a reticon ccd with 27@xmath38m@xmath39 pixels arranged in a 1200@xmath40400 pixel format . the spectrograms cover the wavelength range 5700 to 7000 and have a resolving power of 60,000 . the collapsed one - dimensional stellar spectra have a representative signal - to - noise ratio of @xmath41300 . each stellar observation was followed by an observation of a thorium argon comparison spectrum . the observations at kpno were obtained with the coud feed telescope , coud spectrograph , and a ti ccd detector . all the spectrograms were centered at 6430 , cover a wavelength range of about 80 , and have a resolution of 0.21 or a resolving power of just over 30,000 . the spectrograms typicaly have signal - to - noise ratios of 200 - 250 . the data were processed and wavelength - calibrated in a conventional manner with the iraf package of programs . the spectra of the three systems are double - lined and , at most orbital phases , the secondary lines are well separated from their primary counterparts . in rr lyn there are numerous lines of both the primary and secondary with the secondary lines being distinctly weaker , while in 12 boo the primary and secondary lines are also numerous and are of roughly similar strength . in hr 6169 the primary is represented by only a few lines from ionised species , while the secondary has numerous lines , mostly from neutral species , which are also sharper than those of the primary . in this system , therefore , at the yellow and red wavelengths covered by our observations the secondary lines are much _ better _ for velocity measurement than those of the primary . we now outline the procedures used to measure the mcdonald and kpno radial velocities . for the mcdonald velocities we proceeded as follows : the wavelengths of well - defined primary and secondary lines were measured by fitting gaussian profiles with the splot routine of iraf , the wavelength differences between the measured wavelengths and rest wavelengths @xcite of the lines provided the topocentric radial velocities , telluric o@xmath42 lines near 6300 ( @xmath43 band ) and 6900 ( b band ) were measured in the same way so as to determine the offset between the stellar spectrum and its associated thorium argon comparison spectrum , the stellar topocentric velocities were then corrected by subtracting from them the telluric line offset in velocity form , and , finally , the heliocentric correction led to the heliocentric radial velocities . the mcdonald velocities are , thus , absolute velocities . the kpno radial velocities were determined with the iraf cross - correlation program fxcor @xcite . the iau radial velocity standard stars @xmath44 vir , hr 5694 , and hr 7560 , which have late - f spectral classes , were used as cross - correlation reference stars . their velocities of 4.4 , 54.4 , and 0.0 km s@xmath13 , respectively , were adopted from @xcite . in the kpno spectra , lines in the wavelength region redward of 6445 are not particularly suitable for measurement because most features are blends , and there are a number of modest strength water vapor lines . thus , the radial velocities were determined from lines in the region 6385 to 6445 . unfortunately , this 60 portion of the spectrum is so small that a mismatch , caused by the varying strength of line blends , between the spectral type of the program star and that of the cross - correlation standard can significantly affect the measured velocity . thus , instead of cross - correlating this entire 60 wavelength region , only the wavelength regions around the strongest and least - blended lines were cross - correlated . for our stars with components of late - a or f spectral type the list includes five lines , the fe i lines at 6394 , 6412 , 6421 , and 6431 plus the ca i line at 6439 . the weakness of those neutral lines in the spectra of early - a stars means that the only measurable lines in their kpno spectrograms are the two fe ii lines at 6417 and 6433 , which can provide good velocities if the lines are narrow enough . each cross - corrrelation profile was fitted with a gaussian function . because different techniques were used to measure the mcdonald and kpno radial velocities , the two velocity systems may have slightly different zero points . we will examine this possibility in 6 . table 3 gives the calendar dates , heliocentric julian dates and heliocentric radial velocities for the mcdonald and kpno observations . @xcite identified several luminosity - sensitive and temperature - sensitive line ratios in the 6430 - 6465 region . they employed those critical line ratios and the general appearance of the spectrum as spectral - type criteria . however , for stars that are hotter than early - g spectral class , the line ratios in that wavelength region have little sensitivity to luminosity . thus , for the a and f stars of our three systems , we have used the entire 80 spectral region of our kpno observations to estimate just the spectral classes of the individual components . in the 6430 region most of the strongest features are fe i lines with the addition of three fe ii and three ca i lines . the luminosity class may be determined by computing the absolute visual magnitude with the _ hipparcos _ parallax and comparing that magnitude to evolutionary tracks or a table of canonical values for giants and dwarfs . spectra of our three binaries were compared with the spectra of a number of a and f type stars primarily from the lists of @xcite and @xcite . the reference - star spectra were obtained at kpno with the same telescope , spectrograph , and dectector as our binary - star spectra . to facilitate a comparison , the spectra of the reference stars were rotationally broadened and shifted in radial velocity with a computer program developed by @xcite and @xcite . this spectrum addition enables us to determine a continuum intensity ratio of the binary components at 6430 , a wavelength that is about 0.6 of the way between the central wavelengths of the johnson @xmath4 and @xmath26 bandpasses . if the two stars have very similar spectral types , this intensity ratio is also the luminosity ratio and , thus , can be converted directly into a magnitude difference . however , if the lines of the secondary are intrinsically stronger than those of the primary , as would be the case , for example , if the components were main sequence stars of rather different spectral types , then the continuum intensity ratio results in a minimum magnitude difference . we have determined projected rotational velocities from our red - wavelength kpno spectra with the procedure of @xcite . for each star the full width at half - maximum of several metal lines in the 6430 region was measured and the results averaged . an instrumental broadening of 0.21 was removed from the measured broadening by taking the square root of the difference between the squares of measurements of the stellar and comparison lines , resulting in the intrinsic broadening . the calibration polynomial of @xcite was used to convert this broadening in angstroms into a total line broadening in kilometers per second . for a - type stars this line broadening corresponds to the _ v _ sin _ i _ value . for f - type stars , macroturbulent broadening has been taken into account . following @xcite , for early - f stars a macroturbulence of 5 km s@xmath13 was adopted and removed , while for mid - f and late - f stars values of 4 and 3 km s@xmath13 , respectively , were used . all three systems have eccentric orbits , so we obtained simultaneous solutions of the primary and secondary radial velocities by the lehmann filhes method to determine the orbits . a preliminary to the orbit calculation was the assignment of suitable weights for the various velocities . typically the primary and secondary spectra are of different quality because the components can differ , sometimes markedly , as to strength of lines , line width , and number of available lines , resulting in primary and secondary velocities that are of different weight . also , the mcdonald velocities , because of the higher resolution by a factor of two and the substantially greater wavelength coverage of the mcdonald spectra , resulting in a larger number of available lines , are somewhat more precise than the kpno velocities and therefore have higher weight . for each system there are , thus , four different sets of velocity , each with its own weight . these weights , which are given in table 3 , were chosen so as to approximately equalize the products of the weights and the mean square velocity residuals for the different sets of velocity . once the weights were assigned , the velocities were then solved to determine the spectroscopic orbits . the velocity residuals ( table 3 ) for the orbital solutions , given to two decimal places to facilitate our examination , allow a straightforward check on the consistency of the radial velocities from the two observatories . any systematic difference between the mcdonald and kpno velocities will manifest itself as a corresponding difference between the average residuals , calculated with regard to sign , for the two types of velocity , while if they are consistent then their average residuals , apart from statistical fluctuations , will be identical and equal to zero . a comparison of the residuals for the mcdonald and kpno radial velocities is given in table 4 . although the differences are all smaller than or similar to their estimated errors , they all have the same sign , thus suggesting an offset of order 0.1 0.2 km s@xmath13 between the mcdonald and kpno velocities . we attach the most importance , however , to the primaries of rr lyn and 12 boo because of the six stars in our three systems these two have the most well - determined radial velocities , and in both cases there is no evidence of an offset . this simple check , therefore , indicates the possible presence of an offset of 0.1 0.2 km s@xmath13 between the mcdonald and kpno velocities , but the results are inconclusive . in the absence of clear evidence of an offset , we decided not to adjust either set of velocities . we now look at the results for the individual systems . table 5 gives the spectroscopic orbital elements of rr lyn obtained from the solution of our radial velocities . for comparison , the elements of @xcite and the photometrically - based elements of @xcite are also listed . figure [ fig1 ] shows our radial velocities along with the calculated radial - velocity curves for our orbital elements . although the photometrically determined orbital period is more precise than our spectroscopically measured value , the difference between the two 10-day periods corresponds to less than 1 second . we have retained the spectroscopic period for the determination of our spectroscopic orbit , since adopting the photometric period does not change the orbital elements . examining the solutions in table 5 , one sees that to within the uncertainties all three investigations give the same eccentricity of 0.08 . this agreement is especially pleasing in the case of @xcite because their photometric determination of the eccentricity is completely independent of our spectroscopic one . we note that our determination of @xmath45 ( 1794 @xmath15 06 ) is the first accurate measurement of this element and that to within its uncertainty , it is @xmath46 , which means the major axis of the true orbit and the line of nodes coincide . also of interest , we find that our and kondo s determinations of @xmath47 and @xmath48 are the same to within the uncertainties . we will compare the two @xmath49-velocity determinations shortly . the minimum masses are also in agreement , with our uncertainties being about 10 times smaller than kondo s . as we have seen , rr lyn is also a detached eclipsing binary so the photometric and spectroscopic orbital elements can be combined to provide the sizes and masses of the components as well as the linear separation between the primary and secondary . this is done in table 6 . the new mass determinations of @xmath50 m@xmath7 and @xmath51 m@xmath7 are extremely precise , having uncertainties of only 0.4% and 0.3% , respectively . although our new determinations of the masses and radii of the primary and secondary are more precise than those of @xcite , our values differ only slightly from theirs . the new primary and secondary masses are just 0.037m@xmath7 and 0.017m@xmath7 larger , respectively , than theirs , while the new primary and secondary radii differ from theirs by 0.00 r@xmath7 and 0.01 r@xmath7 , respectively . we will not rediscuss , therefore , the evolutionary state of rr lyn because our new results would have little impact on the conclusions of @xcite in this regard . instead , we note that insofar as our new masses and radii confirm theirs , so we support their determinations of other derived quantities . in particular we confirm their photometric estimate of the system s distance , 73.5 @xmath15 2.8 pc , and its age , @xmath52 billion years . we now look at the question of the third body in the system . as already mentioned , @xcite proposed the presence of a third star in an extremely eccentric orbit ( @xmath53 ) with a period of 39.7 years as an explanation of variations in the times of minima of the primary and secondary eclipses . the proposed star is of low mass , estimated to be 0.10m@xmath7 if its orbit is coplanar with that of the eclipsing pair . thus , it would make a negligible contribution to the spectrum , but its presence would cause a regular variation of the systemic velocity of the eclipsing pair as the center - of - mass of the eclipsing pair orbits the center - of - mass of the entire system . in view of the long interval between the observations of @xcite and ours , it is natural to compare kondo s systemic velocity with ours to see if there is a variation . the difference between our and kondo s @xmath49-velocities is insignificant . from table 5 we see that it is @xmath54 with the estimated uncertainty being set by the uncertainty in kondo s result . the observations thus set an upper limit of @xmath55 km s@xmath13 on any difference between the two @xmath49-velocities . what is the predicted difference that would be caused by the proposed third star ? to answer this question we note that we do _ not _ need to know the ( unknown ) systemic velocity of the entire system ; calculation of the _ orbital motion _ alone , of the center of mass of the eclipsing pair , is sufficient to provide the predicted @xmath49-velocity _ difference _ that we want . @xcite have provided estimates of all the orbital elements needed to calculate the predicted variation of the @xmath49-velocity of the eclipsing pair , with the exception of @xmath56 , and , from the information they give , one can infer @xmath57 km s@xmath13 . we can thus calculate the predicted variation of the @xmath49-velocity of the eclipsing pair due to its motion in the proposed 39.7-year orbit , and this is shown in figure [ fig2 ] . it is apparent from that figure that the predicted difference of @xmath49-velocities is tiny , only @xmath58 km s@xmath13 . the smallness of the difference results from the extreme eccentricity of the proposed orbit , so the velocity variation outside of periastron passage is slight , and because , by chance , kondo s observations and ours are separated by almost exactly one orbit and thus are at the same phase . our null result , therefore , is consistent with the predicted minute change in the @xmath49-velocity of the eclipsing pair of stars caused by the presence of the proposed low - mass third star . of course , the result is also consistent with the absence of such a body , and so we are unable to either confirm or deny its existence . as can be seen in figure [ fig2 ] , the extreme eccentricity of the 39.7-year orbit means that usually the @xmath49-velocity of the eclipsing pair changes very gradually . so , during most of the orbit , radial velocities are almost impervious to the presence or absence of the third body . near periastron it is a different story , however , as the third star swoops toward the eclipsing pair , and their @xmath49-velocity suffers an abrupt change with a total amplitude of 2.5 km s@xmath13 . at the next periastron passage in 2028 one or two precise radial velocities will readily confirm or refute the existence of the third star . in their recent investigation of the 12 boo system @xcite combined their interferometric observations of the visual orbit and complementary , new radial velocities to make an accurate determination of the system s three - dimensional orbit and the masses and radii of its components . they then use these basic parameters to discuss the system s evolutionary status . although our own work duplicates theirs to some extent , our radial velocities are more precise , so our results provide a useful check on the spectroscopic component of their investigation . table 7 lists the spectroscopic orbital elements from the combined interferometric and radial velocity solution of @xcite with our new radial - velocity solution . the two sets of orbital elements are in excellent agreement , and the uncertainties of our elements are usually smaller than those of @xcite . with the exception of the period , for which our observational timespan is quite short , all our elements are improved determinations and for four of them ( @xmath59 , @xmath47 , @xmath48 , and @xmath49 ) the improvement is a factor of two . there is also good agreement between the two @xmath49 velocities , where we find @xmath60 km s@xmath13 , while they find @xmath61 km s@xmath13 , which argues that our and their radial velocities have the same zero point . to make the most of the radial velocities , we also obtained a combined solution of our radial velocities and theirs , making no adjustment to either set of velocities . ( the weights for our velocities in this solution are given in table 3 , while the weights for theirs were 0.1 and 0.05 for the primary and secondary , respectively . ) the elements from this combined spectroscopic solution are also listed in table 7 . they represent an all - round improvement , with the period determination , in particular , being sharpened thanks to the longer timespan , 1987 to 2005 , of the combined observations . figure [ fig3 ] shows our radial velocities and those of @xcite along with the calculated radial velocity curves for the combined solution of the two sets . with the help of the inclination , @xmath62 , from the orbit of @xcite we determine new values of the stellar masses and the semi - major axis of the relative orbit . we find @xmath63m@xmath7 , @xmath64m@xmath7 , and @xmath65 gm . these results certainly confirm those of @xcite , which are @xmath66m@xmath7 , @xmath67m@xmath7 , but are more precise . finally we look at the orbital parallax , @xmath68 , and distance of the system . the orbital parallax is simply @xmath69 ( arcsec)/@xmath69 ( au ) where @xmath69 denotes both the angular and linear sizes , respectively , of the true orbit . the angular size of the orbit , @xmath70 mas , of course , comes from the visual - orbit determination @xcite . we note , however , that the uncertainty in the angular size of the orbit , which is 0.5% and is much bigger than the uncertainties in the determinations of the linear size of the orbit ( 0.1% for theirs and 0.06% for ours ) , dominates the uncertainty in the orbital parallax . thus , although we have improved the determination of the linear size of the orbit , there are no corresponding improvements in the orbital parallax or distance of 12 boo , so we content ourselves with restating the results of @xcite @xmath71mas and a distance to the system of @xmath72pc . table 8 lists the orbital elements of hr 6169 , derived from the solution of our radial velocities , as well as the elements obtained by @xcite . in figure [ fig4 ] our radial velocities and those of @xcite are compared with the radial velocity curves calculated from our elements alone . for hr 6169 , in the yellow and red region of the spectrum where we measured our radial velocities , the only useful lines representing the primary are the two siii lines at 6347 and 6371 and a smattering of very weak feii lines , while the secondary is represented by lines of neutral and ionised species that are both more numerous and sharper . for this reason the _ secondary _ velocities are more precise than those of the primary . it follows that the secondary velocities have much higher weight than those of the primary ( see table 3 ) , the secondary radial - velocity curve is better defined than that of the primary , and @xmath48 is better determined than @xmath47 . comparison of our results with young s reveals some very significant discrepancies , especially for the @xmath49 velocity . at first sight the large difference of @xmath73 km s@xmath13 between our @xmath49 velocity ( @xmath74 km s@xmath13 ) and young s ( @xmath75 km s@xmath13 ) suggests the presence of an unseen third star , which must have changed the @xmath49 velocity of the spectroscopic - binary pair in the 84 years between the two measurements . a glance at figure [ fig4 ] , however , shows there is no need to invoke a third star . instead , a simpler explanation of the difference can be found because of a systematic error in young s @xmath49 velocity , caused by blending of the primary and secondary lines in his spectra . @xcite remarked that in his spectra with smaller primary - secondary velocity separations the primary lines and their secondary counterparts were blended , so that he was only able to measure a single blend velocity , which he adopted as the primary velocity . inspection of figure [ fig4 ] shows that these blended primary velocities cluster around the smaller velocity extremum associated with the nodal passage near apastron . they mostly fall below the @xmath49 velocity and , being dragged upwards and towards it , are too positive and , not being balanced by the few velocities that are dragged downwards , their presence in young s orbit solution made his @xmath49 velocity too positive . similar considerations account for his semi - amplitude of the primary being too small , @xmath76 km s@xmath13 compared with our value of @xmath77 km s@xmath13 . unlike our other two systems , rr lyn , which is an eclipsing system , and 12 boo , for which interferometric observations have been used to determine a visual orbit , hr 6169 has no direct determination of its orbital inclination . thus , at present we can only compute the minimum masses of its two components , which are 2.20 @xmath15 0.01 and 1.64 @xmath15 0.02 m@xmath16 for the primary and secondary , respectively . @xcite classified the combined spectrum of rr lyn as a3/a8/a6 , based on its ca ii k , hydrogen balmer , and metal lines , respectively , providing a general starting point for our determination . the rapid rotation of most a - type stars plus a variety of abundance peculiarities make it difficult to identify a significant number of slowly rotating reference stars for that spectral class . despite the limited choices , identified primarily from the work of @xcite , we obtained a reasonable fit to the spectrum of rr lyn with hr 3354 ( spectral classes of a3/a5/a7 = calcium / hydrogen / metals @xcite ) for the primary plus hr 1613 ( spectral type of a9 v @xcite and [ fe / h ] @xmath78 0.0 @xcite ) for the secondary . while the iron lines of the primary are fitted relatively well by hr 3354 , the calcium lines of this reference star are too weak . our results for the stellar components , however , are similar to those of @xcite , who determined spectral types of a6 iv plus f0 v from their @xmath23 photometry and a comparison with evolutionary tracks . the continuum intensity ratio of our fit is 0.236 , which corresponds to a magnitude difference of 1.6 . this is in rough agreement with the results of @xcite , who used the @xmath79 photometric system @xcite and found a @xmath4-band luminosity ratio of 0.276 , which corresponds to a magnitude difference of 1.40 . our magnitude difference is consistent with the luminosity classes assigned by @xcite . thus , we adopt their estimated spectral types . guided by previous spectral classifications , we used hr 7560 ( f8 v @xcite and mean [ fe / h ] = 0.06 @xcite ) as a proxy for both components , but found the lines of the resulting composite spectrum fit to be too strong . however , with hr 5694 ( f8 iv - v @xcite and mean [ fe / h ] = @xmath180.10 @xcite ) used for both components we found an excellent fit to the spectrum of 12 boo . the continuum intensity ratio of this fit is 0.637 . because the spectral classes of the two stars are essentially identical , this value is also the luminosity ratio at 6430 , a wavelength that is about 0.6 of the way between the center of the johnson @xmath4 and @xmath26 bandpasses . the intensity ratio produces a magnitude difference of 0.5 with an estimated uncertainty of 0.1 . such a difference is in accord with the more precise @xmath80@xmath4 magnitude of 0.48 determined by @xcite . the slightly subsolar iron abundance is also in agreement with the results of @xcite and @xcite , who found [ fe / h ] = @xmath180.03 @xmath15 0.09 and @xmath180.1 @xmath15 0.1 , respectively . from dao photographic plates , @xcite determined magnitude differences for 82 spectroscopic binaries , including hr 6169 . he described the spectrum of this system as consisting of two narrow - lined early - a stars and concluded that both components were dwarfs . his low - weight magnitude difference in the blue region of the spectrum is 0.7 @xmath15 0.4 . unlike our results for 12 boo and rr lyn we have had difficulty finding a good combination of reference stars to represent the spectrum hr 6169 . given the a spectral class of the components , the lines of both stars are quite weak in the 6430 region with all but one of the lines being less than 5% deep in the combined spectrum . the primary component has the stronger fe ii lines and weaker fe i lines and so is earlier in spectral class . from the metal lines our best estimates of those spectral classes are a1 for the primary and a7 for the secondary with a continuum intensity ratio in the 6430 region that corresponds to a minimum magnitude difference of @xmath411 . alternatively , the spectral types may be estimated from the colors of the components . guided by the above results , we found that the @xmath81 colors for stars of spectral types a1 v and a5 v , taken from table ii of @xcite , plus an adopted @xmath4 magnitude difference of 0.9 result in the observed @xmath81 = 0.06 from the _ hipparcos _ satellite @xcite . we have adopted those spectral types and colors in our additional analyses . the two main theories of orbital circularization and rotational synchronization ( e.g. ; @xcite ) disagree significantly on absolute time scales but do agree that synchronization should occur first . from surveys of b and a stars @xcite concluded that synchronization occurs faster than circularization , confirming for early - type stars the theoretical results . the three systems discussed here have orbital periods between 9.6 and 10.6 days and eccentric rather than circular orbits . since two of the three systems contain a - type stars , the observed eccentricities of their orbits are not surprising . @xcite examined 62 spectroscopic binaries with a - type primaries and periods less than 100 days . they concluded that all systems with orbital periods @xmath82 3 days have circular or nearly circular orbits . they also found that many binaries with periods in the range of 3 to 10 days have circular orbits , while all those with periods greater than 10 days have eccentric orbits . thus , the eccentric orbits of rr lyn and hr 6169 appear to be typical of a - type systems with periods near 10 days . @xcite has shown that in an eccentric orbit a star s rotational angular velocity will tend to sychronize with that of the orbital motion at periastron , a condition called pseudosynchronous rotation . with equation ( 42 ) of @xcite we calculated pseudosynchronous periods for the three systems . for rr lyn our _ v _ sin _ i _ values are 14.6 @xmath15 1 and 11.3 @xmath15 2 km s@xmath13 for its primary and secondary , respectively . the uncertainties are estimated , with that for the secondary being larger because of the weakness of its lines . we assume that the orbital and rotational axes are parallel , so the rotational inclination is 874 . since it is so close to 90 , we adopt our _ v _ sin _ i _ values as the equatorial rotational velocities of the components . from table 6 the radii are 2.57 and 1.59 r@xmath16 for the primary and secondary , respectively . those values , combined with the pseudosynchronous period of 9.59 days , lead to rotational velocities of 13.6 and 8.4 km s@xmath13 . thus , the rotation of the primary is consistent with its pseudosynchronous value , while the secondary appears to be rotating slightly faster than pseudosynchronous . our projected rotational velocities for 12 boo , 13.0 @xmath15 1 and 10.3 @xmath15 1 km s@xmath13 for its primary and secondary , respectively , are in good agreement with those determined by @xcite and @xcite . again we assume that the orbital and rotational axes are parallel , so the rotational inclination is 108 . thus , the observed equatorial rotational velocities become 13.7 and 10.8 km s@xmath13 . the radii of the two components @xcite and the pseudosynchronous period of 7.84 days , lead to predicted rotational velocities of 15.9 and 12.0 km s@xmath13 . thus , the primary is rotating slightly slower than its predicted velocity , while the secondary may well be rotating pseudosynchronously . to determine if the components of hr 6169 are pseudosynchronously rotating we must first estimate their radii . from the _ hipparcos _ catalog @xcite the @xmath4 magnitude and @xmath81 color of the hr 6169 system are 6.42 and 0.062 , respectively . the _ hipparcos _ parallax of 6.17 @xmath15 0.087 ( mas ) @xcite corresponds to a distance of 162 @xmath15 24 pc . although at such a distance there may be a modest amount of interstellar reddening , the color of the system is consistent with the spectral type of the primary , and so we have assumed no interstellar reddening . as a result , the parallax , the @xmath4 magnitude , and the adopted @xmath4 magnitude difference of 0.9 ( 7.3 ) were combined to obtain absolute magnitudes @xmath83 = 0.8 @xmath15 0.3 mag and @xmath83 = 1.7 @xmath15 0.3 mag for primary and secondary , respectively . a @xmath81 color of 0.03 for the primary and a @xmath81 of 0.14 for the secondary @xcite were adopted and then used in conjunction with table 3 of @xcite , to obtain the bolometric corrections and effective temperatures of the two components . the resulting luminosities of the primary and secondary are @xmath84 = 43 @xmath15 12 @xmath85 and @xmath86 = 17 @xmath15 5 @xmath85 , respectively , while the radii are @xmath87 = 2.6 @xmath15 0.4 @xmath88 and @xmath89 = 2.1 @xmath15 0.3 @xmath88 , respectively . the uncertainties in the computed quantities are dominated by the parallax and magnitude difference uncertainties plus , to a lesser extent , the effective temperature uncertainty , which is estimated to be @xmath15200 k. because the orbit of hr 6169 has a moderately large eccentricity of 0.414 , the pseudosynchronous period of 4.95 days is less than half of the orbital period . with that period and the above radii , we obtain pseudosynchronous rotational velocities of 26.6 and 21.5 km s@xmath13 for the primary and secondary , respectively . we have determined _ v _ sin _ i _ values of 32 @xmath15 5 and 18 @xmath15 2 km s@xmath13 for the primary and secondary , where the uncertainties are estimated . the minimum masses of the components are rather large , suggesting a relatively high orbital inclination . a comparison with canonical values of absolute magnitudes and masses for dwarfs of the appropriate spectral types @xcite suggests an inclination of about 70 . adopting this value , assuming , as has been done for the other two systems , that the rotational inclination has the same value , produces equatorial velocities of 34 and 19 km s@xmath13 . thus , the primary appears to be rotating slightly faster than its pseudosynchronous velocity , while the secondary may be rotating pseudosynchronously . we have determined new spectroscopic orbits for the bright sb2 systems rr lyn , 12 boo , and hr 6169 . the accuracies of the minimum masses ( @xmath2 and @xmath3 ) range from 0.13% for the secondary of 12 boo to 0.9% for the secondary of hr 6169 , while the accuracies of the linear sizes of the relative orbit ( @xmath90 ) range from 0.05% for 12 boo to 0.25% for hr 6169 . these results show our radial velocities provide well - determined spectroscopic orbits . which do justice to interferometric measurements of 12 boo @xcite and , we expect , will also do justice to prospective interferometric measurements of rr lyn , hr 6169 , and other systems in our program . for rr lyn , an eclipsing binary with a well - determined orbital inclination , we have determined accurate new masses and radii for its components . the semi - major axis of the relative orbit , @xmath91r@xmath92 , plus the distance to the system , @xmath93 pc @xcite , lead to a predicted angular size @xmath94 mas for the true orbit and , because @xmath45 ( @xmath95 ) is @xmath96 to within its measurement error so the semi - major axis lies across the line - of - sight , this will also be the @xmath69 of the apparent orbit . we find that the expected effect of the unseen third star proposed by @xcite on our and kondo s ( 1976 ) radial velocities is so slight as to be well below the threshold of our measurement errors . we have determined a new spectroscopic orbit for 12 boo , which confirms , and is more precise than , that determined by @xcite as part of their measurement of the three - dimensional orbit of this system . our investigation of hr 6169 corrects systematic errors in young s ( 1920 ) orbit and leads to significant revisions of the minimum masses and @xmath97 . at the distance of hr 6169 , @xmath98pc ( perryman et al . 1997 ) , the @xmath90 of 22.05 @xmath15 0.05 gm corresponds to an angular separation of 0.91mas . in conclusion we find that radial velocities observed with small and medium - sized telescopes and measured by traditional methods provide a suitable spectroscopic contribution to the combination of measurements required for determination of three - dimensional orbits . we thank eshwar reddy and chris sneden for some helpful tips and david doss for his patient guidance , which was essential for the successful operation of the 2.1-m telescope at mcdonald and its instrumentation . we also thank richard green , former director of kpno , for continuing to make the coud feed telescope available . we appreciate daryl willmarth s invaluable assistance with the kpno coud feed telescope and our observations . this research at tennessee state university was supported in part by nasa grant ncc5 - 511 and nsf grant hrd-9706268 . matthews , l. d. , & mathieu , r. d. 1992 , in asp conf . 32 , complimentary approaches to double and multiple star research , iau colloquium 135 , ed . h. a. mcalister & w. i. hartkopf ( san francisco : asp ) , 244 lcrccccrc rr lyn & 2291 & 44691 & a3/a8/a6 & 5.53 & 0.238 & 12.01 & 9.95 & 0.08 + 12 boo & 5304 & 123999 & f8iv & 4.82 & 0.541 & 27.27 & 9.60 & 0.19 + ... & 6169 & 149632 & a2v & 6.42 & 0.062 & 06.17 & 10.56 & 0.41 + lccc abt & levy 1976 & 1966 1971 & 20 & 1.3 + de medeiros & udry 1999 & 1987 1992 & 12 & 0.90 + boden et al . 2005 & 1987 2001 & 49 & 0.39 + this paper ( mcdonald and kpno ) & 2002 2005 & 24 & 0.14 + lrrrrrrrrr + 2003 apr 22 & mcd & 52,751.635 & 0.3781 & 33.9 & @xmath99 & 1.000 & 0.100 & @xmath100 & 0.16 + 2004 feb 07 & mcd & 53,042.843 & 0.6598 & 25.3 & @xmath101 & 1.000 & 0.100 & 0.03 & @xmath102 + 2004 feb 08 & mcd & 53,043.803 & 0.7563 & @xmath103 & & 0.000 & & @xmath104 & + 2004 feb 09 & mcd & 53,044.812 & 0.8577 & @xmath105 & 36.5 & 1.000 & 0.100 & 0.11 & @xmath106 + 2004 mar 10 & mcd & 53,074.760 & 0.8691 & @xmath107 & 43.0 & 1.000 & 0.100 & 0.03 & 0.06 + 2004 mar 11 & mcd & 53,075.734 & 0.9670 & @xmath108 & 75.4 & 1.000 & 0.100 & @xmath109 & @xmath110 + 2004 apr 13 & mcd & 53,108.648 & 0.2766 & 2.9 & @xmath111 & 1.000 & 0.100 & 0.12 & @xmath112 + 2004 apr 15 & mcd & 53,110.647 & 0.4776 & 47.8 & @xmath113 & 1.000 & 0.100 & @xmath114 & @xmath106 + 2004 apr 24 & kp & 53,119.647 & 0.3826 & 35.5 & @xmath115 & 0.500 & 0.050 & 0.44 & @xmath116 + 2004 sep 25 & kp & 53,274.019 & 0.9050 & @xmath117 & 58.8 & 0.500 & 0.050 & @xmath118 & 0.21 + 2004 sep 28 & kp & 53,276.991 & 0.2039 & @xmath119 & 6.3 & 0.500 & 0.050 & @xmath120 & @xmath121 + 2004 nov 25 & mcd & 53,334.893 & 0.0260 & @xmath122 & 76.6 & 1.000 & 0.100 & 0.10 & @xmath123 + 2004 dec 03 & mcd & 53,342.913 & 0.8325 & @xmath124 & 24.6 & 1.000 & 0.100 & @xmath125 & 0.15 + 2005 jan 29 & mcd & 53,399.846 & 0.5572 & 45.6 & @xmath126 & 1.000 & 0.100 & 0.07 & 0.40 + 2005 apr 28 & mcd & 53,488.717 & 0.4934 & 48.3 & @xmath127 & 1.000 & 0.100 & @xmath128 & 0.01 + 2005 apr 29 & mcd & 53,489.670 & 0.5892 & 41.1 & @xmath129 & 1.000 & 0.100 & @xmath118 & 0.31 + 2005 apr 30 & kp & 53,490.636 & 0.6863 & 17.7 & @xmath130 & 0.500 & 0.050 & 0.58 & 0.59 + 2005 sep 21 & kp & 53,634.874 & 0.1898 & @xmath131 & 13.6 & 0.500 & 0.050 & @xmath132 & @xmath133 + 2005 oct 18 & mcd & 53,662.011 & 0.9185 & @xmath134 & 64.1 & 1.000 & 0.100 & 0.13 & 0.58 + 2005 oct 20 & mcd & 53,663.896 & 0.1080 & @xmath135 & 54.9 & 1.000 & 0.100 & 0.10 & 0.38 + 2005 oct 20 & mcd & 53,663.990 & 0.1175 & @xmath136 & 51.1 & 1.000 & 0.100 & 0.04 & 0.61 + + 2002 apr 28 & mcd & 52,392.711 & 0.1964 & 76.7 & @xmath137 & 1.000 & 0.500 & @xmath138 & @xmath118 + 2002 apr 29 & mcd & 52,393.750 & 0.3046 & 51.3 & @xmath139 & 1.000 & 0.500 & @xmath125 & @xmath140 + 2003 mar 21 & mcd & 52,719.792 & 0.2512 & 65.6 & @xmath141 & 1.000 & 0.500 & @xmath142 & @xmath143 + 2003 apr 20 & mcd & 52,749.700 & 0.3651 & 33.7 & @xmath144 & 1.000 & 0.500 & 0.06 & 0.10 + 2003 apr 20 & mcd & 52,749.892 & 0.3851 & 27.6 & @xmath145 & 1.000 & 0.500 & 0.00 & 0.18 + 2003 apr 21 & mcd & 52,750.903 & 0.4904 & @xmath146 & 23.3 & 1.000 & 0.500 & @xmath147 & 0.27 + 2003 sep 06 & mcd & 52,888.618 & 0.8289 & @xmath148 & 69.2 & 1.000 & 0.500 & @xmath147 & 0.12 + 2003 sep 07 & mcd & 52,889.586 & 0.9297 & @xmath149 & 29.8 & 1.000 & 0.500 & 0.10 & @xmath150 + 2003 sep 08 & mcd & 52,890.584 & 0.0336 & 52.0 & @xmath151 & 1.000 & 0.500 & 0.09 & @xmath100 + 2004 feb 08 & mcd & 53,043.967 & 0.0034 & 35.0 & @xmath152 & 1.000 & 0.500 & 0.07 & 0.14 + 2004 feb 09 & mcd & 53,044.932 & 0.1039 & 76.7 & @xmath153 & 1.000 & 0.500 & @xmath154 & 0.16 + 2004 apr 13 & mcd & 53,108.880 & 0.7620 & @xmath155 & 75.0 & 1.000 & 0.500 & @xmath142 & 0.00 + 2004 apr 14 & mcd & 53,109.858 & 0.8638 & @xmath156 & 59.9 & 1.000 & 0.500 & 0.15 & @xmath116 + 2004 apr 26 & kp & 53,121.792 & 0.1063 & 77.4 & @xmath101 & 0.500 & 0.250 & 0.16 & 0.21 + 2004 jun 12 & kp & 53,168.659 & 0.9860 & 24.0 & @xmath157 & 0.500 & 0.250 & @xmath120 & 0.69 + 2004 jun 13 & kp & 53,169.752 & 0.0998 & 76.0 & @xmath158 & 0.500 & 0.250 & @xmath138 & 0.34 + 2004 sep 12 & mcd & 53,260.584 & 0.5570 & @xmath159 & 41.6 & 1.000 & 0.500 & 0.01 & 0.11 + 2004 sep 13 & mcd & 53,261.583 & 0.6610 & @xmath160 & 64.7 & 1.000 & 0.500 & 0.11 & @xmath116 + 2005 apr 27 & mcd & 53,487.876 & 0.2220 & 72.3 & @xmath161 & 1.000 & 0.500 & 0.07 & 0.16 + 2005 apr 28 & mcd & 53,488.844 & 0.3228 & 46.3 & @xmath162 & 1.000 & 0.500 & 0.03 & 0.11 + 2005 apr 29 & mcd & 53,489.708 & 0.4128 & 20.0 & 0.3 & 0.000 & 0.000 & 0.74 & 0.69 + 2005 may 01 & kp & 53,491.692 & 0.6193 & @xmath163 & 56.3 & 0.500 & 0.250 & 0.09 & @xmath143 + 2005 may 02 & kp & 53,492.672 & 0.7214 & @xmath164 & 72.9 & 0.500 & 0.250 & @xmath120 & @xmath154 + 2005 jun 11 & kp & 53,532.796 & 0.8990 & @xmath165 & 46.3 & 0.500 & 0.250 & 0.13 & 0.36 + + 2002 apr 28 & mcd & 52,392.790 & 0.2518 & @xmath166 & 20.1 & 0.100 & 1.000 & 0.67 & @xmath167 + 2002 apr 29 & mcd & 52,393.881 & 0.3551 & @xmath168 & 35.1 & 0.100 & 1.000 & @xmath169 & @xmath170 + 2003 mar 21 & mcd & 52,719.891 & 0.2289 & @xmath171 & 14.3 & 0.100 & 1.000 & @xmath172 & 0.13 + 2003 apr 20 & mcd & 52,749.868 & 0.0678 & 35.7 & @xmath173 & 0.100 & 1.000 & @xmath174 & 0.10 + 2003 apr 21 & mcd & 52,750.881 & 0.1637 & & @xmath175 & & 0.000 & & @xmath176 + 2003 sep 06 & mcd & 52,888.661 & 0.2118 & & 8.8 & & 1.000 & & 0.11 + 2003 sep 07 & mcd & 52,889.664 & 0.3068 & & 30.1 & & 1.000 & & @xmath140 + 2004 feb 09 & mcd & 53,044.975 & 0.0150 & 75.0 & @xmath177 & 0.100 & 1.000 & @xmath178 & @xmath179 + 2004 mar 10 & mcd & 53,074.983 & 0.8569 & 14.3 & @xmath101 & 0.100 & 1.000 & 1.96 & @xmath180 + 2004 apr 10 & mcd & 53,105.944 & 0.7889 & & @xmath181 & & 0.000 & & 0.76 + 2004 apr 24 & kp & 53,119.942 & 0.1146 & 0.6 & @xmath182 & 0.025 & 0.500 & @xmath183 & @xmath174 + 2004 apr 26 & kp & 53,121.925 & 0.3024 & @xmath184 & 29.6 & 0.025 & 0.500 & 0.47 & @xmath167 + 2004 jun 14 & kp & 53,170.838 & 0.9345 & 62.4 & @xmath185 & 0.025 & 0.500 & 2.09 & 0.66 + 2004 jun 15 & kp & 53,171.771 & 0.0229 & 74.0 & @xmath186 & 0.025 & 0.500 & 2.95 & @xmath187 + 2004 sep 11 & mcd & 53,259.654 & 0.3456 & @xmath188 & 34.2 & 0.100 & 1.000 & @xmath114 & @xmath110 + 2004 sep 12 & mcd & 53,260.625 & 0.4375 & @xmath168 & 38.2 & 0.100 & 1.000 & 1.26 & @xmath189 + 2004 sep 13 & mcd & 53,261.620 & 0.5318 & @xmath190 & 34.8 & 0.100 & 1.000 & 1.23 & 0.16 + 2004 sep 13 & mcd & 53,261.685 & 0.5379 & @xmath191 & 34.3 & 0.100 & 1.000 & 2.07 & 0.14 + 2005 jan 29 & mcd & 53,400.003 & 0.6369 & @xmath141 & 21.6 & 0.100 & 1.000 & @xmath192 & @xmath178 + 2005 apr 27 & mcd & 53,487.910 & 0.9619 & 75.7 & @xmath193 & 0.100 & 1.000 & 0.00 & @xmath114 + 2005 apr 28 & mcd & 53,488.899 & 0.0556 & 48.0 & @xmath194 & 0.100 & 1.000 & 1.56 & 0.20 + 2005 jun 10 & kp & 53,531.732 & 0.1119 & 6.2 & @xmath195 & 0.025 & 0.500 & 2.55 & 0.25 + 2005 jun 13 & kp & 53,534.900 & 0.4119 & @xmath190 & 38.4 & 0.025 & 0.500 & 3.79 & 0.34 + 2005 jun 14 & kp & 53,535.789 & 0.4961 & @xmath188 & 37.6 & 0.025 & 0.500 & 1.56 & 0.77 + 2005 jun 15 & kp & 53,536.754 & 0.5875 & @xmath184 & 29.6 & 0.025 & 0.500 & 0.04 & 0.43 + rlrrc rr lyn & primary & @xmath196 ( 15 ) & @xmath197 ( 5 ) & @xmath198 + & secondary & @xmath199 ( 15 ) & @xmath200 ( 5 ) & @xmath201 + 12 boo & primary & @xmath202 ( 17 ) & @xmath203 ( 6 ) & @xmath204 + & secondary & @xmath205 ( 17 ) & @xmath206 ( 6 ) & @xmath207 + hr 6169 & primary & @xmath208 ( 13 ) & @xmath209 ( 8) & @xmath210 + & secondary & @xmath211 ( 15 ) & @xmath212 ( 8) & @xmath213 + lrrr @xmath214 ( days ) & 9.945080 & 9.9450740 & 9.945080 + & fixed & @xmath215 & @xmath216 + @xmath217 ( hjd ) & 2,438,048.97 & 2,444,988.49594 & 2,453,334.634 + & @xmath218 & @xmath219 & @xmath220 + @xmath59 & 0.081 & 0.0782 & 0.0793 + & @xmath221 & @xmath222 & @xmath222 + @xmath45 ( deg ) & 176.1 & 185 & 179.4 + & @xmath223 & @xmath224 & @xmath225 + @xmath47 ( km s@xmath13 ) & 65.87 & & 65.65 + & @xmath226 & & @xmath227 + @xmath48 ( km s@xmath13 ) & 83.1 & & 83.92 + & @xmath228 & & @xmath229 + @xmath49 ( km s@xmath13 ) & @xmath230 & & @xmath231 + & @xmath232 & & @xmath233 + @xmath2 ( m@xmath7 ) & 1.88 & & 1.921 + & @xmath234 & & @xmath235 + @xmath3 ( m@xmath7 ) & 1.49 & & 1.503 + & @xmath236 & & @xmath237 + @xmath0 ( @xmath238 km ) & & & 8.950 + & & & @xmath235 + @xmath1 ( @xmath238 km ) & & & 11.441 + & & & @xmath239 + rms residual ( km s@xmath13 ) & & & 0.11 + ( unit weight ) + lr @xmath240 ( deg ) & @xmath241 + @xmath242 & @xmath243 + @xmath244 & @xmath245 + @xmath69 ( @xmath246 ) ( r@xmath7 ) & @xmath247 + @xmath248 ( m@xmath7 ) & @xmath10 + @xmath249 ( m@xmath7 ) & @xmath11 + @xmath250 ( r@xmath7 ) & @xmath8 + @xmath251 ( r@xmath7 ) & @xmath9 + lrrr @xmath214 ( days ) & 9.6045492 & 9.604596 & 9.6045529 + & @xmath252 & @xmath253 & @xmath254 + @xmath217 ( hjd ) & 2,451,238.2729 & 2,452,400.4266 & 2,452,400.4292 + & @xmath255 & @xmath256 & @xmath257 + @xmath59 & 0.19233 & 0.19243 & 0.19268 + & @xmath258 & @xmath259 & @xmath260 + @xmath45 ( @xmath261 ) & 286.67 & 286.88 & 286.87 + & @xmath262 & @xmath263 & @xmath264 + @xmath47 ( km s@xmath13 ) & 67.302 & 67.283 & 67.286 + & @xmath265 & @xmath266 & @xmath267 + @xmath48 ( km s@xmath13 ) & 69.36 & 69.29 & 69.30 + & @xmath268 & @xmath227 & @xmath236 + @xmath49 ( km s@xmath13 ) & 9.551 & 9.588 & 9.578 + & @xmath269 & @xmath239 & @xmath270 + @xmath2 ( m@xmath7 ) & & 1.218 & 1.218 + & & @xmath271 & @xmath271 + @xmath3 ( m@xmath7 ) & & 1.183 & 1.183 + & & @xmath271 & @xmath272 + @xmath0 ( @xmath238 km ) & & 8.720 & 8.720 + & & @xmath221 & @xmath273 + @xmath1 ( @xmath238 km ) & & 8.981 & 8.981 + & & @xmath235 & @xmath274 + rms residual ( km s@xmath13 ) & 0.49 & 0.18 & 0.18 + rms residual ( km s@xmath13 ) & & 0.10 & 0.11 + lrr @xmath214 ( days ) & 10.56 & 10.559435 + & @xmath273 & @xmath275 + @xmath217 ( hjd ) & 2,422,422.236 & 2,453,171.529 + & @xmath276 & @xmath237 + @xmath59 & 0.430 & 0.4140 + & @xmath277 & @xmath278 + @xmath45 ( deg ) & 4.12 & 10.69 + & @xmath279 & @xmath280 + @xmath47 ( km s@xmath13 ) & 62.41 & 71.35 + & @xmath281 & @xmath282 + @xmath48 ( km s@xmath13 ) & 101.36 & 95.46 + & @xmath283 & @xmath284 + @xmath49 ( km s@xmath13 ) & @xmath285 & @xmath286 + & @xmath287 & @xmath288 + @xmath2 ( m@xmath7 ) & 2.19 & 2.197 + & & @xmath277 + @xmath3 ( m@xmath7 ) & 1.35 & 1.642 + & & @xmath289 + @xmath0 ( @xmath238 km ) & 8.18 & 9.431 + & & @xmath269 + @xmath1 ( @xmath238 km ) & 13.28 & 12.618 + & & @xmath290 + rms residual ( km s@xmath13 ) & 4.3 & 0.26 + ( unit weight ) +

we present a microscopic theory of skyrmions in the monolayer quantum hall ferromagnet . it is a peculiar feature of the system that the number density and the spin density are entangled intrinsically as dictated by the w algebra . the skyrmion and antiskyrmion states are constructed as w-rotated states of the hole - excited and electron - excited states , respectively . they are spin textures accompanied with density modulation that decreases the coulomb energy . we calculate their excitation energy as a function of the zeeman gap and compared the result with experimental data .

cond-mat0509262

quantum coherence in the quantum hall ( qh ) system has proved to be a fascinating subject during the past decade@xcite . electron spins are spontaneously polarized even in the absence of the zeeman effect due to the exchange coulomb interaction , leading to the qh ferromagnet . a prominent characteristic is that a charged excitation is a spin texture called skyrmion@xcite . it is a coherent excitation of spins , and its excitation can be confirmed if the number of flipped spins is found to be more than that of a hole excitation@xcite . a conventional way is to measure the increase of the activation energy by tilting samples@xcite , which is roughly proportional to the number of flipped spins due to the zeeman effect . the skyrmion has been known@xcite to be a classical solution to the nonlinear sigma model . indeed , the concept of skyrmion was introduced@xcite into qh ferromagnets first in this context . subsequently a microscopic skyrmion state was considered to carry out a hartree - fock approximation@xcite . in this paper we elaborate this idea and present a microscopic theory of skyrmions at the filling factor @xmath2 , employing the framework of noncommutative geometry@xcite . we also compare the result with the experimental data@xcite . the property of electrons becomes very peculiar when they are confined within the lowest landau level@xcite . the electron position is described solely by the guiding center @xmath3 subject to the noncommutative relation , @xmath4=-i\ell _ { b}^{2}$ ] , with @xmath5 the magnetic length . this noncommutativity is related with the so - called w@xmath0 algebra@xcite . when the spin degree of freedom is introduced , the algebraic property becomes the su(2 ) extension of the w@xmath1 algebra@xcite , which we have named the w@xmath6(2 ) algebra@xcite . it implies an intrinsic entanglement of the number density and the spin density of electrons , which amounts to a spin excitation accompanied with a density modulation . this paper is organized as follows . in section [ secdensiopera ] we formulate the number density and the spin density of electrons confined within the lowest landau level . they form the w@xmath1(2 ) algebra . in section [ sechamil ] we address the question whether there are states having the same electron number but a lower excitation energy compared with the hole state . in section [ secmicroskyrm ] we propose the skyrmion and antiskyrmion states as w@xmath1(@xmath7)-rotated states of the hole - excited and electron - excited states , respectively . they agree with the anzatsfertig94b made in a hartree - fock approximation . since a w@xmath1(@xmath7 ) rotation modulates not only the spin density but also the number density , it decreases the coulomb energy of the excitation . in section [ secwavefunct ] we derive the wave function of the skyrmion state . in particular we examine the case where the wave function is factorizable in the electron coordinates . we call such a skyrmion the factorizable skyrmion . in section [ sechardcore ] we study the system governed by the hard - core interaction instead of the coulomb interaction , where the factorizable skyrmion is an eigenstate of the hamiltonian . however , it is shown that skyrmion excitations are energetically unfavored once the zeeman effect is taken into account . in section [ secintercoulo ] we study the realistic system with the coulomb interaction together with the zeeman interaction . the factorizable skyrmion can not be physical since its zeeman energy diverges . we propose a trial skyrmion which interpolates the hole and the factorizable skyrmion . by minimizing the total energy we determine the state as well as the excitation energy as a function of the zeeman gap . the result is compared successfully with the experimental data@xcite in section [ secexperevide ] . in section ix we re - examine the semiclassical approximation@xcite in the present context of the microscopic theory . it is shown that , though strictly speaking the factorizable skyrmion can not be physical , it still presents a reasonable approximation with an appropriate cutoff of the divergent zeeman energy . section [ secdiscu ] is devoted for the conclusion . electrons in a plane perform cyclotron motion under strong magnetic field @xmath8 and create landau levels . the coordinate @xmath9 of the electron is decomposed into the guiding center @xmath3 and the relative coordinate @xmath10 , @xmath11 , where @xmath12 and @xmath13 with @xmath14 the covariant momentum . in the symmetric gauge the decomposition implies@xmath15where @xmath16 is the magnetic length . when the cyclotron energy gap is large , thermal excitations across landau levels are practically suppressed at sufficiently low temperature . in such a system the electron position is specified solely by the guiding center subject to the noncommutative relation , @xmath17=-i\ell _ { b}^{2}. \label{algebnc}\]]due to this noncommutative relation an electron can not be localized and occupies an area @xmath18 called the landau site . there are @xmath19 landau sites per one landau level , where @xmath20 is the area of the system . it is equal to the number @xmath21 of flux quanta passing through the system , where @xmath22 is the dirac flux quantum . one landau site may accommodate two electrons with up and down spins according to the pauli exclusion principle . the filling factor is @xmath23 , where @xmath24 is the total number of electrons and @xmath25 is the electron density . the system becomes incompressible , leading to the integer qh effect , when the filling factor @xmath26 takes an integer value . in this paper we study the qh system at @xmath2 . thus all electrons are assumed to be confined within the lowest landau level . we define operators @xmath27with ( [ reprediffe ] ) . they obey @xmath28=1 $ ] . the landau site @xmath29 . its wave function is easily calculable,@xmath30where @xmath31 and @xmath32 . we now construct the fock space @xmath33 with the use of the creation operator @xmath34 on this landau site @xmath35 , satisfying @xmath36 , where @xmath37 denotes the spin index , @xmath38,@xmath39 . we expand the two - component electron field @xmath40,@xmath41 in terms of wave functions,@xmath42the physical variables are the number density @xmath43 and the spin density @xmath44 with the pauli matrix @xmath45 . substituting ( [ electinlll ] ) into these density operators , we obtain [ physidensi ] @xmath46 @xmath47we employ the relation @xmath48 and transform them in the momentum space , @xmath49where [ projedensiq ] @xmath50we call @xmath51 and @xmath52 the bare densities . in the coordinate space we have the relation @xmath53 between the physical density @xmath54 and the bare density @xmath55 . the density operators generate the su(2 ) extension of the w@xmath1 algebra,@xmath56 & = \delta _ { mj}\rho ( i , n)-\delta _ { in}\rho ( m , j ) , \notag \\ \lbrack \rho ( m , n),s_{a}(i , j ) ] & = \delta _ { mj}s_{a}(i , n)-\delta _ { in}s_{a}(m , j ) , \notag \\ \lbrack s_{a}(m , n),s_{b}(i , j ) ] & = \frac{i}{2}\varepsilon _ { abc}\left [ \delta _ { mj}s_{c}(i , n)+\delta _ { in}s_{c}(m , j)\right ] \notag \\ & \hspace{0.7cm}+\frac{1}{4}\delta _ { ab}\left [ \delta _ { mj}\rho ( i , n)-\delta _ { in}\rho ( m , j)\right ] , \label{walgebp}\end{aligned}\]]which we have named@xcite the w@xmath1(2 ) algebra . the electron density and the spin density are intrinsically entangled in the algebra , as implies that a spin rotation induces a density modulation . the hamiltonian of the qh system consists of a four - fermion interaction term and the zeeman term , @xmath57 with [ hamilqhf ] @xmath58where @xmath59 is the density excitation operator , @xmath60 and @xmath61 is the zeeman gap with @xmath62 the bohr magneton . in the actual system @xmath63 is given by the coulomb potential , but here we only assume that it represents a repulsive interaction . substituting the density operators ( [ physiandhat ] ) into these we obtain @xmath64 where @xmath65 with @xmath66 the fourier transformation of the potential @xmath67 . we have used the notation @xmath68 where @xmath69 and @xmath70 are the direct and exchange energy parameters , respectively . in the qh state at @xmath2 , the number of electrons is equal to the number of landau sites . one landau site contains one electron due to the repulsive interaction . the system without the zeeman effect ( @xmath71 ) is most interesting . an intriguing feature is that the su(2 ) symmetry is spontaneously broken by the repulsive interaction and all spins are spontaneously polarized , as demonstrated in appendix [ appspontsymme ] . thus the qh system is a ferromagnet , where quantum coherence develops spontaneously . without loss of generality we may take @xmath72 as the ground state . it satisfies @xmath73and@xmath74a hole - excited and electron - excited states are given by@xmath75with their energies@xmath76the main question is whether there are states having the same electron number but a lower excitation energy compared with the hole - excited or the electron - excited state . we consider a w@xmath1(@xmath7)-rotated state of the hole - excited state @xmath77h@xmath78 and the electron - excited state @xmath77e@xmath78 , [ skyrmformua ] @xmath79where @xmath80 are elements of the w@xmath1(2 ) algebra ( [ walgebp ] ) . the coulomb energy of the state depends on @xmath80 , since the w@xmath0(@xmath7 ) rotation modulates not only the spin texture but also the electron density around the hole or electron excitation . the element @xmath81 is to be determined by requiring the excitation energy to be minimized . as we shall see soon , @xmath82 and @xmath83 describe the skyrmion and antiskyrmion states . we use the index ( @xmath84 ) for the skyrmion and ( @xmath85 ) for the antiskyrmion . similarly , @xmath86 describe a multi - skyrmion and a multi - antiskyrmion as w@xmath1(2)-rotated states of a multi - hole state and a multi - electron state . more generally we may consider a wide class of states presented by@xmath87where @xmath88 is a state of the form @xmath89 ^{\nu _ { \mu } ( n)}|0\rangle , \label{purestate}\]]where @xmath90 may take the value either @xmath91 or @xmath92 depending whether the spin state @xmath37 at a site @xmath93 is occupied or not . the class of states ( [ generstate ] ) is quite general though it may not embrace all possible ones . nevertheless all physically relevant states seem to fall in this category . indeed , as far as we know , perturbative excitations are spin waves and nonperturbative excitations are skyrmions in qh systems . the corresponding states belong surely to this category . the electron number of the w@xmath1(2)-rotated state ( [ generstate ] ) is easily calculable,@xmath94since @xmath95 is a casimir operator . we set@xmath96this is the electron number carried by the excitation described by the state @xmath97 . it is an integer since the electron number of the state @xmath97 is the same as that of the state @xmath98 . we examine more in detail the state ( [ skyrmformua ] ) with @xmath99 . a simplest w@xmath1(2 ) rotation mixes only neighboring two sites and is given by the choice of @xmath100 with@xmath101 , \]]where @xmath102 is a real parameter . note that @xmath103 is a hermitian operator belonging to the w@xmath1(2 ) algebra , and @xmath104=0 $ ] . we find@xmath105since the spin - up operator @xmath106 is contained only in @xmath103 . we calculate @xmath107 using the standard technique of deriving the differential equation with respect to @xmath102 . since it satisfies@xmath108together with the initial condition @xmath109 at @xmath110 , we integrate it as @xmath111where we have set @xmath112 and @xmath113 . thus the constraint @xmath114is imposed on @xmath115 and @xmath116 . we now find@xmath117where we have used ( [ wtransa ] ) and @xmath118 . similarly we choose @xmath119 in ( [ skyrmformuae ] ) with @xmath120 .\]]we find@xmath121where@xmath122together with@xmath123since @xmath124 and @xmath125 . the states ( [ microskyrmstate ] ) and ( [ microskyrmstateanti ] ) agree with the skyrmion state and the antiskyrmion state proposed by fertig et al.fertig94b , respectively . the operators @xmath126 and @xmath127 satisfy the standard canonical anticommutation relations,@xmath128with the use of the constrains ( [ condionu ] ) and ( [ condionv ] ) . since these states should approach the ground state asymptotically it is necessary that@xmath129for later convenience we define@xmath130 in accordance with ( [ condionu ] ) and ( [ condionv ] ) . in what follows we calculate explicitly the classical densities@xmath131the basic relations for the skyrmion state are@xmath132they reduce the action of @xmath133 operators to that of @xmath134 , thus allowing to carry out exact calculus . we employ @xmath135 together with ( [ basicformusky ] ) and its conjugate . in this way we come to@xmath136 , s_{x}^{\pm } ( n+1,n ) & = s_{x}^{\pm } ( n , n+1)=\frac{1}{2}u_{\pm } ( n)v_{\pm } ( n ) , \notag \\ s_{y}^{\pm } ( n+1,n ) & = -s_{y}^{\pm } ( n , n+1)=\frac{i}{2}u_{\pm } ( n)v_{\pm } ( n ) . \label{microdensia}\end{aligned}\]]all other components , @xmath137 , etc . , vanish . here and hereafter we omit the superscript `` cl '' . by substituting these into the physical density ( [ physidensi ] ) , we obtain @xmath138 \left ( \frac{|z|^{2}}{2}\right ) ^{n } , \notag \\ \frac{s_{z}^{\pm } ( \mathbf{x})}{\rho _ { 0}}= & \pm \frac{1}{2}% e^{-|z|^{2}/2}\sum_{n=0}\frac{u_{\pm } ^{2}(n)-v_{\pm } ^{2}(n-1)}{n!}\left ( \frac{|z|^{2}}{2}\right ) ^{n } , \notag \\ \frac{s_{x}^{\pm } ( \mathbf{x})}{\rho _ { 0}}= & \frac{x}{\sqrt{2}\ell _ { b}}% e^{-|z|^{2}/2}\sum_{n=0}\frac{u_{\pm } ( n)v_{\pm } ( n)}{n!\sqrt{n+1}}\left ( \frac{|z|^{2}}{2}\right ) ^{n } , \notag \\ \frac{s_{y}^{\pm } ( \mathbf{x})}{\rho _ { 0}}= & \pm \frac{y}{\sqrt{2}\ell _ { b}}% e^{-|z|^{2}/2}\sum_{n=0}\frac{u_{\pm } ( n)v_{\pm } ( n)}{n!\sqrt{n+1}}\left ( \frac{|z|^{2}}{2}\right ) ^{n}. \label{microdensic}\end{aligned}\]]we have calculated the classical densities of the w@xmath1(@xmath7)-rotated states of a hole - excited and electron - excited states , which contain infinitely many variables @xmath139 and @xmath140 . we estimate the energies of these states and minimize them in later sections . we consider the @xmath24-electron system over @xmath141 sites . the wave function of a many - body state @xmath97 is defined by @xmath142=\langle 0|\psi _ { \mu _ { 1}}(\mathbf{x}_{1})\psi _ { \mu _ { 2}}(\mathbf{x}_{2})\cdots \psi _ { \mu _ { n}}(\mathbf{x}_{n})|\mathfrak{s}\rangle , \]]where @xmath143 is given by ( [ electinlll ] ) , or @xmath144with@xmath145for the hole - excited state @xmath77h@xmath146g@xmath78 it is easy to see @xmath147=\prod_{r}^{n}\left ( \begin{array}{@{\,}c } z_{r } \\ 0% \end{array}% \right ) \mathfrak{s}_{\text{ln}}[\mathbf{x}],\]]where @xmath148 $ ] is the slater determinant of the one - body wave functions , @xmath149=\rho _ { 0}^{n/2}\left\vert \begin{array}{cccc } 1 & z_{1 } & \cdots & z_{1}^{n-1 } \\ 1 & z_{2 } & \cdots & z_{2}^{n-1 } \\ \vdots & \vdots & \ddots & \vdots \\ 1 & z_{n } & \cdots & z_{n}^{n-1}% \end{array}% \right\vert e^{-\sum_{r=1}^{n}|z_{r}|^{2}/4}. \label{laughwave}\]]the wave function of the state @xmath77h@xmath78 is factorizable in the electron coordinates apart from the factor @xmath150 $ ] . we proceed to derive the wave function of the skyrmion state ( microskyrmstate ) . first we examine the component with all spins up , to which only the term @xmath151 in @xmath152 contributes,@xmath153=\prod_{n=0}^{n-1}v_{-}(n)\langle 0|\prod_{i=1}^{n}\psi _ { \uparrow } ( % \mathbf{x}_{i})\prod_{n=1}^{n}c_{\uparrow } ^{\dagger } ( n)|0\rangle .\]]apart from the factor @xmath154 this is nothing but the wave function of a hole . thus@xmath153=c_{n}\prod_{n=1}^{n}z_{n}\mathfrak{s}_{\text{ln}}[\mathbf{x } ] , \label{skyrmwaveu}\]]where @xmath155 . next we consider the component with all spins down , which arises only from the term @xmath156 in @xmath107,@xmath157= & \prod_{n=1}^{n}u_{-}(n)\langle 0|\prod_{i=1}^{n}\psi _ { \downarrow } ( \mathbf{x% } _ { i})\prod_{n=1}^{n}c_{\downarrow } ^{\dagger } ( n)|0\rangle \notag \\ = & c_{n}\prod_{n=0}^{n-1}\beta _ { n}\mathfrak{s}_{\text{ln}}[\mathbf{x } ] . \label{skyrmwaved}\end{aligned}\]]here we have set@xmath158with ( [ paramalpha ] ) . comparing ( [ skyrmwaveu ] ) and ( [ skyrmwaved ] ) we remark that the wave function @xmath159 $ ] is obtained by replacing @xmath160 with @xmath161 within the factor @xmath162 of the wave function @xmath163 $ ] for all @xmath93 . the wave function with mixed spin components is similarly derived , where @xmath164 is replaced with @xmath161 for certain indices @xmath93 within the factor @xmath162 . in general we derive@xmath165=c_{n}e^{-\sum_{r=1}^{n}|z_{r}|^{2}/4 } \notag \\ & \times \left\vert \begin{array}{cccc } \left ( \begin{array}{c } z_{1 } \\ \beta _ { 0}% \end{array}% \right ) _ { \mu _ { 1 } } & z_{1}\left ( \begin{array}{c } z_{1 } \\ \beta _ { 1}% \end{array}% \right ) _ { \mu _ { 1 } } & \cdots & z_{1}^{n-1}\left ( \begin{array}{c } z_{1 } \\ \beta _ { n-1}% \end{array}% \right ) _ { \mu _ { 1 } } \\ \left ( \begin{array}{c } z_{2 } \\ \beta _ { 0}% \end{array}% \right ) _ { \mu _ { 2 } } & z_{2}\left ( \begin{array}{c } z_{2 } \\ \beta _ { 1}% \end{array}% \right ) _ { \mu _ { 2 } } & \cdots & z_{2}^{n-1}\left ( \begin{array}{c } z_{2 } \\ \beta _ { n-1}% \end{array}% \right ) _ { \mu _ { 2 } } \\ \vdots & \vdots & \ddots & \vdots \\ \left ( \begin{array}{c } z_{n } \\ \beta _ { 0}% \end{array}% \right ) _ { \mu _ { n } } & z_{n}\left ( \begin{array}{c } z_{n } \\ \beta _ { 1}% \end{array}% \right ) _ { \mu _ { n } } & \cdots & z_{n}^{n-1}\left ( \begin{array}{c } z_{n } \\ \beta _ { n-1}% \end{array}% \right ) _ { \mu _ { n}}% \end{array}% \right\vert . \label{skyrmwavefunct}\end{aligned}\]]this is the wave function of the skyrmion . it is notable that , when all @xmath166 are equal ( @xmath167 ) , or@xmath168it is reduced to @xmath169=\prod_{r}^{n}\left ( \begin{array}{@{\,}c } z_{r } \\ \sqrt{2}\omega% \end{array}% \right ) \mathfrak{s}_{\text{ln}}[\mathbf{x } ] . \label{factoskyrmwave}\]]the wave function is factorizable as in the case of a hole . we call such a skyrmion the factorizable skyrmion . as we shall see in the following section , this is the wave function of the skyrmion in the hard - core model . it is quite difficult to write down the wave function of an antiskyrmion in terms of the analytic variable @xmath170 . we discuss it in appendix secantiskyrm . we first investigate a detailed structure of skyrmions in the system governed by the hard - core interaction,@xcite@xmath171instead of the coulomb interaction . the hamiltonian ( [ spincoulo ] ) reads@xmath172all previous formulas hold with@xmath173and@xmath174we rewrite ( [ contahamil ] ) into the normal ordered form,@xmath175the ground state @xmath77g@xmath78 is given by ( [ grounstate ] ) as an eigenstate of the hamiltonian , @xmath176g@xmath177 . a hole - excited and electron - excited states are given by @xmath77h@xmath146g@xmath178 and @xmath77e@xmath179g@xmath78 . we determine the skyrmion and antiskyrmion states by minimizing the energy of the w@xmath1(@xmath7)-rotated state ( [ skyrmformua ] ) . we deal with the skyrmion state ( [ microskyrmstate ] ) explicitly , but similar formulas follow also for the antiskyrmion state ( [ microskyrmstateanti ] ) . we define the state@xmath180using ( [ basicformusky ] ) we come to @xmath181where we have used @xmath182 . hence @xmath183using ( [ contastepa ] ) we obtain@xmath184 ^{2}\geqslant 0.\end{aligned}\]]the skyrmion energy @xmath185 is minimized when the equality is achieved , as occurs when the factorizable - skyrmion condition ( factocondi ) is satisfied , or@xmath186due to the normalization ( [ condionu ] ) . it follows @xmath187 when the equality holds in ( [ stephca ] ) . hence , @xmath82 is an eigenstate of the hamiltonian with @xmath188,@xmath189similarly , when@xmath190the antiskyrmion state is an eigenstate of the hamiltonian with @xmath191,@xmath192we may summarize them as @xmath193 with @xmath194 . the skyrmion state and the antiskyrmion state are the lowest energy states possessing the electron numbers @xmath188 and @xmath195 , respectively . in conclusion , we have determined the w@xmath1(2)-rotated states ( [ skyrmformua ] ) by minimizing the energy , though their energy is independent of the scale @xmath196 and degenerates with the hole - excited and electron - excited state . the physical densities can be expressed in terms of the kummer function @xmath197 , @xmath198as @xmath199with @xmath200 . we have illustrated the density @xmath201 for typical values of the parameters @xmath196 in fig.[figskydensi ] . ) accompanied with the skyrmion excitation is plotted as a function of the radius ( @xmath202 ) . the heavy curve ( @xmath203 ) is for the hard - core model based on ( [ microdensid ] ) for the choice of @xmath204 , @xmath205 , @xmath92 , @xmath7 , while the heavy curve ( @xmath206 ) represents the density of a hole . the thin curves represent the interpolating formula ( microdensie ) for @xmath207 , @xmath208 , where the parameter @xmath209 is defined by ( [ condionuvb]).,scaledwidth=50.0% ] using the relations@xmath210we are able to summarize the densities as@xmath211with@xmath212and@xmath213where we have set @xmath214 . it is interesting that the physical spin densities @xmath215 are factorizable into the number density @xmath216 and the normalized spin field @xmath217 as in ( [ skyrmconficonta ] ) . the normalized spin field ( [ standskyrm ] ) agrees with the nonlinear - sigma - model skyrmion configuration@xcite . we explore the properties of the density modulation . the skyrmion is reduced to a hole for @xmath218 , where @xmath216 is given by@xmath219and vanishes at @xmath220 . on the other hand , @xmath221 for all @xmath222 with @xmath223we can expand @xmath216 in a power series of @xmath224 as follows . let us set@xmath225by comparing ( [ kummeeqa ] ) with ( [ kummeeqb ] ) , @xmath226 is found to be the kummer function of the form@xmath227with @xmath228 , and satisfies the kummer equation,@xmath229this can be solved by expanding @xmath230 in a power series of @xmath224 , @xmath231 . in particular , the lowest order term is given by setting @xmath232 , whose result is@xmath233hence we have @xmath234or [ skyrmstepe ] @xmath235the densities @xmath236 as well as @xmath237 approach the ground - state values only polynomially . finally we examine what happens when the zeeman interaction is taken into account . the number of spins flipped around a skyrmion is given by @xmath238 which we call the skyrmion spin . substituting ( [ skyrmconficonta ] ) into this , unless @xmath218 , we find @xmath239 to diverge logarithmically due to the asymptotic behavior,@xmath240the zeeman energy @xmath241 is divergent , except for the hole , from the infrared contribution however small the zeeman effect is . the factorizable skyrmion ( skyrmconficonta ) is no longer valid . there exists a skyrmion state which has a finite zeeman energy : see an example of ( [ condionuvb ] ) we use for the coulomb interaction . nevertheless , we can show that the hole state has the lowest energy . the reason reads as follows . the factorizable skyrmion is an eigenstate of the hard - core hamiltonian , @xmath242 with @xmath243 , as in ( contaskyrmenerg ) . accordingly any spin texture @xmath97 possessing the same electron number @xmath244 has a higher energy , @xmath245 . furthermore its zeeman energy is larger than that of the hole , @xmath246 . hence , @xmath247where the equality holds for the hole state . consequently there are no skyrmions in the presence of the zeeman interaction in the system with the hard - core interaction . we next investigate the realistic system ( [ spincoulo ] ) governed by the coulomb hamiltonian @xmath248 with the potential@xmath249with @xmath250 the coulomb energy unit . it is hard to construct the skyrmion state explicitly as an eigenstate . we are satisfied to estimate the excitation energy of a skyrmion by minimizing the expectation value of the hamiltonian . before so doing , it is instructive to estimate the coulomb energy by using the factorizable skyrmion ( [ skyrmconficonta ] ) obtained in the hard - core model . the result is very different from the one in the hard - core model . the coulomb energy is a monotonously decreasing function of the scale parameter @xmath251 , and@xmath252with@xmath253consequently , an infinitely large skyrmion is necessarily excited . however , when the zeeman interaction is introduced , the zeeman energy of the factorizable skyrmion diverges due to the asymptotic behavior ( [ asympsz ] ) . namely the factorizable skyrmion can not be a quasiparticle in the realistic coulomb system . it is necessary to consider a skyrmion not factorizable as in ( skyrmconficonta ) . making a slight generalization of the parametrization ( [ skyrmanzatl ] ) we search for a skyrmion possessing a finite energy even in the presence of the zeeman effect . we propose an anzats , @xmath254the parameter @xmath209 presents a smooth interpolation between the hole ( @xmath206 ) and the factorizable skyrmion ( @xmath203 ) . this anzats automatically satisfies the condition ( [ condiuvatm ] ) for @xmath255 . substituting ( [ condionuvb ] ) into ( [ microdensic ] ) we obtain@xmath256where@xmath257it is easy to see that various densities approach the ground - state values exponentially fast for all @xmath258 . using ( [ skyrmstepb ] ) , we can rewrite these as@xmath259 we have illustrated the density @xmath201 for typical values of the parameters @xmath196 and @xmath209 in fig.[figskydensi ] . it is instructive to expand them in power series of @xmath224 . setting@xmath260and writing down similar equations to ( [ kummeeq ] ) , we obtain the lowest order term as @xmath261 e^{-r^{2}/\beta ^{2 } } , \label{skyrmstepfa } \\ \frac{s_{z}^{\pm } ( \mathbf{x})}{\rho _ { 0}}= & \frac{1}{2}-\left [ \frac{% \lambda ^{2}}{r^{2}+\lambda ^{2}}+o(\ell _ { b}^{2})\right ] e^{-r^{2}/\beta ^{2}}.\end{aligned}\ ] ] the zeeman energy remains finite due to a rapid decrease to the ground - state value . in calculating the expectation value of the coulomb hamiltonian we appeal to the decomposition formula , @xmath262 dictating that the coulomb energy consists of the direct energy @xmath263 and the exchange energy @xmath264 . we have already used it on a case - by - case basis in our previous papers.ezawax03b,ezawax04b,ezawax05b in appendix [ secdecomformu ] we derive the formula for the class of states ( [ generstate ] ) . here the direct and exchange energies read [ decomdx ] @xmath265 , \label{decomx}\end{aligned}\]]in terms of the bare densities , where @xmath266 with @xmath267 and @xmath268 is the modified bessel function . . the heavy solid curve is obtained by the numerical analysis based on the density formulas ( [ densicoulo ] ) . the thin line is the hole excitation energy . it is seen that a skyrmion is excited for @xmath269 but a hole is practically excited for @xmath270 . we have also given the excitation energy calculated analytically based on the simplified formula ( [ simplresul ] ) , which is represented by the dotted curve . the two curves are practically identical for @xmath271 the inset is an enlargement of the figure near @xmath272 , where three curves meet.,scaledwidth=40.0% ] we are able to determine the parameters @xmath196 and @xmath209 so as to minimize the sum of the coulomb and zeeman energies , @xmath273 , as a function of the zeeman gap @xmath274 . calculating numerically @xmath275 as a function of @xmath276 and @xmath209 for a given value of @xmath274 , we determine the values of @xmath196 and @xmath209 which minimizes @xmath275 . in this way we obtain the skyrmion excitation energy @xmath277 as a function of @xmath274 . we have plotted the excitation energy @xmath275 as a function of the normalized zeeman gap @xmath278 in fig.[figskyenespinth ] , where @xmath279 is the coulomb unit . a skyrmion excitation is characterized by a coherent excitation of spins . hence an evidence of the skyrmion excitation is given if @xmath280 , where @xmath281 for a hole or an electron excitation . it is a remarkable fact that skyrmions have already been observed experimentally in qh systems@xcite . let us show the number of flipped spins is given by@xmath282the excitation energy @xmath283 is the sum of the coulomb energy @xmath284 and zeeman energy @xmath285 , which depend on a set of parameters ( @xmath209,@xmath196 ) denoted collectively by @xmath209 for simplicity . the quantity to minimize with respect to @xmath209 is @xmath286at the minimum we obtain @xmath287from which we solve out @xmath288 and substituting it back into ( [ spinnumbestepb ] ) , [ spinnumbestepd ] @xmath289 + \delta _ { \text{z}}n_{\text{spin}}\left [ t_{0}(\delta _ { \text{z}% } ) \right ] , \\ n_{\text{spin}}(\delta _ { \text{z}})= & n\left [ t_{0}(\delta _ { \text{z}})% \right ] .\end{aligned}\]]using ( [ spinnumbestepa ] ) it is easy to verify that @xmath290 which is ( [ spinnumbeenerg ] ) . of a skyrmion - antiskyrmion pair is compared with experimental data at @xmath291 . the data are taken from schmeller et . al .. schmeller95l the theoretical curve is based on the formula ( activenergformu ) with @xmath292 . we have taken @xmath293 for sample si1 . the offset @xmath294 increases as the mobility decreases . the skyrmion spin is one half of the slope of the activation - energy curve , @xmath295 , where @xmath296 is taken in units of @xmath297 . the number @xmath298 depends sensitively on the normalized zeeman gap @xmath299 for small @xmath299 . the thin line is for the hole excitation energy . it is seen that a hole is practically excited for @xmath300.,scaledwidth=45.0% ] to compare our theoretical result with experimental data , it is necessary to take into account two points so far neglected . first , what we observe experimentally is the thermal activation energy of a skyrmion - antiskyrmion pair . but this activation takes place in the presence of charged impurities . the existence of charged impurities reduces the activation energy considerably . we include an offset parameter @xmath301 to treat this effect phenomenologically.@xcite second , we have so far assumed an ideal two - dimensional space for electrons . this is not the case . electrons are confined within a quantum well of a finite width . this modifies the coulomb energy considerably . the coulomb energy becomes smaller than what we have assumed . it is quite difficult to make a rigorous analysis of the coulomb energy in an actual quantum well . we simulate the effect by including the reduction factor @xmath302 . we consider the excitation energy of a skyrmion - antiskyrmion pair since it is an observable quantity . it is simply twice of the skyrmion excitation energy . taking into account these two points , instead of ( spinnumbestepb ) we set the activation energy as @xmath303where @xmath304 . repeating the same steps as for ( [ spinnumbestepd ] ) we come to [ spinnumbestepe ] @xmath305with the use of @xmath306 and @xmath307 derived in ( [ spinnumbestepd ] ) . experimental data@xcite were obtained in three samples ( fig.figskyene56 ) : a single heterointerface ( si1 ) and two gasa single quantum wells ( qw1 and qw2 ) with widths of 20 nm and 14 nm . the sample si1 has a much wider thickness of the two - dimensional sheet . in fig.[figskyene56 ] , from top to bottom the sample mobilities are @xmath308 , @xmath309 , @xmath310 and @xmath311 @xmath312 , respectively . we have compared our theoretical result ( [ activenergformu ] ) successfully with the experimental data by making appropriate choices of the reduction factor @xmath302 and the offset parameter @xmath294 in fig.[figskyene56 ] . we have used @xmath313 for all samples . it would imply the thickness independence of the excitation energy , suggesting that a skyrmion has a certain fixed size in the third direction . however this problem is yet to be explored . on the other hand we have used different values of @xmath294 for each . we have set @xmath314 for sample si1 , and have found that @xmath294 increases as the mobility decreases . this is reasonable since both the mobility and the offset @xmath315 represent impurity effects . the formulas we have used to estimate the skyrmion excitation energy are quite complicated . the modulations of the electron density and the spin density , being expressed only in infinite power series as in ( densicoulo ) , are very difficult to handle with . skyrmion excitations have been observed also in bilayer qh systemsmurphy94l , sawadax03pe , terasawa04 , where it is practically impossible to repeat the calculations carried out here because too many dynamical variables are involved . we have already applied a semiclassical approximation to those systems@xcite . in this section we compare the semiclassical result with the microscopic result . in the semiclassical approximation we can make the spin - charge separation of the electron field based on the composite - boson picture,@xcite @xmath316 where the u(1 ) filed @xmath317 carries the electric charge while the cp@xmath318 field @xmath319 carries the spin . it leads to the factorization of the spin field @xmath320 into the number density @xmath321 and the normalized spin field @xmath322,@xmath323with@xmath324the cp@xmath318 field of the skyrmion is given by@xcite@xmath325together with the normalized spin field ( [ standskyrm ] ) , and the wave function is factorizable as in ( [ factoskyrmwave ] ) . we may also derive the soliton equation@xcite@xmath326which determines the density modulation around the charge excitation . we evaluate the excitation energy in the lowest order of @xmath224 by assuming a large - scale skyrmion . it corresponds to the commutative limit in the microscopic scheme . the coulomb energy is given by ( [ decomdx ] ) . to calculate the direct energy ( [ decomd ] ) we take the leading term in the solution of the soliton equation ( [ soliteq]),@xmath327which agrees with the leading term in the density modulation ( skyrmstepea ) in the microscopic theory . the direct coulomb energy ( decomd ) reads@xmath328where @xmath329 . in calculating the exchange energy ( [ decomx ] ) , since the exchange potential @xmath330 is short - ranged , we make the derivative expansion and take the lowest order term . it is to make an approximation in ( [ decomx]),@xmath331with ( [ spinstiff ] ) . furthermore we approximate @xmath332 in the exchange energy in the lowest order of @xmath333 . then the exchange energy turns out to be the nonlinear sigma model,@xmath334^{2}.\]]this is bounded from below,@xmath335the lower bound , @xmath336 , is saturated by the skyrmion configuration ( [ standskyrm ] ) . we note that the hamiltonian @xmath337 describes a spin wave . the coherence length @xmath134 is given by@xmath338 in the presence of the zeeman effect . we cut off the divergence of the zeeman energy . first , the skyrmion excitation occurs within the coherent domain with the coherence length @xmath134 . second , the skyrmion size is proportional to the scale parameter @xmath339 . the zeeman energy is approximated as @xmath340where we have cut off the upper limit of the integration at @xmath341 with a phenomenological parameter @xmath342 . substituting these into ( [ skyrmenergd ] ) , we obtain@xmath343 \epsilon _ { % \text{c}},\]]where we have introduced another phenomenological parameter @xmath344 to reduce the exchange energy , since the true exchange energy is smaller than @xmath345 . minimizing this with respect to @xmath339 analytically we obtain @xmath346with@xmath347we have plotted the result in fig.[figskyenespinth ] . it reproduces excellently the previous numerical result for @xmath348 with the choice of @xmath349 and @xmath350 . we may practically use it even in the region where a hole is excited . we have thus confirmed the validity of the semiclassical approximation . in this paper we have presented a microscopic theory of skyrmions in qh ferromagnets . we have shown that the skyrmion is a w@xmath1(2)-rotated state of a hole - excited state . because of an intrinsic entanglement between the electron density and the spin density implied by the w@xmath1(2 ) algebra , a w@xmath1(2 ) rotation modulates not only the spin configuration but also the electron density around a hole , thus decreasing the coulomb energy . similarly , the antiskyrmion is a w@xmath1(2)-rotated state of an electron - excited state . there is a simple type of skyrmion state characterized by the fact that its wave function is factorizable in the electron coordinates . we call it the factorizable skyrmion . it corresponds to the nonlinear - sigma - model skyrmion previously derived in the semiclassical approximation . we have analyzed the skyrmion state in the realistic coulomb system with the zeeman interaction . by minimizing the excitation energy we have estimated the activation energy of a skyrmion - antiskyrmion pair . the result is found to explain the experimental data@xcite remarkably well . we are grateful to the hospitality of theoretical physics laboratory , riken , where a part of this work was done . on of the authors ( zfe ) is supported in part by grants - in - aid for scientific research from ministry of education , science , sports and culture ( nos . 13135202,14540237 ) . the other author ( gt ) acknowledges a research fellowship from japan society for promotion of science ( nos . we analyze the problem of spontaneous symmetry breaking due to a repulsive interaction between electrons in the @xmath2 qh system . we assume that every landau site is occupied by one electron . the problem is to show that the spin polarized state is the lowest - energy state though the hamiltonian ( [ spincoulo ] ) involves no spin variables . we start with a proof , which makes the physical mechanism of spontaneous symmetry breaking clear . according to the decomposition formula ( decomformua ) the energy of the four - fermion interaction hamiltonian ( spincoulo ) is a sum of the direct and exchange energies ( [ decomdx ] ) , or @xmath351 with [ decomdxx ] @xmath352 , \end{aligned}\]]where @xmath353 and @xmath354 for a repulsive interaction . here , @xmath355 and @xmath356 . the key observation is that , though the hamiltonian ( [ spincoulo ] ) involves no spin variables , the energy of a state does . it is important that both the energies are positive semidefinite . the direct energy @xmath357 is insensitive to spin orientations , and it vanishes for the homogeneous electron distribution since @xmath358 . the exchange energy @xmath264 depends on spin orientations . the spin texture is homogeneous when the spin is completely polarized , where @xmath359 . furthermore , @xmath360 due to the homogeneous electron distribution . for such a spin orientation the exchange energy also vanishes since @xmath361 in ( [ appenssba ] ) . on the other hand , @xmath362 if the spin is not polarized completely since @xmath363 contains nonzero momentum components . consequently the spin - polarized state has the lowest energy , which is zero . hence the exchange interaction is the driving force of spontaneous symmetry breaking . we may present another proof , which is mathematically more formal . first of all , the state @xmath364 is an eigenstate of the hamiltonian ( [ spincoulo ] ) with the zero energy , @xmath365g@xmath177 . it is one of the ground states . we shall prove the following : ( a ) any globally w@xmath1(2)-rotated state is degenerate with @xmath77g@xmath78 ; ( b ) any locally w@xmath1(2)-rotated state has a positive energy . a spin rotated state of @xmath77g@xmath78 is @xmath366g@xmath178 with@xmath367where @xmath368 is a hermitian matrix , @xmath369^{\ast } = \xi _ { a}(n , m)$ ] . we calculate how the rotation @xmath370 affects the polarized spin for an infinitesimal parameter @xmath371 . using@xmath372we get @xmath373only relevant transformations are generated by @xmath374 and @xmath375 , since @xmath376 does not affect the spin polarization . besides we are interested in transformations rotating spins without moving electrons from site to site . then the parameter is reduced to @xmath377 . the energy induced by an infinitesimal transformation is @xmath378 \xi _ { a}(m)\xi _ { a}(n),\]]where we have set @xmath379g@xmath380g@xmath177 . the question is the positive definiteness of the quantity @xmath381 \xi _ { k}\xi _ { n } , \label{appstepspona}\]]where @xmath382 is the symmetric matrix . without loss of generality we assume the site index to run up to a finite value @xmath383 , which will eventually be taken to infinity . the analysis of ( [ appstepspona ] ) is reduced to the analysis of the eigenvalues of @xmath384 . introducing the complex valued function @xmath385we write @xmath386hence , @xmath387 is positive definite , and all of its eigenvalues are positive . now we refer to gerschgorin s theorem@xcite , which originally deals with complex matrices . formulating for a real matrix with positive elements @xmath388 the theorem reads as follows . let @xmath389 be the numbers defined as @xmath390then all of the eigenvalues of @xmath391 lie in the union of the segments @xmath392 $ ] . in our case we have @xmath393so that the segments appear as @xmath394 $ ] . on the other hand , we have shown that the eigenvalues of @xmath384 are positive . consequently , irrespective of the value of @xmath395 , we conclude that they lie within the segment @xmath396 $ ] , meaning that the eigenvalues @xmath397 of the quadratic form ( [ appstepspona ] ) satisfy the condition @xmath398the eigenstate of the lowest eigenvalue @xmath399 is given by @xmath400,@xmath401 \xi _ { n}=\xi \sum_{n}\left [ \varepsilon _ { x}\delta _ { kn}-v_{kn}\right ] = 0,\]]which corresponds to a global rotation of electron spins , while for an arbitrary local rotation we have @xmath402 . it implies the degeneracy of the ground states only under a global rotation , leading to a spontaneous symmetry breaking of the rotational symmetry . an antiskyrmion is a w@xmath1(@xmath7)-rotated state of the electron - excited state . we consider the antiskyrmion state ( microskyrmstateanti ) , @xmath403 % \left [ u_{1}^{+}c_{\uparrow , 1}^{\dag } + v_{1}^{+}c_{\downarrow , 2}^{\dag } % \right ] \notag \\ \cdots & \left [ u_{n-1}^{+}c_{\uparrow , n-1}^{\dag } + v_{n-1}^{+}c_{\downarrow , n}^{\dag } \right ] c_{\uparrow , n}^{\dag } |0\rangle \label{antistepa}\end{aligned}\]]with @xmath404 . we note that the state comprises @xmath405 electrons over @xmath141 sites . here and hereafter , for notational simplicity , we set @xmath406 , @xmath407 and so on . in contrast with the skyrmion state ( [ microskyrmstate ] ) , the antiskyrmion state @xmath408 does not lead to a wave function with any reasonable structure like ( [ skyrmwavefunct ] ) . the expression for general @xmath24 is technically difficult to write down . as an example we present the one for @xmath409 . up to multiplicative factors it appears as @xmath410 , \\ \mathfrak{s}_{\downarrow \downarrow \downarrow \uparrow } ^{+}= & ( z_{1}-z_{2})(z_{1}-z_{3})(z_{2}-z_{3})z_{4}^{2 } , \\ \mathfrak{s}_{\downarrow \downarrow \downarrow \downarrow } ^{+}= & 0 , \label{antistep6}\end{aligned}\ ] ] where no reasonable order is seen . in this respect skyrmions and antiskyrmions might be regarded to possess rather different properties . actually , this is not so since the above scheme of dealing with antiskyrmions is not quite satisfactory because of the following observation . the skyrmion state ( [ microskyrmstate ] ) comprises @xmath24 electrons , and the wave function ( [ skyrmwavefunct ] ) describes these electrons in terms of its @xmath24 complex arguments . on the other hand , the antiskyrmion wave function comprises @xmath411 complex arguments , which at first sight seems to match the number of electrons in the corresponding fock state . however , since two electrons out of those @xmath411 are fixed at @xmath412 and @xmath413 , they can not be transferred to the neighboring sites . therefore , though constructed of @xmath411 electrons , the antiskyrmion configuration comprises only @xmath414 site - transferrable electrons , which is exactly the same as in the skyrmion state . in other word , the skyrmion and antiskyrmion fock states comprise the equal number of the degrees of freedom , since the two fixed electrons carry no degrees of freedom at all . nevertheless , the antiskyrmion wave function involves @xmath411 complex arguments , which does not match the number of site - transferable electrons . there is nothing wrong there , and the answer to this mismatch is that the components of antiskyrmion wave function are not independent , but subject to certain functional relations . the most transparent manifestation of this statement are the relations ( antistep1 ) and ( [ antistep6 ] ) . though the scheme is not wrong by itself , it is inconvenient to work with . all these difficulties disappear if we develop a dual picture based on the electron - hole symmetry . so far the antiskyrmion fock state is built up via creating electrons in the vacuum state . the dual picture deals with the same state via removing electrons from the completely filled system of @xmath141 sites . namely , up to the overall sign , the state ( [ antistepa ] ) can be rewritten as @xmath415 \left [ v_{1}^{+}c_{\uparrow , 1}-u_{1}^{+}c_{\downarrow , 2}\right ] \notag \\ & \cdots \left [ v_{n-1}^{+}c_{\uparrow , n-1}-u_{n-1}^{+}c_{\downarrow , n}% \right ] \prod_{\mu n}c_{\mu n}^{\dagger } |0\rangle .\end{aligned}\ ] ] we now introduce the hole annihilation operators @xmath416 and the hole vacuum @xmath417 , @xmath418with @xmath419 and @xmath420 . now the antiskyrmion fock state can be presented as@xmath421 \left [ v_{1}^{+}\tilde{c}_{\downarrow , 1}^{\dagger } -u_{1}^{+}\tilde{% c}_{\uparrow , 2}^{\dagger } \right ] \notag \\ & \cdots \left [ v_{n-1}^{+}\tilde{c}_{\downarrow , n-1}^{\dagger } -u_{n-1}^{+}% \tilde{c}_{\uparrow , \hspace*{0.25mm}n}^{\dagger } \right ] |\tilde{0}\rangle , \end{aligned}\]]where the analogy with the skyrmion state is manifest . in the same way we introduce the hole field operator as @xmath422 , or equivalently@xmath423 . now the wave function of the antiskyrmion state comprising @xmath411 electrons contains @xmath24 variables @xmath424 . so the unphysical degrees of freedom associated with two non - transferable electrons of ( antistepa ) do not appear at all , since those two electrons are accounted within the ground state @xmath417 . using the analogy with ( [ skyrmwavefunct ] ) it is straightforward to write down the @xmath24-body wave function of antiskyrmion configuration . we obtain @xmath425=c_{n}e^{-\sum_{r=1}^{n-1}|z_{r}|^{2}/4 } \notag \\ & \times \left\vert \begin{array}{cccc } \bar{z}_{1}^{0}\left ( \begin{array}{c } \bar{z}_{1 } \\ \tilde{\kappa}_{0}% \end{array}% \right ) _ { \mu _ { 1 } } & \bar{z}_{1}^{1}\left ( \begin{array}{c } \bar{z}_{1 } \\ \tilde{\kappa}_{1}% \end{array}% \right ) _ { \mu _ { 1 } } & \cdots & \bar{z}_{1}^{n-1}\left ( \begin{array}{c } \bar{z}_{1 } \\ \tilde{\kappa}_{n-1}% \end{array}% \right ) _ { \mu _ { 1 } } \\ \bar{z}_{2}^{0}\left ( \begin{array}{c } \bar{z}_{2 } \\ \tilde{\kappa}_{0}% \end{array}% \right ) _ { \mu _ { 2 } } & \bar{z}_{2}^{1}\left ( \begin{array}{c } \bar{z}_{2 } \\ \tilde{\kappa}_{1}% \end{array}% \right ) _ { \mu _ { 2 } } & \cdots & \bar{z}_{2}^{n-1}\left ( \begin{array}{c } \bar{z}_{2 } \\ \tilde{\kappa}_{n-1}% \end{array}% \right ) _ { \mu _ { 2 } } \\ \vdots & \hspace*{-1mm}\hspace*{-1mm}\vdots & \ddots & \vdots \\ \bar{z}_{n}^{0}\left ( \begin{array}{c } \bar{z}_{n } \\ \tilde{\kappa}_{0}% \end{array}% \right ) _ { \mu _ { n } } & \bar{z}_{n}^{1}\left ( \begin{array}{c } \bar{z}_{n } \\ \tilde{\kappa}_{1}% \end{array}% \right ) _ { \mu _ { n } } & \cdots & \bar{z}_{n}^{n-1}\left ( \begin{array}{c } \bar{z}_{n } \\ \tilde{\kappa}_{n-1}% \end{array}% \right ) _ { \mu _ { n}}% \end{array}% \right\vert \label{antistepb}\end{aligned}\]]as in ( [ skyrmwavefunct ] ) . we prove the decomposition formula ( [ decomformua ] ) , according to which the four - fermion interaction energy consists of the direct energy and the exchange energy . we evaluate the energy of a state @xmath97 in the class of states ( [ generstate]),@xmath426where the hamiltonian is given by ( [ spincoulo ] ) . we assume that the electron field carries the su(n@xmath427 ) isospin index , @xmath428 . for the sake of accuracy we first consider a system with a finite number of landau sites ( @xmath429 ) and take the limit @xmath430 in final expressions . it is convenient to combine the isospin and site indices into a multi - index @xmath431 , where the multi - index runs over the values @xmath432 . the w@xmath1(@xmath433 ) algebra is identical to the algebra u@xmath434 in the limit @xmath430 , and the transformation rules for the fermion operators appear as @xmath435where @xmath436 is an @xmath437 unitary matrix , @xmath438 . here and hereafter the repeated index implies the summation over it . we first calculate the two - point averages by the state @xmath439 in ( [ generstate ] ) . using ( [ dfa ] ) we get @xmath440for the state @xmath88 given by ( [ purestate ] ) we have @xmath441which eventually leads to @xmath442carrying out analogous manipulations in four - point averages we get @xmath443we can use @xmath444for the state @xmath88 . substituting ( [ dfc ] ) into ( [ dfb ] ) and accounting ( [ dfd ] ) we summarize as @xmath445where the direct and exchange terms are easily recognized . here , we have @xmath446 in terms of decoupled spin and site indices , where @xmath447 and @xmath448 are defined by ( [ classdensi ] ) with @xmath449 the gell - mann matrix , @xmath450 . in deriving the formula for @xmath264 we have accounted @xmath453 and also the relation@xmath454 . \label{dff}\end{aligned}\]]this relation is derived as follows . we deal with the quantity @xmath455where we have used ( [ dfd ] ) and @xmath456 since @xmath457 or @xmath92 . comparing this with ( [ dfd ] ) we conclude @xmath458substituting ( [ dfe ] ) into this we find@xmath459this amounts to ( [ dff ] ) in the momentum space .

we present the discovery of debris systems around three solar mass stars based upon observations performed with the _ spitzer space telescope _ as part of a legacy science program , `` the formation and evolution of planetary systems '' ( feps ) . we also confirm the presence of debris around two other stars . all the stars exhibit infrared emission in excess of the expected photospheres in the 70 m band , but are consistent with photospheric emission at m . this restricts the maximum temperature of debris in equilibrium with the stellar radiation to k. we find that these sources are relatively old in the feps sample , in the age range 0.7 3 gyr . based on models of the spectral energy distributions , we suggest that these debris systems represent materials generated by collisions of planetesimal belts . we speculate on the nature of these systems through comparisons to our own kuiper belt , and on the likely planet(s ) responsible for stirring the system and ultimately releasing dust through collisions . we further report observations of a nearby star hd 13974 ( 11 pc ) that is indistinguishable from a bare photosphere at both 24 m and 70 m . the observations place strong upper limits on the presence of any cold dust in this nearby system ( ) .

astro-ph0506434

studying the formation of our own solar system and observing the frequency of similar systems associated with other stars are two ways in which we seek to understand our origins . through remote observation and direct exploration , we have developed a much clearer understanding of our solar system . however many questions about the processes involved in the initial formation and subsequent evolution towards the present configuration can not be addressed directly . therefore , we need to study other stars to help place our solar system in context . there are two major zones of debris in the solar system : the asteroid belt at 2 @xmath4 4 au composed of rocky material that is ground up by collisions to produce most of the zodiacal dust cloud and the kuiper belt ( kb ) that consists of small bodies orbiting beyond neptune s orbit at 30 @xmath4 50 au . since their discovery over a decade ago ( jewitt & luu 1993 ) , kuiper belt objects have played an increasingly important role in understanding the formation and evolution of our planetary system ( e.g. malhotra 1993 ; kenyon & bromley 2004 ) . while direct detection of _ in situ _ debris from collisions amongst kuiper belt objects has yet to be confirmed , of order 10% of the solar system s far - ir luminosity could be emitted by kuiper belt dust ( e.g. backman et al . searching for kb - like debris around other sun - like stars as a function of age will determine both the frequency of such systems and provide important insight into the formation of our solar system . most debris disks were found by their mid - to - far infrared emission in excess of the expected photosphere ( e.g. , auman et al . 1984 ; backman & paresce 1993 ; mannings & barlow 1998 ; habing et al . 1999 ; silverstone 2000 ; spangler et al . 2001 , decin et al . 2003 , zuckerman & song 2004 ) . the majority of these debris disks are also associated with hot , luminous stars , since observatories such as iras and iso did not have the necessary sensitivities to detect debris disks around lower - luminosity solar - type stars at distances beyond a few parsecs . the increased sensitivity afforded by the _ spitzer space telescope _ ( _ spitzer _ ; werner et al . 2004 ) has the potential to identify and investigate debris systems that were not detectable with previous observatories . using data from the formation and evolution of planetary systems ( feps ) _ spitzer _ legacy program , we have conducted a search for outer dust disks dominated by temperatures characteristic of the kuiper belt ( @xmath8 k ) . the earliest results from our validation observations are presented in meyer et al . ( 2004 ) , where we identified such debris disks ( exo - kbs ) surrounding the 30 myr - old sun - like star hd 105 and the 0.4 - 1 gyr old star hd 150706 . herein we present five more sun - like stars that exhibit characteristics of exo - kbs , and discuss their properties in the context of the evolution of our own solar system . we also present one star that has a firm detection at 70 @xmath0 m , but the measurement is consistent with the photospheric emission from the star . in this case , we place unprecedentedly low upper limits on the presence of kb - like debris in this system . in 2 , we describe the observations and data reduction . in 3 , we discuss the methodology for identifying the five exo - kuiper belt candidate stars from the feps sample , and we present the resulting spectral energy distributions ( seds ) . we outline our interpretation of the seds in terms of physical models in 4 . we discuss the implications of our results in 5 , and summarize our findings in the feps program is described by meyer et al . ( 2004 , 2005 ) and a detailed description of the data acquisition and data reduction for the feps program is given in the feps v1.1 explanatory supplement ( hines et al . 2004 ) . in this section , we present a condensed description as it applies to the five targets investigated herein . observational details are listed for each star in table [ obslog ] . imaging data were obtained with the multi - band imaging photometer for _ spitzer _ ( mips : rieke et al . 2004 ) in the 24 @xmath0 m and 70 @xmath0 m bands using small field photometry mode with 2 10 cycles of 3 and 10 second integration times respectively ; each of such dataset is a data collection event ( dces ) . after initial processing by the _ spitzer _ science center ( ssc ) version s10.5.0 pipeline to provide reconstructed pointing information , we used the mips data analysis tool ( dat ) software @xcite to process the data . for the 24 @xmath0 m data , read-2 correction , dark subtraction , droop correction , electronic nonlinearity correction , scan mirror dependent flat fields , cosmic ray rejection , and distortion correction were applied . we produced both individual dce frames and a mosaic image using all dce frames . outlier rejection of each pixel was performed for a mosaic image . the typical number of total dce frames per source used for photometry was 28 for 2 cycles . for the 70 @xmath0 m data , dark subtraction , illumination correction , and cosmic ray identification / rejection were applied . the electronic nonlinearity of the detector was corrected . @xcite note in their investigation that this nonlinearity is @xmath9 1 % , much smaller than absolute calibration uncertainties of 20 % ( see the _ spitzer _ observers manual 4.6 : som 4.6 , and the mips data handbook 2004 ) . a `` time - filtering '' algorithm as described by @xcite was used to eliminate time - dependent gain drifts . all 70 @xmath0 m flux measurements were performed on the distortion corrected final mosaic image . we used the idl - based software package idp3 ( schneider & stobie 2002 ) to perform aperture photometry on the 24@xmath0 m and 70@xmath0 m data . idp3 was developed by the instrument definition team for the near infrared camera and multi - object spectrometer ( nicmos ) , and has been further optimized for the _ spitzer _ irac and mips data . object fluxes were measured using standard aperture photometry techniques . for the 24 @xmath0 m photometry , we adopted a 147 aperture radius and a background annulus from 294 to 417 to perform aperture photometry on the distortion corrected individual dce frames . for the 70 @xmath0 m photometry we used a target aperture radius of 297 , and a background annulus of 396 792 on the single mosaic image . aperture correction factors , listed in the som 4.6 , were then applied . for hd 8907 , we used a smaller target aperture ( 197 ) with the same background annulus at 70@xmath0 m to avoid contamination by a nearby source . we also placed a mask on the nearby source to minimize any contribution from the wings of the psf to the background annulus . the aperture correction for this smaller aperture was derived relative to the standard aperture using 70um observations from other bright and isolated feps stars . the uncertainty in this small - aperture correction is @xmath10 1.2% . the 24@xmath0 m flux uncertainties are reported in table 2 , and include both the standard deviation of the mean flux density for individual dces , which is an estimate of the internal precision of our measurement , and the absolute calibration uncertainty . the 70@xmath0 m internal uncertainty is estimated by the rms pixel - to - pixel dispersion inside the background annulus measured on the mosaic image and scaled to the area of the target aperture . the total uncertainty for both bands is the combination of the internal uncertainty combined ( i.e. , added in quadrature ) with the calibration uncertainty . the calibration uncertainties for 24 @xmath0 m and 70 @xmath0 m photometry are 10 % and 20 % respectively ( see the _ spitzer _ observers manual 4.6 and mips data handbook 2004 ) . note that the the 70 @xmath0 m internal uncertainties are much smaller than absolute calibration uncertainty , and the internal signal - to - noise is very high for our detected objects . it is our knowledge of the absolute flux density in physical units that retains the large uncertainties in table [ fluxes ] . infrared array camera ( irac : fazio et al . 2004 ) observations at 3.6 , 4.5 and 8.0@xmath0 m were also obtained for each object using the 32@xmath1132 pixel subarray mode with an effective integration time of 0.01 sec per image ( frame time of 0.02 s ) . sixty four images of the object were obtained at four different positions in a random dither pattern on the array . this gives a total of 256 images ( 2.56 sec total integration time ) per band . we used the basic calibrated data ( bcd ) products produced by the ssc s10.5.0 data pipeline as described in the som 4.6 . aperture photometry was performed with idp3 using a 3 pixel radius aperture centered on the target . the background was estimated using the 820 pixels that lie beyond an 10 pixel radius relative to the star ( i.e. , all pixels beyond the target radius ) . the background flux was normalized to the area of the target aperture and subtracted from the summed target flux . the final source flux is the mean of the 256 measures , corrected from a 3 pixel radius to the 10 pixel radius used for the irac instrumental absolute flux calibration . the internal uncertainty was estimated as the standard deviation of the 256 independent measurements . as for mips , the total uncertainty in the physical flux density was constructed by adding the internal and absolute uncertainties in quadrature absolute flux calibration uncertainties of 10 % in all three bands . low - resolution ( @xmath12 ) spectra were obtained with the infrared spectrograph ( irs : houck et al . 2004 ) from @xmath13 , using `` high - accuracy '' blue peak - up to place the source in the spectrograph slit . integration times per exposure were 6 sec over the short - low wavelength range ( 7.4 @xmath4 14.5 @xmath0 m ) , and 14 sec over the long - low wavelength range ( 14.0 @xmath4 38.0 @xmath0 m ) . the spectra for wavelengths beyond @xmath14 m are not reliable ( houck et al . 2004 ) , so have been omitted . the spectra were obtained at two nod positions in staring mode for averaging and estimating the noise . the bcds resulting from the ssc pipeline s10.5.0 were further processed within the smart software package ( higdon et al . we used the _ droopres _ data products , which are intermediate products produced before stray - light and flat - field corrections have been applied . spectra were extracted assuming point source profiles with a fiducial width of 5 @xmath4 6 pixels in the center of the orders . residual `` sky '' emission was subtracted using adjacent pixels . random errors were calculated from the difference between the two independent spectra , then added in quadrature with an estimated 14 % uncertainty in absolute flux calibration . a more detailed discussion of the irs data reduction and extraction is presented in bouwman et al . as part of the ongoing analysis of the feps database , we have looked for stars that are detected at 70 @xmath0 m but show no significant excesses above the stellar photospheres for @xmath15 . this criterion restricts the equilibrium temperature and inner radius of a significant dust component to @xmath3 k and @xmath16 10 au for dust in thermal equilibrium around a star with 1 l@xmath17 . three other parallel investigations based on feps observations will concentrate on stars that have : 1 ) young optically - thick disks detected by irac , which focus on determining the evolutionary time scale of the inner disks ( silverstone et al . 2005 ) ; 2 ) strong emission at 24 but weak at 70 indicating material at temperatures approaching those of our solar system s asteroid belt and terrestrial planets ( hines et al . 2005 ) ; and 3 ) infrared excess indicating a broad range of temperatures characteristic of both terrestrial zones and kuiper belts ( bouwman et al . 2005 ) . the focus of this study is on stars that are dominated by excess infrared emission suggestive of kuiper belt like disks . five stars were selected from the feps data obtained between may 28 , 2004 ( mips campaign 8) and sep . 25 , 2004 ( mips campaign 13 ) , that have @xmath18 3 @xmath19 excess at 70 @xmath0 m , but are consistent with photospheric emission ( @xmath20 3 @xmath19 excess ) @xmath1 33 @xmath0 m . three of the objects are newly discovered debris systems ( hd 6963 , hd 145229 , & hd 206374 ) , while two of the objects were previously identified as having an infrared excess ( hd 8907 & hd 122652 ; respectively , silverstone 2000 , zuckerman & song 2004 ) from iras observations . we note that one object , hd 8907 , is one of 10 stars with known excess sources , which was included in the feps targets for the purposes of a gas emission line search . a sixth star , hd 13974 , was also detected at 70 @xmath0 m with high signal - to - noise , but in this case the 70 @xmath0 m flux density is consistent with emission from the photosphere to within 1 @xmath19 uncertainty . _ spitzer _ photometry for the five stars is presented in table [ fluxes ] , and the seds are shown in figures [ 4seds ] and [ hd13974 ] . mid - infrared spectra are also displayed for all of the objects . we use the `` average weighted '' wavelengths of irac and mips bands as suggested by the _ spitzer _ observer s manual ( som ) . color corrections have not been applied . the magnitude of these corrections are expected to be smaller than the present absolute calibration uncertainties ( see the irac and mips data handbooks ) . the expected photospheric emission for each star was determined by fitting kurucz model atmospheres with convective overshoot to published optical photometry including , if available , johnson @xmath21 , strmgren @xmath22 , tycho @xmath21 , hipparcos @xmath23 , cousins @xmath24 and 2mass @xmath25 measurements . predicted magnitudes in each filter were computed by multiplying the kurucz model with the combined system response of the filter , the atmospheric transmission ( for ground - based observations ) , and the spectral response function of the detector as outlined in a series of papers by ( * ? ? ? * ; * ? ? ? * and references therein ) . the johnson-@xmath26 and strmgren-@xmath27 filters were excluded from the model fits because the observed photometry shows large deviations from the model values explained in part by chromospheric activity common in young late - type stars . the best - fit kurucz model was computed in a least squares sense with the effective temperature and normalization constant as free parameters . the metallicity was fixed at [ fe / h ] = 0.0 and the surface gravity at log g cm s@xmath28 = 4.5 . in addition , each of the stars is within the local bubble ( see , e.g. , welsh et al . 1998 ) , and the visual extinction was fixed at @xmath290 mag . the stellar parameters used in the kurucz models , and additional properties of the stars including their distance and estimated ages , are listed in table [ stars ] . the deviation from the stellar photospheres at @xmath30 m is readily apparent for hd 6963 , hd 8907 , hd 122652 , hd 145229 , and hd 206374 ( fig . [ 4seds ] ) . the 70 emission from hd 13974 is consistent with the stellar photosphere ( fig . [ hd13974 ] ) within the uncertainty of 70 @xmath0 m flux . figure [ histogram ] shows the significance of the 70 @xmath0 m excesses relative to the kurucz model atmospheres for the five excess stars compared with the distribution for the other stars in the feps v1.1 data release . figure [ ccd ] also illustrates the excess in a color - color space that is sensitive to the color temperature of the excess ( 24@xmath0m / ks vs. 70@xmath0m / ks ) . large filled circles in figure [ ccd ] are sources with 70 detections , and small open circles with arrows are 1 @xmath19 upper limits for non - detections . this plot has the virtue of also capturing the lack of strong 24 @xmath0 m excesses in these targets . note that the 24 @xmath0 m excess of hd 12039 is well separated from the other targets ; this star is unique within the feps program in exhibiting only warm debris dust , and is discussed in detail in @xcite . the location of hd 13974 in this plot agrees well with 1 @xmath19 upper limits of 70 non - detections . in the next section we use the observed seds to constrain the properties of the detected debris systems . for hd 13974 the 70 detection is consistent the photosphere , placing stringent limits on any kb - like excess . for those objects that were not detected at 70@xmath0 m , we are still able to place upper limits on the presence of a debris system . we model our debris disks assuming optically thin dust in thermal equilibrium with the stellar radiation field . in this case , the temperature of a dust grain with a given chemical composition and grain size depends on the radial distance to the central star only . the fact that the seds do not significantly deviate from the photosphere for @xmath31 places the maximum temperature for the detected dust grains at @xmath32 k. this suggests a minimum equilibrium distance from the stars @xmath33au , assuming grains similar in size or larger than those found in our own zodiacal dust cloud ( radius @xmath34 @xmath0 m , reach et al . 2003 ) . the dust grains are subject to radiation pressure and poynting - robertson ( p - r ) drag , and the action of both mechanisms places limits on the time that the dust remains in the system . very small grains will be ejected from the system by radiation pressure , while larger grains will suffer p - r drag and spiral toward the parent star . a detailed prescription for computing `` blowout '' sizes and p - r drag timescales is given in burns et al . ( 1979 ) for the solar system , and are derived for a range of extrasolar systems by artymowicz ( 1988 ) and backman & paresce ( 1993 ) . particles smaller than @xmath35 m have a blown out time @xmath2010@xmath36 yrs . we consider grains to be no longer important contributors to the sed when they travel a factor of 4 farther from the star than their position of creation in the main dust belt . at these distances , the grain is 0.5 @xmath37 initial temperature , and emits at a twice the wavelength . we consider such a grain to no longer be contributing to the main excess sed , and in effect has left the system . grains blown out from points of origins a few @xmath37 10 au from a solar type star will leave a system by this definition in at most only a few@xmath3710@xmath36 years . particles larger than @xmath10 1 will be subject to slow p - r inward drift and will be destroyed on timescales of 10@xmath38 yrs for particle sizes of @xmath101 @xmath4 10 ( timescale linearly proportional to particle size ) starting from distances of order @xmath39 @xmath40 au ( timescale proportional to @xmath41 ) . this time scale is very short compared to the age of the systems ( see table [ stars ] ) , which suggest that the debris in these systems is being replenished by a parent population of objects , a putative kuiper belt . from these arguments , we can build simple models of the debris systems . below we develop models for five of the objects that have limited photometry beyond 24 @xmath0 m . for hd 8907 , where we have more photometric constraints on the sed , we explore a more comprehensive model . blackbody models of the detected debris disks were based on excess color temperature ( @xmath42 ) , estimated from the two shortest broadband wavelengths with @xmath18 1 @xmath19 excess ( including the synthetic band at 33 using irs long - low data ) blackbody grains by definition absorb and emit radiation efficiently at all relevant wavelengths . for each object , the solid angle , @xmath43 , subtended by the emitting material as seen from the central star was calculated from @xmath42 and excess @xmath44 using the planck formula . this value for @xmath43 in turn was converted to : ( 1 ) total grain emitting cross - section area , @xmath45 , using the system distance , ( 2 ) grain luminosity using @xmath45 and @xmath42 in the stefan - boltzmann formula for blackbody luminosity , and ( 3 ) grain mass assuming a constant per - grain mass density was chosen to represent generic silicate fragments of asteroid - like planetesimals , c.f . , mean value of densities determined for 10 asteroids @xmath102.4 g @xmath46 ; http://aa.usno.navy.mil/hilton/asteroid_masses.htm . ] of 2.5 g @xmath46 for solid `` astronomical silicate '' ( draine & lee 1984 ) and grain radius 10 for efficient emission at 70 . the 10 radius grain is the smallest size that acts as a blackbody at 70 . the uncertainty in blackbody temperature estimate ( @xmath42 ) is about 10% . in all cases , the flux densities between 33 @xmath0 m and 70 @xmath0 m are increasing , suggesting that we are observing the wien side of the dust excess sed , thus @xmath42 is interpreted as an estimate of the maximum dust temperature . the corresponding minimum distance from the star of the emitting material @xmath47 was found from the estimated maximum @xmath42 and the relevant @xmath48 using the formula for blackbody grains in backman and paresce ( 1993 ) . the relationship between grain temperature , position ( @xmath49 ) , and primary star luminosity ( equation 3 of backman & paresce 1993 ) for grains larger than @xmath50 m , that emit efficiently at @xmath51 m , and have negligible albedo , is @xmath52 where @xmath48 is in unit of @xmath53 , 3.9@xmath3710@xmath54 erg s@xmath55 . the model fit parameters and masses are presented in table [ results ] , and figure [ 4seds ] shows simple model seds . the blackbody models also allow estimates of the mass in the radiating grains . because ( 1 ) these calculations made use of a maximum dust temperature and corresponding inner radius ( @xmath47 ) , also ( 2 ) lack of photometric measurements or even useful upper limits beyond the apparent peak of excess emission ( 70 @xmath0 m ) prevents useful characterization of the dust outer boundary radius , and finally ( 3 ) grains larger than the minimum size assumed here would radiate the same blackbody sed but would have a higher ratio of mass to surface area , the values of @xmath56 and @xmath57 in table [ results ] characterize the portion of the sed observable out to 70 ; future observations at longer wavelengths could conceivably indicate more luminosity and dust mass . we note that debris disk models comprised of grains slightly larger than the blowout size ( @xmath10 1 ) can not be ruled out based on simple fits to the sed . such a model ( based on a temperature vs. position function given in equation 5 of backman and paresce 1993 ) requires an @xmath47 of @xmath10 270 au for a solar luminosity star and a dust mass @xmath10 3 larger than the lower limit given above . we comment on the likelihood of this model in section 5.4 . the simple black body grain model yields @xmath58 k , the inner radius @xmath47 = 48 au , a dust to stellar luminosity ratio ( log @xmath56 ) @xmath59 , and a dust mass log ( @xmath60 ) @xmath61 for 10 @xmath0 m radius blackbody grains with grain density 2.5 g @xmath46 . the detection of significant excess emission from hd 8907 at several wavelengths allows us to conduct a more thorough analysis of that system . therefore we modeled the sed of hd 8907 using the dust disk models of wolf & hillenbrand ( 2003 , wh03 ; 2005 ) . we used the levenberg - marquardt algorithm to solve the least - squares problem , giving a best - fit dust disk model ( marquardt 1963 ; markwardt 2003craigm / idl / fitting.html ] ) . a detailed description of the fitting algorithm will be given in rodmann et al . due to lack of mineralogical features in the irs spectrum , we assumed the optical properties of astronomical silicates ( draine & lee 1984 ; laor & draine 1993 ; weingartner & draine 2001 ) . we assumed the density profile to be , corresponding to a disk with a constant surface density . the power law exponent of the grain size distribution was set to the canonical value @xmath62 , characteristic for a size distribution initially produced by a collisional cascade ( dohnanyi 1969 ; tanaka et al . 1996 ) . for the maximum grain size ( @xmath63 ) and the @xmath64 , we assumed arbitrary values of and . grains larger than 1 mm contribute little to the infrared emission of the disks , and the outer radius is poorly constrained in the absence of sub - millimeter / millimeter data . we used an upper limit at @xmath65 mm ( carpenter et al . 2005 ) to constrain flux of the disk models at mm wavelengths . we then fit for three parameters simultaneously in a @xmath66 sense : @xmath47 , @xmath67 , and m@xmath68 in grains smaller than 1 mm . the first two parameters influence the shape of the sed and the wavelength at which the disk emission begins to depart from the stellar photosphere ; the latter only scales the dust reemission to match the peak of the infrared emission . figure [ hd8907 ] shows the sed of hd 8907 with the wh03 model result using astronomical silicate and grain radii range from 6 @xmath4 1 mm . the result suggests @xmath47 @xmath10 42.5 au , with dust mass of 1.7@xmath69 @xmath70 . the larger dust mass compared to the mass estimate from the simple blackbody model is from the inclusion of larger grains , which contribute the bulk of the mass even though they do not contribute to the bulk of the infrared emission . the 1 @xmath19 uncertainty level for each parameters used in our modeling are : @xmath71(@xmath47 ) @xmath10 20 au ( 50% ) ; @xmath71(@xmath67 ) @xmath10 3 ( 50% ) ; and @xmath71(@xmath57 ) @xmath10 3.3 @xmath37 10@xmath72 @xmath70 ( 20% ) . these parameter uncertainties are estimated from the covariance matrix ( inverse of the @xmath66 curvature function ) by the levenberg@xmath4marquardt algorithm . the selection of debris disks for this investigation is purposely biased against objects with debris at temperatures warmer than @xmath32 k. however , in our own solar system we observe a warm component associated with debris from the asteroid belt . such a warm component , albeit faint , has also been verified in other debris systems , e.g. , around an a - star hr 4796a ( e.g. , koerner et al . 1998 ) , and recently with _ spitzer _ imaging of fomalhaut ( stapelfeldt et al . 2004 ) . we do not have the luxury of high resolution imaging of the five sun - like stars presented herein , so we have to rely on the seds to place upper limits on the warm dust components . table [ model2 ] presents upper limits to the amount of warm dust located interior to the outer disk inner boundaries ( @xmath47 ) in each system . these were evaluated by assuming that dust in these locations would be drifting in from the denser outer source zones via p - r radiation drag and thus would extend with constant surface density from @xmath47 inward to a radius correspond to the vaporization temperature of silicates at @xmath74 1500 k at @xmath100.1 au from the star . for each system a surface density ( @xmath75 ) upper limit was calculated for dust between sublimation radius , @xmath76(@xmath771500 k ) and @xmath47 of the outer disk model such that the summed seds of inner dust model plus outer dust model plus stellar photosphere was less than the upper limit to the observed sed , i.e. , observations plus 1@xmath19 uncertainty including calibration , as defined above . this was then converted to dust mass limits assuming @xmath7810@xmath0 m grains with material density of 2.5 g @xmath46 as for the blackbody models of the cool outer disks . these masses are of order 10@xmath79 m@xmath80 , the mass of a single asteroid only a few hundred km in diameter , and are generally 2@xmath43 orders of magnitude below the lower limits for the masses we derive for the outer cool dust . note that our _ spitzer _ observations of hd 13974 , which shows photospheric emission up to 70 @xmath0 m , yield an upper limit on warm dust surface density equivalent to only about 20 times that of the zodiacal cloud in our solar system ( i.e. , 20 `` zodis '' ) . a primary goal of the feps project is to place our solar system in context with other debris disk systems ( meyer et al . 2005 ) . to date , most disk systems identified with _ iras _ and _ iso _ have been associated with either stars more luminous than the sun ( typically a - stars ) , or very young systems ( @xmath81 30 myrs ) . a recent census of sun - like stars with kb - like disks that have been identified with _ iras _ and _ iso _ has been presented by decin et al . ( 2003 ) and zuckerman & song ( 2004 ) , but these studies were only able to identify the brightest systems and did not cover a large range of ages that encompass stars approaching the age of our sun . the feps sample is a coherent effort to understand the debris systems around sun - like stars over a large range of ages from 3 myrs 3 gyrs . we are beginning to assemble a sufficiently large sample of objects over a range of ages to start to answer fundamental questions about the evolution of dust disks surrounding sun - like systems with _ spitzer _ data . the age determination of all of the stars in the full feps sample is discussed in hillenbrand et al . age bins given for the five stars ( table [ stars ] ) in this study are inferred from the level of chromospheric and coronal activity . the chromospheric activity is indicated by caii h&k emission . values of the fractional luminosity emitted in the h&k lines , @xmath82 , were derived by soderblom ( 2000 , private communication ) . here we adopt the calibration of donahue ( 1993 ) between log @xmath82 and age . coronal activity , on the other hand , is indicated by x - rays . the fractional x - ray luminosity , @xmath83 , in general tracks @xmath84 ; for hd 8907 in particular , relative youth is indicated by both activity indicators . we use coarse age bins rather than specific ages due to the uncertainty in the age estimates . the total mass of the kb in the solar system is highly uncertain , however some recent studies ( e.g. , luu & jewitt 2003 , teplitz et al . 1999 ) suggest its total mass is in the range @xmath100.01 @xmath4 1.0 m@xmath80 , mostly in large planetesimal bodies . from cobe observations at wavelengths 150 @xmath0 m and 240 @xmath0 m , backman et al . ( 1995 ) estimated upper limits on kb dust luminosity of @xmath85 10@xmath79 . this corresponds to a kb infrared dust mass limit of @xmath8610@xmath87 assuming grains with @xmath88 10 @xmath0 m and @xmath89 = 2.5 g @xmath46 as in the present paper s models of exo - kbs . for comparison , moro - martn & malhotra ( 2003 ) estimate a kb mass of about 4@xmath3710@xmath79m@xmath90 in particles with sizes between 2.4 @xmath0 m and 160 @xmath0 m based on flux of kb dust detected drifting toward jupiter s orbit ( landgraf et al . we see that the systems considered in this paper contain more massive and luminous dust ensembles than does the kb . but what would our own kb have looked like at the younger evolutionary stages represented by these five objects ? backman et al . ( 2005 ) have devised a simple evolutionary model of the kb based on the model of the current kb in backman et al . ( 1995 ) that can be used to estimate what our system would have looked like at the ages , distances and stellar luminosities of our targets . the model assumes the planetesimal population in the kb of these solar systems extends from @xmath47 @xmath91 40 au to @xmath64 @xmath91 50 au at the ages of our targets , after migration of the outer planets and kb had been completed ( malhotra 1993 ; levison et al . 2004 ) , and the influence by neptune ( at 30 au ) had eroded the belt substantially inward of the 3:2 resonance at r @xmath92 au . the model also assumes that the starting mass of the kb was 10 m@xmath80 , the minimum necessary to yield a high enough density to build the observed large kb bodies accretionally in the allowed time span @xcite . the model evolution is not very sensitive to the starting mass because a purely collisional system evolves asymptotically toward a state in which the collision timescale is approximately equal to the age of the system . for example , a wide range of starting masses evolve in this model to a kb with a mass of about 0.5 m@xmath80 at age 4.5 gyr . this means that , in a dissipative system such as this , evidence of the original state is mostly erased and can not be inferred by modeling _ backwards_. a parent body size of 10 km and maximum fragment size of 5 km were parameters tuned to yield a kb dust population at an age 4.5 gyr consistent with present limits on kb dust emission ( backman et al . 1995 ; teplitz et al . 1999 ) and inference of the kb dust production rate from dust impact rates in the outer solar system ( landgraf et al . the dust distribution and thermal emission were calculated for 30 logarithmically spaced size bins between 1 @xmath0 m and 1 mm diameter . an equilibrium spatial distribution of dust was calculated , as in backman et al . ( 1995 ) , balancing the dust between production in planetesimal collisions , inward drift of dust via p - r radiation drag , and destruction by mutual collisions of dust yielding fragments smaller than the blowout size . the dust distribution was modeled both within the planetesimal zone at 40 @xmath4 50 au and in a p - r induced zodiacal " dust cloud extending inward and assumed completely truncated at @xmath64 @xmath91 30 au . results of the kb evolutionary model runs are presented in table [ model3 ] , where columns 3 and 4 are the predicted and observed 70 @xmath0 m excess fluxes of each source . remarkably , although the solar system evolutionary model was tuned only to represent our kb , the results are within a factor of only 2 @xmath4 3 from the targets observed 70 @xmath0 m excesses , except for the hd 13974 . the solar system kb evolution model predicts a 70 excess of @xmath10 21 mjy for the sun observed from @xmath5 30 pc at an age of @xmath934.6 @xmath94 yrs ( on top of 5 mjy photosphere ) . this predicted excess flux is less than any of the 5 detected excess sources , e.g. , about 30 % of the dust flux at 70 observed for hd 145229 with 0.96 l@xmath17 observed at 33 pc . therefore the solar system present - day kb model has about 30 % of the dust mass of the hd 145229 cold dust . these calculations indicate the possibility that the target systems represent snap shots of the history of our own solar system suggesting that systems like our own might be common among g stars in the galactic disk . hd 13974 is a short period binary system ( period = 10 days , duquennoy & mayor 1988 ) with a g0v primary ( duquennoy & mayor 1988 ) and a companion with spectral type between g9v ( duquennoy & mayor 1988 ) and k4v ( hummel et al . 1995 ) . hd 13974 at 11 pc is the only star in our feps sample to date with a 70 @xmath0 m detection that is consistent with `` bare '' photospheres within the calibration and kurucz model uncertainties ( figure [ hd13974 ] ) . a two - component kurucz model was fitted to the observed sed with derived temperatures of 6215 k and 4493 k for the primary and secondary components respectively . the semi - major axis of these stars are @xmath95 @xmath91 1.45 @xmath96 0.03 @xmath37 10@xmath97 km and @xmath98 @xmath91 1.62 @xmath96 0.04 @xmath37 10@xmath97 km ( duquennoy & mayor 1988 ) . such a close ( @xmath99 1 au ) binary star system should not have a dramatic effect on the evolution of the circumbinary disk at distances greater than roughly twice the semi - major axis ( e.g. , jensen et al . 1996 ; artymowicz & lubow 1994 ) . adopting the blackbody grain model we have used the 1@xmath19 uncertainty in the observed flux density at 70 @xmath0 m compared with the predicted kurucz model flux to place an upper limit on the mass of dust log ( @xmath60)@xmath100 at @xmath101 55 k. the lower limit on @xmath47 is about 28 au . the fractional dust luminosity log ( @xmath56 ) @xmath102 of this system is interesting , as it is similar to the estimates for the current solar system kuiper belt . the solar system kb evolutionary model predicts 70 @xmath0 m excess flux density of hd 13974 at its age ( 1 @xmath4 3 gyr ) to be about 300 mjy , while the observed excess flux density limit is @xmath20 20 mjy ( table [ model3 ] ) . this system may not contain kb bodies , or perhaps there is not a perturbing planet like neptune to stir up the system and cause collisional cascades . for the five systems in which 70 @xmath0 m excesses have been detected , the comparison between the disk mass estimate ( @xmath57 in table [ results ] ) , and the upper limit to the amount of warm dust ( @xmath103 in table [ model2 ] ) located interior to the inner boundary of the dust ( @xmath47 in table [ results ] ) , indicates that the dust depletion inside @xmath47 is significant . because grains not directly ejected by radiation pressure tend to spiral toward the star due to p - r drag , a central depletion would be filled in by dust on p - r timescales ( table 5 ) much shorter than the ages of these systems , unless some other mechanism intervenes to eliminate grains . we argue that ice sublimation and grain blowout " both fail to explain the @xmath47 locations in these systems , and that a possible explanation for the dust spatial distribution is a sizable planet in each system limiting the inward drift of the grains by ejecting them out of the system via gravitational scattering . an upper limit of the location ( @xmath76 ) at which grains will sublimate can be calculated by assuming that they are composed of water ice with sublimation temperature @xmath104100 k. the maximum grain temperatures ( @xmath42 ) observed in these systems presented in table [ results ] are all well below the ice sublimation temperature . the presence of grains large enough to radiate efficiently at 70 @xmath0 m implies that their temperature will scale as @xmath105 @xmath106 @xcite . the radius of ice sublimation @xmath76 will be @xmath107 of @xmath47 for @xmath42 in the range @xmath108 k. and thus sublimation is not a likely explanation for the inner depletion of the disk . the uncertainties in the temperature estimate are about 10% for individual sources . can these central zones that are relatively free of dust be explained as the result of dust grains in the outer rings being controlled by mutual collisions rather than p - r drag ? thus the inner edge of the dust distribution might be explained without invoking the existence of a planet to consume inward drifting grains . in their study of the age dependence of vega - like excesses , dominik & decin ( 2003 ) argued that in dense enough disks , the collision time scale is much shorter than the p - r drag time scale . if grain collisions actually dominate p - r drag as the dominant dynamical process then grains could be ground to the blowout size , and would not drift radially . in that case , the inner edge to the dust distribution could represent the inner edge of the source bodies . the dust collision timescales ( @xmath109 ) for the disks can be estimated as @xmath110/9@xmath75 , where @xmath110 is the orbital period at @xmath47 and @xmath75 is the fractional surface density of the disks that is of order the value of @xmath56 ( backman & paresce 1993 ; backman 2004 ) . the timescales for particles to drift via p - r drag over distances equal to 10% of the @xmath47 values in table [ results ] ( i.e. , an estimate of the time for grains to move from populated to relatively unpopulated regions ) are about log @xmath111 5.5 to 6.0 years . the grain - grain collision timescales for the same systems are about log @xmath112 5.0 to 5.5 years . for ratios of this sort between @xmath113 and @xmath109 , a substantial fraction of grains would survive enough collision lifetimes to be able to drift from the disk into a central void . note also that without knowing more about the disks ( especially , their vertical extents ) we do not know if grain - grain collisions would be at high enough speeds to destroy the grains . thus we conclude that p - r drag is important in controlling the structures of these systems , especially in that a planetary barrier would be a plausible explanation for a central depletion . could these cold disks arise from grains generated recently from planetesimals located near @xmath47 and then ejected by radiation pressure from the system , again without needing a planet to explain the inner boundary of the disk ? we note that these debris belts can be self - stirred without needing a planet to explain the collisional cascade itself ( e.g. , also kenyon & bromley 2004 ) . spitzer _ observations of vega at 24 , 70 , and 160 @xmath0 m show that the vega debris disk has an inner boundary at @xmath10 86 au , and extends to distances much larger than those observed at submm and mm wavelengths ( su et al . the authors suggest that the grains in the vega far - ir disk are produced close to the inner disk boundary and are unbound , flowing away from the star . in this scenario , the inner gap in the disk would simply represent a limit to the location of the parent bodies , rather than the location of a perturbing planet . however , vega s luminosity to mass ratio ( l / m ) is 24 in solar units , which makes even relatively large grains ( like the 18 grains ) to be unbound , with @xmath114 0.5 , if the grain s porosity is @xmath1150.56 . ( @xmath116 is the ratio of the radiation pressure force to the gravitational force ) . in contrast , for the g star systems we consider here with l / m @xmath10 1 , the 10 silicate grains that dominate the dust emission at 70 have @xmath117 0.02 . therefore , we argue that grains in the five disks discussed here are bound and as such will drift toward the star due to p - r drag . one explanation for the inner empty zones is that one or more massive planets dynamically deplete the dust generated by an outer belt of planetesimals ( e.g. , liou & zook 1999 , moro - martn & malhotra 2003 , 2005 ) . if we assume a single planet in a circular orbit about each of the five stars , then this mechanism can account for the large dust depletion factors inferred for the inner parts of these dust disks if the planet is @xmath115 1 m@xmath118 ( moro - martn and malhotra 2005 ) . the inner edge of the dust disk would be located between 0.8 @xmath119 and 1.25 @xmath119 , where @xmath120 is the semi - major axis of the planet . therefore , a possible explanation is that the dust disk surrounding each of these five systems harbors a planet with an orbital radius of approximately 10 @xmath4 20 au , together with a belt of dust - production planetesimals exterior to the planet s orbit , and no asteroid belt with significant dust production interior to the planet s orbit . as noted in 4.1 , the data are also consistent with a debris disk comprised of 1 sized grains , inconsistent with the assumptions required for blackbody emission discussed above , yet larger than the blowout size . in such a model , the timescale for mutual collisions is more than 100 times shorter than the timescale for p r drag to significantly affect the orbital radius . this suggests that a ring such as inferred from the model could be maintained through mutual collisions with dust removal dominated by radiation pressure blowout of the smallest particles @xmath20 1 @xmath0 m on short timescales . however , this requires the formation of a sun like star lacking an inner disk , with a remnant planetesimal belt capable of generating dust through mutual collisions only outside of 200 au with dust only at the smallest possible stable particle size . while the presence of a low mass companion could in principle explain such a large gap in the circumstellar environment , no wide stellar companions are known in these systems , and we do not expect many brown dwarfs ( due to the observed lack of such systems ) or giant planets ( given current formation scenarios ) at these radii . we believe that models requiring larger grains ( @xmath10 10 ) at smaller radii ( @xmath10 30 au ) with the possible presence of a planet ( or planetary system ) interior to 30 au are favored as they are consistent with the structure and extent of circumstellar disks observed around pre main sequence stars during the epochs when planetary systems are thought to form . we hope to distinguish between these two hypotheses in the future through direct imaging of the outer edge of the disks in scattered light and placing strong constraints on the presence of very low mass companions at large separations ( e.g. with nicmos coronagraphy on hst and ground based ao observations ) . we have presented _ spitzer _ observations of five sun - like stars that possess kuiper belt - like debris systems ( hd 6963 , hd 8907 , hd 122652 , hd 145229 , and hd 206374 ) . of these two were previously suggested to have debris disks ( hd 8907 and hd 122652 ) , and three are newly discovered . the fractional luminosity ( @xmath56 ) and ages of these newly identified disks illustrates the potential of the feps program to measure fainter debris systems , which will allow a more detailed census of their nature and evolution . the five excess sources have seds that are consistent with photospheric models out to 33@xmath0 m , but show clear excesses at 70 @xmath0 m , which was the selection criterion . we find that these stars are all `` old '' ( four sources are in our 1 @xmath4 3 gyr age bin , while one , hd 145229 , is in the 0.3 @xmath4 1 gyr age bin ) . as seen in figure [ histogram ] , the improved sensitivity of _ spitzer _ allows us to detect debris disk systems that are much fainter than those detected by iras and iso . the overall impression is that kb - like systems detectable by spitzer and considered in this paper are less massive and more distant than systems detected with iras and iso . \3 . another star , hd 13974 , has a mips 70 @xmath0 m flux consistent with photospheric emission within 1 @xmath19 total uncertainty . the upper limit of log(@xmath56 ) is @xmath102 , similar to that of inferred for the solar systems kb . \5 . a solar system kb evolution model predicts _ spitzer _ 70 @xmath0 m fluxes ( table [ model3 ] ) from hypothetical planetesimal assemblages around our target stars that are within factors of @xmath102 @xmath4 3 of the observed fluxes . we infer that these systems have outer remnant planetesimal belts that are consistent in scale and starting masses to our kuiper belt . the absence of a disk around the @xmath122 gyr old star hd 13974 suggests that either this object does not contain the parent bodies that produce infrared - emitting debris , or perhaps the debris has been cleared out already . we placed upper limits on warm dust masses interior to @xmath47for each of these systems , and showed that the depletion of the disk @xmath20@xmath47 is significant . we commented on several possible causes for @xmath47 . we speculate that the @xmath47 of exo - kbs presented in this study could be explained by the existence of one or more jupiter mass planets at 10 @xmath4 20 au from each star . we thank to the rest of the feps team members and _ spitzer _ science center help desk . we have used the simbad database . this work is based [ in part ] on observations made with the spitzer space telescope , which is operated by the jet propulsion laboratory , california institute of technology under nasa contract 1407 . feps is pleased to acknowledge support through nasa contracts 1224768 , 1224634 , and 1224566 administered through jpl . s.w . was supported by the german research foundation ( dfg ) through the emmy noether grant wo 857/2 - 1 . eem is supported by a clay fellowship from the smithsonian astrophysical observatory . mpia team is supported through the european planet network . ardila , d. r. , golimowski , d. a. , krist , j. e. et al . 2004 , apj , 617 , l147 artymowicz , p. 1988 , apj , 335 , l79 artymowicz , p. & lubow , s. h. 1994 , apj , 421 , 651 aumann , h. h. , et al . 1984 , apj , 278 , l23 backman et al . 2005 , in prep . backman , d. e. , dasgupta , a. , & stencel , r. e. 1995 , apj , 450 , l35 backman , d. e. , & paresce , f. 1993 , in protostars and planets iii , ed . e. levy & j. i.lunine ( tucson : univ . arizona press ) , 1253 beichman et al . 2005 , apj , in press bouwman , j. , et al . 2005 , in prep . burns , j. a. , lamy , p. l. , & soter , s. 1979 , icarus , 40 , 1 carpenter , j. m. , wolf , s. , schreyer , k. , launhardt , r. , henning , t. 2005 , aj , 129 , 1049 chen , c. h. , jura , m. , gordon , k. d. , & blaylock , m. 2005 , apj , in press cohen , m. , megeath , s. t. , hammersley , p. l. , martin - luis , f. , & stauffer , j. 2003a , aj , 125 , 2645 cohen , m. , wheaton , w. , & megeath , s. t. 2003b , aj , 126 , 1090 decin , g. , dominik , c. , waters , l. b. f. m. , & waelkens , c. 2003 , apj , 598 , 636 dominik , c. , & decin , g. 2003 , apj , 598 , 626 donahue , r.a . 1993 , ph.d . thesis , university of new mexico dohnanyi , j. w. 1969 , j. geophys . res , 74 , 2531 draine , b.t . & lee , h .- 1984 , apj , 285 , 80 duquennoy , a. & mayor , m. 1988 , a&a , 195 , 129 . g. g. , et al . 2004 , apjs , 154 , 10 gordon , k. d. , et al . 2004 , in `` optical , infrared , and millimeter space telescopes '' , ( spie vol . 5487 ) , ed . j. c. mather , 177 gordon , k. d. , et al . 2005 , pasp , in press greaves , j. s. , wyatt , m. c. , holland , w. s. , & dent , w. r. f. 2004 , mnras , 351 , l54 habing , h. , dominik , c. , jourdain de muizon , m. , et al . 1999 , nature , 401 , 456 higdon , s. j. u. et . 2004 , pasp , 116 , 975 hines et al . 2004 , feps explanatory supplement v1.1 , ssc , pasadena hines , d. c. , 2005 , submitted hillenbrand , l. a. , et al . 2005 , submitted halbwachs et al . 2003 , a&a 397 , 159 houck . j. r. , et al . 2004 , apjs , 154 , 18 hummel , c. a. , armstrong , j. t. , buscher , d. f. , mozurkewich , d. , quirrenbach , a. , & vivekanand , m. 1995 , aj , 110 , 376 jensen , e. l. n. , koerner , d. w. , & mathieu , r. d. 1996 , aj , 111 , 2431 jewitt , d. & luu , j. 1993 , nature , 362 , 730 kenyon , s. j. , & bromley , b. c. 2004 , apj , 602 , l133 koerner , d. w. , ressler , m. e. , werner , m. w. , & backman , d. e. 1998 , apj , 503 , l83 landgraf et al . 2002 , aj , 123 , l285 laor , a. & draine , b.t . 1993 , apj , 402 , 441 levison , h. f. , morbidelli , a. , & dones , l. 2004 , aj , 128 , l2553 liou , j .- c . & zook , h. a. 1999 , aj , 118 . 580 luu , j. x. , & jewitt , d. c. 2002 , ara&a , 40 , 63 malhotra , r. 1993 , nature , 365 , 819 malhotra , r. 1995 , aj , 110 , 420 mamajek , e. e. , meyer , m. r. , & liebert , j. 2002 , aj , 124 , 1670 mannings , v. & barlow , m. j. 1998 , apj , 497 , 330 marquardt , d.w . 1963 , journal of the society for industrial and applied mathematics , vol . 431 - 441 metchev , s. a. , hillenbrand , l. a. , & meyer , m. r. 2004 , apj , 600 , 435 moro - martn , a. & malhotra , r. 2003 , aj , 125 , 2255 moro - martn , a. & malhotra , r. 2005 , aj , submitted meyer , m. r. et al . 2004 , apjs , 154 , 422 meyer , m. r. et al . 2005 , in preparation reach , w. t. , morris , p. , boulanger , f. , okumura , k. 2003 , icarus , 164 , 384 rieke , g. h. , et al . 2004 , apjs , 154 , 25 rodmann , j. et al . 2005 , in preparation schneider , g. , becklin , e. e. , smith , b. a. , weinberger , a. j. , silverstone , m. , & hines , d. c. aj , 121 , 525 schneider , g. & stobie , 2002 , asp conf . 281 , astronomical data analysis software and system xi , ed . d. a. bohlender , d. durand , & t. h. handley ( san francisco : asp ) version 2.9 silverstone , m.d . 2000 , ph . d. thesis silverstone , m.d . et al . 2005 , submitted spangler , c. , sargent , a. i. , silverstone , m. d. , becklin , e. e. , & zuckerman , b. 2001 , apj , 555 , 932 stapelfeldt , k. r. et al . 2004 , apjs , 154 , 458 stern , s. a. & colwell , j. e. 1997 , aj , 114 , 841 su , k. y. l. et al . 2005 , apj , in print tanaka , h. , inaba , s. , nakazawa , k. 1996 , icarus , 123 , 450 teplitz , v. l. , stern , s. a. , anderson , j. d. , rosenbaum , d. , scalise , r. i. , & wentzler , p. 1999 , apj , 516 , 425 weingartner , j. c. & draine , b. t. 2001 , apj , 548 , 296 welsh , b. y. , crifo , f. , & lallement , r. 1998 , , 333 , 101 werner , m. w. , et al . 2004 , apjs , 154 , 1 williams , j. p. , najita , j. , liu , m. c. , bottinelli , s. , carpenter , j. m. , hillenbrand , l. a. , meyer , m. r. , & soderblom , d. r. 2004 , apj , 604 , 414 wolf , s. , & hillenbrand , l. a. 2003 , apj , 596 , 603 ( wh03 ) wolf , s. , & hillenbrand , l. a. 2005 , comp . comp , in press wolf , s. , & voshchinnikov , n.v . 2004 , comp . comm . , 162 , 113 zuckerman , b. & song , i. 2004 , ara&a , 42 , 685 lcccccc hd 6963 & 8/06/04 & 5387776 & 8/07/04 & 5395968 & 8/18/04 & 5396224 + hd 8907 & 7/25/04 & 5361920 & 1/02/05 & 5362176 & 8/23/04 & 5362432 + hd 13974 & 8/14/04 & 5410816 & 8/07/04 & 5411072 & 8/27/04 & 5411328 + hd 122652 & 6/29/04 & 5427712 & 6/06/04 & 5427968 & 6/22/04 & 5428224 + hd 145229 & 7/19/04 & 5387264 & 7/17/04 & 5387520 & 8/06/04 & 5387776 + hd 206374 & 7/04/04 & 5414656 & 6/06/04 & 5414912 & 7/11/04 & 5415168 + lcccccc hd 6963 & 1263 @xmath96126 & 729 @xmath9673 & 255 @xmath9626 & 31 @xmath963 & 19 @xmath96 3 & 32 @xmath967 + hd 8907 & 1985 @xmath96199 & 1186 @xmath96119 & 402 @xmath9640 & 49 @xmath965 & 41 @xmath96 6 & 231 @xmath9646 + hd 13974 & 15274 @xmath961527 & 8970 @xmath96897 & 3079 @xmath96308 & 360 @xmath9636 & 186 @xmath96 28 & 44 @xmath9610 + hd 122652 & 1307 @xmath96131 & 771 @xmath9677 & 265 @xmath9627 & 34 @xmath963 & 21 @xmath96 3 & 67 @xmath9614 + hd 145229 & 1167 @xmath96117 & 695 @xmath9670 & 237 @xmath9624 & 29 @xmath963 & 22 @xmath96 3 & 56 @xmath9612 + hd 206374 & 1445 @xmath96145 & 842 @xmath9684 & 292 @xmath9629 & 34 @xmath963 & 18 @xmath96 3 & 22 @xmath966 + lllcrccc hd 6963 & 01:10:41.91 & @xmath12342:55:54.50 & 27 & 13 & k0 & 5517 & 0.55 + hd 8907 & 01:28:34.35 & @xmath12342:16:03.70 & 34 & 13 & f8 & 6250 & 2.05 + hd 13974 & 02:17:03.23 & @xmath12334:13:27.32 & 11 & 13 & g0/(g9@xmath4k4 ) & 6215/4493 & 1.18 + hd 122652 & 14:02:31.63 & @xmath12331:39:39.09 & 37 & 13 & f8 & 6157 & 1.55 + hd 145229 & 16:09:26.63 & @xmath12311:34:28.25 & 33 & 0.31 & g0 & 5893 & 0.96 + hd 206374 & 21:41:06.19 & @xmath12326:45:02.25 & 27 & 13 & g6.5 & 5580 & 0.74 + lcrrrrrcl hd 6963 & 10 & 18 & @xmath124 & @xmath125 & 57 & 33 & g0 & this study + hd 8907 & 10 & 48 & @xmath126 & @xmath127 & 48 & 34 & f8 & this study , silverstone ( 2000 ) + hd 13974 & 10 & @xmath11528 & @xmath128 & @xmath129 & 55 & 22 & g0v & this study , photosphere + hd 122652 & 10 & 31 & @xmath130 & @xmath131 & 56 & 37 & f8 & this study + hd 145229 & 10 & 24 & @xmath132 & @xmath133 & 56 & 33 & g0 & this study + hd 206374 & 10 & @xmath11520 & @xmath134 & @xmath135 & 57 & 27 & g6.5 & this study + lccrccrccc hd 6963 & 0.35 & 1.1 & 18 & 7.0 & 6.5 & irs 33 & @xmath136 + hd 8907 & 0.85 & 2.7 & 48 & 6.5 & 6.0 & irs 33 & @xmath137 + hd 13974 & 0.58 & 1.9&@xmath11528 & 7.0 & 6.5 & mips 70 & @xmath138 + hd 122652 & 0.68 & 2.2 & 31 & 6.8 & 6.3 & irs 33 & @xmath139 + hd 145229 & 0.49 & 1.6 & 24 & 7.0 & 6.5 & irs 33 & @xmath140 + hd 206374 & 0.43 & 1.4&@xmath11520 & 6.9 & 6.4 & irs 33 & @xmath141 + lccc hd 6963 & 1.03.0 & 70 & 30 + hd 8907 & 1.03.0 & 170 & 225 + hd 13974 & 1.03.0 & 300 & @xmath2020 + hd 122652 & 1.03.0 & 20 & 65 + hd 145229 & 0.31.0 & 70 & 55 + hd 206374 & 1.03.0 & 58 & 22 + m fluxes ) divided by @xmath19(random uncertainty ) . four excess sources ( filled histogram ) are easily noticed in this histogram . note that hd 8907 shows very strong excess flux at 70 @xmath0 m , which was detected by iras and iso . hd 6963 , hd 122652 , hd 145229 , and hd 206374 discovered by feps program are nearer to the rest of the stars without ir excess . ] m / ks vs. 24@xmath0m / ks . small circle with left - arrow symbols are 1@xmath19 upper limits of non - detections , and large filled circles marked with numbers are the source presented in this paper . hd 12039 ( a source with a dusty ring at around 5au , see hines et al . 2005 ) is the source with the largest 24@xmath0m / ks value in this plot . ]

recently , a color decomposition using structure constants was introduced for purely gluonic tree amplitudes , in a compact form involving only the linearly independent subamplitudes . we give two proofs that this decomposition holds for an arbitrary number of gluons . we also present and prove similar decompositions at one loop , both for pure gluon amplitudes and for amplitudes with an external quark - antiquark pair . = 0.1 cm = -2.5 cm = -2.0 cm = 23.2truecm = 16.7truecm 30 pt plus 1000pt minus 100pt # 1 # 1= to [ section ] eqnnum @eqnnum = eqnnum # 1@eqnnum @eqnnum@font(#1)currentlabel#1 @eqnnum @eqnnum definecounterequationaddtoresetequationsection @prefix . # 1_#1 _ # 1eq . ( [ # 1 ] ) # 1eq . ( [ # 1 ] ) # 1#2eqs . ( [ # 1 ] ) and ( [ # 2 ] ) # 1#2eqs . ( [ # 1 ] ) and ( [ # 2 ] ) # 1fig . [ # 1 ] # 1fig . [ # 1 ] # 1#2figs . [ # 1 ] and [ # 2 ] # 1#2figs . [ # 1 ] and [ # 2 ] # 1section [ # 1 ] # 1section [ # 1 ] # 1#2sections [ # 1 ] and [ # 2 ] # 1appendix [ # 1 ] # 1appendix [ # 1 ] # 1#2appendices [ # 1 ] and [ # 2 ] # 1#2appendices [ # 1 ] and [ # 2 ] # 1table [ # 1 ] # 1table [ # 1 ] @=11 p hep - ph/9910563 slac pub8294 + dftt 53/99 + october , 1999 + 2.cm * new color decompositions for gauge amplitudes + at tree and loop level * + vittorio del duca + + lance dixon + + fabio maltoni + + _ submitted to nuclear physics b _ ' '' '' width 3.6inaddress after november 1 , 1999 : theory division , cern , ch 1211 geneve 23 , switzerlandresearch supported by the us department of energy under grant de - ac03 - 76sf00515

hep-ph9910563

the computation of multi - parton scattering amplitudes in qcd is essential for quantitative predictions of multi - jet cross - sections at high - energy colliders . the feynman diagrams for such an amplitude can generate a thicket of complicated color algebra , tangled together with expressions composed of kinematic invariants . an extremely useful way to disentangle the color and kinematic factors is via _ color decompositions _ @xcite in terms of chan - paton factors @xcite , or traces of @xmath4 matrices in the fundamental representation , @xmath5 . the traces are multiplied by purely kinematical coefficients , called partial amplitudes or subamplitudes . originally motivated by the representation of gluon amplitudes in open string theory , such decompositions have been widely studied and applied at both the tree level @xcite and the loop level @xcite . two major advantages of the trace - based color decompositions are that ( a ) subamplitudes are gauge - invariant , and ( b ) an important ` color - ordered ' class of them receive contributions only from diagrams with a particular cyclic ordering of the external partons ; these subamplitudes therefore have much simpler kinematic properties than the full amplitude . although the standard color decompositions are very effective , in some cases they are not quite optimal . for example , the decomposition of the purely gluonic tree amplitude is overcomplete . explicitly , the @xmath6-gluon amplitude is written in terms of @xmath7 single - trace color structures , as @xmath8 where @xmath9 are the subamplitudes . the cyclic invariance of the trace has been used to fix its first entry , and the sum is over the set of non - cyclic permutations of @xmath6 elements , @xmath10 , where @xmath11 . however , the @xmath7 subamplitudes appearing in the equation are not all independent . besides cyclic and reflection invariances , which they inherit from the traces , subamplitudes also obey a @xmath12 decoupling identity ( also called a dual ward identity , or subcyclic identity ) @xcite , @xmath13 this identity can be derived by setting @xmath14 equal to the unit matrix and collecting terms containing the same trace ; this gives the amplitude for a @xmath12 ` photon ' and @xmath15 @xmath4 gluons , which must vanish . more general identities can be derived by assigning the @xmath4 generators for the external gluons to commuting subgroups such as @xmath16 @xcite . kleiss and kuijf found a linear relation between subamplitudes @xcite ( see ) , which is consistent with all of these identities , and which reduces the number of linearly independent subamplitudes from @xmath7 to @xmath17 . in contrast , the conventional color factors for gluonic feynman diagrams are composed of structure constants @xmath18 . in this case , the @xmath12 and generalized decoupling identities are all manifest ; for example , the @xmath12 generator commutes with all other generators , so all structure constants containing it are zero . on the other hand , a particular string of contracted structure constants will not appear _ ab initio _ with a gauge - invariant kinematic coefficient . to pass from the @xmath18-based decomposition to the trace - based decomposition , one substitutes @xmath19 and simplifies and collects the various traces . in the process , gauge invariance of the partial amplitudes is gained , but properties such as @xmath12 decoupling are no longer manifest . another disadvantage of trace - based decompositions at tree level is that the color structure of amplitudes in the multi - regge kinematics is obscured @xcite . the trace - based decomposition is also not optimal at the loop level . at one loop , the standard color decomposition @xcite for @xmath6-gluon amplitudes includes double trace structures , @xmath20 @xmath21 , in addition to single traces of the type that appear at tree level . the subamplitudes multiplying the single trace structures give the leading contributions in the large @xmath22 limit , and they are color - ordered . the double - trace subamplitudes , corresponding to subleading - in-@xmath22 contributions , can be written in terms of permutations of the leading ( color - ordered ) subamplitudes @xcite in a formula reminiscent of the tree - level kleiss - kuijf relation ( see ) . because the permutation formula is rather complicated , its implementation in a numerical program can be rather slow . similar formulae hold for one - loop amplitudes with two external quarks and @xmath23 gluons @xcite . recently , a new color decomposition for the @xmath6-gluon tree amplitude , in terms of structure constants rather than traces , has been presented @xcite , whose kinematic coefficients are just color - ordered subamplitudes . because of this property , the new decomposition retains all the advantages of trace - based decompositions , yet avoids the disadvantages mentioned above . in particular , the tree - level @xmath6-gluon decomposition is automatically given in terms of the @xmath17 independent subamplitudes , in a form that is very convenient for analyzing the high - energy limit . the purpose of this paper is to derive the new tree - level decomposition in two different ways , and to present similar @xmath18-based color decompositions at one loop , whose kinematic coefficients are again color - ordered subamplitudes . the leading and subleading - in-@xmath22 contributions combine neatly into one expression in the new decompositions . also , gluons circulating in the loop are put on the same footing as fermions in the loop . parton - level cross - sections require amplitude squares or interferences which are summed over all external color labels . we provide general expressions for color - summed cross - sections in terms of the new color decompositions . in we give explicit evaluations for quantities encountered in cross - section computations through @xmath24 . the paper is organized as follows . in we give two proofs of the new color decomposition for the @xmath6-gluon tree amplitude conjectured in ref . @xcite , and provide accordingly the square of the tree amplitude summed over colors . in we present ( and prove ) new color decompositions for one - loop @xmath6-gluon amplitudes and one - loop amplitudes with an external quark - antiquark pair plus @xmath23 gluons . in addition , we compute the color - summed interference terms between tree amplitudes and one - loop amplitudes , and the square of one - loop amplitudes , which are relevant respectively for next - to - leading order ( nlo ) and next - to - next - to - leading order ( nnlo ) calculations of jet production rates . the color - summed interference terms and squares are given in terms of color matrices ; appendix a provides many of those required for jet rate computations up to @xmath24 . finally , we outline how a new color decomposition for one - loop amplitudes with external photons , gluons and a quark - antiquark pair can be obtained from the amplitudes with only gluons and a @xmath25 pair . in we summarize our present understanding of the color decomposition of tree - level and one - loop amplitudes , and comment briefly on possible extensions to multi - loop amplitudes . the new color decomposition for the @xmath6-gluon tree amplitude is . hence our structure constants @xmath18 are larger than the conventional ones by a factor of @xmath26 . ] @xcite @xmath27 where @xmath28 is an @xmath4 generator in the adjoint representation . this color decomposition is analogous to the standard decomposition for the tree amplitude with a quark - antiquark pair and @xmath23 gluons @xcite , @xmath29 the only difference between the two is the representation used for the color matrices , namely the adjoint representation for the @xmath6-gluon amplitude and the fundamental representation for the amplitude containing a @xmath30 pair . is written in terms of the @xmath17 subamplitudes where legs @xmath31 and @xmath6 are adjacent . this is precisely the basis of linearly independent subamplitudes which is singled out by the kleiss - kuijf relation . in fact , we shall show that is equivalent to the kleiss - kuijf relation @xcite , which can be written as , @xmath32 here @xmath33 , @xmath34 is the number of elements in the set @xmath35 , the set @xmath36 is @xmath35 with the ordering reversed , and @xmath37 is the set of ` ordered permutations ' ( also called mergings @xcite ) of the @xmath23 elements of @xmath38 that preserve the ordering of the @xmath39 within @xmath40 and of the @xmath41 within @xmath36 , while allowing for all possible relative orderings of the @xmath39 with respect to the @xmath41 . we wish to show that the new color decomposition ( [ gluondecompnew ] ) is equivalent to inserting the kleiss - kuijf relation ( [ kkequation ] ) into the standard color decomposition ( [ treegluoncolorold ] ) . we first substitute for the structure constants appearing in , @xmath42)$ ] , and use the identity @xmath43\ldots\right]\right]\right ) . \label{fpro}\ ] ] we want to identify all terms that contain a trace of the form @xmath44 because these should give rise to @xmath45 . since there are @xmath23 commutators , there are @xmath46 terms on the right - hand side of , but only @xmath47 of them have exactly @xmath48 @xmath49 matrices appearing to the right of @xmath50 . the @xmath51 indices on the matrices to the right of @xmath50 must come in reversed order compared to how they appear in the @xmath18 string , but they can appear in that string in any relative order with respect to the @xmath51 that end up to the left of @xmath50 . thus , for any ordered permutation @xmath52 , the subamplitude @xmath53 appears in the new decomposition ( [ gluondecompnew ] ) accompanied by the desired trace ( [ desiredtrace ] ) ; a relative sign of @xmath54 comes from the commutators . collecting all such ordered permutations , we obtain the kleiss - kuijf formula for @xmath45 , , thus establishing its equivalence to the new color decomposition ( [ gluondecompnew ] ) . hence we can derive the new @xmath6-gluon color decomposition via its connection to the kleiss - kuijf relation . the latter relation was checked up to @xmath55 in ref . it can be proved for all @xmath6 using the same techniques that were used to prove an analogous one - loop formula @xcite , below . consider first the set of color - ordered feynman diagrams with only three - point vertices , which are ` multi - peripheral ' with respect to @xmath31 and @xmath6 , by which we mean that all other external legs connect directly to the line extending from @xmath31 to @xmath6 ; i.e. there are no non - trivial trees branching off of this line . label the legs on one side of the @xmath31@xmath6 line by @xmath39 , and those on the other side by @xmath56 . these diagrams contribute to both sides of the kleiss - kuijf relation ( [ kkequation ] ) in the proper way : there is no relative ordering requirement on the @xmath39 with respect to the @xmath56 on the left - hand side of the relation , and the ordering is summed over on the right - hand side ; the sign factor @xmath57 comes from the antisymmetry of color - ordered three - point vertices ( in feynman gauge ) . next consider diagrams with non - trivial trees attached to the @xmath31@xmath6 line . they work in exactly the same way as the multi - peripheral diagrams if the leaves ( external legs ) of each tree all belong to the same set , either @xmath40 or @xmath35 . however , if a tree contains leaves from both sets , then the diagram does not appear on the left - hand side of , so one must show that it cancels out of the permutation sum on the right - hand side . for trees with only three - point vertices , this can be done using the antisymmetry of the vertices . for other cases the cancellations are slightly more complicated ; one way of establishing them is via the generalized decoupling identities @xcite mentioned in the introduction @xcite . ( 200,60)(0,0 ) ( 10,15)[r]@xmath31 ( 10,15)(190,15 ) ( 30,15)(30,30 ) ( 30,40)[b]@xmath58 ( 60,15)(60,30 ) ( 60,40)[b]@xmath59 ( 90,15)(90,30 ) ( 115,20)[b]@xmath60 ( 140,15)(140,30 ) ( 170,15)(170,30 ) ( 170,40)[b]@xmath61 ( 193,15)[l]@xmath6 there is another way to prove . for this argument it is convenient to use a graphical notation for color factors made out of structure constants , in which @xmath18 is represented by a three - vertex and @xmath62 ( an index contraction ) is represented by a line @xcite . then the color factors appearing in are associated with the multi - peripheral color diagrams shown in . the color factor for a generic @xmath6-gluon feynman diagram is not of this form . however , we can use the jacobi identity @xmath63 shown graphically in , to put it into this form @xcite . represents the color factor for a generic feynman diagram graphically , as a tree structure with only three - point vertices . it also shows how it can be simplified into multi - peripheral form by repeated application of the jacobi identity . ( if the line running from @xmath31 to @xmath6 is in the fundamental representation , then the same steps , but using @xmath64 , lead directly to the color decomposition ( [ quarkdecompold ] ) for a quark - antiquark pair and @xmath23 gluons . ) ( 150,60)(0,0 ) ( 10,10)(20,20 ) ( 10,50)(20,40 ) ( 20,20)(20,40 ) ( 20,20)(30,10 ) ( 20,40)(30,50 ) ( 40,30)[l]@xmath65 ( 60,20)(70,30 ) ( 60,40)(70,30 ) ( 70,30)(90,30 ) ( 90,30)(100,20 ) ( 90,30)(100,40 ) ( 110,30)[l]@xmath66 ( 130,20)(140,30 ) ( 130,40)(148,34 ) ( 152,32.6666)(160,30 ) ( 140,30)(160,30 ) ( 160,30)(170,20 ) ( 140,30)(170,40 ) ( 400,95)(0,0 ) ( 10,40)[r]@xmath31 ( 10,40)(40,40 ) ( 25,25)(25,40 ) ( 40,25)(40,40 ) ( 40,25)(30,15 ) ( 40,25)(50,15 ) ( 40,40)(50,45 ) ( 50,45)(60,40 ) ( 50,45)(50,60 ) ( 50,60)(40,65 ) ( 50,60)(60,65 ) ( 40,65)(25,65 ) ( 40,65)(40,80 ) ( 60,65)(75,65 ) ( 60,65)(60,80 ) ( 60,40)(60,25 ) ( 60,40)(75,40 ) ( 78,40)[l]@xmath6 ( 50,52.5)(12,0,360)3.1 ( 110,45)[l]@xmath67 ( 150,40)[r]@xmath31 ( 150,40)(230,40 ) ( 165,25)(165,40 ) ( 180,25)(180,40 ) ( 180,25)(170,15 ) ( 180,25)(190,15 ) ( 190,40)(190,80 ) ( 190,65)(175,65 ) ( 205,40)(205,80 ) ( 205,65)(220,65 ) ( 215,40)(215,25 ) ( 233,40)[l]@xmath6 ( 260,45)[l]@xmath66 ( 295,40)[r]@xmath31 ( 295,40)(375,40 ) ( 310,25)(310,40 ) ( 325,25)(325,40 ) ( 325,25)(315,15 ) ( 325,25)(335,15 ) ( 335,65)(335,80 ) ( 335,65)(320,65 ) ( 335,40)(350,65 ) ( 335,65)(341,55 ) ( 344,50)(350,40 ) ( 350,65)(350,80 ) ( 350,65)(365,65 ) ( 360,40)(360,25 ) ( 378,40)[l]@xmath6 so far we have established that a color decomposition of the form ( [ gluondecompnew ] ) exists , but we have not yet shown that the kinematic coefficients are equal to the partial amplitudes @xmath68 . however , we can do this rather simply by equating the new decomposition ( but with unknown kinematic coefficients ) to the standard one , , and contracting both sides with @xmath69 , i.e. summing over all @xmath51 , @xmath70 , for some permutation @xmath71 . because the kinematic coefficients contain no @xmath22 dependence , we may retain only the leading contraction terms as @xmath72 . but it is well known that single traces for different cyclic orderings are orthogonal at large @xmath22 @xcite . similarly , shows that only one of the multi - peripheral color factors in survives the contraction , because the other @xmath73 permutations give rise to nonplanar , and hence @xmath22-suppressed , color topologies . thus the contraction selects a unique term from either side of the equation , corresponding to the same ordering of the color indices @xmath51 , and with the same weight at large @xmath22 , namely @xmath74 . this proves that the coefficients in the new color decomposition ( [ gluondecompnew ] ) are indeed the subamplitudes @xmath75 . ( 450,60)(0,0 ) ( 3.432,20)(116.568,20 ) ( 15,20)(15,49.8431 ) ( 30,20)(30,55.981 ) ( 45,20)(45,59.047 ) ( 60,20)(60,60 ) ( 75,20)(75,59.047 ) ( 90,20)(90,55.981 ) ( 105,20)(105,49.8431 ) ( 60,30)(30,60)(0 ) ( 60,30)(30,270.1,269.9 ) ( 130,30)[l]@xmath76 ( 253.432,20)(366.568,20 ) ( 265,20)(280,55.981 ) ( 280,20)(295,59.047 ) ( 295,20)(286,28.9529)(282,32.9320)(276,38.9006 ) ( 272,42.8797)(265,49.8431 ) ( 310,20)(310,60 ) ( 325,20)(325,59.047 ) ( 340,20)(340,55.981 ) ( 355,20)(355,49.8431 ) ( 310,30)(30,60)(0 ) ( 310,30)(30,270.1,269.9 ) ( 380,30)[l]@xmath77 because we showed that the new color decomposition is equivalent to the kleiss - kuijf relation , our second proof of the new decomposition can alternatively be viewed as a proof of the kleiss - kuijf relation . interestingly , this proof does not rely on any kinematic properties of color - ordered feynman diagrams ; only group - theoretic properties such as the jacobi identity and the large-@xmath22 behavior of color traces were used . the decomposition ( [ gluondecompnew ] ) has a color - ladder structure which naturally arises in the configurations where the gluons are strongly ordered in rapidity , i.e. in the multi - regge kinematics @xcite . ( indeed , this is how was first discovered @xcite . ) in contrast , some laborious work is necessary to obtain the color - ladder structure from the trace - based decomposition ( [ treegluoncolorold ] ) in the multi - regge kinematics @xcite . let gluons @xmath78 be the particles produced in a scattering where gluons 1 and @xmath6 are the incoming particles . the available final - state phase space may be divided into @xmath17 simplices , in each of which the produced gluons are ordered in rapidity . let us take the simplex defined by the rapidity ordering @xmath79 , and consider its sub - simplex with strong rapidity ordering @xmath80 . then the crucial point to note is that the dominant subamplitudes in in the multi - regge limit are @xmath81 , where @xmath40 and @xmath35 are both increasing sequences , whose union is @xmath82 ( @xmath36 is @xmath35 in reversed order ) . in other words , @xmath83 contains the sequence @xmath82 . if @xmath48 is the number of elements in @xmath35 , then for each @xmath48 there are @xmath47 such choices of @xmath40 , @xmath35 . summing over @xmath84 , there are @xmath46 orderings in total @xcite . using the identity ( [ fpro ] ) and the fact that in the multi - regge kinematics @xmath85 one obtains the color - ladder structure of , with just one allowed string of structure constants @xmath18 . the procedure can be repeated for all of the @xmath17 simplices , thus generating the @xmath17 strings of @xmath18 s of . the tree - level partonic scattering cross - section is given by the square of the tree amplitude @xmath86 , summed over all colors . this expression can be written either of two ways , @xmath87 where is obtained from , while is obtained from , and the subscript @xmath88 on @xmath89 now refers to the subamplitude @xmath9 evaluated for the @xmath90 permutation @xmath91 in @xmath92 or @xmath73 , respectively . in , the color matrix @xmath93 is @xmath94^ * , \label{cdef}\end{aligned}\ ] ] whereas in the matrix is @xmath95^*. \label{ctilde}\ ] ] in refs . @xcite , the kleiss - kuijf relation was used to calculate @xmath96 from @xmath93 . from we see that it can be calculated directly in terms of structure constants . in we give @xmath96 for @xmath97 . although the expression ( [ square2 ] ) seems to have fewer terms than , the sparseness of the matrices is also important for determining which expression has the fastest numerical evaluation @xcite . for an evaluation that is good only to leading order in @xmath22 the leading color approximation , or lca it is best to use and the large-@xmath22 orthogonality of different single traces , to obtain @xmath98 \ , , \label{square1largen}\end{aligned}\ ] ] where @xmath99 up to @xmath100 , the lca is exact , and so is also superior to for an exact evaluation . for @xmath101 , where the lca has subleading corrections , it is best to take the leading terms from , and the subleading terms from ( these are the terms proportional to @xmath102 in ) @xcite . for @xmath103 , ref . @xcite employed the linear dependences in the kleiss - kuijf relation to find an even more compact form for the subleading terms than using . ( see ref . @xcite for another approach to numerically evaluating multi - parton amplitudes beyond the lca . ) the standard color decomposition for one - loop @xmath6-gluon amplitudes in @xmath4 gauge theory with @xmath104 flavors of quarks is @xcite @xmath105}(\si_1,\ldots,\si_n ) \nonumber\\ & + & \hskip-4 mm \sum_{c=2}^{\lfloor n/2\rfloor+1 } \sum_{\si \in s_n / s_{n;c } } \tr ( \lambda^{a_{\si_1}}\cdots \lambda^{a_{\si_{c-1 } } } ) \tr ( \lambda^{a_{\si_c } } \cdots \lambda^{a_{\si_n } } ) a_{n;c } ( \si_1,\ldots,\si_n ) \nonumber\\ & + & \nf \sum_{\si \in s_n/{\bbb z}_n } \tr ( \lambda^{a_{\si_1}}\cdots \lambda^{a_{\si_n } } ) a_{n;1}^{[1/2]}(\si_1,\ldots,\si_n ) \biggr ] \ , , \label{loopcolor}\end{aligned}\ ] ] where @xmath106 are the subamplitudes , @xmath107 is the subset of @xmath108 that leaves the double trace structure invariant , and @xmath109 is the greatest integer less than or equal to @xmath110 . the superscript @xmath111 $ ] denotes the spin of the particle circulating in the loop . the subamplitudes @xmath112}$ ] are color - ordered , and give rise to the leading contributions to the cross - section in the large-@xmath22 limit . the subleading subamplitudes @xmath113 can be obtained from the leading ones @xmath114}_{n;1}$ ] through the permutation sum @xcite @xmath115}_{n;1}(\si_1,\ldots,\si_n ) \ , , \label{sublanswer}\ ] ] where @xmath116 , @xmath117 , and @xmath118 is the set of all permutations of @xmath119 with @xmath6 held fixed that preserve the cyclic ordering of the @xmath39 within @xmath40 and of the @xmath41 within @xmath35 , while allowing for all possible relative orderings of the @xmath39 with respect to the @xmath41 . in a dual representation of one - loop amplitudes , the subleading subamplitudes @xmath120 come from the annulus diagram where gluons belonging to @xmath40 and @xmath35 are emitted from the inner and outer boundaries of the annulus , respectively . thus their color properties are very similar to those of the tree subamplitudes @xmath45 if legs @xmath31 and @xmath6 are ` sewn together ' , i.e. contracted with @xmath121 . indeed , with this interpretation , has a very similar structure to the kleiss - kuijf relation ( [ kkequation ] ) . these remarks suggest that the new tree - level color decomposition should have a one - loop analog , where the strings of structure constants that appear are ` ring diagrams ' instead of multi - peripheral diagrams , as depicted in a. ( 350,120)(0,0 ) ( 0,60)[r](a ) ( 80,60)(40,0,360 ) ( 25,60)(40,60 ) ( 120,60)(135,60 ) ( 80,100)(80,115 ) ( 80,20)(80,5 ) ( 114.641,80)(127.631,87.5 ) ( 100,94.641)(107.5,107.631 ) ( 60,94.641)(52.5,107.631 ) ( 45.359,80)(32.369,87.5 ) ( 45.359,40)(32.369,32.5 ) ( 60,25.359)(52.5,12.369 ) ( 100,25.359)(107.5,12.369 ) ( 114.641,40)(127.631,32.5 ) ( 200,60)[r](b ) ( 224,60)[r]@xmath122 ( 339,60)[l]@xmath48 ( 280,60)(40,0,360 ) ( 280,100)(280,115 ) ( 280,20)(280,5 ) ( 314.641,80)(327.631,87.5 ) ( 300,94.641)(307.5,107.631 ) ( 260,94.641)(252.5,107.631 ) ( 245.359,80)(232.369,87.5 ) ( 245.359,40)(232.369,32.5 ) ( 260,25.359)(252.5,12.369 ) ( 300,25.359)(307.5,12.369 ) ( 314.641,40)(327.631,32.5 ) ( 280,60)(40,180,360 ) ( 225,60)(240,60 ) ( 320,60)(335,60 ) the new one - loop @xmath6-gluon color decomposition , expressed in terms of adjoint generator matrices @xmath123 , is @xmath124}(\si_1,\ldots,\si_n ) + 2 \nf \tr ( \lambda^{a_{\si_1 } } \cdots \lambda^{a_{\si_n } } ) a_{n;1}^{[1/2]}(\si_1,\ldots,\si_n ) \bigr ] . \label{newloopcolor}\end{aligned}\ ] ] here @xmath125 is the group of non - cyclic permutations , and @xmath126 is the reflection : @xmath127 . thus the number of linearly independent subamplitudes is @xmath128 . the terms proportional to @xmath104 in follow from using only the reflection identity satisfied by the @xmath129 : @xmath112}(n,\ldots,2,1 ) = ( -1)^n \ , a_{n;1}^{[j]}(1,2,\ldots , n)$ ] . to prove that the remaining terms in are correct , we follow closely the second proof of . using the jacobi identity , the generic one - loop @xmath18 color factor can be turned into ring diagrams . the manipulations are exactly the same as those shown in , once legs @xmath31 and @xmath6 are joined together to form a loop . this establishes , but with an unknown coefficient for @xmath130 . next we equate this expression to the standard color decomposition ( [ loopcolor ] ) , and contract both sides with @xmath131 . again a unique term survives on each side of the equation , at leading order in @xmath22 , and this establishes that the coefficient is correctly given by @xmath132}(\si_1,\ldots,\si_n)$ ] . ( if we had not removed the reflections @xmath126 from , then both a ring diagram and its reflection would have contributed to the contraction ; this would have led to a factor of @xmath133 in the coefficient of @xmath130 . ) puts the gluon and fermion loops manifestly on the same footing . the contributions of the subleading - color subamplitudes @xmath113 are neatly packaged into the ring - diagram color structures . in addition , all of the generalized @xmath12 decoupling identities are automatically incorporated . finally , shows explicitly the difference between the tree - level and the one - loop color decomposition of the gluon sector for any number of external legs . for instance , if we take the amplitude for a higgs boson plus three gluons , both and the gluon - loop contribution to have as a color factor @xmath134 ; the actual difference in the color structure is concealed in the different properties of the subamplitudes @xmath135 and @xmath136}$ ] . in contrast , and show explicitly the difference in color structure . one - loop amplitudes contribute to nlo qcd cross - sections through their interference with tree amplitudes . carrying out the color - summed nlo interference terms for amplitudes color - decomposed as in is straightforward : @xmath137^ * { \large\vert}_{\rm nlo } } \nonumber\\ & & = 2\ , \sum_{\rm colors } \re \left ( \a^\tree ( \a^\oneloop)^ * \right ) \label{nlosquare}\\ & & = 2 ( g^2)^{n-1}\ , \re \sum_{i=1}^{(n-2 ) ! } \sum_{j=1}^{(n-1)!/2 } a_i^\tree \left [ \hat{c}_{ij } ( a_j^{[1]})^ * + 2 \ , \nf \ , \hat{d}_{ij } ( a_j^{[1/2]})^ * \right ] \ , , \nonumber\end{aligned}\ ] ] where @xmath138^ * \ , , \nonumber\\ \hat{d}_{ij } & = & \sum_{\rm colors } \left(p_i \{f^{a_2 } \cdots f^{a_{n-1}}\ } \right)_{a_1 a_n } \left[\tr ( \lambda^{a_1 } p_j \ { \lambda^{a_2 } \cdots \lambda^{a_n } \ } ) \right]^ * , \label{chat}\end{aligned}\ ] ] with @xmath91 the @xmath90 permutation in @xmath73 and @xmath139 the @xmath140 permutation in @xmath141 . nnlo production rates include the virtual contributions , @xmath142^ * { \large\vert } _ { \rm nnlo } = \sum_{\rm colors } \left [ \ , 2\ , \re \left ( \a^\tree ( \a^\twoloop)^ * \right ) + |\a^\oneloop|^2 \,\right]\ , .\label{nnlosquare}\ ] ] for the square of one - loop @xmath6-gluon amplitudes on the right - hand side of , we obtain @xmath143 } ( a_j^{[1]})^ * + 4 \ , \nf \ , \re \ , \left [ \bar{d}_{ij } a_i^{[1]}(a_j^{[1/2]})^ * \right ] \nonumber\\ & & \hskip 2.5 cm + \ , 4 \ , \nf^2 \ , c_{ij } a_i^{[1/2 ] } ( a_j^{[1/2]})^ * \biggr]\ , , \label{1loopsq}\end{aligned}\ ] ] where @xmath144^ * , \nonumber\\ \bar{d}_{ij } & = & \sum_{\rm colors } \tr \left(f^{a_1 } p_i \{f^{a_2 } \cdots f^{a_n}\}\right ) \left[\tr ( \lambda^{a_1 } p_j \ { \lambda^{a_2 } \cdots \lambda^{a_n } \ } ) \right]^ * , \label{cbar } \end{aligned}\ ] ] @xmath93 is given in , and @xmath91 is the @xmath90 permutation in @xmath141 . in we give the explicit values for the color matrices @xmath145 , @xmath146 , @xmath147 , @xmath148 and @xmath93 that are required for cross - section computations up to @xmath24 . a similar analysis can be applied to one - loop amplitudes with an external quark - antiquark pair plus @xmath23 gluons . these amplitudes have the standard color decomposition @xcite , @xmath149 where the color structures @xmath150 are defined by @xmath151 and @xmath152 is the subgroup of @xmath73 that leaves @xmath150 invariant . the leading - color subamplitudes in can be expressed in terms of color - ordered ( ` primitive ' ) amplitudes as @xmath153}(1_\qb,2_q,3,\ldots , n ) - { 1\over n_c^2 } a_n^{r,[1]}(1_\qb,2_q,3,\ldots , n ) \nonumber\\ & & \quad + { \nf \over n_c } a_n^{l,[1/2]}(1_\qb,2_q,3,\ldots , n ) , \label{qqleadanswer}\end{aligned}\ ] ] whereas the subleading contributions are given by the permutation formula @xcite , @xmath154}(\si(1_\qb,2_q,3,\ldots , n ) ) - { \nf \over n_c } a_n^{r,[1/2]}(\si(1_\qb,2_q,3,\ldots , n ) ) \biggr ] \ , , \label{qqsublanswer}\end{aligned}\ ] ] where @xmath155 , @xmath156 , with @xmath157 held fixed . in , the quantities @xmath158}$ ] and @xmath159}$ ] are defined in terms of color - ordered feynman diagrams , much like the @xmath6-gluon subamplitudes @xmath112}$ ] . however , because an external @xmath160 pair is present now , one also has to specify which way the line running from the antiquark to the quark turns when it enters the loop , left or right ; this accounts for the additional index , @xmath161 or @xmath162 . again the index @xmath111 $ ] denotes the spin @xmath163 of a particle circulating around the loop , for those graphs in the primitive amplitude where the external @xmath160 line does not enter the loop . the reflection identity , @xmath164}(1_\qb,3,4,\ldots,2_q,\ldots , n-1,n ) = ( -1)^n \ , a_n^{l,[j]}(1_\qb , n , n-1,\ldots,2_q,\ldots,4,3 ) , \label{qqreflection}\ ] ] relates the left and right subamplitudes . there are two types of subleading - color contributions in , distinguished by whether they contain a factor of @xmath104 or not . although both types are generated by cyclicly ordered permutations , their origins are quite different . in particular , the subleading @xmath104 terms arise from the @xmath165 term in the @xmath4 fierz identity , @xmath166 where @xmath167 , @xmath168 are strings of generator matrices @xmath169 . such manifest @xmath165 terms are not generated by strings of structure constants . ( in fact , @xmath165 terms are generally suppressed when @xmath18s are present , because @xmath165 factors come from explicitly projecting out the @xmath12 factor in @xmath170 , while structure constants accomplish this projection automatically . ) because of this fact , the freedom to include structure constants in a color decomposition does not seem to lead to any simpler representation of the subleading @xmath104 terms in , and we shall be content to simplify the no-@xmath104 terms terms in the @xmath171 subamplitudes arise from projecting out a @xmath12 factor , they can also be viewed as subamplitudes for a fictitious @xmath172 theory containing two fermion representations , say @xmath173 ( the external @xmath160 pair ) and @xmath174 ( the fermion in the loop ) . this description makes it a bit more transparent how the color structure @xmath175 @xmath176 multiplying these subamplitudes corresponds to their definition . but it does not lead to a distinct color decomposition . ] . the new color decomposition for one - loop amplitudes with an external @xmath160 pair is @xmath177}(1_\qb,\si_{p+1},\ldots,\si_n,2_q,\si_3,\ldots,\si_p ) \label{loopcolorqqnew}\\ & + & { n_f\over n_c } \ , \sum_{j=1}^{n-1 } \sum_{\si\in s_{n-2}/s_{n;j } } \gr^{(\qb q)}_{n;j } ( \si_3,\ldots,\si_n ) a_{n;j}^{[1/2 ] } ( 1_{\qb},2_q;\si_3,\ldots,\si_n ) \biggr]\ , , \nonumber\end{aligned}\ ] ] where for @xmath178 the product of generators in the adjoint representation reduces to the identity , @xmath179 ; the color structures @xmath150 are defined in ; and @xmath180}(1_\qb,2_q,3,\ldots , n ) & = & a_n^{l,[1/2]}(1_\qb,2_q,3,\ldots , n ) , \label{newqqsubamp } \\ a_{n;j}^{[1/2]}(1_\qb,2_q;3,\ldots , j+1;j+2,j+3,\ldots , n ) & = & ( -1)^j \sum_{\si\in\cop\{\alpha\}\{\beta\ } } a_n^{r,[1/2]}(\si(1_\qb,2_q,3,\ldots , n))\ , , \nonumber\end{aligned}\ ] ] with @xmath40 and @xmath35 as in . the new types of color structures appearing in are shown in b. most of the subamplitudes @xmath181}$ ] do not contribute at leading order in the large @xmath22 limit . ( only those of the form @xmath181}(1_\qb,\ldots,2_q)$ ] do . ) therefore the technique of contracting with a suitable color structure at large @xmath22 can not be used to prove . instead we follow the first line of reasoning that we used at tree - level ; that is , we substitute for the structure constants in terms of @xmath49 matrices and collect terms . the result agrees with , after substituting in for the leading and subleading subamplitudes and making use of the reflection identity ( [ qqreflection ] ) . the contribution of one - loop amplitudes with an external quark - antiquark pair to nlo qcd cross sections parallels the @xmath6-gluon case , . carrying out the color - summed nlo interference terms for amplitudes color - decomposed as in , and omitting the @xmath104 terms , we obtain , @xmath182^ * { \large\vert}_{\rm nlo } } \nonumber\\ & & = 2\ , \sum_{\rm colors } \re \left ( \a^\tree ( \a^\oneloop)^*_{{\rm no}-\nf } \right ) \label{nloqqsquare}\\ & & = 2 ( g^2)^{n-1}\ , \re \sum_{p=2}^n \sum_{i , j=1}^{(n-2 ) ! } \hat{m}_{ij}^{(p ) } a_i^\tree(1_\qb,2_q , i_3,\ldots , i_n ) \left[a_j^{r,[1]}(1_\qb , j_{p+1},\ldots , j_n,2_q , j_3,\ldots , j_p)\right]^ * \ , , \nonumber\end{aligned}\ ] ] where @xmath183^\dagger\ , , \label{mhat}\ ] ] with @xmath91 the @xmath90 permutation in @xmath73 , and @xmath139 the @xmath140 permutation in @xmath73 at fixed @xmath184 . the partitions @xmath184 span a matrix - valued vector @xmath185 , which we give in appendix a for @xmath186 , as required for cross - section computations up to @xmath24 . for @xmath100 , we have checked that , with @xmath185 given in , agrees with the no-@xmath104 terms of ref . the @xmath104 terms for @xmath186 have been computed in refs . @xcite and @xcite , respectively . finally , we note that by converting gluons into photons we can obtain multi - photon amplitudes at tree and loop level . in order to convert a gluon into a photon we make use of the embedding of @xmath12 in @xmath187 . we replace the quark - gluon vertex factor @xmath188 with the quark - photon vertex factor @xmath189 , where @xmath190 is the quark electric charge and the factor @xmath26 is due to our choice of normalization of the @xmath49 matrices . since the identity matrix @xmath191 commutes with the @xmath49 matrices , all possible attachments of the photon to the quark line contribute to the same color structure . thus subamplitudes involving a photon can be obtained by summing over permutations of gluon subamplitudes in which the photon assumes all possible places in the ordering . photons do not couple to gluons . in a @xmath18-based decomposition this is apparent : when the index @xmath102 corresponds to the identity matrix @xmath191 , the @xmath187 generator @xmath192 in the adjoint representation vanishes . in this paper we presented new color decompositions for gluon tree amplitudes , , and for the gluon - loop sector of one - loop amplitudes , ; we still use the @xmath49-based decomposition for quark - gluon tree amplitudes , , and for the quark - loop sector of one - loop amplitudes . thus it might seem that we can bring no new insight to the calculation of multi - photon amplitudes . however , that is not the case for one - loop amplitudes with an external @xmath160 pair : in the no-@xmath104 terms in a photon can still attach to the external quark line , i.e. to the string of @xmath49 matrices ; however , it manifestly can not attach to the string of @xmath193 matrices . this warrants that , like in and at least for the gluon - loop sector , the ensuing amplitude involving a photon is directly decomposed in terms of independent subamplitudes . we immediately see that subamplitudes of type @xmath181}(1_\qb,\si_3,\ldots,\si_n,2_q)$ ] , corresponding in to partitions with @xmath194 , do not contribute , because there is no @xmath5 matrix to replace by the identity . ( the @xmath195 matrices correspond to internal gluons and so should not be replaced . ) in the @xmath49-based decomposition @xcite , the vanishing contribution of these subamplitudes is realized only after summing over all permutations . thus the new color decomposition we obtain for @xmath196 one - loop amplitudes , by converting a gluon into a photon in , and summing over all the possible attachments of the photon to the quark line , is more compact than the one obtained from the @xmath49-based decomposition . for @xmath100 we verified that explicitly by computing the @xmath197 one - loop amplitude from , and comparing it to ( and checking its equivalence with ) the same amplitude , but @xmath49-decomposed @xcite . in ref . @xcite a new color decomposition was presented for the @xmath6-gluon tree amplitude ; it was shown to be equivalent to the standard one up to @xmath103 and was conjectured to be valid for arbitrary @xmath6 . in the present paper we have proven this conjecture . we have also presented , and proven , analogous color decompositions for one - loop @xmath6-gluon amplitudes and one - loop amplitudes with an external quark - antiquark pair plus @xmath23 gluons . the new color decompositions have the advantage of being written in terms of just the independent subamplitudes . this makes the decomposition of an amplitude more compact . it is particularly useful for computing the color - summed interference terms between tree amplitudes and one - loop amplitudes , and the square of one - loop amplitudes , which are relevant respectively for nlo and nnlo calculations of jet production rates . the color - summed interference terms and squares are given in terms of color matrices ; we have provided in the appendix explicit values required for cross - section computations up to @xmath24 . ( 420,170)(0,0 ) ( 0,85)[r](a ) ( 50,120)(20,0,360 ) ( 50,100)(50,140 ) ( 67.3205,130)(80.3109,137.5 ) ( 67.3205,110)(80.3109,102.5 ) ( 32.6795,130)(19.6891,137.5 ) ( 32.6795,110)(19.6891,102.5 ) ( 50,50)(20,0,360 ) ( 50,30)(50,70 ) ( 67.3205,60)(80.3109,67.5 ) ( 67.3205,40)(80.3109,32.5 ) ( 30,50)(17.5,50 ) ( 50,50)(62.5,50 ) ( 130,85)[r](b ) ( 210,85)(50,0,360 ) ( 210,35)(210,135 ) ( 253.3013,110)(270.6218,120 ) ( 235,128.3013)(245,145.6218 ) ( 270,85)1 ( 267.9555,69.4709)1 ( 261.9615,55)1 ( 235,41.6987)(245,24.3782 ) ( 166.6987,110)(149.3782,120 ) ( 185,128.3013)(175,145.6218 ) ( 150,85)1 ( 152.0444,69.4709)1 ( 158.0385,55)1 ( 185,41.6987)(175,24.3782 ) ( 210,110)(230,110 ) ( 220,95)1 ( 220,85)1 ( 220,75)1 ( 210,60)(230,60 ) ( 330,85)[r](c ) ( 380,85)(20,0,360 ) ( 380,65)(380,105 ) ( 394.1421,99.1421)(404.7487,109.7487 ) ( 394.1421,70.8579)(404.7487,60.2513 ) ( 400,85)(415,85 ) ( 360,85)(345,85 ) we expect that structure - constant - based decompositions will also be useful for general multi - loop scattering amplitude calculations , once they have been derived . in a particular case they have already proven useful : two - loop four - gluon amplitudes in @xmath198 supersymmetric yang - mills theory were computed via their unitarity cuts and were presented in a trace - based formalism @xcite , but they can be more compactly expressed in an @xmath18 basis @xcite . in the latter basis only two types of color structures appear ; these are represented graphically in a by the planar and nonplanar double box diagrams . the kinematic object multiplying each structure is simply the scalar integral ( i.e. , the @xmath199 feynman diagram ) for the graph with the same topology . some of this simplicity is undoubtedly due to @xmath198 supersymmetry . however , group theory alone restricts the possible multi - loop color structures considerably . at one loop , we used the jacobi identity to remove nontrivial trees from the set of @xmath6-gluon color structures . we can similarly remove all such trees from multi - loop @xmath6-gluon color structures . the remaining color factors are illustrated in b ( for the two - loop case ) . in the one - loop case , this reduction exhausted the use of the jacobi identity ; in the multi - loop case this is no longer true . for example , in the two - loop four - gluon amplitude the nonplanar double box diagram in a could be removed in favor of the planar double box diagram in a and another planar diagram , shown in c. yet this would be an unwise move , at least in the @xmath198 supersymmetric case , in the sense that the nonplanar scalar double box integral would then have to be associated with planar color factors . thus the jacobi identity should be wielded more judiciously in the multi - loop case , and we leave this to future work . finally , we remark that even if a general , multi - loop @xmath18-based color decomposition is not yet available , the results in this paper can still be employed in the context of specific multi - loop gluonic calculations based on unitarity cuts @xcite . for example , the unitarity cuts of two - loop amplitudes are given in terms of tree - level and one - loop amplitudes . color decomposing these amplitudes as in this paper leads directly , after sewing the color factors on each side of the cut , to an @xmath18-based decomposition for the cut of the two - loop amplitude . this decomposition is gauge invariant on the cut , by virtue of the gauge invariance of the tree and one - loop amplitudes . the only potential difficulty is that to reconstruct the amplitude from its cuts , one has to find a single function that consistently matches all the cuts , and the color factors may have to be manipulated at this point . still , this seems likely to be a more efficient than a corresponding approach using a trace - based decomposition of tree and one - loop amplitudes , at least for computing terms that are subleading in the number of colors . at first sight it may seem like a step backward to abandon trace - based color decompositions of gluonic scattering amplitudes in favor of ones based on structure constants . however , we have seen in this paper that with an appropriate choice of color factors , one does not really have to give up any desirable properties of the trace - based formalism ( such as gauge invariance ) , at least at the tree and one - loop level . we look forward to multi - loop extensions of this work , and their application to more precise jet rate computations . .3 cm * acknowledgments * we thank zvi bern and michelangelo mangano for useful conversations and comments on the manuscript . we first give the color matrices appearing in the color - summed tree - level cross - section , defined in . for @xmath200 they are @xmath201 and for @xmath100 , @xmath202 where @xmath203 is defined in . the above results agree with ref . @xcite , up to an overall factor due to the different choice of normalization of the matrices in the fundamental representation . the color matrix @xmath96 for @xmath204 shows the @xmath205 dependence outlined in . for @xmath206 , @xmath96 is a @xmath207 matrix . rather than give the entire matrix , we note that its essential information is contained in its first row , i.e. in the matrix elements of @xmath208 with the 24 permutations . defining @xmath209 , we have @xmath210 where @xmath211 . the terms proportional to @xmath102 agree with eq . ( a.10 ) of ref . @xcite . the color matrices for the tree - level one - loop interference , defined in , are for @xmath200 @xmath212}_{4;1})_j= \left ( \begin{array}{c } ( 1 , 2 , 3 , 4)\\ ( 1 , 2 , 4 , 3)\\ ( 1 , 3 , 2 , 4 ) \end{array } \right ) \ ; , \label{chat5}\end{aligned}\ ] ] and for @xmath100 , @xmath213 & & \hat{d}_{ij}={\cal c}_5(n_c ) \cdot \left ( \begin{array}{cccccccccccc } 1 & -1 & 0 & -1 & 0 & 1 & 0 & 0 & b & -b & b & b\\ 0 & -1 & 1 & -1 & 1 & 0 & b & b & b & 0 & 0 & b\\ 0 & 0 & b & -1 & -b & 1 & 1 & -1 & 0 & 0 & b & b\\ b & b & b & 0 & 0 & 1 & 0 & -1 & 1 & 1 & b & 0\\ b & -1 & 0 & 0 & 1 & -b & b & 0 & b & 1 & 1 & 0\\ b & 0 & b & b & 1 & 0 & b & -1 & 0 & 1 & 0 & 1 \end{array } \right ) \ ; , \\ & & ( a^{[1]}_{5;1})_j= \left ( \begin{array}{c } ( 1 , 2 , 3 , 4 , 5)\\ ( 1 , 2 , 3 , 5 , 4)\\ ( 1 , 2 , 4 , 3 , 5)\\ ( 1 , 2 , 4 , 5 , 3)\\ ( 1 , 2 , 5 , 3 , 4)\\ ( 1 , 2 , 5 , 4 , 3)\\ ( 1 , 3 , 2 , 4 , 5)\\ ( 1 , 3 , 2 , 5 , 4)\\ ( 1 , 3 , 4 , 2 , 5)\\ ( 1 , 3 , 5 , 2 , 4)\\ ( 1 , 4 , 2 , 3 , 5)\\ ( 1 , 4 , 3 , 2 , 5 ) \end{array } \right ) \ ; , \label{a51def}\end{aligned}\ ] ] where @xmath211 , @xmath214 . although the matrices @xmath145 and @xmath146 have a relatively small dimension , they are not particularly sparse . we can improve the sparseness considerably by using the tree - level @xmath12 decoupling identity , , in every column where three identical entries appear , thus trading three terms for one . of course , more than @xmath17 different tree subamplitudes will now appear . using also the tree - level reflection identity , we arrive at the formula @xmath215^ * { \large\vert}_{\rm nlo } } \label{nlo5final } \\ & & = 4 g^8 { \cal c}_6(n_c ) \ , \re \sum_{i=1}^{(n-1)!/2 } \biggl [ a_i^\tree \bigl ( a_i^{[1 ] } + { \nf \over n_c } \ , a_i^{[1/2 ] } \bigr)^ * + { 1\over n_c^2 } a_{(24135)\cdot i}^\tree \bigl ( 12 a_i^{[1 ] } + 2 \ , { \nf \over n_c } \ , a_i^{[1/2 ] } \bigr)^ * \biggr ] \ , , \nonumber\end{aligned}\ ] ] where @xmath216 is the permutation @xmath217 composed with the @xmath90 permutation . in comparing this formula to eq . ( 11 ) of ref . @xcite , we see that the @xmath218 terms can be removed , at the price of multiplying the remaining @xmath219-suppressed pure - glue ( no @xmath104 ) terms by a factor of six . this form for the nlo five - gluon cross - section has been noted independently by w. kilgore @xcite . for the nnlo terms in with @xmath200 we have @xmath220 @xmath221 @xmath222 where @xmath223 are defined below . equivalently , @xmath224 } \big\vert^2 + 4 { \nf\over n_c } \ , \re \ , a_i^{[1 ] } ( a_i^{[1/2]})^ * + 4 { \nf^2\over n_c^2 } ( 1-b ) \ , \big\vert a_i^{[1/2 ] } \big\vert^2 \bigr ) \nonumber\\ & & \hskip 0.5 cm + \ , a \big\vert \sum_{i=1}^3 a_i^{[1 ] } \big\vert^2 + 4 { \nf\over n_c } b \ , \re \ , \bigl ( \sum_{i=1}^3 a_i^{[1 ] } \bigr ) \bigl ( \sum_{i=1}^3 a_i^{[1/2 ] } \bigr)^ * + 4 { \nf^2\over n_c^2 } \ , d \ , \big\vert \sum_{i=1}^3 a_i^{[1/2 ] } \big\vert^2 \ , \biggr]\ , . \label{nnlo4final}\end{aligned}\ ] ] the color matrix - valued vector @xmath225 for the tree - level one - loop interference for nlo processes with an external quark - antiquark pair , defined in , is for @xmath200 , @xmath226 @xmath227 where @xmath228 . for @xmath100 we obtain , @xmath229 @xmath230 where @xmath231 . in both , for each value of @xmath184 , the permutations @xmath139 of the gluons in the loop amplitude are understood to appear in the same order as the tree amplitude permutations @xmath232 . old qqgg(g ) matrices : the color matrix - valued vector @xmath225 for the tree - level one - loop interference for nlo processes with an external quark - antiquark pair , defined in , is for @xmath200 , @xmath233 @xmath234 where @xmath228 . for @xmath100 we obtain , @xmath235 @xmath236 where @xmath231 . in both , for each value of @xmath184 , the permutations @xmath139 of the gluons in the loop amplitude are understood to appear in the same order as the tree amplitude permutations @xmath232 . p. cvitanovi , p.g . lauwers and p.n . scharbach , nucl . * b186 * , 165 ( 1981 ) . berends and w.t . giele , nucl . * b294 * , 700 ( 1987 ) ; + m. mangano , s. parke and z. xu , nucl . * b298 * , 653 ( 1988 ) . d.a . kosower , b .- h . lee and v.p . nair , phys . * 201b * , 85 ( 1988 ) ; + m. mangano , nucl . * b309 * , 461 ( 1988 ) ; + d. zeppenfeld , int . j. mod . * a3 * , 2175 ( 1988 ) ; + d.a . kosower , nucl . b315 * , 391 ( 1989 ) . m.l . mangano and s.j . parke , phys . * 200 * , 301 ( 1991 ) . z. bern and d.a . kosower , nucl . * b362 * , 389 ( 1991 ) . paton and h .- chan , nucl . b10 * , 516 ( 1969 ) . z. bern , l. dixon and d.a . kosower , ann . nucl . part . sci . * 46 * , 109 ( 1996 ) [ hep - ph/9602280 ] . berends and w.t . giele , nucl . * b306 * , 759 ( 1988 ) . f.a . berends and w.t . giele , nucl . * b313 * , 595 ( 1989 ) . r. kleiss and h. kuijf , nucl . phys . * b312 * , 616 ( 1989 ) . v. del duca , phys . rev . * d52 * , 1527 ( 1995 ) [ hep - ph/9503340 ] . z. bern , l. dixon , d.c . dunbar and d.a . kosower , nucl . phys . * b425 * , 217 ( 1994 ) [ hep - ph/9403226 ] . z. bern , l. dixon and d.a . kosower , nucl . * b437 * , 259 ( 1995 ) [ hep - ph/9409393 ] . v. del duca , a. frizzo and f. maltoni , preprint hep - ph/9909464 . fadin , e.a . kuraev and l.n . lipatov , zh . eksp . . fiz . * 71 * , 840 ( 1976 ) [ sov . jetp * 44 * , 443 ( 1976 ) ] . v. del duca , phys . rev . * d48 * , 5133 ( 1993 ) [ hep - ph/9304259 ] . berends , w.t . giele and h. kuijf , nucl . phys . * b333 * , 120 ( 1990 ) . f. caravaglios , m.l . mangano , m. moretti and r. pittau , nucl . * b539 * , 215 ( 1999 ) [ hep - ph/9807570 ] . z. kunszt , a. signer and z. trcsnyi , nucl . phys . * b411 * , 397 ( 1994 ) [ hep - ph/9305239 ] . v. del duca , w. kilgore and f. maltoni , preprint hep - ph/9910253 . z. bern , j.s . rozowsky and b. yan , phys . lett . * b401 * , 273 ( 1997 ) [ hep - ph/9702424 ] . z. bern , l. dixon , d.c . dunbar , m. perelstein and j.s . rozowsky , nucl . b530 * , 401 ( 1998 ) [ hep - th/9802162 ] . z. bern , l. dixon and d.a . kosower , phys . lett . * 70 * , 2677 ( 1993 ) [ hep - ph/9302280 ] . w. kilgore , private communication .

the scalar and electromagnetic fields produced by the geodesic and uniformly accelerated discrete charges in de sitter spacetime are constructed by employing the conformal relation between de sitter and minkowski space . special attention is paid to new effects arising in spacetimes which , like de sitter space , have _ spacelike _ conformal infinities . under the presence of particle and event horizons , purely retarded fields ( appropriately defined ) become necessarily singular or even can not be constructed at the future light cones of the at which the sources the universe . we construct smooth ( outside the sources ) fields involving both retarded and advanced effects , and analyze the fields in detail in case of ( i ) scalar monopoles , ( ii ) electromagnetic monopoles , and ( iii ) electromagnetic rigid and geodesic dipoles . pacs : 04.20.-q , 04.40.-b , 98.80.hw , 03.50.-z

gr-qc0107078

the de sitter s 1917 solution of the vacuum einstein equations with a positive cosmological constant @xmath0 , in which freely moving test particles accelerate away from one another , played a crucial role in the acceptance of expanding standard cosmological models at the end of the 1920s @xcite . it reappeared as the basic arena in steady - state cosmology in the 1950s , and it has been resurrected in cosmology again in the context of inflationary theory since the 1980 s @xcite . de sitter spacetime represents the of cosmological models with @xmath1 @xcite . since de sitter space shares with minkowski space the property of being maximally symmetric but has a nonvanishing constant positive curvature and nontrivial global properties , it has been widely used in numerous works studying the effects of curvature in quantum field theory and particle physics ( see , e.g. , ref . @xcite for references ) . recently , its counterpart with a constant negative curvature , anti de sitter space , has received much attention again from quantum field and string theorists ( e.g. ref . @xcite ) . these three maximally symmetric spacetimes of constant curvature also played a most important role in gaining many valuable insights in mathematical relativity . for example , both the particle ( cosmological ) horizons and the event horizons for geodesic observers occur in de sitter spacetime , and the cauchy horizons in anti de sitter space ( e.g. ref . the existence of the past event horizons of the worldlines of sources producing fields on de sitter background is of crucial significance for the structure of the fields . the existence of the particle and event horizons is intimately related to the fact that de sitter spacetime has , in contrast with minkowski spacetime , two _ spacelike _ infinities past and future at which all timelike and null worldlines start and end @xcite . since the pioneering work of penrose @xcite it has been well known that minkowski , de sitter , and anti de sitter spacetimes , being conformally flat , can be represented as parts of the ( conformally flat ) einstein static universe . however , the causal structure of these three spaces is globally very different . the causal character of the conformal boundary @xmath2 to the physical spacetime that represents the endpoints at infinity reached by infinitely extended null geodesics , depends on the sign of @xmath0 . in minkowski space , these are _ null _ hypersurfaces future and past null infinity , @xmath3 and @xmath4 . in de sitter space , both @xmath3 and @xmath4 are _ spacelike _ ; in anti de sitter space the conformal infinity @xmath2 is not the disjoint union of two hypersurfaces , and it is timelike . towards the end of his 1963 les houches lectures @xcite , penrose discusses briefly the zero rest - mass free fields with spin @xmath5 in cosmological ( not necessarily de sitter ) backgrounds . at a given point @xmath6 , not too far from @xmath4 , say , the field can be expressed as an integral over quantities defined on the intersection of the past null cone of @xmath6 and @xmath4 ( ) plus contributions from sources whose worldlines intersect the past null cone . however , the concept of at @xmath4 depends on the position of @xmath6 if @xmath4 is spacelike @xcite . if there should be no incoming radiation at @xmath4 with respect to all , all components of the fields must vanish at @xmath4 . imagine that spacelike @xmath4 is met by the worldlines of discrete sources . then there will be points @xmath6 near @xmath4 whose past null cones will not cross the worldlines see fig . [ fig : penrose ] . the field at @xmath6 should vanish if an incoming field is absent . this , however , is not possible since the part of the field of the sources can not vanish there ( as follows from gauss s law ) . penrose @xcite thus concludes that `` if there is a particle horizon , then purely retarded fields of spin @xmath7 do not exist for general source distributions . '' and advanced fields . ] ( the restriction on @xmath5 follows from the number of arbitrarily specifiable initial data for the field with spin @xmath5 see ref . penrose also emphasized that the result depends on the definition of advanced and retarded fields , and `` the application of the result to actual physical models is not at all clear cut ... . '' this observation was reported in a somewhat more detail at the meeting on `` the nature of time '' @xcite , with an appended discussion ( in which , among others , bondi , feynman , and wheeler participated ) but technically it was not developed futher since 1963 . in a much later monograph penrose and rindler @xcite discuss ( see p. 363 in vol . ii ) the fact that the radiation field is defined when @xmath2 is spacelike than when it is null , but no comments or references are given there on the absence of [ fig : penrose ] one of the purposes of this paper is to study the properties of fields of pointlike sources the de sitter universe across spacelike @xmath4 . we thus provide a specific physical model on which penrose s observation can be demonstrated and analyzed . we assume the sources and their fields to be weak enough so that they do not change the de sitter background . in de sitter space we identify retarded ( advanced ) fields of a source as those which are in general nonvanishing only in the future ( past ) domain of influence of the source . as a consequence , purely retarded ( advanced ) fields have to vanish at the past ( future ) infinity . adopting this definition we shall see that indeed purely retarded fields produced by pointlike sources can not be smooth or even do not exist . we find this general conclusion to be true not only for charges ( monopoles and dipoles ) producing electromagnetic fields @xmath8 but , to some degree , also for scalar fields @xmath9 produced by scalar charges . in general , purely retarded fields of monopoles and dipoles become singular on the past horizons ( ) of the particle s worldlines . a singularity at the particle s creation light cone can be understood similarly to a cauchy horizon instability inside a black hole ( see , e.g. , ref . @xcite ) ; an observer crossing the creation light cone sees an infinitely long history of the source in a finite interval of proper time . in the scalar field case ( not considered by penrose ) no constraints exist and retarded fields can be constructed . however , we shall see that the strength of the retarded field ( the gradient of the field ) of a scalar monopole has a @xmath10-function character on the creation light cone so that , for example , its energy - momentum tensor can not be evaluated there . in the electromagnetic case it is not even possible to construct a purely retarded field of a single monopole one has to allow additional sources on the creation light cone to find a consistent retarded solution vanishing outside the future domain of influence of the sources . in both our somewhat different explanation of the nonexistence of purely retarded fields of general sources , and in penrose s original discussion , the main cause of difficulties is the spacelike character of @xmath4 and the consequential existence of the past horizons , respectively , it was only after we constructed the various types of fields produced by sources on de sitter background and analyzed their behavior when we noticed penrose s general considerations in ref . our original motivation has been to understand fields of accelerated sources , and in particular , the electromagnetic field of uniformly accelerated charges in de sitter spacetime . the question of electromagnetic field and its radiative properties produced by a charge with hyperbolic motion in minkowski spacetime has perhaps been one of the most discussed of classical electrodynamics if not of all classical physics in the 20th century . here let us only notice that the december 2000 issue of annals of physics contains the series of three papers ( covering 80 pp . ) by eriksen and grn @xcite , which study in depth and detail various aspects of `` electrodynamics of hyperbolically accelerated charges '' ; the papers also contain many ( though not all ) references on the subject . the electromagnetic field of a uniformly accelerated charge along the @xmath11 axis , say , is symmetrical not only with respect to the rotations around the axis , but also with respect to the boosts along the axis . now spacetimes with boost - rotation symmetry play an important role in full general relativity ( see , e.g. , ref . @xcite , and references therein ) . they represent the only explicitly known exact solutions of the einstein vacuum field equations , which describe moving accelerated singularities or black holes emitting gravitational waves , and which are asymptotically flat in the sense that they admit global , though not complete , smooth null infinity @xmath4 . their radiative character is best manifested in a nonvanishing bondi s news function , which is an analog of the radiative part of the poynting vector in electrodynamics . the general structure of all vacuum boost - rotation symmetric spacetimes with hypersurface orthogonal killing vectors was analyzed in detail in ref . one of the best known examples is the @xmath12-metric , describing uniformly accelerated black holes attached to conical singularities ( or ) along the axis of symmetry . there exists also the generalization of the @xmath12-metric including a nonvanishing @xmath0 @xcite . it has been used to study the pair creation of black holes @xcite ; its interpretation as uniformly accelerating black holes in a de sitter space has been discussed recently @xcite . however , no general framework is available to analyze the whole class of boost - rotation symmetric spacetimes , which are asymptotically approaching a de sitter ( or anti de sitter ) spacetime as it is given in ref . @xcite for @xmath13 . before developing such a framework in full general relativity , we wish to gain an understanding of fields produced by ( uniformly ) accelerated sources in a de sitter background . this has been our original motivation for this work . although it has been widely known and used in various contexts that there exists a conformal transformation between de sitter and minkowski spacetimes , this fact does not seem to be employed for constructing the fields of specific sources . in the following we make use of this conformal relation to find scalar and electromagnetic fields of the scalar and electric charges in de sitter spacetime . the plan of the paper is as follows . in section [ sc : confinv ] , we will analyze the behavior of scalar and electromagnetic field equations with source terms under general conformal transformations . few points contained here appear to be new , like the behavior of scalar sources in a general , @xmath14-dimensional spacetime , but the main purpose of this section is to review results and introduce notation needed in subsequent parts . in section [ sc : minkdscompact ] , the compactification of minkowski and de sitter spacetimes and their conformal properties are discussed . again , all main ideas are known , especially from works of penrose . but we need the detailed picture of the complete compactification of both spaces and explicit formulas connecting them in various coordinate systems , in order to be able to appropriate motion of the sources and their fields from minkowski into de sitter spacetime . the worldlines of uniformly accelerated particles in de sitter space are defined , found , and their relation to the corresponding worldlines in minkowski space under the conformal mapping is discussed in section [ sc : pointpart ] . in general , a single worldline in minkowski space gets transformed into two worldlines in de sitter space . in section [ sc : symfield ] , by using the conformal transformation of simple boosted spherically symmetric fields of sources in minkowski spacetime , we construct the fields of uniformly accelerated monopole sources in de sitter spacetime . in particular , with both the scalar and electromagnetic fields , we obtain what we call they are analytic everywhere outside the sources and can be written as a linear combination of retarded and advanced fields from both particles . from the symmetric fields we wish to construct purely retarded fields that are nonvanishing only in the future domain of influence of particles worldlines . for the scalar field , this is accomplished in section [ sc : sfretfield ] . we do find the retarded field , but its strength contains a @xmath15-function term located on the particle s past horizon ( creation light cone ) . in section [ sc : emretfield ] , the retarded electromagnetic fields are analyzed for free ( unaccelerated ) monopoles ( subsection [ ssc : freemon ] ) , for ( subsection [ ssc : rigdip ] ) , consisting of two close , uniformly accelerated charges of opposite sign , and for ( subsection [ ssc : geoddip ] ) , made of two free opposite charges moving along geodesics . in subsection [ ssc : emconstr ] the role of the constraints , which electromagnetic fields and charges have to satisfy on any spacelike hypersurface , is emphasized . these constraints in de sitter space with compact spatial slicings require the total charge to be zero . as is well known , there can be no net charge in a closed universe ( see , e.g. , ref . however , we find out that the constraints imply even _ local _ conditions on the charge distribution if @xmath4 is spacelike and purely retarded fields are only admitted . in the case of an unaccelerated electromagnetic monopole , we discover that the solution resembling retarded field represents not only the monopole charge but also a spherical shell of charges moving with the velocity of light along the creation light cone of the monopole . the total charge of the shell is precisely opposite to that of the monopole . retarded fields of both rigid and geodesic dipoles blow up along the creation light cone since , by restricting ourselves to the fields nonvanishing in the future domain of influence , we the field lines produced by the dipoles into their past horizon ( creation light cone ) . we do not discuss the radiative character of the fields obtained . the problem of radiation is not a straightforward issue since the conformal transformation does not map an infinity onto an infinity and , thus , one has to analyze carefully the falloff ( ) of the fields along appropriate null geodesics going to future , respectively , past spacelike infinity . a detailed discussion of the radiative properties of the solutions found here and of some additional fields will be given in a forthcoming publication @xcite . a brief discussion in section [ sc : summary ] concludes the paper . some details concerning coordinate systems on de sitter space are relegated to the appendix . conformal rescaling of metric is given by a common spacetime dependent conformal factor @xmath16 : @xmath17 an equation for a physical field @xmath18 is called conformally invariant if there exists a number @xmath19 such that @xmath20 solves a field equation with metric @xmath21 , if and only if @xmath18 is a solution of the original equation with metric @xmath22 . it is well known ( see , e.g. , ref . @xcite ) that ( i ) the wave equation for a scalar field @xmath23 can be generalized in a conformally invariant way to curved @xmath14-dimensional spacetime geometry by a suitable coupling with the scalar curvature @xmath24 , and ( ii ) the vacuum maxwell s equations are conformally invariant in four dimensions with conformal weight @xmath25 of covariant components ( 2-form ) of electromagnetic field @xmath26 , but they fail to be conformally invariant for dimensions @xmath27 . the behavior of the above equations with sources is not so widely known ( cf . @xcite for the electromagnetic case ) . it is , however , easy to see that the wave equation for the scalar field @xmath23 with the scalar charge source @xmath28 , @xmath29\sf = \sfs\commae\ ] ] where in @xmath14 dimensions @xmath30 , @xmath24 is the scalar curvature , and @xmath31 is the dalembertian constructed from the covariant metric derivative @xmath32 , under the conformal rescaling eq . goes over into the equation of the same form @xmath33\hat\sf = \hat\sfs\commae\ ] ] provided that @xmath34 and @xmath35 and @xmath36 are the metric covariant derivative and scalar curvature associated with the rescaled metric @xmath21 ( see , e.g. , eqs . ( d.1)(d.14 ) in ref . @xcite ) . next , it is easy to demonstrate that in four dimensions maxwell s equations with a source given by a four - current @xmath37 , @xmath38 } = 0 \quad\text{or}\quad \emf_{\alpha\beta}=\stcnx_\alpha \ema_\beta - \stcnx_\beta \ema_\alpha\commae \end{gathered}\ ] ] are conformally invariant if the vector potential does not change , so that @xmath39 and the current behaves as follows : @xmath40 since the levi - civita tensor transforms as @xmath41 the following quantities are conformally invariant : @xmath42 therefore , maxwell s equations with a source can be written using the external derivative as @xmath43 where only conformally invariant quantities appear . the continuity equation for the electromagnetic current is also conformally invariant : @xmath44 thanks to eq . and the conformal property of the four - dimensional volume element @xmath45 : @xmath46 it is interesting to notice , however , that as a consequence of the invariance of the electromagnetic potential under a conformal rescaling , the lorentz gauge condition is not conformally invariant : @xmath47 implies @xmath48 a remarkable property arises in four - dimensional spacetimes : in both the scalar and electromagnetic case the total charge distributed on a three - dimensional spacelike hypersurface is conformally ( pseudo)invariant . this follows form the conformal invariance of spatial charge distributions . denoting @xmath49 a future - oriented unit 1-form normal to the hypersurface @xmath50 , we get @xmath51 the three - dimensional volume element is given by @xmath52 , where three - metric @xmath53 is the restriction of the four - metric @xmath54 to the hypersurface @xmath50 . under the conformal transformation , @xmath55 a charge distribution is defined as a charge density multiplied by this volume element . hence , the scalar charge distribution reads @xmath56 and @xmath57 we see that it is conformal invariant except for a change of sign if the conformal factor @xmath58 is negative . as seen from eq . , this fails to be true for @xmath27 . in the following we consider only the case @xmath59 . the electromagnetic charge distribution is given by @xmath60 again , regarding eqs . , , and , we get @xmath61 thus , the electromagnetic charge is invariant even under conformal transformation with a negative conformal factor @xmath58 . similarly , we define the electric field with the three - dimensional volume element included : @xmath62 which represents the momentum conjugated to the potential @xmath63 ( cf . , e.g. , ref . @xcite ) . with the definition , @xmath64 is conformally invariant . gauss s law simply reads @xmath65 where @xmath66 is three - metric covariant derivative and @xmath67 is a region in @xmath50 . the conformal structure of minkowski and de sitter spacetimes and their conformal relation to the regions of the einstein static universe is well known and has been much used ( see , e.g. , refs . @xcite for basic expositions ) . however , the complete compactified picture of both spaces and their conformal structure do not appear to be described in detail in the literature , although all main ideas are contained in various writings by penrose ( e.g. refs . since we shall need some details in explicit form when analyzing the character of the fields of sources in de sitter spacetime and their relation to their counterparts in minkowski spacetime , we shall now discuss the compactification and conformal properties of these spaces . recall first that flat euclidean plane @xmath68 can be compactified by adding a point at infinity so that the resulting space is a two - sphere @xmath69 with a regular homogenous metric @xmath70 , conformally related to the euclidean metric : @xmath71 where @xmath72 is a constant parameter with the dimension of length , @xmath73 , and @xmath74 . notice that @xmath75 is not regular at @xmath76 , where the conformal factor @xmath77 . the group of conformal transformations of @xmath68 acts on the compact manifold @xmath69 . analogously , one can construct compactified minkowski space @xmath78 ( see ref . @xcite , section 9.2 and references therein ) on which the 15-parameter conformal group acts . one starts with the standard penrose diagram of minkowski space and makes an identification of @xmath79 and @xmath80 by identifying the past and future endpoints of null geodesics as indicated in fig . [ fig : mink ] . the future and past timelike infinities , @xmath81 , and the spatial infinity , @xmath82 , are also identified into one point . the topology of @xmath78 is @xmath83 ( this is not evident from first sight , but see ref . @xcite ) . [ fig : mink ] rescaled minkowski space ( without the identification ) can be drawn in a two - dimensional diagram as a part of the einstein static universe , which is visualized by a cylindrical surface imbedded in @xmath84 see , e.g. , refs . however , in fig . [ fig : ds ] we illustrate the compactified minkowski space @xmath78 by a _ three - dimensional _ diagram as a part of the einstein universe represented by a _ solid _ cylinder in @xmath84 . this is achieved in the following way : in fig . [ fig : mink ] the minkowski spacetime ( with one dimension suppressed ) is illustrated as a region bounded by two cones joined base to base . now we take a two - dimensional cut @xmath85 and we compactify it by dividing it into four regions , i iv , as indicated in fig . [ fig : mink ] . we cut out regions iii and iv and place them regions ii and i so that they are joined along their corresponding null boundaries ( e.g. , points @xmath86 , @xmath87 and @xmath88 , @xmath89 become identical ) . now the segment @xmath90 has to be identified with @xmath91 and @xmath92 with @xmath93 they correspond to a single segment in fig . [ fig : mink ] . similarly , boundaries @xmath94 and @xmath95 are identified , and as a result a compact manifold is formed . consider then all posible cuts @xmath85 , i.e. , @xmath85 around the @xmath96 , and make the same identifications as we just described . now all boundary lines as @xmath94 and @xmath95 have to be identified ( notice that all these points were on the segment @xmath97 in fig . [ fig : mink ] ) . the resulting four - dimensional compact manifold is represented in the three - dimensional fig . [ fig : ds ] . from the construction described , it follows that the top and bottom bases of the solid cylinder are identified and each of the circles on the cylindrical surface , as , e.g. , @xmath98 , should be considered as a single point . the inside these circles are thus two - spheres , i.e. , without suppressing one dimension three - spheres in @xmath78 . [ fig : ds ] now it is important to realize that fig . [ fig : ds ] can be understood as a part of the einstein static cylinder , which also represents the _ compactified de sitter space_. in de sitter space two bases are future and past spacelike infinities . they are not usually identified in the standard two - dimensional penrose diagram of de sitter spacetime ( see , e.g. , ref . @xcite ) , as @xmath80 and @xmath79 are not identified in the standard penrose diagram of minkowski space . manifold @xmath78 represents the compactification of both minkowski and de sitter space . similarly as @xmath69 , representing the compactification of @xmath68 , can be equipped with a regular metric @xmath99 mentioned above , @xmath78 can be equipped with the regular metric @xmath100 where dimensionless coordinates @xmath101 , spacelike hypersufaces @xmath102 , and @xmath103 are identified by means of null geodesics , @xmath104 , and the constant @xmath72 has dimension of length . the metric is the well known metric of the einstein universe , in which case @xmath105 where @xmath0 is the cosmological constant . in order to see this explicitly , write the minkowski metric in standard spherical coordinates , @xmath106 and introduce coordinates @xmath101 by @xmath107 inversely @xmath108 so that @xmath109 let us notice that by requiring the ranges @xmath101 , we fix the branch of @xmath110 in eq . . further , observe that for @xmath111 , relations imply negative @xmath112 we shall return to this point in a moment . in the case of de sitter space with the metric @xmath113 we put @xmath114 or @xmath115 so that @xmath116 in this way we obtain explicit forms of the conformal rescaling of both spaces into the metric of the einstein universe : @xmath117 where @xmath118 is given by eq . . as in the simple case of conformal relation of @xmath68 to @xmath69 , the conformal factors @xmath119 and @xmath120 vanish at infinities of minkowski , respectively , de sitter space . as a consequence of eqs . we also find the conformal relation between minkowski and de sitter space : @xmath121 where @xmath119 , @xmath120 are given by eqs . and . the conformal factor @xmath58 has the simplest form when expressed in terms of the minkowski time @xmath122 : @xmath123 the conformal transformation is not regular at the infinity of minkowski , respectively , de sitter space because @xmath58 diverges , respectively , vanishes there . we shall return to this point at the end of this section . first , however , we have to describe the coordinate systems employed in relating particular regions i iv in figs . [ fig : mink ] and [ fig : ds ] . relations and can be used automatically in region i only . in other regions , ranges of coordinates have to be specified more carefully . in the following we always require @xmath124 . then , if @xmath125 , we find that relations and imply negative @xmath112 in region iv . also , if we consider events with @xmath126 , @xmath127 ( region iii ) , we notice that as a consequence of eqs . and with @xmath124 we get @xmath128 . relations and can be made meaningful in all regions i iv if we allow negative @xmath112 , @xmath129 and adopt the following convention : at a fixed value of time coordinate @xmath122 , respectively , @xmath130 , the points symmetrical with respect to the origin of spherical coordinates have opposite signs of the radial coordinate , i.e. , points with given @xmath131 , respectively @xmath132 , are identical with @xmath133 , respectively @xmath134 . the way in which regions i iv are covered by the particular ranges of coordinates is explicitly illustrated in figs . [ fig : coord](a)[fig : coord](c ) in the appendix , where our convention is described in more detail . in the appendix , various useful coordinate systems in de sitter space are given . first , we shall frequently employ coordinates @xmath132 which are simply related ( by eqs . and ) to the standard coordinates @xmath135 , @xmath136 , @xmath137 , @xmath138 covering nicely the whole de sitter hyperboloid . next , relations and can be viewed as the definition of another coordinate system @xmath131 on de sitter space ( with the metric being given by eq . ) . let us remind that for fixed @xmath137 , @xmath139 values @xmath140 ( commonly assumed in de sitter space ) correspond to @xmath127 for @xmath141 ( region i ) and to @xmath142 for @xmath111 ( region iv ) . further , one frequently uses static coordinates @xmath143 , @xmath144 associated with the static killing vector of de sitter spacetime eqs . in the appendix . finally , it will be useful to introduce the null coordinates ( cf . @xmath145 when employing null coordinates we shall consider only @xmath140 ( @xmath146 , @xmath147 would if @xmath128 ) ; the ranges of @xmath148 and @xmath149 are thus given by the choice @xmath101 . this leaves the standard ( minkowski ) meaning of @xmath150 , @xmath151 in region i ; however , @xmath150 and @xmath151 exchange their usual ( minkowski ) role in region iv ( @xmath152 ) because here @xmath142 . with this choice , the coordinates @xmath146 , @xmath147 and @xmath150 cover de sitter space continuously , in particular the horizon @xmath153 ( @xmath154 on this horizon see fig . [ fig : coord](e ) ) . from eqs . and we get simple relations @xmath155 which explicitly verify that local null cones ( local causal structure ) are unchanged under conformal mapping . nevertheless , it is well known that the global causal structure of the minkowski and de sitter space is different . this is reflected in the fact that , as mentioned above , the conformal transformation between the two spaces is not regular everywhere . in particular , relation shows that points at null infinity with @xmath154 in minkowski space go over into regular points with @xmath156 in de sitter space , whereas spacelike hypersurface @xmath157 goes into spacelike infinities @xmath158 in de sitter space . in the next sections , when we shall generate solutions for the scalar and electromagnetic fields for given sources in de sitter space by employing the conformal transformation from minkowski space , we have to check the behavior of the new solutions at points where the transformation is not regular . before turning to the construction of the fields produced by specific sources , let us emphasize that in all the following expressions for fields in de sitter spacetime only positive @xmath129 can be considered . however , the results contained in sections [ sc : pointpart ] and [ sc : symfield ] are valid also for @xmath128 provided that the convention described above is used . , we require @xmath140 . the right - hand sides of expressions , , , and would have to be multiplied by a factor @xmath159 to be also valid for @xmath128 . similar changes would also be necessary in other equations but these contain null coordinates that have not been defined for @xmath128 . ] in this section we study the correspondence of the worldlines of uniformly accelerated particles under the conformal mapping and between minkowski and de sitter spacetimes . let a particle have four - velocity @xmath160 , @xmath161 , so that its acceleration is @xmath162 , @xmath163 . we say that the particle is uniformly accelerated if the projection of @xmath164 into the three - surface orthogonal to @xmath160 vanishes : @xmath165 here the projection tensor @xmath166 and @xmath167 . multiplying eq . by @xmath168 , we get @xmath169 so that @xmath170 this definition of uniform acceleration goes over into the standard definition used in minkowski space @xcite . it implies that the components of a particle s acceleration in its instantaneous rest frames remain constant . of course , as a special case , a particle may have zero acceleration when it moves along the geodesic . consider a particle moving with a constant velocity @xmath171 along the @xmath11 axis ( @xmath172 , @xmath173 ) of the inertial frame in minkowski spacetime with coordinates @xmath131 so that it passes through @xmath174 at @xmath175 . transformations , map its worldline into _ two _ worldlines in de sitter spacetime , given in terms of parameter @xmath176 , its proper time in minkowski space , or in terms of @xmath177 , its proper time in de sitter space , as follows : @xmath178 in these expressions , the @xmath110 has values in for @xmath179 . ] @xmath180 ; @xmath181 , and @xmath182 for the worldline starting and ending with @xmath183 ( denoted by @xmath184 in fig . [ fig : worldlines ] ; plus sign in the last equation in ) , whereas @xmath185 for the second worldline , starting and ending with @xmath186 ( denoted by @xmath187 in fig . [ fig : worldlines ] ; minus sign in ) . one thus gets two worldlines in de sitter space from two of one worldline in minkowski space . [ fig : worldlines ] these two worldlines are uniformly accelerated with the constant magnitude of the acceleration equal ( up to the sign ) to @xmath188 an intuitive understanding of the acceleration is gained if we introduce standard _ static _ coordinates @xmath189 in de sitter space ( see the appendix ) . the two worldlines described by eq . in coordinates @xmath190 are in the static coordinates simply given by @xmath191 . ( as seen from eq . , for a given @xmath130 , @xmath129 , the same @xmath144 corresponds to @xmath129 and @xmath192 . ) owing to the caused by the presence of a cosmological constant , fundamental geodesic observers with fixed @xmath129 ( i.e. , fixed @xmath136 ) are one from the other in proportion to their distance . their initial implosion starting at @xmath102 ( @xmath193 ) is stopped at @xmath194 ( @xmath195 at the of the de sitter hyperboloid ) and changes into expansion . a particle having constant @xmath196 , thus a constant proper distance from an observer at @xmath197 ( or at @xmath186 , @xmath198 ) , has to be accelerated towards that observer . the acceleration of particle @xmath184 points towards the observer at @xmath183 , whereas that of particle @xmath187 points towards the observer at @xmath186 ( fig . [ fig : worldlines ] ) . notice that the two uniformly accelerated worldlines are causally disconnected ; no retarded or advanced effects from the particle @xmath184 can reach the particle @xmath187 and vice versa . this is analogous to two particles symmetrically located along opposite parts of say the @xmath11 axis and uniformly accelerated in opposite directions in minkowski space . the worldlines of uniformly accelerated particles in minkowski space are the orbits of the boost killing vector . analogously , in de sitter space the killing vector @xmath199 also has the character of a boost . another type of simple worldline in de sitter space arises from transforming the worldlines of a particle at rest at @xmath200 ( @xmath201 ) , @xmath173 , @xmath172 in minkowski space . it transforms to two worldlines in de sitter space given by @xmath202 with @xmath124 , and @xmath203 for @xmath204 and @xmath205 for @xmath206 . thus , a geodesic in minkowski space goes over into two geodesics in de sitter space . in fig . [ fig : worldlines ] these are the worldlines @xmath207 and @xmath208 . as the last example , let us just mention that the worldlines of two particles uniformly accelerated in minkowski space get transformed into two geodesics @xmath209 , @xmath210 in de sitter space . two uniformly accelerated particles described by worldlines were obtained by the conformal transformation from the worldline of a particle moving with a uniform velocity @xmath211 in minkowski space . hence , their fields can be constructed by the conformal transformation of a simple boosted spherically symmetric field . in the case of scalar field , eqs . , , and then lead to the field @xmath212 where @xmath213 is spherical coordinate in an inertial frame in which the particle is at rest at the origin . as emphasized earlier , we have to examine the field at the null hypersurface @xmath153 , where the conformal transformation fails to be regular . we find that field is indeed not smooth there . the limit of @xmath23 as @xmath153 is approached from the region @xmath214 differs from the limit from @xmath215 ; @xmath23 has a jump at @xmath153 , although , as can be checked by a direct calculation , this discontinuous field satisfies the scalar wave equation . however , a field analytic everywhere outside the sources can be obtained by an analytic continuation of the field from the domain @xmath214 to the domain @xmath215 . we discover that the new field in @xmath215 differs from eq . just by a sign . therefore it simply corresponds to the charge of the particle in @xmath215 , which is opposite to that implied by conformal transformation . it is easy to see that , due to the conformal transformation , the sign of the charge on the worldline with @xmath215 is opposite to the original charge @xmath216 because the conformal factor @xmath217 for @xmath215 . hence , the field which is analytic represents the field of two uniformly accelerated particles with the _ same _ scalar charge @xmath216 , which move along two worldlines given by . in section [ sc : sfretfield ] we shall see that this field can be written as a linear combination of retarded and advanced fields from both particles . we call it the . regarding eq . , in which @xmath213 is first expressed in terms of the original minkowski coordinates @xmath131 , and then using the transformation , we find the field as a function of @xmath132 : @xmath218 where @xmath219 is the static radial de sitter coordinate ( see the appendix ) . as could have been anticipated , the field is static in the static coordinates since the accelerated particles are at rest at @xmath196 , @xmath220 . ( recall that we need two sets of such coordinates to cover both worldlines but the coordinate @xmath144 is well defined in the whole de sitter spacetime cf . the appendix . ) however , it is dynamical in the coordinates @xmath132 , or in the standard coordinates @xmath221 , covering in contrast to the static coordinates the whole de sitter spacetime . in order to construct the electromagnetic field produced by uniformly accelerated particles in de sitter spacetime , we start , analogously to the scalar field case , from the boosted coulomb field in minkowski space . the potential 1-form is thus simply @xmath222 where the prime again denotes the coordinates in an inertial frame in which the particle is at rest . since electromagnetic field described by its covariant component is conformally invariant , the field _ is _ automatically a solution of maxwell s equations in de sitter space . however , like in the scalar field case , we have to examine its character at @xmath153 . we discover that the potential does _ not _ , in fact , solve maxwell s equations there ( in contrast to the scalar field , which is discontinuous on @xmath153 , but satisfies the scalar wave equation ) . this result can be understood when we realize that by the conformal transformation the sign of the electric charge in contrast to the scalar charge does _ not _ change at the worldline with @xmath215 so that the total electric charge is @xmath223 . a nonzero total charge in de sitter spacetime , however , violates the constraint , as we shall see in subsection [ sc : emretfield].[ssc : emconstr ] . in fact , it is well known that in a closed universe the total electric charge must be zero due to the gauss s law ( e.g. ref . @xcite ) . as with the scalar field , we still can construct a field smooth everywhere outside the sources by analytic continuation of the field obtained in the region @xmath214 across @xmath153 into whole spacetime . similar to the scalar case , the resulting field in @xmath215 corresponds just to the opposite charge , so that now , in the electromagnetic case , the total charge is indeed zero . the electromagnetic field can be written as a combination of retarded and advanced fields from both charges , as will be shown in subsection [ sc : emretfield].[ssc : freemon ] . the potential describing this has a simple form in the static coordinates : @xmath224 where is taken here as in eq . . ] @xmath225 as noticed in section [ sc : confinv ] , the lorentz gauge condition is not conformally invariant , so that the potential need not satisfy the condition , although the original coulomb field does . expressing the static radial coordinate as @xmath219 , and similarly @xmath143 ( see the appendix ) , we can find the potential in global coordinates @xmath132 , respectively , @xmath221 . since the resulting form is not simple , we do not write it here , but we give the electromagnetic field explicitly in both the static and global coordinates . in static coordinates it reads @xmath226 where @xmath227 is given by eq . . in @xmath132 coordinates we explicitly find @xmath228 summarizing , the field represents the time - dependent electromagnetic field of two particles with charges @xmath229 , uniformly accelerated along the worldlines with accelerations @xmath230 . the field is analytic everywhere outside the charges . in the static coordinates the charges are at rest at @xmath196 and their static field is given by eq . . the symmetric scalar field solution , representing two uniformly accelerated scalar charges , is non - vanishing in the whole de sitter spacetime . as mentioned before , and will be proved at the end of this section ( see eqs . and ) , this field is a combination of retarded and advanced effects from both charges . a retarded field of a point particle should in general be nonzero only in the future domain of influence of a particle s worldline , i.e. , at those points from which past causal curves exist which intersect the worldline . hence , the retarded field of the uniformly accelerated charge , which starts and ends at @xmath183 ( see fig . [ fig : worldlines ] ) , should be nonvanishing only at @xmath231 . it is natural to try to construct such a field by restricting the symmetric field to this region , i.e. , to ask whether the field @xmath232 where @xmath233 is the usual heaviside step function , is a solution of the field equation . the field does , of course , satisfy the scalar field wave equation eq . at @xmath234 since @xmath235 does , and also at @xmath236 since @xmath237 is a solution of outside a source . thus we have to examine the field only at @xmath238 , i.e. , at of the particle s worldline , also referred to as the past event horizon of the worldline @xcite . the field strength 1-form implied by eq . becomes @xmath239 an explicit calculation shows that @xmath240 therefore , the conformally invariant scalar wave equation ( with @xmath241 ) has the form @xmath242\sf_\ret = \bigl(\bigl[\dalamb-\frac16\sccr\bigr]\sf_\sym\bigr ) \,\stepfc(\tlu ) = \sfs_\sym\,\stepfc(\tlu ) = \sfs_{\mon1}\commae\ ] ] where @xmath243 denotes the monopole scalar charge starting and ending at @xmath183 . hence , we proved that the field , where @xmath235 is given by eq . , satisfies the field eq . everywhere , including the past event horizon of the particle . analogously , we can make sure that @xmath244 has its support in the future domain of influence of the monopole particle @xmath187 , starting and ending at @xmath186 , and is thus the advanced field of source @xmath245 . from the results above it is not difficult to conclude that the symmetric field can be interpreted as arising from the combinations of retarded and advanced potentials due to both particles @xmath184 and @xmath187 , in which the potentials due to one particle can be taken with arbitrary weights , and the weights due the other particle then determined by @xmath246 where @xmath247 is an arbitrary constant factor . in particular , choosing @xmath248 , the field @xmath249 is the symmetric field from both particles . this freedom in the interpretation is exactly the same as with two uniformly accelerated scalar particles in minkowski spacetime ( see ref . @xcite , section iv b ) . a remarkable property of the retarded field is that the field strength has a term proportional to @xmath250 , i.e. , it is singular at the past horizon . since the energy - momentum tensor of the scalar field is quadratic in the field strength , it can not be evaluated at @xmath238 . the at the can be understood on physical grounds similarly as the instability of cauchy horizons inside black holes ( e.g. , ref . @xcite ) ; an observer crossing the pulse along a timelike worldline will see an infinitely long history of the source within a finite proper time . the character of the shock is given by the pointlike nature of the source . if , for example , a scalar charge has typical extension @xmath251 at @xmath194 , i.e. , at the moment of the minimal size of the de sitter universe ( @xmath195 ) , and the extension of the charge in @xmath129 coordinate would be roughly the same at the @xmath252 , the corresponding shock would be smoothed around @xmath238 with a width @xmath253 . however , the _ proper _ extension of the charge at @xmath252 would be infinite in that case . let us note that the retarded field could also be computed by means of the retarded green s function . in our case of the conformally invariant equation for a scalar field , the retarded green s function in de sitter space is localized on the future null cone , as it is in the minkowski space . ( it is interesting to note that in the case of a minimal , or more general coupling , the scalar field does not vanish inside the null cone . ) thanks to this property we can understand a in the field on the creation light cone : the creation light cone is precisely the future light cone of the point at which the source the spacetime , i.e. , it is the boundary of a domain where we can obtain a contribution from the retarded green s function integrated over sources . in this section we shall analyze the electromagnetic fields of free or accelerated charges with monopole and also with a dipole structure . we shall pay attention to the constraints which the electromagnetic field , in contrast to the scalar field , has to satisfy on any spacelike hypersurface . let us start with an unaccelerated monopole at rest at the origin of both coordinate systems used , i.e. , at @xmath197 . with @xmath254 , the potential and the field simplify to @xmath255 and @xmath256 let us restrict the potential to the and its interior by defining @xmath257 the field in null coordinates @xmath146 , @xmath147 then reads @xmath258 so that the left - hand side of maxwell s equations becomes @xmath259{\tlu } + 2 ( 1-\cos\tlv)\cvect[\alpha]{\tlv } \bigr)\ ; \deltafc(\tlu ) \\ & - \frac{\emq}{4\pi\alpha^4 } \cvect[\alpha]{\tlv}\ ; \deltafc^\prime(\tlu ) \period \end{split}\ ] ] here @xmath260 is the current produced by the charge at @xmath183 . additional terms on the right - hand side of eq . , localized on the null hypersurface @xmath238 , clearly show that the restricted field does not correspond to a single point source . the terms of this type did not arise in case of the scalar field discussed in the previous section . we can try to add a field localized on @xmath238 , which would cancel the additional terms . although we shall see in the following subsection that this can not be achieved , it is instructive to add , for example , the field @xmath261 which cancels the second term on the right hand side of the field . thus denoting , @xmath262 we find that with @xmath263 , maxwell s equations become @xmath264{\tlv}\period\ ] ] hence , the field @xmath265 does not represent only the unaccelerated monopole charge but also a spherical shell of charges moving outwards from the monopole with the velocity of light along the @xmath238 . the total charge of the shell is precisely opposite to the monopole charge so that the total charge of the system is zero . we shall return to this point in the following subsection , now let us add yet two comments . it is interesting that , in contrast to the scalar field strength , the electromagnetic field @xmath263 is not singular at @xmath238 . apparently , the effects of the monopole and the charged shell compensate along @xmath238 in such a way that even the energy - momentum tensor of the field is finite there . second , if the field @xmath265 , corresponding to the retarded field from the charge @xmath266 at @xmath183 and the outgoing charged shell is superposed with the analogous field corresponding to the advanced field from the charge @xmath267 at @xmath186 and the ingoing charged shell , the fields corresponding to charged shells localized on @xmath238 cancel each other and the field @xmath268 ( eq . ) with @xmath254 is obtained . the same compensation occurs for two uniformly accelerated charges ( @xmath269 ) considered in section [ sc : symfield ] , as it follows from the symmetry . therefore , the symmetric electromagnetic field can be interpreted as arising from the combinations of retarded and advanced potentials due to both charges @xmath184 and @xmath187 in the same way as was the case for the symmetric scalar field ; relation remains true if @xmath23 s are replaced by @xmath270 s . the appearence of a shell with the total charge exactly opposite to that of the monopole discussed above has deeper reasons . it is a consequence of the constraints , which any electromagnetic field , and charge distributions have to satisfy on a spatial hypersurface and of the fact that spatial hypersurfaces , including past and future infinities , in de sitter spacetime , are compact . integrating the constraint equation @xmath271 ( see eq . for the definition of @xmath64 ) over a compact cauchy hypersurface @xmath50 , we convert the integral of the divergence on the left - hand side to the integral over a which , however , does not exist for a compact @xmath50 . hence , as it is well known , the total charge on a compact hypersurface ( in any spacetime , not only de sitter ) must vanish : @xmath272 therefore , the field @xmath263 constructed in represents the monopole field plus the additional source localized on the past horizon of the monopole that leads to the total zero charge . this enables the monopole electric field lines to end on this horizon . a stronger , even _ condition on the charge distribution in de sitter spacetime ( or , indeed , in any spacetime with spacelike past infinity @xmath4 ) arises if we admit purely retarded fields only . here we define as those that _ vanish at @xmath4_. then , however , the constraint directly implies that at @xmath4 the charge distribution vanishes : @xmath273 in the les houches lectures in 1963 penrose @xcite gave a general argument showing that if @xmath4 is spacelike and the charges meet it in a discrete set of points , then there will be incosistencies if an incoming field is absent ( see in particular fig . 16 in ref . @xcite , cf . [ fig : penrose ] , see also ref . penrose also remarked that an alternative definition of advanced and retarded fields might be found that leads to different results , and that the application of the result to physical models is not clear . we found nothing more on this problem in the literature since penrose s observation in 1963 . our work appears to give the first explicit model in which this issue can be analyzed . as the first example of a simple source satisfying both the constraint and the local condition required by the absence of incoming radiation , we consider a rigid dipole . to construct an elementary rigid dipole , we place point charges @xmath274 and @xmath275 on the worldlines with @xmath276 and @xmath277 , fixed in the static coordinates , and take the limit @xmath278 . the constant dipole moment is thus given by @xmath279 . the resulting symmetric field can easily be deduced from the symmetric fields of electric monopoles : . ] @xmath280 as in section [ sc : symfield ] , @xmath254 corresponds to two worldlines and the symmetric solution describes the fields of _ two _ dipoles , one at rest at @xmath183 , the other at @xmath186 . to construct a purely retarded field of the dipole at @xmath183 , we first restrict the symmetric field to the inside of the of the dipole , analogously as we did with the monopole charge in subsection [ sc : emretfield].[ssc : freemon ] . writing ( cf . ) @xmath281 we now get @xmath282 so that no additional term like that in eq . arises ; however , expressing the left - hand side of maxwell s equations as in eq . , we find @xmath283{\tlv}\period\ ] ] here @xmath284 is the current corresponding to the rigid dipole at @xmath183 . similarly to the case of the monopole , there is an additional term on the right - hand side of eq . , localized on the null cone @xmath238 , indicating that the field represents , in addition to the dipole , an additional source located on the horizon @xmath238 . in contrast to the monopole case , however , this source can be compensated by adding to the potential the term @xmath285 in this way we finally obtain the purely retarded field of the dipole with dipole moment @xmath286 , located at @xmath197 , in the form @xmath287 where @xmath288 , @xmath289 , @xmath290 are given by eqs . , , and . it is easy to check that @xmath291 is satisfied . regarding the retarded field , we see that it is , in contrast to the symmetric field , singular on the ( past event horizon ) @xmath238 . this is not surprising ; in order to obtain a purely retarded field , we the field lines produced by the dipole into the horizon . next we consider dipoles consisting of two free charges moving along the geodesic @xmath292 , @xmath293 , @xmath294 in the minkowski space which , as discussed at the end of section [ sc : pointpart ] ( see eq . and the worldlines @xmath207 , @xmath208 in fig . [ fig : worldlines ] ) transforms into geodesics of the conformally related de sitter space . we call two free opposite charges a . we start again by constructing first the symmetric field . two elementary geodesic dipoles located at @xmath183 and @xmath186 can be obtained by placing point charges @xmath295 on the worldlines @xmath296 , @xmath277 and taking the limit @xmath278 . as with monopoles , to find the symmetric field we conformally transform the field of a standard rigid minkowski dipole ( bewaring the signs for @xmath297 and @xmath126 so that the field is analytic outside the sources ) . the dipole moment in minkowski space @xmath298 is constant , the corresponding value in de sitter spacetime , however , depends now on time : @xmath299 as it follows from the transformation relations and . in terms of @xmath300 we find the symmetric field of geodesic dipoles located at @xmath183 and @xmath186 to read @xmath301 proceeding as with the rigid dipole , we generate the retarded field of only one geodesic dipole at @xmath183 by restricting the symmetric field by the step function : @xmath302 since @xmath303{\vartheta}+ \cos\vartheta\sin\tlv\,\cvect[\alpha]{\tlv}\bigr)\\ & \qquad\qquad-\frac{2 \emf}{4\pi\alpha^6}\ , \cos\vartheta\,(1-\cos\tlv)\ ; \deltafc^\prime(\tlu)\,\cvect[\alpha]{\tlv}\commae \end{split}\ ] ] where @xmath304 , an additional source is present at the horizon @xmath238 ; it can be compensated by adding an additional field with potential , for example , given by @xmath305 the total retarded field of the geodesic dipole is then given as follows : @xmath306 it indeed satisfies maxwell s equations : @xmath307 by using the conformal relation between de sitter and minkowski space , we constructed various types of fields produced by scalar and electromagnetic charges moving with a uniform ( possibly zero ) acceleration in de sitter background . one of our main conclusions has been the explicit confirmation and elucidation of the observation by penrose that in spacetimes with a spacelike past infinity , which implies the existence of a particle horizon , purely retarded fields do not exist for general source distributions . in subsection [ sc : emretfield].[ssc : geoddip ] we constructed the field of the geodesic dipole which , at first sight , appears as being a purely retarded field . can thus purely retarded fields of , for example , two opposite monopoles , be constructed by distributing the elementary geodesic dipoles along a segment @xmath308 , @xmath309 , so that neighboring opposite charges cancel out , and only two monopoles , at @xmath183 and @xmath310 , remain ? then , however , the local constraint on the charge distribution would clearly be violated ! the solution of this paradox is in the fact that the field _ is not a purely _ retarded field ; if we calculate ( by using distributions ) the initial data leading to the field strength , we discover that the electric field strength contains terms proportional to the @xmath15-function at @xmath102 , @xmath183 so that the field does not vanish at @xmath4 . hence the field of the dipoles distributed along the segment does _ not _ vanish at @xmath4 and thus can not be considered as a purely retarded field of the two monopoles at @xmath183 and @xmath311 . we thus arrive at the conclusion that a purely retarded field of even two opposite charges ( so that the global constraint of a zero net charge _ is _ satisfied ) can not be constructed in the de sitter spacetime unless the charges the universe at the same point at @xmath4 and the _ local _ constraint is satisfied . if we allow nonvanishing initial data at @xmath4 , the resulting fields can hardly be considered as by applying the superposition principle we can consider a greater number of sources , and our arguments of the insufficiency of purely retarded fields ( based on the global and local constraints ) can clearly be generalized also to infinitely many discrete sources . our discussion of the fields of dipoles indicates that interesting situations may arise . the absence of purely retarded fields in de sitter spacetime or , in fact , in any spacetime with space - like @xmath4 is , of course , to be expected to occur for higher - spin fields as well . in particular , there has been much interest in the primordial gravitational radiation . since de sitter spacetime is a standard arena for inflationary models , the generation of gravitational waves by ( test ) sources in de sitter spacetime has been studied in recent literature ( see ref . @xcite , and references therein ) . we plan to analyze the linearized gravity on de sitter background in the light of the results described above . the fact that particles are expected to have only positive gravitational mass will apparently prevent any purely retarded field to exist . first , however , we shall present the generalization of the well known born s solution for uniformly accelerated charges in minkowski spacetime to the generalized born solution representing uniformly accelerated charges in de sitter spacetime @xcite . as it was demonstrated in the present - paper , to be is quite a different matter than to be . the authors would like to thank the albert - einstein institute , golm , where part of this work was done , for its kind hospitality . j.b . also acknowledges the sackler fellowship at the institute of astronomy in cambridge , where this work was finished . we were supported in part by grants nos . gar 202/99/026 and gauk 141/2000 of the czech republic and charles university . we describe here coordinate systems employed in the main text . there exists extensive literature on various useful coordinates in de sitter spacetime for standard reference see ref . @xcite , for more recent reviews , containing also many references , see , for example , refs . @xcite . de sitter spacetime has topology @xmath312 . it is best visualized as the four - dimensional hyperboloid imbedded in flat five - dimensional minkowski space ; it is the homogeneous space of constant curvature spherically symmetric about any point . the following coordinate systems are constructed around any fixed ( though arbitrary ) point . two of the coordinates are just standard spherical angular coordinates @xmath313 and @xmath314 on the orbits of the killing vectors of spherical symmetry with the homogeneous metric which is @xmath315 in the transformations considered below , these coordinates remain unchanged and thus are not written down . the axis is fixed by @xmath277 . next , we have to introduce time and radial coordinates labeling the orbits of spherical symmetry . the coordinate systems defined below differ only in these time and radial coordinates and , therefore , we essentially work with a two - dimensional system . radial coordinates commonly take positive values and coordinate systems are degenerate for their value zero . as discussed in section [ sc : minkdscompact ] below eq . , the relations between various regions i iv of minkowski and de sitter spaces are conveniently described if we allow radial coordinates to attain negative values . we adopt the convention that at a fixed time the points symmetrical with respect to the origin of spherical coordinates have the opposite sign of the radial coordinate . hence , the points with @xmath131 are identical with @xmath133 , analogously for the coordinates @xmath132 introduced below . let us characterize this convention in more detail . imagine that on our manifold ( either compactified minkowski space or de sitter space ) two coordinate maps @xmath316 and @xmath317 are introduced , which for any fixed point are connected by the relations @xmath318 where @xmath319 and @xmath320 . _ both _ maps cover the _ whole _ spacetime manifold . now we consider a two - dimensional cut given by @xmath321 , @xmath322 , with @xmath323 , @xmath324 changing . this represents the history of a half - line @xmath325 in fig . [ fig : mink ] , illustrated by regions i and iii , say . the history of a half - line @xmath326 , obtained by the smooth extension of @xmath325 through the origin , illustrated by regions ii and iv , is covered by the coordinates @xmath316 with the _ same _ angular coordinates @xmath327 , @xmath328 but @xmath329 ( which , in our convention , is identical to @xmath330 , @xmath331 , @xmath332 ) . in exactly the same way we may introduce coordinates @xmath333 , with @xmath334 covering regions i and iv in fig . [ fig : ds ] , whereas @xmath335 cover regions ii and iii . in fig . [ fig : coord ] , the two - dimensional cuts ( with angular coordinates fixed ) through de sitter space ( or through the compactified space @xmath78 ) are illustrated . both regions covered by @xmath334 and by @xmath335 are included ( the right and left parts of the diagrams ) . however , in the figures , as well as in the main text , we do not write down subscripts `` @xmath336 '' and `` @xmath337 '' at the coordinates , since the sign of the radial coordinate specifies which map is used . de sitter spacetime can be covered by coordinates @xmath338 , @xmath136 ( @xmath339 , @xmath340 ) in which the metric has the form @xcite : @xmath341 it is useful to rescale the coordinate @xmath338 to obtain coordinates @xmath342 : @xmath343 in these coordinates , de sitter space is explicitly seen to be conformal to the part of the einstein static universe . coordinate lines are drawn in fig . [ fig : coord](b ) . another coordinate system used in our work are inertial coordinates @xmath122 , @xmath112 of conformally related minkowski space . we call them coordinates : @xmath344 coordinate lines are drawn in fig . [ fig : coord](c ) . commonly used are coordinates @xmath143 , @xmath144 , related to the time - like killing vector @xmath199 of de sitter spacetime : @xmath345 as it is well - known , these coordinates do not cover the whole spacetime but only the domain with @xmath141 and @xmath346 . the boundary of this domain is the killing horizon . the coordinate @xmath144 can be extended smoothly to the whole spacetime but it is not unique globally . it is also useful to rescale coordinate @xmath144 to obtain the coordinates @xmath347 , @xmath348 : @xmath349 coordinate lines for the static coordinates are drawn in fig . [ fig : coord](d ) . finally , three sets of coordinates @xmath350 , @xmath351 , and @xmath352 are defined by @xmath353 from here we find @xmath354 the metric in these coordinates reads @xmath355 the corresponding coordinate lines are illustrated in fig . [ fig : coord](e ) . k. maeda , cosmic no - hair conjecture , in _ proceedings of the fifth m. grossman meeting on general relativity , perth , australia , 1988 _ , edited by d. g. blair , m. j. buckingham , and r. ruffini ( world scientific , singapore , 1989 ) . j. bik , selected solutions of einstein s field equations : their role in general relativity and astrophysics , in _ einstein s field equations and their physical implications _ , edited by b. g. schmidt , lecture notes in physics vol . 540 , ( springer , berlin , 2000 ) , pp . 1126 , e - print : gr - qc/0004016 .

schlesinger transformations are algebraic transformations of a fuchsian system that preserve its monodromy representation and act on the characteristic indices of the system by integral shifts . one of the important reasons to study such transformations is the relationship between schlesinger transformations and discrete painlev equations ; this is also the main theme behind our work . in this paper we show how to write an elementary schlesinger transformation as a discrete hamiltonian system w.r.t . the standard symplectic structure on the space of fuchsian systems . we then show how schlesinger transformations reduce to discrete painlev equations by considering two explicit examples , d- ( or difference painlev v ) and d- .

1302.2972

in the theory of ordinary linear differential equations on a complex domain , and in particular in the theory of fuchsian systems , an important characteristic of the equation is its configuration of singularities and the characteristic indices at these singular points . associated to this data is the notion of the monodromy representation of the equation . roughly speaking , this representation describes how the fundamental solution matrix of the equation changes under analytic continuation along the closed paths around the singular points and it gives a significant insight into the global behavior of solutions of the equation . the fruitful idea of deforming the equation by moving the location of critical points into different configurations , or by changing the characteristic indices , without changing its monodromy representation , goes back to b. riemann , but the actual foundations of this theory of _ isomonodromic deformations _ in the fuchsian case were laid down in the works of r. fuchs @xcite , l. schlesinger @xcite , and r. garnier @xcite . an extension of the theory to the non - fuchsian case was done relatively recently in the series of papers by m. jimbo , t. miwa , and k. ueno , @xcite . at present , the theory of isomonodromic deformations is a very active research field with deep connections to other areas such as the theory of integrable systems , classical theory of differential equations , and differential and algebraic geometry . in this work we focus on the relationship between isomonodromic deformations and painlev equations . we need to distinguish between continuous and discrete isomonodromic deformations . in the continuous case we move the location of singular points and the resulting schlesinger _ equations _ reduce to painlev - type nonlinear _ differential _ equations . on the other hand , if we deform the characteristic indices , the isomonodromy condition requires that the indices change by integral shifts , and so the resulting dynamic is discrete and is expressed in the form of _ difference _ equations called schlesinger _ transformations_. similarly to the continuous case , these transformations reduce to discrete analogues of painlev - type equations and the study of this correspondence has been a major research topic in the field of discrete integrable systems over the last twenty years , @xcite . we need to remark here that schlesinger transformations correspond to _ difference _ painlev equations , but the other two types of _ discrete _ painlev equations , _ @xmath2-difference _ and _ elliptic - difference _ painlev equations can also be considered in a modification of this approach . discrete painlev equations share many properties with the differential painlev equations , e.g. , the existence of special solutions such as algebraic solutions or solutions that can be expressed in terms of special functions , affine weyl group symmetries , and the geometric classification of equations in terms of rational surfaces . however , while it is well - known how to write differential painlev equations in the hamiltonian form @xcite , the discrete hamiltonian structure for discrete painlev equations is at present missing . the aim of the present paper is to study such discrete hamiltonian structure for schlesinger transformations . in particular , we present an explicit formula for the discrete hamiltonian function of an elementary schlesinger transformation expressed in the same coordinates as the hamiltonian function of the continuous schlesinger equations considered by jimbo , miwa , mri , and sato in @xcite . note that this explicit formula gives a convenient tool for computing schlesinger transformations and consequently for the derivation of discrete painlev equations , since the usual computation of a schlesinger transformation from the compatibility condition between the fuchsian equation and the deformation equation is often quite complicated . we also present two explicit examples of discrete painlev equation computed in this framework . the first example that we consider is an isomonodromic deformation of a two - dimensional fuchsian system with three finite singular points . a nice feature of this example is that it allows us to compare , in the same setting and using the same coordinates , continuous deformations that are described by the painlev - vi equation @xmath3 , and discrete deformations that are described by the difference painlev - v equation d-@xmath4 , which corresponds to the bcklund transformations of @xmath3 . recall that in terms of the surface of initial conditions d-@xmath4 is also denoted as d-@xmath5 . in the second example we consider a discrete painlev equation of the type d-@xmath6 that does not have a continuous painlev counterpart . in @xcite one of the authors posed a problem of representing this equation using schlesinger transformations of some fuchsian system . in @xcite p. boalch described the fuchsian system whose schlesinger transformations are described by d-@xmath7 without doing the explicit coordinate computation . also note that d. arinkin and a. borodin clarified the correspondence between d-@xmath7 and the _ difference _ fuchsian equation in @xcite . here we compute the d-@xmath7 example explicitly using the discrete hamiltonian and use it to illustrate some interesting features , as well as some difficulties , that occur when we study discrete painlev equations using this approach . the text is organized as follows . in the next section we briefly set up the context of the problem and describe in more details some of the main objects . in section [ sec : schlesinger_transformations ] we derive the equations of the elementary schlesinger transformations and in section [ sec : the_generating_function ] we show how to write them in the hamiltonian form . section [ sec : examples ] is dedicated to the d-@xmath5 and d-@xmath7 examples , and the final section is conclusions and discussion of the resul . * acknowledgements * : this work started at the _ discrete integrable systems _ program at the _ isaac newton institute for mathematical sciences _ in cambridge , uk . the authors are very grateful to the institute for its hospitality and the stimulating working environment . was supported by grant - in - aid no . 24540205 of the japan society for the promotion of science . part of this work was done when a.d . visited h.s . and t.t in tokyo , and he thanks the _ tokyo university of marine science and technology _ and the _ university of tokyo _ for their hospitality . would also like to acknowledge the generous travel support by the university of northern colorado provost fund which was very important for the success of this collaboration . schlesinger transformations that we study originate from the deformation theory of fuchsian equations . recall that a _ fuchsian equation _ ( or a _ fuchsian system _ ) is a matrix linear differential equation on the riemann sphere @xmath8 such that all of its singular points are _ regular singular points_. we consider a generic case when a fuchsian equation can be written in the the _ schlesinger normal form _ , i.e. , when the coefficient matrix is a rational function with at most simple poles at some ( distinct ) points @xmath9 ( and possibly at the point @xmath10 ) , @xmath11 here @xmath12 are constant @xmath13 matrices . we also put @xmath14 and so this equation is regular at infinity iff @xmath15 . [ sub : equations ] [ assume : diagonal ] from now on we make an additional semi - simplicity assumption that the coefficient matrices @xmath16 are _ diagonalizable_. geometrically , schlesinger ( or isomonodromic ) dynamic takes place on the space of coefficients @xmath17 of the fuchsian equation ( [ eq : fuchs-1 ] ) , considered modulo gauge transformations . here we briefly outline the description of this space following @xcite . first , given @xmath17 , we separate the data about locations of singular points @xmath18 ( that we think of as parameters of the dynamic ) and the residue matrices @xmath16 at those points . thus , we define @xmath19 to be the set of all fuchsian equations with possible singularities at the points @xmath18 . further , since schlesinger dynamic either preserves ( in the continuous case ) or shifts ( in the discrete case ) the eigenvalues of @xmath16 , we treat the eigenvalues as the parameters of the dynamic as well . thus , the appropriate phase space is a quotient of the fiber of the eigenvalue map by gauge transformations . local coordinates on the phase space are called _ accessory parameters_. note that the space @xmath20 of accessory parameters is quite complicated and its dimension depends on the spectral type of @xmath17 that we define next . spectral type of @xmath17 encodes the degeneracy of eigenvalues of the coefficient matrices @xmath16 via partitions @xmath21 , @xmath22 , where @xmath23 is the matrix size and @xmath24 denotes the multiplicities of the eigenvalues of @xmath16 . the _ spectral type _ @xmath25 of @xmath17 is then defined to be the collection of these partitions for all indices ( including @xmath26 ) : @xmath27 it can be used to classify fuchsian systems up to isomorphisms and the operations of addition and middle convolution introduced by n. katz @xcite , as in the recent work by t. oshima @xcite . [ assume : rank - reduce ] since for a fuchsian system we can use scalar local gauge transformation of the form @xmath28 , where @xmath29 is a solution of the _ scalar _ equation @xmath30 to change the residue matrices by @xmath31 , we can assume , without loss of generality , that one of the eigenvalues @xmath32 . thus , we always assume that the eigenvalue @xmath33 of the highest multiplicity @xmath34 is zero and we denote the corresponding subset of @xmath35 of _ reduced _ fuchsian equations by @xmath36 . this amounts to choosing a representative in the quotient space of all fuchsian equations by the group of local scalar gauge transformations . in view of * assumptions [ assume : diagonal ] * and * [ assume : rank - reduce ] * , @xmath16 is similar to a diagonal matrix @xmath37 , where @xmath38 . omitting the zero eigenvalues , we put @xmath39 denote by @xmath40 the set of all possible diagonal matrices of the spectral type @xmath25 with the highest degeneracy eigenvalues at _ finite _ points set to @xmath41 and omitted , as above . thus , @xmath42 and @xmath43 . then we have an eigenvalue map @xmath44 from the set of reduced fuchsian equations to the set of characteristic indices , where @xmath45 is the subset of @xmath46 satisfying the _ fuchs relation _ ( or the trace condition ) @xmath47 with these definitions , given @xmath48 , the fiber of the eigenvalue map is @xmath49 . we still need to take into account the global similarity transformations . it is convenient to do this in two steps . first , we use such transformations to _ normalize _ our equation at infinity by reducing @xmath50 to a particular form , e.g. , make it diagonal , and then take the further quotient by the stabilizer subgroup @xmath51 of the group @xmath52 of global similarity transformations . thus , for a given @xmath48 , we _ fix _ @xmath50 such that @xmath53 and denote by @xmath54 the subset of all fuchsian equations in @xmath55 satisfying the condition @xmath56 . then , finally , our phase space is @xmath57 when @xmath50 is diagonal , which is the case that we are mostly interested in , the dimension of this space @xmath20 of accessory parameters of the spectral type @xmath25 is given by the formula @xmath58 to study the hamiltonian structure of schlesinger equations , and also of schlesinger transformations , it is convenient to consider a larger decomposition space that is defined as follows . in view of * assumption [ assume : diagonal ] * , there exist full sets of _ right _ eigenvectors @xmath59 , @xmath60 , and _ left _ eigenvectors @xmath61 , @xmath62 , ( here we use the @xmath63 symbol to indicate a _ row_-vector ) . in the matrix form , omitting vectors with indices @xmath64 that are in the kernel of @xmath16 , we can write @xmath65 with @xmath66 defined by ( [ eq : theta - i ] ) . but this means that we have a decomposition @xmath67 provided that @xmath68 , where the last condition is related to the normalization ambiguity of the eigenvectors . thus , given @xmath16 , we can construct , in a non - unique fashion , a corresponding decomposition pair @xmath69 . the space of all such pairs for all finite indices @xmath70 , without any additional conditions , we call the _ decomposition space_. we denote it as @xmath71 since it is convenient to write an element @xmath72 of this space as a list of @xmath73 pairs @xmath74 . this space can be equipped with the natural symplectic structure where we take the matrix elements of @xmath75 and @xmath76 as canonical coordinates , i.e. , we take our symplectic form to be @xmath77 [ rem : coords ] to avoid symbol overcrowding we slightly abuse the notation and use the same symbols , e.g. , @xmath75 or @xmath59 , to denote both the canonical coordinate system in the decomposition space and an actual point in the space ; the exact meaning of the symbol is always clear from the context . the idea of using the decomposition space @xmath78 to study the hamiltonian structure of isomonodromic deformations goes back to the paper by m. jimbo , t. miwa , y. mri , and m. sato @xcite . since then it has been used by many other researchers studying isomonodromic deformations , most notably by j. harnad , see , e.g. , @xcite . [ rem : transf ] there are two natural actions on the decomposition space @xmath78 . first , the group @xmath52 of global gauge transformations of the fuchsian system induces the following action . given @xmath79 , we have the action @xmath80 which translates into the action @xmath81 . we refer to such transformations as _ similarity transformations_. it is often necessary to restrict this action to the subgroup @xmath51 preserving the form of @xmath50 . second , for any pair @xmath69 the pair @xmath82 determines the same matrix @xmath16 for @xmath83 . the condition @xmath84 restricts @xmath85 to the stabilizer subgroup @xmath86 of @xmath87 . in particular , when all @xmath33 are distinct , @xmath85 has to be a diagonal matrix . we refer to such transformations as _ trivial transformations_. these two actions obviously commute with each other . we are now ready to give an alternative description of the space of accessory parameters @xmath88 . given the pair @xmath89 as in the previous section , let @xmath90 the following diagram illustrates the relationship between different spaces defined in these two sections : consider now isomonodromic deformations of a fuchsian system . for the continuous case , the most interesting situation is when we take the deformation parameters to be locations of critical points . thus , we let @xmath91 ( i.e. , @xmath92 ) , and consider a family of fuchsian equations on the function @xmath93 , @xmath94 the compatibility conditions for these deformation equations are called _ schlesinger equations_. these are partial differential equations on the coefficient matrices @xmath95 and they have the form @xmath96 } { u_{i } - u_{j } } \quad ( i\neq j),\qquad \frac { \partial \mathbf{a}_{i } } { \partial u_{i } } = - \sum_{j\neq i}\frac { [ \mathbf{a}_{i},\mathbf{a}_{j } ] } { u_{i } - u_{j } } .\ ] ] in @xcite jimbo _ et.al _ showed that schlesinger equations can be written as a hamiltonian system on the decomposition space @xmath78 , @xmath97 with the hamiltonian @xmath98 schlesinger equations admit reductions to painlev - type nonlinear differential equations , which remains valid on the level of the hamiltonians as well ; we review an example of @xmath3 in section [ sec : examples ] . the main question that we consider is what happens to this picture in the _ discrete _ case . the discrete counterparts of schlesinger equations are schlesinger transformations . those are rational transformations preserving the singularity structure and the monodromy data of the fuchsian system ( [ eq : fuchs-1 ] ) , except for the integral shifts in the characteristic indices @xmath33 . the study of such transformations again goes back to schlesinger @xcite , and this and the more general case of irregular singular points was considered in great detail in @xcite . schlesinger transformations have the form @xmath99 , where @xmath100 is a specially chosen rational matrix function called the _ multiplier _ of the transformation . the coefficient matrix @xmath17 of our fuchsian system then transforms to @xmath101 that is related to @xmath17 by the equation @xmath102 similarly to the continuous case , schlesinger transformations induce discrete painlev - type dynamic on the space of accessory parameters . this dynamic is very complicated , but it becomes much simpler when considered on the decomposition space , which allows us to represent it as a discrete hamiltonian system using the same canonical coordinates as @xcite , as in figure [ fig : sch - trans ] : unfortunately , at present we do not have a discrete version of the hamiltonian reduction procedure from the decomposition space to the space of accessory parameters , this is an important problem for future research . in this paper we focus on the _ elementary _ schlesinger transformations that only change two of the characteristic indices by unit shifts , i.e. , @xmath103 and @xmath104 . for the examples section , we also assume @xmath105 and @xmath106 have no multiplicities . in @xcite such elementary transformation are denoted by @xmath107 . in section [ sec : the_generating_function ] we construct a function @xmath108 of the decomposition space @xmath78 that is a discrete hamiltonian function , in the sense explained in the next section , for such elementary schlesinger transformation . in developing a discrete version of the hamiltonian formalism , one approach is to discretize the continuous dynamic preserving its integrability properties , see the recent encyclopedic book by y. suris , @xcite , but it requires underlying continuous dynamical system . an alternative and more direct procedure is to develop a discrete version based on the variational principles . in this case , the lagrangian formalism is more natural , but it is possible to extend it to include the discrete version of the hamiltonian formalism as well . this approach has its origin in the optimal control theory ( see , e.g. , b. jordan and e. polak @xcite , and j. cadzow @xcite ) and mechanics ( see e.g. , j. logan @xcite and s. maeda @xcite ) . in the theory of integrable systems it was first used in the foundational works by a. veselov @xcite and a. veselov and j. moser @xcite . a lot of recent work in this field is motivated by the development of very effective _ symplectic integrators _ , see an excellent review paper by j. marsden and m. west @xcite ( and the references therein ) , as well as more recent works by s. lall and m. west @xcite and a. bloch , m. leok , and t. oshawa @xcite that emphasize the hamiltonian aspects of the theory . below we give a _ very _ brief outline of this approach closely following @xcite , we refer to this and other publications above for details . let @xmath109 be our configuration space . then , in the discrete case , the _ lagrangian _ @xmath110 is a function on the _ state space _ @xmath111 . for the discrete time parameter @xmath112 , a trajectory of motion is a map @xmath113 , @xmath114 , or , equivalently , a sequence @xmath115 . the action functional on the space of such sequences is given by @xmath116 . using the variational principle @xmath117 we obtain _ discrete euler - lagrange _ equations that have the form @xmath118 where @xmath119 ( resp . @xmath120 ) denote vectors of partial derivatives w.r.t . the first ( resp . second ) sets of local coordinates on the state space @xmath111 . these equations then implicitly determine the map ( or more precisely , the correspondence ) @xmath121 , which then defines the discrete lagrangian flow @xmath122 on the state space via @xmath123 . this flow is symplectic w.r.t . the discrete lagrangian symplectic form @xmath124 @xmath125 , where one - forms @xmath126 are defined by @xmath127 if we introduce _ right _ and _ left discrete legendre transforms _ @xmath128 and the _ momenta _ variables @xmath129 , @xmath130 by @xmath131 we see that @xmath132 and @xmath133 , where @xmath134 and @xmath135 are the standard liouville and symplectic forms on @xmath136 respectively . we can then define the discrete hamiltonian flow @xmath137 by @xmath138 , i.e. , @xmath139 . but then the equations @xmath140 mean that @xmath141 is just the generating function ( of type one , see @xcite for the terminology ) of the canonical transformation @xmath142 . we can then define the _ right _ and _ left discrete hamiltonian functions _ as generating functions of this canonical transformation of type two and three respectively . right discrete hamiltonian _ is @xmath143 and then the map @xmath137 is given ( implicitly ) by the _ right discrete hamiltonian equations _ @xmath144 similarly , _ left discrete hamiltonian _ @xmath145 gives _ left discrete hamiltonian equations _ @xmath146 for completeness , we mention that the generating function of type four , @xmath147 is nothing but the lagrangian function on the space of momenta satisfying the discrete euler - lagrange equations @xmath148 for elementary schlesinger transformations @xmath107 the multiplier matrix @xmath100 has a very special simple form , and in this section we describe some useful properties of such matrices that we need for our computations . consider a rational matrix function @xmath149 because of their role in representing general matrix functions as a product of factors of this form , such matrices are sometimes called elementary divisors @xcite . assuming that @xmath150 , which is the case we need , we can consider instead of @xmath151 a corresponding rank - one projector @xmath152 , @xmath153 . [ lem : el - divs ] let @xmath154 then we have the following a. basic properties , @xmath155 b. the _ vanishing rule _ , @xmath156 c. and the _ exchange rule _ , @xmath157 part ( i ) is a simple direct computation , see also @xcite . next , notice that for a rank - one projector @xmath152 , @xmath158 and taking the trace proves the _ vanishing rule_. finally , @xmath159 which proves the _ exchange rule_. in this section we derive equations of an elementary schlesinger transformation @xmath107 in terms of the coordinates on the decomposition space @xmath78 . we now take @xmath100 to be of the form considered in lemma [ lem : el - divs ] . then we have the following result . [ lem : schlesinger ] consider transformation ( [ eq : sch - transf ] ) with @xmath160 then a. the poles @xmath161 of @xmath101 coincide with the poles @xmath162 of @xmath17 if and only if , for some choice of indices @xmath163 , @xmath164 , @xmath165 , and @xmath166 , we have @xmath167 where @xmath168 means _ proportional_. thus , such @xmath100 defines the elementary transformation @xmath169 . b. for @xmath100 satisfying ( [ eq : sch - r - matrix - params ] ) , the residue matrices of @xmath17 and @xmath101 are connected by the following equations : @xmath170 from the schlesinger transformation equations @xmath171 and @xmath172 we see that @xmath173 we see that @xmath174 . let @xmath175 and @xmath176 . then @xmath177 and @xmath178 . since both @xmath100 is regular and invertible at @xmath18 for @xmath179 , taking residues of ( [ eq : sch - transf - a ] ) at @xmath18 gives ( [ eq : ai - generic ] ) . since @xmath100 is regular ( but degenerate ) at @xmath180 , we can take a residue at @xmath180 of the first equation in ( [ eq : sch - transf - a ] ) to get the second equation in ( [ eq : ai - special ] ) . similarly , taking the residue of the second equation in ( [ eq : sch - transf - a ] ) at @xmath181 we get the second equation in ( [ eq : ai - special ] ) . further , at @xmath181 the first equation in ( [ eq : sch - transf - a ] ) becomes @xmath182 the @xmath183-terms give @xmath184 and so we see that @xmath185 has to be an eigenvector of @xmath186 and @xmath187 has to be an eigenvector of @xmath188 . choose an index @xmath165 and let @xmath189 and @xmath190 , then @xmath191 . taking the residue at @xmath181 gives @xmath192 but this equation follows from @xmath193 and @xmath194 which is an immediate consequence of ( [ eq : ai - generic ] ) . repeating this argument for @xmath180 and the second equation in ( [ eq : sch - transf - a ] ) implies that @xmath195 , @xmath196 , and @xmath197 , for some choice of the index @xmath166 , which completes the proof . we can extend elementary schlesinger transformations to the decomposition space as follows . [ cor : schlesinger - lift ] let @xmath17 and @xmath101 be connected by an elementary schlesinger transformation @xmath198 given by @xmath100 of the form ( [ eq : r - matrix - generic ] ) satisfying ( [ eq : sch - r - matrix - params ] ) . then the corresponding points @xmath72 and @xmath199 in the decomposition space are related as follows : a. if @xmath200 , @xmath201 is invertible and @xmath202 b. for the special indices @xmath163 and @xmath203 we have @xmath204 multiplying first equation in ( [ eq : ai - generic ] ) on the right by @xmath75 gives @xmath205 , and so the matrix @xmath206 is the matrix of eigenvectors of @xmath207 , @xmath208 . the remaining equations are proved similarly . recall that the matrices @xmath209 and @xmath210 are degenerate , @xmath211 and @xmath212 , see ( [ eq : el - divs - basic ] ) , so the proportionality coefficients in ( [ eq : bc - special ] ) at @xmath213 , @xmath214 , @xmath215 , and @xmath216 are zero . the main objective of this section is to obtain an explicit formula for the discrete hamiltonian function @xmath217 that generates schlesinger transformations dynamic on the decomposition space . in this case the equations ( [ eq : hplus - gen - eqns ] ) take the form ( recall remark [ rem : coords ] ) @xmath218 where we use the identifications @xmath219 . first we introduce some notation . let @xmath220 and for a multi - index @xmath221 , we put @xmath222 [ thm : gen - func ] let @xmath223 then , if @xmath100 is the matrix defining the elementary transformation @xmath224 from @xmath17 to @xmath101 , @xmath225 , @xmath226 are the corresponding points in the decomposition space @xmath78 , and @xmath227 , we have @xmath228 we prove this theorem using a sequence of lemmas . the generic situation is very simple . [ lem : gen - func - reg ] let @xmath100 be any matrix of the form ( [ eq : r - matrix - generic ] ) with @xmath229 , and let @xmath230 , with @xmath67 , @xmath231 , and @xmath232 . then @xmath233 indeed , in view of corollary [ cor : schlesinger - lift ] , @xmath234 , and the normalization @xmath235 gives @xmath236 the other equation is proved in exactly the same way . the following corollary is then immediate . [ cor : generic ] equations ( [ eq : dynamics ] ) in theorem [ thm : gen - func ] hold for all indices @xmath237 with @xmath200 and for @xmath238 for @xmath239 . moreover , for @xmath200 we have @xmath240 for the index @xmath241 , since @xmath242 and @xmath243 , a different approach is needed to find expressions for @xmath214 and @xmath244 . we do it in lemma [ lem : gen - func - beta ] . let us now consider the indices @xmath163 and @xmath203 defining the transformation . in this case @xmath100 becomes either singular or degenerate and some regularization is necessary . we do it using the equations that cut out the locus @xmath245 . [ lem : gen - func - alpha ] in the setting of theorem [ thm : gen - func ] with @xmath246 given by ( [ eq : r - alpha ] ) , we have @xmath247 from ( [ eq : sch - r - matrix - params ] ) , @xmath248 . this and the normalization @xmath249 gives ( [ eq : b - am - bar ] ) , ( [ eq : c - am ] ) follows similarly from @xmath250 . this also gives an additional orthogonality conditions @xmath251 so @xmath246 is well - defined . we now need to solve the equations @xmath252 for @xmath253 and @xmath254 for @xmath255 . the matrix @xmath256 has rank @xmath257 with the null - spaces spanned by @xmath258 and @xmath259 . we can invert it on the orthogonal complements of those subspaces using @xmath246 . indeed , @xmath260 thus , for @xmath261 , @xmath262 , which gives ( [ eq : b - aj - bar ] ) ; ( [ eq : c - aj ] ) is similar . [ lem : gen - func - beta ] in the setting of theorem [ thm : gen - func ] with @xmath263 given by ( [ eq : s - full ] ) , we have @xmath264 using ( [ eq : aisbc ] ) and the residue at infinity condition @xmath265 , we get @xmath266 thus , @xmath267 which completes the proof of the lemma ( and of theorem [ thm : gen - func ] ) , and also gives the expressions for @xmath214 and @xmath244 . even though the generating function @xmath268 is defined on a dense open subspace of @xmath199 , in view of the orthogonality conditions ( [ eq : orth - cond ] ) , to get the well - defined map from @xmath269 to @xmath270 we need to restrict it to the subspace @xmath271 in this section we consider , using our discrete hamiltonian framework , two explicit examples of discrete schlesinger transformations that can be reduced to discrete painlev equations . we also try to illustrate in this section the role played by the geometry of painlev equations in not only determining the type of the equation , but also in studying the relationship between different explicit forms of equations of the same type . recall that , according to the geometric theory of painlev equations developed in @xcite , both continuous and discrete painlev equations can be classified by the type of their space of initial conditions , introduced by okamoto @xcite in the continuous case . this space can be obtained by successively blowing up the projective plane @xmath272 at nine ( possibly infinitely - close ) points that lie on a unique ( possibly singular ) cubic curve . the type of the resulting surface @xmath273 is then given by the configurations of irreducible components @xmath274 of its _ unique _ anti - canonical divisor @xmath275 or , equivalently , by the root sub - lattice @xmath276 spanned by @xmath274 in the orthogonal complement @xmath277 of @xmath275 in the picard lattice @xmath278 . the orthogonal complement @xmath279 of @xmath276 in @xmath277 then corresponds to the symmetries of the equation its weyl group acts as automorphisms , called the _ cremona transformations _ , of @xmath278 preserving @xmath280 and the translational part of the weyl group corresponds to discrete painlev equations . we also want to emphasise the following point , see also @xcite . given a discrete painlev equation , we can determine its type by constructing its space of initial conditions . however , there is no converse process , although there are some particularly nice forms ( that we call _ standard _ ) of equations of each type , see @xcite . further , there can be many _ non - equivalent equations _ ( i.e. , equations that can not be transformed into each other by some change of coordinates ) of the same type such equations can correspond to different translation directions in the affine weyl group of symmetries of @xmath273 . and even when the two different equations are equivalent , finding the actual change of coordinates that will transform one equation into the other can be a very challenging problem . in each case , understanding the geometry and the blow - up structure of the space of initial conditions can be of great help in determining the relationship between two different equations of the same type , as we illustrate by the two explicit examples below . for the first example , we consider the fuchsian equation of matrix size @xmath281 with @xmath282 finite poles and rank - one condition @xmath283 for the residue matrices at those poles . it is well - known that continuous isomonodromic deformations of this system are described by the classical continuous painlev equation @xmath3 , the type of the space of initial conditions is @xmath284 , and the discrete painlev v equation d-@xmath285 d-@xmath286 appears as bcklund transformations of @xmath3 . thus , it is reasonable to expect that an elementary schlesinger transformation of this system is described by d-@xmath4 , and we explicitly show that this is indeed the case . the second example is a more interesting one . in @xcite one of the authors ( h.s . ) posed a problem of writing down discrete painlev equations of type @xmath287 , @xmath288 and @xmath289 as schlesinger transformations of some fuchsian equations . one difficulty here is that these equations do not correspond to any continuous painlev differential equations . this problem was solved theoretically by boalch in @xcite , where it was shown that the moduli spaces of monodromy groups of some fuchsian equations are analytically isomorphic to the spaces of initial conditions for these discrete painlev equations . in particular , boalch showed that the fuchsian system with matrix size @xmath290 with @xmath291 finite poles ( and so there are no continuous deformation parameters ) and rank - two residue matrices , @xmath292 , has as its symmetry the affine weyl group of type @xmath293 , and so the type of the corresponding surface is either @xmath294 ( for the _ multiplicative _ or _ @xmath2-difference _ case ) or @xmath295 ( for the _ additive _ or _ difference _ case ) . since we consider the additive case , we expect that discrete schlesinger transformations of this system should have type d-@xmath296 . however , an elementary discrete schlesinger transformation , when written in coordinates , becomes very complicated and in fact it can not be reduced to a standard form of d-@xmath296 , as written in @xcite . using the geometry of the space of initial conditions , we see that this is an example of a situation when the standard form of the equation corresponds to a direction that is a combination of two different directions given by different elementary schlesinger transformations , and then we explicitly show how to combine them into a single schlesinger transformation that corresponds to the standard form of d-@xmath296 , this is a new result . the geometric approach that we use requires finding correspondences between the rational surfaces that appear as desingularizations of different coordinate forms of two difference equations of the same type . such correspondence gives us a blow - down structure that can then be used to find a change of coordinates that relates the two equations that we consider , see the recent preprint @xcite by a. carstea and t. takenawa for a more detailed introduction into this technique . for this example we consider a fuchsian system of the spectral type @xmath298 , i.e. , we have @xmath282 ( finite ) poles and the matrix size is @xmath281 . in view of assumption [ assume : rank - reduce ] , it is possible to make the rank of the residue matrices at the finite poles equal to one . further , using the mbius transformations on @xmath299 , we can map @xmath300 to @xmath301 . we put @xmath302 and from now on we use @xmath303 instead of the indices @xmath304 , so @xmath305 , etc . then @xmath306 and we have the following _ riemann scheme _ and the corresponding _ fuchs relation _ : @xmath307 in the rank - one case there is no need for the second index , and so we put @xmath308 , @xmath309 . then the spectral decomposition is simply @xmath310 with @xmath311 . consider the elementary transformation @xmath312 , i.e. , @xmath313 , @xmath314 and @xmath315 . its multiplier matrix and the generating hamiltonian function are @xmath316 this function generates the following dynamic equations : @xmath317 to illustrate the general situation , we now show how to solve these equations to define a map from @xmath269 to @xmath270 in this relatively simple example . this map is defined up to trivial transformations and becomes unique once we specify some normalization condition . indeed , equation ( [ eq : dpv - c1 ] ) implies that @xmath318 , where @xmath319 is some proportionality constant . thus , we can rewrite the multiplier matrix as @xmath320 . equation ( [ eq : dpv - c0 ] ) can be rewritten as @xmath321 , and so @xmath322 , cf . ( [ eq : bc - generic ] ) . finally , expression for @xmath323 can either be obtained by a direct computation or , simpler , from the equation @xmath324 . we then substitute these expressions into equations ( [ eq : dpv - b1][eq : dpv - bt ] ) and simplify . the final map is given by the following set of equations ( with @xmath325 , @xmath326 , and @xmath319 arbitrary constants corresponding to trivial transformations ) : @xmath327 let us now look at the schlesinger dynamic on the space of accessory parameters . normalizing the eigenvectors @xmath328 and @xmath329 , we can write @xmath330 here @xmath331 and so @xmath332 and @xmath333 . the reason for introducing @xmath334 is to simplify the relations between parameters @xmath335 , @xmath336 , and @xmath337 that follow from the trace conditions and the equation @xmath338 . these relations , written in terms of @xmath335 , @xmath339 , and @xmath340 , are @xmath341 where the last equation also follows from ( [ eq : dpv - rel22 ] ) and the fuchs relation . thus , everything is parameterized by @xmath342 , @xmath343 , @xmath344 , and @xmath345 subject to one additional constraint @xmath346 . one additional degree of freedom is related to the global gauge action by constant non - degenerate diagonal matrices that allows us to normalize one of the @xmath335s , but it is convenient to keep this freedom for computations . finally , note that the normalization parameter @xmath335 should be considered as a part of either @xmath328 or @xmath329 , where the actual choice depends on the context . same coordinates were used in @xcite to describe the isomonodromic dynamic in the continuous case and the reduction to @xmath3 . with @xmath335 attached to @xmath329 , the standard symplectic form on the decomposition space reduces as follows , see * proposition 4 * in @xcite for a more general statement : @xmath347 where we used ( [ eq : dpv - rel22 ] ) to eliminate @xmath348 . it is convenient to introduce new symplectic coordinates @xmath349 and @xmath350 . this gives , using ( [ eq : dpv - rel12][eq : dpv - rel11 ] ) , @xmath351 for the continuous deformations , the only deformation parameter is @xmath352 and the corresponding hamiltonian , given by equation ( [ eq : cont - ham ] ) , under this reduction becomes @xmath353 since @xmath354 where @xmath355 is the hamiltonian function for @xmath3 . unfortunately , such a reduction procedure does not work in the discrete case and we need to work directly with the evolution equations . we give a brief outline of such computations and show the reduction to d-@xmath4 . it is convenient to attach the normalization parameters @xmath356 to @xmath357 and @xmath335 to @xmath329 ( i.e. , in what follows , @xmath357 denotes the normalized vector , but in the formulas describing the dynamics we use @xmath358 , and so on ) . it turns out that it is possible to separate the `` normalization '' dynamics of @xmath335 and the `` directional dynamics '' of @xmath339 and @xmath340 , the latter one is given by the _ painlev equations_. equation ( [ eq : dpv - b1 ] ) becomes @xmath359 note that @xmath360 , and so @xmath361 , as it should be . equation ( [ eq : dpv - c1 ] ) becomes @xmath362 using this equation in the computation of the trace , @xmath363 , we again see that the trace condition is satisfied , @xmath364 . important consequences of ( [ eq : dpv - b1-p ] ) and ( [ eq : dpv - c1-p ] ) are @xmath365 we show shortly that the first equation generates the `` normalization '' dynamics . equation ( [ eq : dpv - b0 ] ) becomes @xmath366 the traces in this formula are , using ( [ eq : dpv - b1-p ] ) and ( [ eq : dpv - c1-p ] ) , @xmath367 the equality of the normalized vectors then gives @xmath368 which gives the following important relation @xmath369 as before , the normalization condition reduces to @xmath315 . it is possible to show that the remaining evolution equations do not give any additional information . let us now write everything in the symplectic coordinates @xmath370 and @xmath2 . equation ( [ eq : beta - gamma ] ) gives the equation for the normalization dynamic , @xmath371 multiplying ( [ eq : dpv - b0-p ] ) by @xmath372 , we get @xmath373 finally , combining ( [ eq : dpv - b1-p ] ) and ( [ eq : dpv - rel21 ] ) gives @xmath374 dividing by @xmath375 and doing some re - arranging , we get @xmath376 we now claim that after a certain change of variables equations ( [ eq : dpv - eq1-pq ] , [ eq : dpv - eq2-pq ] ) transform into the standard form for d-@xmath4 . although this can be seen directly without too much difficulty , we prefer to use the geometry of the rational surface corresponding to the equation to explain how to find this coordinate transformation , illustrating some of the ideas that we need for the next , more complicated , example . let us briefly review the geometric description of the difference painlev v equation from the point of view of the theory of rational surfaces @xcite . we think about it as a birational map @xmath378 with parameters @xmath379 or , equivalently , as an automorphism of a field @xmath380 , @xmath381 where @xmath382 and where @xmath383 and @xmath384 are given by the equation that is usually called the discrete painlev - v equation @xmath385 we refer to it as the standard form of d-@xmath4 . we lift this map to an _ isomorphism _ @xmath386 between rational surfaces @xmath387 and @xmath388 obtained from @xmath389 via a sequence of blowups that resolve the indeterminacies of the map @xmath390 , @xmath391_{\hat{\varphi}}^{\simeq } \ar@{-->}[d ] & x_{\bar{\mathbf{a } } } \ar@{-->}[d ] \\ \mathbb{p}^{1}\times \mathbb{p}^{1 } \ar@{-->}[r]_-{\varphi } & \mathbb{p}^{1}\times \mathbb{p}^{1 } } } .\ ] ] decomposing @xmath390 as a sequence of two maps , @xmath392 it is easy to see that the indeterminate points of @xmath393 are @xmath394 and @xmath395 and the indeterminate points of @xmath396 are @xmath397 , @xmath398 , @xmath399 , and @xmath400 . applying @xmath401 we obtain that the indeterminate points of @xmath390 are @xmath402 , @xmath403 , @xmath404 , @xmath405 , @xmath406 , and @xmath407 . in exactly the same way we find that the indeterminate points of @xmath408 are @xmath409 . as usual , we denote a local coordinate description of a blowup at a point @xmath410 by @xmath411 where @xmath412 and @xmath413 , then the exceptional divisor is given by @xmath414 . for example , extending the map @xmath390 at the blowup at @xmath402 , we use @xmath415 and @xmath416 . then @xmath417 and @xmath418 become well - defined and we see that the exceptional divisor @xmath419 is mapped to the line @xmath420 . on the level of divisor classes on @xmath278 we get the map @xmath421 , where @xmath422 is the class of the total transform of the line @xmath423 . proceeding along the same lines and noticing that we need to blow - up at two additional points @xmath424 on @xmath425 and @xmath426 on @xmath427 ( here we give the coordinates of the points in the @xmath428 charts ) , we get the following blowup diagram describing the construction of the surface @xmath429 corresponding to our equation : the picard lattice of @xmath430 is generated by the total transforms @xmath422 and @xmath431 of the coordinate lines and the classes of the exceptional divisors @xmath432 , @xmath433 the anti - canonical divisor @xmath434 uniquely decomposes as a positive linear combination of @xmath435-curves @xmath274 , @xmath436 where @xmath437 the configuration of these @xmath435-curves ( on which the blow - up points are located ) is described by the affine dynkin diagram of type @xmath284 , , which is why we say that the difference painlev v equation has the type d-@xmath377 . the corresponding divisors @xmath274 generate a sub - lattice @xmath438 in @xmath278 . symmetries of @xmath273 are given by the affine weyl group of _ cremona transformations _ generated by the _ complementary _ root sub - lattice @xmath279 in @xmath439 acting as automorphisms of @xmath278 . in this case , @xmath440 , where @xmath441{dpv - a - conf}},\ ] ] which is again of the type @xmath284 . finally , we can compute the action of @xmath442 on @xmath278 to be @xmath443 and so the induced action @xmath442 on the sub - lattice @xmath279 is given by _ translation _ : @xmath444 we now demonstrate how to reduce equations ( [ eq : dpv - eq1-pq ] , [ eq : dpv - eq2-pq ] ) to equations ( [ eq : dpv - standard ] ) . first , we resolve the indeterminacies of @xmath445 to get an isomorphism @xmath446 . the surface @xmath447 , with @xmath448 denoting the collection of parameters of the map , @xmath449 , is constructed by the following blow - up diagram : . here the coordinates of the blow - up points are given w.r.t . the following local coordinate systems : @xmath450 note that the resulting configuration of exceptional divisors looks quite different from the previous case and moreover , @xmath451 . the reason for this is that the surface @xmath452 is not _ relatively minimal _ for @xmath453 , see also @xcite there are two pairs of @xmath454-curves that are exchanged by @xmath453 : @xmath455 and @xmath456 , and so blowing down any of these pairs will still result in an isomorphism between the corresponding surfaces . we blow down @xmath457 and @xmath458 , which eliminates @xmath459 and @xmath460-curves and gives us the diagram on the left . this diagram now looks very similar to the configuration we had for the standard form of d-@xmath4 ( on the right ) , [ cols="^,^ " , ] . thus , we have showed that an elementary schlesinger transformation of this fuchsian system is given by the discrete painlev equation of type d-@xmath296 . however , equation ( [ eq : dpa-21 * -sch ] ) is very complicated , and we again consider whether it is possible to transform this equation into a standard form . following murata @xcite and sakai @xcite , we take the standard form of the d-@xmath296 to be the following . we consider d-@xmath296 to be a birational map @xmath378 with parameters @xmath461 @xmath462 @xmath463 , and @xmath383 and @xmath384 are given by the equation @xmath464 note that equation ( [ eq : dpa2-st ] ) is the asymmetric version of equation ( 1.3 ) in @xcite . we get the following blowup diagram describing the construction of the surface @xmath465 corresponding to our equation : @xmath466{dpa - standard - blowup}}\label{diag : dpa - standard - blowup}\ ] ] as in section [ ssub : geometry_of_d_p__v _ ] , @xmath467 , the anti - canonical divisor @xmath468 decomposes as @xmath469 , where @xmath470 the symmetry sub - lattice @xmath471 , where @xmath472{dpa - conf - std}},\ ] ] which is of the type @xmath473 . we compute the action of @xmath442 on @xmath278 to be @xmath474 and so the induced action @xmath442 on the sub - lattice @xmath279 is given by the following translation : @xmath475 to match the surface @xmath452 described by the blow - up diagram ( [ diag : dpa - conf ] ) with the surface @xmath465 described by diagram ( [ diag : dpa - standard - blowup ] ) , we look for the blow - down structure in @xmath476 , i.e. , we look for rational classes @xmath477 , @xmath478 , @xmath479 in @xmath476 such that @xmath480 and the resulting configuration matches diagram ( [ diag : dpa - standard - blowup ] ) . comparing the @xmath435-curves on both diagrams , @xmath481 we get the following identification ( it is easy to see that it satisfies the genus and the rationality conditions ) : @xmath482 to complete the correspondence it remains to define the base coordinates @xmath483 and @xmath484 of the linear systems @xmath485 and @xmath486 that will map the exceptional fibers of the divisors @xmath487 to the points @xmath488 such that @xmath489 and @xmath490 are on the line @xmath491 , @xmath492 and @xmath493 are on @xmath494 , and @xmath495 are on the line @xmath496 . since the pencil @xmath485 consists of all curves on @xmath272 passing through @xmath497 , @xmath498 we can define the projective base coordinate as @xmath499 $ ] , with @xmath203 being some parameter . similarly , @xmath500 and @xmath501 $ ] . then @xmath502,\qquad & g_{1}(\pi_{7 } ) & = g_{1}(p_{1 } ) = [ 1+\alpha:1],\\ f_{1}(\pi_{6 } ) & = f_{1}(p_{8 } ) = [ 1:0],\qquad & g_{1}(\pi_{8 } ) & = g_{1}(p_{1 } ) = [ 1+\alpha:1].\end{aligned}\ ] ] thus , we take @xmath503 ( and looking at the other points we notice that it is also convenient to choose @xmath504 ) , and so we provisionally put @xmath505 in these coordinates equation ( [ eq : quadricq - xyz ] ) of the quadric @xmath506 becomes @xmath507 and so the points @xmath495 lie on the line @xmath508 . hence , if we make the final choice of the base coordinates to be @xmath509 the points @xmath510 will be on the line @xmath511 . specifically , we have : @xmath512 thus , we immediately get the identification between the parameters in the riemann scheme of our fuchsian system and the parameters @xmath336 in the standard form of the painlev equation : @xmath513 then @xmath514 and the induced action of the standard painlev dynamic on the riemann scheme is @xmath515 thus , these two transformations correspond to the different translation directions in the symmetry root sub - lattice of the surface @xmath452 and so are not equivalent . indeed , we compute the action of @xmath516 of an elementary schlesinger transformation @xmath517 on the classes @xmath477 , @xmath478 , and @xmath487 to be @xmath518 and compare with the standard dynamic @xmath442 given by ( [ eq : dpa - hf - std][eq : dpa - e8-std ] ) to see this explicitly : @xmath519 where in both cases @xmath520 . in fact , we can represent the standard painlev dynamic as a composition of two different schlesinger transformations , combined with automorphisms of our fuchsian system , as follows . let @xmath521 be the map that exchanges the @xmath522-th and the @xmath523-th eigenvectors of @xmath16 . if the eigenvalues @xmath33 and @xmath524 are non - zero , this map is just a permutation on the decomposition space @xmath525 . for this example we need @xmath526 , but it also defines a mapping on @xmath525 , since @xmath527 and @xmath528 . combining @xmath526 with the scalar gauge transformation @xmath529 , where @xmath530 , we get the action @xmath531 on the spectral data ( as encoded in the riemann scheme ) and the decomposition space : @xmath532 where we had to take into account the normalization condition @xmath533 . then , looking at the action on the riemann scheme , we see that @xmath534 using * mathematica * , we can verify directly that this composition dynamic satisfies equation ( [ eq : dpa2-st ] ) . to summarize , we gave an explicit expression for the generating function for elementary schlesinger transformations in terms of the coordinates on the decomposition space for the coefficients of the underlying fuchsian system . we also argued that this function should be thought of as a discrete hamiltonian of the dynamic generated by schlesinger transformations , thus continuing to establish a parallel between the theories of continuous and the discrete isomonodromic transformations . we used this discrete hamiltonian to compute some explicit examples , including the derivation of the discrete painlev equation of type d-@xmath1 in this framework , thus answering an earlier question of sakai about the representability of this equation in terms of schlesinger transformations . we also emphasized the role played by the geometry of rational surfaces corresponding to the equations in comparing different equations of the same type . d. arinkin and a. borodin , _ moduli spaces of @xmath535-connections and difference painlev equations _ , duke math . j. * 134 * ( 2006 ) , no . 3 , 515556 , http://arxiv.org/abs/math/0411584[`arxiv:math/0411584 [ math.ag ] ` ] . mr mr2254625 ( 2007h:14047 ) philip boalch , _ quivers and difference painlev equations _ , groups and symmetries , crm proc . lecture notes , vol . 47 , amer . soc . , providence , ri , 2009 , pp . 2551 , http://arxiv.org/abs/0706.2634[`arxiv:0706.2634 [ math.ag ] ` ] . mr 2500553 ( 2011g:39028 ) alexei borodin , _ isomonodromy transformations of linear systems of difference equations _ , ann . of math . ( 2 ) * 160 * ( 2004 ) , no . 3 , 11411182 , http://arxiv.org/abs/math/0209144[`arxiv:math/0209144 [ math.ca ] ` ] . mr mr2144976 adrian stefan carstea and tomoyuki takenawa , _ a note on minimization of rational surfaces obtained from birational dynamical systems _ , preprint , http://arxiv.org/abs/1211.5393[`arxiv:1211.5393 [ math.ds ] ` ] . anton dzhamay , _ factorizations of rational matrix functions with application to discrete isomonodromic transformations and difference painlev equations _ , j. phys . a : math * 42 * ( 2009 ) , no . 45 , 10pp . , http://arxiv.org/abs/0904.1971[`arxiv:0904.1971 [ math - ph ] ` ] . r. garnier , _ sur les quations diffrentielles du troisime order do nt lintgral gnrale est uniforme et sur une classe dquations nouvelles dodre suprieur do nt lintgrale gnrale a ses points critiques fixes _ , ann . sci . cole norm . * 29 * ( 1912 ) , 1126 . b. grammaticos , a. ramani , and y. ohta , _ a unified description of the asymmetric @xmath536 and @xmath537 equations and their schlesinger transformations _ , j. nonlinear math * 10 * ( 2003 ) , no . 2 , 215228 , http://arxiv.org/abs/nlin/0310050[`arxiv:nlin/0310050 [ nlin.si ] ` ] . mr 1976382 ( 2004c:34273 ) j. harnad , _ dual isomonodromic deformations and moment maps to loop algebras _ , comm . math . * 166 * ( 1994 ) , no . 2 , 337365 , http://arxiv.org/abs/hep-th/9301076[`arxiv:hep-th/9301076 ` ] . mr mr1309553 ( 96e:58073 ) michio jimbo , tetsuji miwa , and kimio ueno , _ monodromy preserving deformation of linear ordinary differential equations with rational coefficients . i. general theory and @xmath538-function _ , phys . d * 2 * ( 1981 ) , no . 2 , 306352 . mr mr630674 ( 83k:34010a ) michio jimbo and hidetaka sakai , _ a @xmath2-analog of the sixth painlev equation _ phys . * 38 * ( 1996 ) , no . 2 , 145154 , http://arxiv.org/abs/chao-dyn/9507010[`arxiv:chao-dyn/9507010 ` ] . mr mr1403067 ( 97m:39002 ) jrgen moser and alexander p. veselov , _ discrete versions of some classical integrable systems and factorization of matrix polynomials _ , comm . * 139 * ( 1991 ) , no . 2 , 217243 . ( 92g:58054 ) mikio murata , _ new expressions for discrete painlev equations _ , funkcial . * 47 * ( 2004 ) , no . 2 , 291305 , http://arxiv.org/abs/nlin/0304001[`arxiv:nlin/0304001 [ nlin.si ] ` ] . mr 2108677 ( 2005j:39014 ) tomoki ohsawa , anthony m. bloch , and melvin leok , _ discrete hamilton - jacobi theory _ , siam j. control optim . * 49 * ( 2011 ) , no . 4 , 18291856 , http://arxiv.org/abs/0911.2258[`arxiv:0911.2258 [ math.oc ] ` ] . mr 2837576 ( 2012g:70028 ) v. g. papageorgiou , f. w. nijhoff , b. grammaticos , and a. ramani , _ isomonodromic deformation problems for discrete analogues of painlev equations _ lett . a * 164 * ( 1992 ) , no . 1 , 5764 . mr mr1162062 ( 93c:58092 ) to3em , _ problem : discrete painlev equations and their lax forms _ , algebraic , analytic and geometric aspects of complex differential equations and their deformations . painlev hierarchies , rims kkyroku bessatsu , b2 , res . inst . math . ( rims ) , kyoto , 2007 , pp . 195208 . mr 2310030 ( 2008c:33020 )

bose - einstein condensates of dilute gases are well - suited for investigations of vortex dynamics and turbulence in quantum fluids , yet there has been little experimental research into the approaches that may be most promising for generating states of two - dimensional turbulence in these systems . here we give an overview of techniques for generating the large and disordered vortex distributions associated with two - dimensional quantum turbulence . we focus on describing methods explored in our bose - einstein condensation laboratory , and discuss the suitability of these methods for studying various aspects of two - dimensional quantum turbulence . we also summarize some of the open questions regarding our own understanding of these mechanisms of two - dimensional quantum turbulence generation in condensates . we find that while these disordered distributions of vortices can be generated by a variety of techniques , further investigation is needed to identify methods for obtaining quasi - steady - state quantum turbulence in condensates .

1303.4764

dilute - gas bose - einstein condensates ( becs ) present unique opportunities for studying quantum fluid dynamics due to the wide range of experimental tools available for probing and manipulating condensates , and the quantitative accuracy available with theoretical and numerical approaches @xcite . experimentally , magnetic and laser fields can be arranged to create a wide variety of trapping configurations and geometries . superfluidity hallmarks such as persistent currents and quantized vortices @xcite may be precisely generated and their dynamics probed , and numerous measurement techniques allow for observations of density and quantum phase distributions . regarding simulations of superfluid dynamics in becs , common numerical approaches utilize the gross - pitaevskii equation ( gpe ) , a nonlinear schrdinger equation that incorporates a mean - field approximation to represent the bec with an order parameter , the familiar macroscopic condensate wavefunction @xcite . numerous variants of the gpe such as those that model finite - temperature environments extend the predictive accuracy of simulations @xcite . taken together , these features allow for highly synergetic experimental , numerical , and theoretical explorations of superfluid dynamics . in this respect , bose - einstein condensates may be ideally suited for investigating quantum turbulence and the dynamics of interacting quantized vortices in compressible quantum fluids ; for recent reviews of this subject see refs . @xcite and @xcite and the references therein . in superfluids , vorticity appears in the form of superfluid - free vortex cores about which there is quantized flow circulation @xcite . whether in superfluid helium or trapped becs , quantum turbulence is often phenomenologically described as a tangled distribution of these quantized vortices throughout the fluid @xcite , and recent work with superfluid helium provides fascinating visual evidence for these vortex distributions @xcite . in becs , quantum turbulence is relevant to early investigations on the formation of large vortex lattices @xcite and routes to vortex nucleation by dynamical instabilities @xcite , and has been experimentally observed and explicitly addressed in these contexts @xcite . more recent experiments have been principally dedicated to the study of quantum turbulence phenomena and the related vortex distributions @xcite . images of vortex tangles in becs have also been experimentally obtained @xcite . yet while becs readily permit vortex observations , other aspects of turbulence remain significant experimental challenges within the bec field . notably , direct measurement of kinetic energy spectra is an open problem , whereas such measurements of liquid helium have allowed direct comparison of classical and quantum fluid turbulent energy spectra @xcite . the different approaches and systems for studying quantum turbulence are therefore complementary , with bec research playing a potentially important role in understanding the broad subject of turbulence , particularly in the context of compressible quantum fluids . more specifically , becs are now beginning to contribute to the development of an understanding of _ two - dimensional _ ( 2d ) turbulence @xcite in compressible quantum fluids . in the basic manifestation of 2d classical turbulence ( 2dct ) , energy and enstrophy , the enstrophy @xmath0 is a measure of the vorticity @xmath1_z$ ] about the @xmath2 axis , where @xmath3 is the fluid velocity in the plane normal to @xmath2 . ] are injected into a fluid at a small length scale , and patches of vorticity merge and lead to the development of large - length - scale flow . this is opposite to the dynamics of the familiar energy cascade of three - dimensional ( 3d ) turbulence in a classical fluid , in which large - scale flows continuously decay to smaller - scale flows that are eventually dissipated by viscous damping . inverse _ energy cascade of 2dct has been studied for decades and numerous excellent reviews exist on 2dct @xcite . in order to study two - dimensional _ quantum _ turbulence ( 2dqt ) in a bec , one might envision subjecting the bec to continuous injection of energy and vorticity . dissipation of energy may also exist , perhaps by thermal damping that could induce vortices to exit the bec at its boundary , or perhaps due to vortex - antivortex annihilation . a balance between forcing and dissipation can be envisioned to lead to a statistically steady degree of turbulence in a bec . we might picture such a state as an approximately constant mean number of vortices in the bec , since enstrophy in a quantum fluid is proportional to the number of vortices @xcite . with the likelihood of future experimental methods for observing vortex dynamics and measuring energy spectra , we further imagine a scenario in which 2dqt can be experimentally and numerically studied , linked with new analytical approaches , and compared with 2dct phenomena . the prospects for the development of a new understanding of at least this one aspect of the exceptionally challenging topic of turbulence are tantalizing motivations for pursuit of 2dqt research . unfortunately there are a number of experimental challenges that must first be overcome . significantly , an isolated , trapped bec subjected to continuous injection of energy would eventually become depleted of atoms as the system rethermalized . it appears that a bec experimentalist s primary hope in this respect is to develop methods for studying 2dqt that are quasi - steady , where steady forcing and dissipation rates are balanced over finite timescales during which the bec is not significantly depleted of atoms . experimentally , the path to studying decaying 2dqt is perhaps much simpler , since this topic is more concerned with the dynamics of a bec after forcing has stopped . while it should not be surprising that disordered vortex distributions in a highly oblate bec are relatively simple to obtain by a variety of experimental techniques , as we discuss in this article , how best to obtain a quasi - steady state of forced 2dqt remains an open problem . we have begun to empirically tackle only the forcing aspect of this problem , although simulations are also beginning to address this issue @xcite . several of our methods for driving large numbers of vortices into highly oblate becs are described below . we generally associate these states with 2dqt in the sense that the disordered vortex distributions may be considered a 2d phenomenological equivalent of the vortex tangle that exists in 3d quantum turbulence . we do not provide any evidence that such disordered vortex distributions conform to particular kinetic energy spectra or display vortex dynamics that may be expected in fully developed 2dqt ; discussions of links between vortex distributions , vortex dynamics and aggregates , and kinetic energy spectra in bec 2dqt can be found elsewhere @xcite . we also note that in the highly oblate trapping limit , bec confinement in one spatial direction is much stronger than the trapping in the two orthogonal directions . this limit is not strictly necessary for studying 2dqt , but vortex excitations in the form of kelvin waves are suppressed and vortices can be accurately approximated as having point - like dynamics in a plane within this limit @xcite . in the following , we first describe some of the key phenomenological features of 2dct , and summarize only a few of the theoretical advances in 2dqt , some of which have been aimed at uncovering the similarities and differences between the classical and quantum systems . this article is not intended as a review of the state of theoretical understanding of 2dqt in becs ; for discussions of theoretical aspects of 2dqt we direct the reader to refs . @xcite and the references contained therein . the study of 2dqt , particularly in compressible quantum fluids such as becs , is still young and a full range of characteristics remains to be explored . our aim is to give an overview of the types of observations that may be desirable in 2dqt bec experiments , then to describe our experimental approach to creating and probing 2dqt in highly oblate becs . we present a range of tools available for exciting the bec using laser beams and magnetic fields . the bulk of the remainder of this article is devoted to describing various specific methods for driving vortices into a highly oblate bec , and we show example images of becs in such excited states . we conclude by briefly summarizing some of the open experimental challenges in generating and probing 2dqt in becs . from a phenomenological standpoint , lesieur describes classical turbulence of continuum flows as having non - deterministic flow details , rapid mixing relative to molecular diffusion , and flow characteristics distributed among and interacting over a wide range of length scales @xcite ; see also sommeria s similar characterization of turbulence @xcite . such a description encompasses 2d turbulence , where flow dynamics vary primarily over two spatial coordinates ( @xmath4 and @xmath5 ) ; flow in a third direction ( @xmath2 ) may exist uniformly throughout the @xmath4-@xmath5 plane or there may be negligible flow along @xmath2 , hence a 3d fluid system may display 2d turbulence . building from this general description , the further characteristics of 2d turbulence are markedly different from the characteristics of 3d turbulence . as noted earlier , one such difference is the direction of energy flux between length scales : in 2d flows , small - scale forcing can lead to the development of large - scale flows . given their disordered and unpredictable nature , and challenges in discerning details of flow dynamics , turbulent flows are often characterized by the statistical properties of the system such as the kinetic energy density @xmath6 over wavenumber @xmath7 , which describes the distribution of kinetic energy among the length scales of the system . in 1967 , kraichnan @xcite found the existence of two inertial ranges associated with distinct kinetic energy spectra power laws , joined by the wavenumber associated with energy and enstrophy forcing . the first range extends from the forcing length scale to larger length scales ( or smaller wavenumbers ) , and displays a kinetic energy spectrum @xmath8 . this spectral distribution has the same form discovered by kolmogorov in 1941 @xcite in the context of 3d fluids , but in 2d this spectrum corresponds to an inverse energy cascade . the second range extends from the forcing scale to smaller length scales ( or larger wavenumbers ) and the kinetic energy spectrum approximately corresponds to @xmath9 . this two - component spectrum is a classic feature of 2d turbulence , and is further described in reviews of 2d turbulence @xcite . physically the inverse energy cascade corresponds to the aggregation of vorticity such that small - scale forcing leads to the growth of large - scale vorticity until flows are of the order of the size of the system . at this scale dissipation of some form occurs , possibly due to frictional damping at the container s walls . prior to the development of 2d turbulence theory in the 1960s @xcite , onsager @xcite approached turbulent flow in 2d inviscid fluids by focusing on the dynamics of point - like centers of vorticity , or point vortices . a collection of point vortices may be constructed to approximately represent classical continuum flows , or may quite accurately describe vortex distributions in superfluids . in fact , in the same conference proceeding in which onsager described his theory of vortex states in 2d turbulent flows , he also first postulated the existence of quantized vorticity in superfluids , and implied that superfluids could be ideal testing grounds for his theoretical arguments of 2d turbulence @xcite . onsager argued that a system with a large number of vortices would maximize entropy when vortices of the same sense of circulation form aggregates , giving more room in a finite - volume phase space for the remainder of the vortex cores to distribute themselves and move with fewer constraints . onsager s predictions , which were the first hints of the existence of an inverse energy cascade , have yet to be directly confirmed experimentally in a superfluid . however , numerical simulations of 2dqt involving forcing in becs are indeed finding evidence for the existence of vortex aggregates that resemble onsager s predictions @xcite . based on the above descriptions of 2d turbulence , direct experimental challenges for the study of 2d turbulence in quantum fluids include : ( i ) reaching a quasi - steady state of forced 2dqt ; ( ii ) observation of the aggregation of like - circulation quantized vortices in a superfluid in a state of forced 2dqt ; and ( iii ) direct measurement of the kinetic energy spectra associated with such turbulent states . these challenges all depend on the existence of means to generate 2dqt , so we now turn to the subject of vortices in highly oblate becs , and then focus on describing methods to nucleate large disordered distributions of vortices in these compressible quantum fluids . a dilute - gas bec is often adequately described in terms of an order parameter , or macroscopic wavefunction , of the form @xcite @xmath10 where @xmath11 is the atomic density distribution of the bec and @xmath12 is the quantum phase profile . the dynamics of the wavefunction @xmath13 are governed in the mean - field approximation by the gross - pitaevski equation ( gpe ) @xmath14 where @xmath15 characterizes the interaction strength between atoms , @xmath16 is the s - wave scattering length of interatomic interactions , @xmath17 is the atomic mass , and @xmath18 is the potential due to the trap and other external perturbations . from a hydrodynamic perspective , a dilute - gas bec can behave as a tiny droplet of superfluid , where the local velocity of superfluid flow is proportional to the local gradient of the phase : @xmath19 consequently , the velocity field of a bec is irrotational , @xmath20 , except at singularities of the quantum phase @xmath21 . these singularities are the superfluid - free cores of the vortices , about which fluid circulation is quantized @xcite . a singly quantized vortex is associated with a 2@xmath22-phase winding about the core , and vortex positions and circulations can therefore be experimentally determined using atom interferometry techniques @xcite , or by sequential images that reveal the dynamics of the vortices @xcite . most single - shot images of becs , however , only reveal the location of the vortices and not their circulation . vortex cores in a bec have a diameter on the order of the healing length @xmath23 @xcite , which is typically sub - micron and much less than the wavelength of imaging light . release of the bec from the trap followed by ballistic expansion of the atom cloud enables the vortices to expand to diameters large enough to be optically resolvable , and this method is the standard approach to observing vortices in a bec . in a 3d trap , vortices may also bend with respect to the imaging axis @xcite , making them even more difficult to observe . in two dimensions however , these kelvin wave excitations are suppressed @xcite . as an example of the clarity that can be obtained in vortex detection , vortices in a highly oblate bec are shown in fig . [ vortexfig ] ; the vortices are the circular dark regions within the bec fluid , shown in grayscale with lightness of shade corresponding to integrated column density after ballistic expansion from the trap . in addition to isolated vortex cores , vortex dipoles may be found in highly oblate or 2d quantum fluids . these structures consist of two vortex cores of opposite circulation in close proximity . vortex aggregates , consisting of clusters of vortices of the same circulation , have also been experimentally observed in the form of multi - quantum vortex dipoles @xcite . in the context of 2dqt in a bec , however , to date only numerical simulations have observed vortex aggregates and their dynamics @xcite . in becs with a vector order parameter , other types of vortex structures may also be found ; see ref . @xcite for a summary of vortex experiments in becs from their initial study in 1999 through 2010 , including vortices in degenerate fermi gases and multi - component becs . in the hydrodynamic regime , where quantum pressure can be neglected , and with vortices that are spaced far enough apart from each other that their core shapes are only very weakly distorted , the gpe formalism can be formally linked to the navier - stokes equation that plays a key role in classical fluid dynamics @xcite . by incorporating into the gpe damping due to atomic scattering with thermal atoms , a reynold s number can also be defined @xcite , although this has yet to be tested experimentally ; see ref . @xcite for a discussion of the gpe navier - stokes equation correspondence . given this correspondence , it should not be surprising that the concepts of turbulence may be investigated using becs and the variants of the gpe . much like the vortex tangle that characterizes quantum turbulence in three dimensions @xcite , we would like to similarly physically characterize a state of 2dqt in a bec . we consider the simple picture of a disordered distribution of numerous vortex cores . such a distribution would satisfy the phenomenological criteria of turbulence stated earlier . first , a large disordered distribution of vortices would have non - deterministic and chaotic flow dynamics @xcite . second , numerical simulations have shown that such distributions can rapidly distribute vorticity @xcite or particles @xcite , even if these are initially concentrated in a small region , thereby satisfying the rapid mixing criterion of turbulence . finally , such a distribution of vortices can display a large range of inter - vortex separations with no characteristic length scale ; in contrast , a vortex lattice has a well - defined length scale of the spacing of the vortices in the lattice @xcite . we therefore interpret a disordered 2d distribution of numerous vortex cores as a phenomenological picture that corresponds to 2dqt . such a picture can be made more quantitative , for example by invoking the methodology and results of onsager @xcite or vortex position statistics , as has been investigated numerically @xcite . nevertheless , our general experimental goal is to create states that consist of many vortices distributed throughout highly oblate becs with no discernible or well - defined arrangement . as noted above , a primary factor distinguishing 2d and 3d turbulence is the suppression of significant flow dynamics in one spatial direction , relative to the flow dynamics in the plane normal to that direction . in the case of quantum turbulence , this corresponds to the suppression of kelvin waves @xcite . it may then be possible to study 2dqt in 3d becs , and it should be noted that quasi-2d becs @xcite or even highly oblate becs are not strictly needed for studies of 2dqt . on the other hand , even in highly oblate becs , vorticity can decay due to vortex - antivortex annihilation . it is therefore possible that enstrophy need not be conserved and that a direct energy cascade may appear , as has been recently noted in numerical simulations of decaying 2dqt @xcite . in a practical sense however , we expect that 2dqt is more readily achievable in highly oblate geometries . our experimental focus is therefore on generating large disordered distributions of vortices in highly oblate becs . radius of @xmath24 m at the location of the bec . the coils for the dc component of the top trap , represented as dark gray bars , are located above and below the bec cell . the coils for the ac component , represented as light gray bars , are located on either side of the cell normal to the @xmath4-axis , and the @xmath5-axis ( not shown ) . a lens below the cell collects light for the vertical imaging system.,width=384 ] to obtain the highly oblate @xmath25rb becs desired for our studies of vortex and 2dqt generation methods , we use a conventional magnetic time - averaged orbiting potential ( top ) trap @xcite overlaid with a 1090-nm red - detuned laser light sheet propagating along the @xmath4-axis as illustrated in and ( a ) . a cylindrical lens focuses the laser light in the @xmath2 direction only , with a waist at the position of the bec such that the laser has horizontal and vertical @xmath26 radii of @xmath27 . this red - detuned optical potential provides the additional axial ( @xmath2 ) confinement needed to flatten the bec while having a minimal effect on the radial ( @xmath28 ) confinement . we begin our experiments with evaporative cooling of atoms in the @xmath29 hyperfine state of @xmath25rb in the purely magnetic top trap . prior to the final evaporation stage , we ramp on the trapping laser to a power of @xmath30 w over 4 s , and then cool the system below the bec transition temperature @xmath31 nk in the combined magnetic and optical potential . the potential well that confines the bec is an axially symmetric harmonic trap with radial and axial trap frequencies of @xmath32 , @xmath33 . the becs formed in the combined magnetic and optical potential have up to @xmath34 atoms with a system temperature of @xmath35 nk and thomas - fermi radii of @xmath36 . the chemical potential is @xmath37 , and our becs are consequently far from the quasi-2d limit and well into the 3d regime . nevertheless , the @xmath38 aspect ratio generated by the tight axial confinement suppresses vortex bending and tilting , resulting in a system characterized by approximately 2d fluid dynamics @xcite . figures [ exp_fig2](b ) and [ exp_fig2](c ) show images of a condensate held in the combined magnetic and optical trap . for comparison , ( d ) is an _ in situ _ image of the bec in the purely magnetic top trap with an aspect ratio of @xmath39 . side views of the bec are obtained with phase contrast imaging along the horizontal ( @xmath5 ) imaging axis and views along the vertical ( @xmath2 ) direction are typically obtained with absorption imaging . vortex cores in our highly oblate becs have a diameter on the order of the healing length , @xmath40 m for our system . this is well below the resolution of our imaging system and it is currently beyond our ability to detect vortices with _ in situ _ images . in order to optically resolve the individual vortex cores of highly oblate becs , we first turn off the top trap and allow the bec to expand for 10 ms in the optical potential . the optical trapping potential is then removed , and the bec freely expands for an additional @xmath41 . during the entire expansion procedure , a separate magnetic field is applied with a gradient along the @xmath2 direction that cancels the downward gravitational force on the cloud . gives a representative absorption image after expansion , demonstrating that vortex cores are indeed resolved . in the remainder of this article , all images that show vortices are absorption images taken with the vertical imaging system after a total of @xmath42 ms of expansion . hz and @xmath43 hz , respectively . ( c ) bec initial state with the obstacle located at @xmath44 m relative to the bec center . ( d),(e ) the maximum repulsive potential energy of the obstacle is @xmath45 ( where @xmath46 is the bec chemical potential ) and is held constant during evaporative cooling . it is ramped down linearly as the trap translates ; relative to the trap center , the beam moves from position @xmath44 m to @xmath47 m over a variable sweep time @xmath48 . the bec is then held in the harmonic trap for a variable time @xmath49 prior to expansion and imaging . figure and caption from ref . [ 46 ] , t. w. neely , e. c. samson , a. s. bradley , m. j. davis , and b. p. anderson , phys . lett . * 104 * , 160401 ( 2010 ) . copyright 2010 by the american physical society . , width=432 ] in many of the experiments described in the following sections , we make use of an additional , focused blue - detuned gaussian laser beam with a variable power and @xmath26 beam radius , directed axially through the bec as illustrated in . this beam can be used to perturb and excite condensates through a variety of methods , which are described below . we now turn to various experiments and tests of experimental methods , including relevant reviews of previous work in our lab on driving vortices into highly oblate becs , and discussion of previously unpublished results on generation of 2dqt in highly oblate becs . as described in neely _ et al _ @xcite , our group observed the deterministic formation and dynamics of vortex dipoles in a bec by slightly stirring the bec with a laser beam . in this experiment , vortices were nucleated as the bec was pushed around and past a blue - detuned laser beam that served as an impenetrable obstacle . in the frame of the moving bec , the beam was swept through the middle of the bec , and the vortex dipoles formed in the wake of the laser beam . a vortex dipole consists of a vortex and an antivortex two vortices of opposite circulation and tends to exist and move as a single excitation with linear momentum as long as the two vortices remain far from other vortices or the boundaries of the superfluid . a vortex dipole can be generated from sound , hence a moving obstacle can excite a dipole in a compressible superfluid . similarly the vortices of a dipole can recombine , annihilating one another and generating a pulse of acoustic energy ; see ref . @xcite for a recent discussion of the role and consequences of compressibility in 2dqt . in this experiment , the relative motion between the bec and the blue - detuned laser beam occurred with a constant velocity , as illustrated in . becs were created in a highly oblate harmonic trap as described above in and shown in figs . [ dipoles_fig1](a ) and [ dipoles_fig1](b ) . the focused blue - detuned gaussian laser beam with a @xmath26 radius of @xmath50 m was directed axially through the bec 20 @xmath51 m to the left of the bec center as shown in ( c ) . a magnetic bias field was used to translate the harmonic trap minimum horizontally along the @xmath4-axis at a constant velocity until the laser beam was located 14 @xmath51 m to the right of the center of the bec . at the same time the intensity of the beam was ramped linearly from a maximum of @xmath52 down to @xmath53 as shown in figs . [ dipoles_fig1](d ) and [ dipoles_fig1](e ) . here @xmath37 is the chemical potential of the bec in the highly oblate harmonic trap . at the end of the sweep , the bec was held in the harmonic trap for varying hold times @xmath54 before expanding and imaging . m - square images . ( b ) 62-@xmath51m - square images from numerical calculations of the gpe obtained for conditions similar to the data of sequence ( a ) , but for a temperature of @xmath55 ( i.e. , no damping was used ) . the orbital period is @xmath56 1.2 s , and the apparent vortex core size is smaller in the simulations because we show in - trap numerical data . ( c ) black and dark gray small circles show average positions of the vortices @xmath57 and @xmath58 ( as fractions of the bec radius ) of each of the two vortices from 5 sequences of experimental data identical to that of sequence ( a ) . the larger circle around each average position point represents the standard deviation of the vortex positions at that specific hold time , and is calculated from the 5 images obtained at that time step . a continuous dipole trajectory from sequence ( b ) is rescaled to the thomas - fermi radius of the expanded experimental images , and superimposed as solid lines on the experimental data ; no further adjustments are made for this comparison . figure and caption adapted from ref . [ 46 ] , t. w. neely , e. c. samson , a. s. bradley , m. j. davis , and b. p. anderson , phys . lett . * 104 * , 160401 ( 2010 ) . copyright 2010 by the american physical society . ] neely _ et al _ found a critical velocity necessary for observing vortex dipole nucleation of 170 - 190 @xmath51 m/s . the repeatability and coherence of this process allowed the dynamics of the vortex dipole to be mapped using images of becs from multiple experimental runs , with each image taken at increasing values of @xmath54 . as shown in , the vortex dipole orbited the bec with a period of @xmath59 s. gpe simulations of the procedure provided visual and quantitative comparisons with the experimentally observed vortex dynamics ; see ( c ) . hall s group at amherst college has also observed a variety of vortex dipole dynamics , using vortices spontaneously trapped in a bec during the phase transition @xcite and a detection method that permits multiple imaging of vortex positions in a single bec @xcite . sweeping the bec past the barrier at velocities much faster than the first critical velocity generated multiply quantized vortex dipoles . at a coarse scale , multiply charged dipoles behave like pairs of highly charged vortices of opposite circulation . the loci of vorticity orbits the condensate just as in the case of a singly charged dipole , albeit with a shorter period indicative of faster fluid flow . a doubly charged vortex dipole , for example , orbited with a period of @xmath60 s. as shown in , at finer scales these loci of vorticity are actually aggregates of singly quantized vortices of the same circulation . both singly and multiply charged dipoles exhibited lifetimes up to many seconds indicating that vortex - antivortex recombination may be suppressed and that vortex dipoles may be meta - stable in highly oblate becs . the generation of acoustic energy and the nucleation of vortex dipoles due to a moving obstacle are associated with injection of kinetic energy ( both compressible and incompressible ) into the condensate . the nucleation of vortices is also associated with the injection of enstrophy into the fluid , with the net enstrophy being proportional to the number of vortices @xcite . we may view this basic laser sweep mechanism as a possible tool for the generation of 2dqt in a bec , however in the experiment described above , the deterministic dynamics of the vortex dipoles indicate that a state of 2dqt was not observed . the length scale associated with incompressible energy and enstrophy injection is most readily interpreted as the spatial separation between the vortices of the dipole at the instant of dipole nucleation ; see ref . @xcite for further discussion of this point . with continuous laser stirring within the bec , one may envision the generation of a 2dqt state , and indeed such a state has been observed as described in . see also the relevant simulations of refs . @xcite and @xcite that involve an obstacle repeatedly sweeping through a bec . whether or not a bec with many vortices and antivortices would generally show traits associated with an inverse energy cascade or a direct energy cascade may strongly depend on specific experimental circumstances . in particular , vortex - antivortex recombination is one mechanism for dissipation of enstrophy in 2d fluid flow and might inhibit the formation of an inverse energy cascade , especially for decaying 2dqt @xcite . the balance between vortex generation and vortex annihilation may influence observed characteristics of turbulence , such as energy spectra , kinetic energy flux , and aggregation of vorticity . finally , the vortex aggregates observed in this experiment are directly generated by the relative motion between the laser obstacle and the bec , and we do not interpret their deterministic formation as a consequence of 2d turbulence as proposed by onsager @xcite . nevertheless , their long lifetimes ( at least one orbital period ) and steady alignment with the tight ( @xmath2 ) trapping axis may be taken as an indication that vortex aggregates in a highly oblate bec can be long - lived , and vortex - antivortex annihilation rates may be slow enough to reach regimes of constant or increasing net enstrophy . in this sense , the observation of vortex aggregates in the experiment of neely _ et al _ gives hope for future experimental observation of vortex aggregates in 2dqt bec experiments . our top trap combines a dc quadrupole magnetic field having variable axial gradient @xmath61 with an ac bias field of magnitude b@xmath62 that rotates in the @xmath4-@xmath5 plane with frequency @xmath63 khz . the time - dependent bias field with @xmath4 and @xmath5 components @xmath64 results in the minimum of the quadrupole field orbiting in a circle within the @xmath4-@xmath5 plane . for an atom with magnetic dipole moment @xmath51 and mass @xmath17 , the time average of the combined magnetic fields results in a harmonic potential with trap frequencies determined by @xmath65 ^{1/2 } \\ \omega_\mathrm{z } = \bigg [ \frac{\mu { b_\mathrm{z}^\prime}^2}{m b_0 } ( 1-\eta^2)^{3/2 } \bigg ] ^{1/2}\end{aligned}\ ] ] where the factors involving @xmath66 account for the vertical sag due to gravity ( with acceleration @xmath67 ) of the top trap minimum from that of a static quadrupole field . such traps are now in common use ; see ref . @xcite for a detailed description of top traps . modulations of the magnetic bias field and hence the shape of the trapping potential can excite various modes in the bec . for example , by deforming the trapping potential and rotating the spatial deformation axis , surface waves can be excited that decay to numerous vortices @xcite . in this section , we describe the vortex distributions and their lifetimes that result when highly oblate becs are subjected to modulations of the magnetic component of the trapping field . in this experiment , we modulated the amplitude of the magnetic bias field @xmath68 in eqs . ( [ eqn6 ] ) and ( [ eqn7 ] ) by adding a small sinusoidal modulation to @xmath68 of the form @xmath69 here @xmath70 is the amplitude of the bias field modulation and @xmath71 is the frequency of the modulation . this results in a small sinusoidal oscillation of the radial trap frequency @xmath72 about the static value , but the trap remains axially symmetric . the axial trapping frequency @xmath73 is primarily determined by the optical potential and therefore remains approximately constant . our aim in this experiment was to determine if a weak symmetric modulation would drive vorticity into the bec such that there was approximately no net angular momentum transfer , i.e. , such that if any vortices were generated , approximately equal numbers of vortices of both circulations would be created . in our procedure , we began with a highly oblate bec in the combined magnetic and optical harmonic trap . we then modulated the amplitude of the rotating magnetic bias field at frequency @xmath74 and amplitude @xmath75 for varying modulation time @xmath76 . after the modulation we held the bec in the static harmonic trap for varying hold times @xmath54 , then expanded and imaged , looking for the presence of vortices . = 0.2 s to 1.5 s. vortex cores start to appear at @xmath76 = 0.4 s and and by @xmath76 = 0.6 s we see a disordered vortex distribution . ( b ) images taken for increasing hold times after modulating for @xmath76 = 1.5 s. ] \(a ) shows images taken at increasing modulation times from @xmath76 = 0.2 s to 1.5 s. vortex cores started to appear at @xmath76 = 0.4 s and by @xmath76 = 0.6 s we observed disordered vortex distributions . vortex cores appear to be generated at the outer edge of the condensate and make their way into the center . in the next step of the experiment , we modulated the trap as described for @xmath76= 1.5 s before removing the modulation and returning the bec to the unmodulated harmonic trap . we added increasing post - modulation hold times @xmath49 and then expanded and imaged the bec . as shown in ( b ) the vortex distribution became better resolved with increasing @xmath49 and decayed to states with only a few cores remaining after @xmath54 = 2 s. by @xmath54 = 8 s no vortex cores remained . the condensate was completely destroyed for @xmath76 = 10 s. in a similar experiment , we modulated the trapping potential of a highly oblate annular trap . to do this , we modulated the strength of the rotating bias field of the top trap in the same manner as described in . we added to the harmonic trap a focused blue - detuned gaussian laser beam , directed axially through the bec in the configuration shown in . example images of such a bec are shown in figs . [ modb_fig2](a ) and [ modb_fig2](b ) . prior to imaging , the blue - detuned laser barrier was ramped off slowly enough to avoid significantly perturbing the condensate , allowing the bec to expand from a harmonic trap for imaging . with @xmath77 and @xmath78 , numerous vortices were quickly nucleated over just a few ms , a significant change from the harmonic trap . as shown in ( c ) vortex cores were observed to be distributed throughout the bec , with the first cores appearing at least an order of magnitude earlier than in the purely harmonic trap . compared to the results of , many more cores were visible using this modulation technique in the annular bec , and the cores have higher contrast . we speculate that the presence of the blue - detuned barrier acted as a vortex dipole nucleation site located within the bec , and that the vortex dipoles quickly dissociated and formed a state of 2dqt . again , the modulation process was designed to be axially symmetric apart from experimental errors in beam positioning and trap roundness , and should have nominally added no net angular momentum to the condensate . = 0 s to show that the hole created by the blue - detuned beam has completely filled between beam ramp down and imaging . for the annular trapping geometry vortices appear after much shorter modulation times than for the harmonic trapping case . ] with additional modulation time , the bec was driven into a state of high excitation in which individual vortex cores were not easily resolvable or countable . such a state is shown in the rightmost image of ( c ) . this process shows that too much modulation drives the bec into a highly excited far - from - equilibrium state , a signature that it may be possible to identify a weaker modulation strength such that vortex nucleation rates and vortex decay rates due to annihilation and thermal damping at the system boundary would be balanced for a given temperature . finding this optimum balance is a goal of future research . the straightforward experimental method of expansion imaging used in our experiments does not easily permit measurement of the circulation of vortices . while the interferometric or dynamic methods mentioned above can be used for this purpose , these have not yet been applied to turbulent states of becs . in order to look for the clustering of vortex cores of like - circulation , an alternative is to examine statistics of vortex distributions containing only large numbers of cores of identical circulation . one possible approach to generate such distributions is by rotating the trapping potential , as has been utilized in numerous experiments of becs with different aspect ratios , such as refs . @xcite ; see also ref . @xcite for an overview of numerous other experiments that utilized this technique . our approach follows such previous work , and is implemented by squeezing and rotating the magnetic trapping field as first implemented by hodby _ _ et al__@xcite . in order to rotate and add angular momentum to the atomic cloud it is necessary to break the symmetry of the trapping potential . in our case this is done by adding an additional bias field to the one described by eqs . ( [ eqn6 ] ) and ( [ eqn7 ] ) so that now @xmath79 where @xmath80 , and @xmath71 is the frequency of the modulating bias field . if @xmath81 the bias field causes the time - averaged trap to have stationary ellipsoidal potential energy surfaces in the @xmath4-@xmath5 plane with the ratio of the minor and major axes determined by @xmath82 and @xmath68 ; for our case this ratio is @xmath60 . increasing @xmath82 with respect to @xmath68 increases this ellipticity . if @xmath83 the ellipse rotates in the @xmath4-@xmath5 plane with frequency @xmath84 . in the top trap , choosing @xmath85 excites the quadrupole mode which decays to a collection of quantized vortex cores of the same circulation . the minimum energy configuration of the rotating bec consists of a large regular lattice of vortices @xcite . a representative vortex lattice formed by spinning the bec in our purely magnetic top trap ( i.e. , without the red - detuned trapping potential ) is shown in . our investigations of highly oblate becs subjected to trap deformation and rotation started with a bec in the combined magnetic and optical harmonic trap . we then applied the magnetic field ellipticity and spun in the highly oblate harmonic trap for time @xmath86 at frequency @xmath87 with ellipticity @xmath82 . after spinning we returned to the symmetric harmonic trap and held for time @xmath54 while the bec shape deformations damped out . finally , we let the condensate expand and we imaged the cloud . hz for a variable spin time @xmath88 and then holding for 2.5 s after the spin prior to expansion . open circles represent a single set of data taken with @xmath89 , gray squares represent @xmath90 , and black triangles represent @xmath91 . both the peak number of vortex cores and @xmath88 corresponding to the peak number of vortex cores seem to be dependent on ellipticity but all three ellipticities show decreasing vortex number for longer spin times . ] by varying spin frequencies , we found a resonance condition for generating large numbers of vortices at @xmath92 hz . here @xmath93 is consistent with previous observations of exciting the quadrupole mode in 3d harmonic traps at @xmath85 @xcite . we measured the number of vortex cores versus spin time for different ellipticities at this resonance frequency . shows three sets of data corresponding to ellipticities of @xmath94 , @xmath95 and @xmath96 . all data points were generated by spinning for a variable spin time @xmath88 and then holding for @xmath97 = 2.5 s after the spin and before expansion . both the peak number of vortex cores and @xmath88 corresponding to the peak number of vortex cores appear to be dependent on ellipticity with peak vortex number occurring at a much shorter spin time for @xmath91 and @xmath98 than for @xmath89 . all three ellipticities show decreased vortex numbers for longer spin times on the order of @xmath88= 8 s indicating that vortex cores may be annihilating or leaving the system before crystalizing into a lattice . m - square expansion images of an oblate bec for varying @xmath86 ( indicated above images ) at increments of 100 ms . spin frequency and ellipticity were held constant at @xmath99 hz and @xmath100 . ( a ) and ( c ) becs were expanded and imaged directly after spinning , @xmath54 = 0 . ( b ) and ( d ) becs were held for 400 ms after spinning , @xmath54 = 400 ms , then expanded and imaged . ] visually tracks the evolution of the condensate during spinning . figures [ spin_fig2](a ) and [ spin_fig2](c ) show successive images at 100 ms intervals for @xmath86 = 0.1 to 1.0 s. the becs were imaged directly after spinning with no hold in the axially symmetric harmonic trap ( @xmath101 ) prior to expansion . figures [ spin_fig2](b ) and [ spin_fig2](d ) show successive images at 100 ms intervals for @xmath86 = 0.1 to 1.0 s with an additional 400 ms hold time in the axially symmetric harmonic trap prior to expansion and imaging . the hold time in the symmetric harmonic trap seems important for the nucleation of vortices as vortex cores appear after @xmath88 = 400 ms and @xmath97 = 400 ms , but do not appear until @xmath88 = 1.0 s for the case of @xmath97 = 0 . m - square expansion images of vortex distributions in the highly oblate bec for varying @xmath54 . spin time , frequency , and ellipticity were held constant at @xmath86 = 1.0 s , @xmath99 hz and @xmath100 . images were taken for variable @xmath54 from 0 to 2.5 s. multiple images for @xmath102 s are shown to give a representative sample of the variation in vortex distribution . ] m - square expansion images of vortex distributions in the highly oblate bec for the varying hold times @xmath54 shown from 3 to 8 s. by this time , bec shape deformations have largely damped out . spin time , frequency , and ellipticity were held constant at @xmath86 = 1.0 s , @xmath99 hz and @xmath100 . ] figures [ spin_fig3 ] and [ spin_fig4 ] show additional distributions of vortex cores nucleated by spinning and relaxing over subsequent hold times . here @xmath103 is held constant and the becs are imaged after increasing values of @xmath97 . vortices appear to be nucleated on the outer edge of the bec . presumably there is net angular momentum added to the condensate , so we expect the vortex cores to be predominately of the same circulation , at least once the system has relaxed to a meta - stable configuration . however , it may be the case that the net circulation is large , but that numerous vortices of the opposite circulation are also generated and present in these images , at least prior to the point at which a disordered distribution is uniformly distributed throughout the bec . if vortices are indeed of the same circulation in images such as the ones in the bottom row of , then such states may be candidates for experimental measurement of vortex power - law distributions in a bec , as has been analytically described for homogeneous becs @xcite . as observed , large , disordered vortex distributions can be generated by exciting collective modes of the bec . the shape deformations damp more quickly than the number of vortex cores , leaving open the possibility of finding parameters where studies of decaying 2dqt could be performed before the system spins down to a state with no vortices . spinning introduces a net angular momentum into the condensate and should result predominately in vortex cores of the same circulation , making this a possible system in which to observe vortex aggregates without time - resolved dynamics measurements . nevertheless , this vortex excitation technique does not appear to satisfy the particular goal of continuous forcing , although it appears to be a candidate for studies of decaying 2dqt . it is interesting to note that even for long hold times we find no evidence of a vortex lattice in the highly oblate bec . thermalization times for a lattice in a highly oblate bec may be beyond the lifetime of our becs ; see ref . @xcite for further discussion of this issue . in the experiments described in this section , we investigated the response of a bec to time - dependent perturbations of the intensity of a focused blue - detuned laser beam that pierced the bec . in all cases in this section , the relative position between the beam and the bec was stationary . our aim in these experiments was to locally excite the bec as an empirical probe of the existence of thermal counterflow @xcite in a region where the bec was locally depleted of atoms . while vortices were observed in all methods examined , the mechanisms for vortex nucleation remain unclear and merit further experimental and numerical investigation . in this experiment we nucleated vortices by subjecting the bec to a short pulse of blue - detuned laser light . we began by forming a highly oblate bec in the purely harmonic trap . we then instantaneously turned on a blue - detuned gaussian beam for a pulse time of 7 ms , short compared to the radial harmonic oscillator period of 125 ms . the focused blue - detuned gaussian beam had a @xmath26 radius of 10 @xmath51 m and was directed axially through the center of the condensate , as shown in . the power in the blue - detuned beam was chosen such that the optical potential generated by the beam was approximately equal to the chemical potential of the bec . after the laser pulse we held the bec in the harmonic trap for varying hold times @xmath54 before expansion and imaging . ( a ) shows an _ in situ _ absorption image of the bec with the blue - detuned beam on , clearly penetrating the condensate . the laser pulse generated a shock wave that can be observed as the central high density region in the first image in ( b ) . subsequent images show the condensate for longer hold times . vortices appear to enter the condensate from the outer boundary as the shock wave propagates . vortices eventually leave the system , most likely through vortex - antivortex recombination and thermal damping . apart from the shock wave , this perturbation created vortices with little residual excitation of the bec . @xcite performed experimental and numerical studies of the dispersive shockwaves generated when a bec was subjected to a short pulse of laser light , however their experiments were performed with a prolate bec and they do not report observations of vortex cores generated as a result of the laser pulse . in this experiment we formed a highly oblate bec in the purely harmonic trap , then turned on and sinusoidally modulated the intensity of a focused blue - detuned gaussian beam for a varying time @xmath76 . after the modulation we held the bec in the purely harmonic trap for varying time @xmath54 , then expanded and imaged . the blue - detuned beam had a @xmath26 radius of 10 @xmath51 m and was directed axially through the center of the condensate , as shown in . during the modulation time the optical potential @xmath104 generated by the beam followed @xmath105 for @xmath106 , with maximum repulsive potential energy @xmath107 , and frequency @xmath71 . for all other times , the beam was turned completely off . we fixed the modulation time at integer multiples of the modulation period , @xmath108 , such that the intensity of the blue - detuned beam always started and ended at zero , and we did not have to be concerned with ramping off the blue - detuned beam for imaging . as in the laser pulse technique , the blue - detuned beam acted as a perturbation to the confining potential . shows a sequence of images for varying values of @xmath54 after modulating for @xmath109 ms , with @xmath110 ( where @xmath37 ) , and frequency @xmath111 hz . this was approximately twice the radial trap frequency . an _ in situ _ image of the bec in the harmonic trap with the blue - detuned beam aligned in the center is shown in the leftmost image of . m - square absorption images acquired after the hold times @xmath54 indicated . the condensate undergoes large breathing oscillations in the trap , and this oscillation leads to a periodic variation in expanded bec radius between @xmath112 m and @xmath113 m . ] m - square absorption images acquired after the hold times @xmath54 indicated . the first image is an _ in situ _ image of the bec in the harmonic trap with the blue - detuned beam partially penetrating the condensate ; note that the optical potential strength is well below the bec chemical potential . ] while this method generated a large number of vortex cores , we also observed large - scale breathing oscillations in the radial dimension of the condensate . the radius of the expanded cloud oscillated between @xmath56 40 @xmath51 m and @xmath56 180 @xmath51 m with a period of @xmath56 70 ms and eventually damped out with an exponential decay time constant @xmath114 500 ms . as shown in , a disordered distribution of vortex cores remained in the condensate even after the bulk oscillations subsided , with @xmath115 cores remaining for @xmath54 = 2 s. it is not entirely surprising that we induced bulk fluid oscillations , given that we were forcing the bec at a frequency that was twice the radial trapping frequency . however , further study is needed for positively identifying the vortex generation mechanism in this experiment . m - square absorption images . the top row of images were taken with the bec in the harmonic trap and the maximum height of the blue - detuned potential at @xmath110 . the locations of the laser beam correspond to the positions designated by the arrow . the bottom row of images were acquired after @xmath54 = 40 ms followed by the expansion procedure . note the correlation between beam position and the position of the vortex cores in each vertical pair of images . ] m - square absorption images . ( a ) taken with the bec in the harmonic trap and the blue - detuned potential on at a strength of @xmath116 . in ( b ) and ( c ) , expansion images were taken after @xmath54 = 40 ms . for these images @xmath110 . note that the vortex cores are localized along the long axis of the blue - detuned beam and that they appear to be nucleating in pairs . ] in the sequence of images shown in , it is unclear where vortex nucleation occurs . to further explore the nucleation mechanism , we aligned the blue - detuned beam near the edge of the bec , and modulated for @xmath117 ms , equivalent to one sinusoidal pulse of the laser light , with @xmath110 , and frequency @xmath111 hz . after modulation , we held the bec in the purely harmonic trap for @xmath118 ms , then expanded and imaged the bec . as shown in , the vortex cores that resulted from the modulation appear to form near the location of the focused laser beam . m - square absorption images acquired after the hold times @xmath54 indicated . the condensate undergoes bulk oscillations that are out of phase in @xmath4 and @xmath5 . vortex cores appear to nucleate at the location of the beam and then move throughout the bec . ] lastly , we replaced the axially symmetric @xmath119 focused blue - detuned gaussian beam with a blue - detuned light sheet , focused along the @xmath4-axis with a @xmath26 radius @xmath120 m , spatially extended along the @xmath5 direction , and directed axially ( along @xmath2 ) through the center of the bec . the width of the beam along the @xmath5-axis was much larger than the diameter of the condensate so that beam extended beyond the edge of the condensate in the @xmath5-direction . ( a ) is a vertical absorption image of the unexpanded highly oblate bec with the elongated blue - detuned beam partially penetrating the condensate . we modulated at @xmath111 hz for @xmath76 = 62.5 ms and @xmath110 , held for @xmath54 , then expanded and imaged the bec . figures [ 660mod_fig4](b ) and [ 660mod_fig4](c ) were taken after @xmath54 = 40 ms . here the vortices appear to be nucleated along the long axis of the elongated beam . the vortex cores shown in these images are not completely resolved but we suspect that the vortices are being nucleated as dipoles in a similar manner to the breakdown of a soliton due to the snake instability in a bec@xcite . alternatively , these features may be acoustic precursors to vortex dipole formation @xcite . as with the axially symmetric blue - detuned potential , we observed bulk excitations in the condensate . shows a sequence of images taken with @xmath111 hz , @xmath76 = 62.5 ms , and @xmath116 , for varying hold times @xmath54 . the oscillations in @xmath4 and @xmath5 are now out of phase by @xmath121 . vortex cores appear to be generated at the location of the beam and move to the outer boundary . in particular the images corresponding to @xmath122 ms show vortices aligned with the long axis of the beam , but by @xmath54 = 50 ms the vortices are located along the outer boundary of the condensate with no cores in the center . later images show more disordered distributions of cores . again , the mechanism for nucleation is not clear . in all of these methods , intensity modulations of a blue - detuned laser beam were observed to be effective for nucleating vortices in a bec . although the mechanisms for vortex nucleation remain unclear , we observe correlations between the position of the laser beam and the site of vortex generation within the bec . with careful parameter selection , it may be possible to use this method for controlled vortex generation rates . however , one must also be careful not to significantly excite shape oscillations , as these would make studies of 2dqt difficult . perhaps with further adjustment of the modulation rates and times , shape oscillations could be minimized , and methods of this sort could be utilized for 2dqt studies . rather than using intensity modulation of a blue - detuned laser beam , the experiment described in neely _ et al _ @xcite used small - scale stirring of the bec with a blue - detuned gaussian beam to generate disordered vortex distributions in a highly oblate annular trapping potential . it was also observed that vortices coalesced into large - scale flow in an annular trap , but here we review only the vortex distributions observed via stirring . lighter grayscale shades indicate larger column densities . ( e ) illustration of stirring , the arrow shows the trajectory of the harmonic trap center relative to the larger fluid - free region created by the laser barrier . ( f ) _ in situ _ image of the bec 10 s after stirring ; vortices are not observable , necessitating an expansion stage to resolve them . ] this experiment was performed with the bec in an annular trap created with a focused blue - detuned gaussian beam , with a @xmath26 radius of 23 @xmath51 m , directed axially through the center of the bec . as shown in the blue - detuned beam penetrated the bec with a barrier height of @xmath123 where @xmath37 . prior to stirring , the bec was held at a temperature @xmath124 , where @xmath125 nk was the critical temperature for the bec phase transition . at @xmath126 magnetic bias coils were used to move the center of the harmonic trap in a 5.7-@xmath51m - diameter , off - center circle about the stationary , blue - detuned barrier . at @xmath127 s , at the end of the stirring motion , the center of the harmonic trap again coincided with the blue - detuned beam , and the bec was held in this annular trap for varying hold times up to @xmath128 s. at @xmath54 = 1.17 s the temperature of the bec was reduced to @xmath129 to decrease thermal damping rates and vortex - antivortex recombination . at the end of the hold time the blue - detuned beam was ramped off over 250 ms and the bec was expanded and imaged . m - square experimental column - density absorption images acquired at the hold times @xmath54 indicated above the images . each bec undergoes ballistic expansion immediately after the central barrier ramp - down in order to resolve the vortex cores . each image is acquired from a separate experimental run . ] experimental images acquired at varying hold times are shown in . immediately after stirring , a large disordered distribution of vortex cores was observed . by @xmath54 = 0.33 s , vortices had begun to coalesce on the central potential barrier , as indicated by the large - vorticity hole in the center of the images . large - scale superflow developed over increasing hold times , as indicated by the growing large area of vorticity at the center of the bec , and was observed to persist up to 50 s. the initial vortex distribution generated by the stirring varied significantly from shot to shot but consistently evolved into a large - scale flow . the damped gpe was also used to numerically simulate the bec stirring and subsequent dynamics ; see ref . @xcite for details . in addition to the evolution of the large - scale flow observed in both experiment and numerics , analysis of the numerical simulations yielded two other results indicative of 2d turbulence . first , coherent vortex structures consisting of two cores of the same circulation intermittently formed and lasted for long timescales , tantalizing evidence for onasger - type vortex aggregates . second , analysis of the incompressible kinetic energy spectrum gave @xmath8 for @xmath130 and @xmath9 for @xmath131 where @xmath132 is the wavenumber associated with injection of kinetic energy into the bec . in this case the injection mechanism was energy transfer between the sound field associated with the compressible component of the fluid excited by stirring , and the vorticity field associated with the incompressible component . further relationships between stirring mechanisms and observed vortex distributions and dynamics are found in refs . @xcite and @xcite . with the development of new experimental techniques , we hope that such information will soon also be experimentally discernible . as it stands , such small - scale stirring with a blue - detuned laser beam appears to be a promising mechanism for further studies of 2dqt , perhaps even with multiple beam sites or with longer stirring times . we have found that it is relatively straightforward to nucleate large , disordered distributions of vortex cores in highly oblate becs , that is , to generate 2dqt in a bec . we interpret the disordered nature of these distributions as analogous to the vortex tangle characteristic of 3d quantum turbulence . many of the vortex distributions that we have observed are long - lived , making them potentially useful for studies of 2dqt . however , methods such as modulating a blue - detuned beam in the center of the bec excite bulk modes of the fluid where the entire condensate undergoes shape oscillations . in many cases , the vortex nucleation mechanisms are not completely clear . possible mechanisms for vortex generation in the laser modulation experiments include counterflow between the thermal and superfluid components @xcite , vacuum bubble cavitation as described by berloff and barenghi @xcite , or perhaps other nonlinear fluid dynamics processes at the boundaries of the quantum fluid . while further numerical and experimental investigations are necessary to determine the origins of the vortices , we envision many of these highly disordered vortex distributions as starting points for further studies of 2dqt . although we have described a number of vortex generation mechanisms , a primary challenge for experimentalists is to characterize 2dqt in a bec , and the characteristics observed may well depend on the vortex generation mechanism used . experimental observation of an inverse energy cascade , vortex aggregation , and kinetic energy spectra are of primary interest . such characteristics might or might not appear via the methods discussed in this article ; this remains to be determined . vortex dynamics and circulation measurements will likely prove to be among the most significant developments remaining to be developed for 2dqt studies in a bec . understanding the temperature dependence , and more generally the role of dissipation , are also of high importance . finally , reaching a regime of quasi - steady - state 2dqt is a highly desired goal , one that may well be reached utilizing methods described in this article , presumably with a suitable balance of dissipation . although many of the methods we describe in this article represent only the beginnings of investigations into new methods for the study of 2dqt , we hope that the descriptions of our observations so far inspire the development of new experimental and numerical studies of vortex generation mechanisms . further experimental and theoretical work in the field of 2dqt promises to provide an exciting new research direction for compressible quantum fluids . + + we gratefully acknowledge the support of the us national science foundation grant phy-0855467 . kew gratefully acknowledges support from the department of energy office of science graduate fellowship program ( doe scgf ) , made possible in part by the american recovery and reinvestment act of 2009 , administered by orise - orau under contract no . de - ac05 - 06or23100 . we also thank ashton bradley and ewan wright for helpful discussions . b. blakie , a. s. bradley , m. j. davis , r. j. ballagh , and c. w. gardiner , dynamics and statistical mechanics of ultra - cold bose gases using c - field techniques , _ adv . in phys . _ * 57 * 0 ( 5 ) , 0 363 ( 2008 ) . j. a. seman , e. a. l. henn , r. f. shiozaki , g. roati , f. j. poveda - cuevas , k. m. f. magalhes , v. i. yukalov , m. tsubota , m. kobayashi , k. kasamatsu , and v. s. bagnato , route to turbulence in a trapped bose - einstein condensate , _ las . * 8 * 0 ( 9 ) , 0 691 ( 2011 ) . m. caracanhas , a. l. fetter , s. r. muniz , k. m. f. magalhes , g. roati , g. bagnato , and v. s. bagnato , self - similar expansion of the density profile in a turbulent bose - einstein condensate , _ j. low temp . phys . _ * 166 * 0 ( 1 - 2 ) , 0 49 ( 2012 ) . j. sommeria , _ two - dimensional turbulence _ , in eds . m. lesieur , a. yaglom , and f. david , _ new trends in turbulence _ , les houches - ecole dete de physique theorique . edp sciences - springer , berlin ( 2001 ) . s. j. rooney , p. b. blakie , b. p. anderson , and a. s. bradley , suppression of kelvon - induced decay of quantized vortices in oblate bose - einstein condensates , _ phys . rev . a_. * 84 * 0 ( 2 ) , 0 023637 ( 2011 ) . a. n. kolmogorov , the local structure of turbulence in incompressible viscous fluid for very large reynolds numbers , _ dokl . nauk . sssr_. * 30 * , 0 301 ( 1941 ) . reprinted and translated in : _ proc . london a _ * 434 * ( 1890 ) , 9 , ( 1991 ) . s. inouye , s. gupta , t. rosenband , a. p. chikkatur , a. grlitz , t. l. gustavson , a. e. leanhardt , d. e. pritchard , and w. ketterle , observation of vortex phase singularities in bose - einstein condensates , _ phys . lett . _ * 87 * 0 ( 8) , 0 080402 ( 2001 ) . d. v. freilich , d. m. bianchi , a. m. kaufman , t. k. langin , and d. s. hall , real - time dynamics of single vortex lines and vortex dipoles in a bose - einstein condensate , _ science_. * 329 * 0 ( 5996 ) , 0 1182 ( 2010 ) . s. middelkamp , p. j. torres , p. g. kevrekidis , d. j. frantzeskakis , r. carretero - gonzlez , p. schmelcher , d. v. freilich , and d. s. hall , guiding - center dynamics of vortex dipoles in bose - einstein condensates , _ phys . rev . a_. * 84 * 0 ( 1 ) , 0 011605(r ) ( 2011 ) . v. bretin , p. rosenbusch , f. chevy , g. v. shlyapnikov , and j. dalibard , quadrupole oscillation of a single - vortex bose - einstein condensate : evidence for kelvin modes , _ phys . lett . _ * 90 * 0 ( 10 ) , 0 100403 ( 2003 ) . e. hodby , g. hechenblaikner , s. a. hopkins , o. m. marag , and c. j. foot , vortex nucleation in bose - einstein condensates in an oblate , purely magnetic potential , _ phys . lett . _ * 88 * 0 ( 1 ) , 0 010405 ( 2001 ) . t. m. wright , r. g. ballagh , a. s. bradley , p. b. blakie , and c. w. gardiner , dynamical thermalization and vortex formation in stirred two - dimensional bose - einstein condensates , _ phys . rev . a_. * 78 * 0 ( 6 ) , 0 063601 ( 2008 ) . m. a. hoefer , m. j. ablowitz , i. coddington , e. a. cornell , p. engels , and v. schweikhard , dispersive and classical shock waves in bose - einstein condensates and gas dynamics , _ phys . rev . a_. * 74 * 0 ( 2 ) , 0 023623 ( 2006 ) . b. p. anderson , p. c. haljan , c. a. regal , d. l. feder , l. a. collins , c. w. clark , and e. a. cornell , watching dark solitons decay into vortex rings in a bose - einstein condensate , _ phys . rev . lett . _ * 86 * 0 ( 14 ) , 0 2926 ( 2001 ) .

bicategories of spans are characterized as cartesian bicategories in which every comonad has an eilenberg - moore object and every left adjoint arrow is comonadic .

0910.2996

let @xmath0 be a category with finite limits . for the bicategory @xmath1 , the locally full subbicategory @xmath2 determined by the left adjoint arrows is essentially locally discrete , meaning that each hom category @xmath3 is an equivalence relation , and so is equivalent to a discrete category . indeed , a span @xmath4 has a right adjoint if and only if @xmath5 is invertible . the functors @xmath6 provide equivalences of categories which are the effects on homs for a biequivalence @xmath7 since @xmath0 has finite products , @xmath2 has finite products _ as a bicategory_. we refer the reader to @xcite for a thorough treatment of bicategories with finite products . each hom category @xmath8 is _ isomorphic _ to the slice category @xmath9 which has binary products given by pullback in @xmath0 and terminal object @xmath10 . thus @xmath1 is a _ precartesian bicategory _ in the sense of @xcite . the canonical lax monoidal structure @xmath11 for this precartesian bicategory is seen to have its binary aspect given on arrows by @xmath12 and its nullary aspect provided by @xmath13 the terminal object of @xmath14 . both of these lax functors are readily seen to be pseudofunctors so that @xmath1 is a _ cartesian bicategory _ as in @xcite . the purpose of this paper is to characterize those cartesian bicategories @xmath15 which are biequivalent to @xmath1 , for some category @xmath0 with finite limits . certain aspects of a solution to the problem are immediate . a biequivalence @xmath16 provides @xmath17 so that we must ensure firstly that @xmath18 is essentially locally discrete . from the characterization of bicategories of relations as locally ordered cartesian bicategories in @xcite one suspects that the following axiom will figure prominently in providing essential local discreteness for @xmath18 . [ frob]_frobenius:_a cartesian bicategory @xmath15 is said to satisfy the _ frobenius _ axiom if , for each @xmath19 in @xmath15 , @xmath19 is frobenius . indeed frobenius objects in cartesian bicategories were defined and studied in @xcite where amongst other things it is shown that if @xmath19 is frobenius in cartesian @xmath15 then , for all @xmath20 , @xmath21 is a groupoid . ( this theorem was generalized considerably in @xcite which explained further aspects of the frobenius concept . ) however , essential local discreteness for @xmath18 requires also that the @xmath21 be ordered sets ( which is automatic for locally ordered @xmath15 ) . here we study also _ separable _ objects in cartesian bicategories for which we are able to show that if @xmath19 is separable in cartesian @xmath15 then , for all @xmath20 , @xmath21 is an ordered set and a candidate axiom is : [ sepax]_separable:_a cartesian bicategory @xmath15 is said to satisfy the _ separable _ axiom if , for each @xmath19 in @xmath15 , @xmath19 is separable . in addition to essential local discreteness , it is clear that we will need an axiom which provides _ tabulation _ of each arrow of @xmath15 by a span of maps . since existence of eilenberg - moore objects is a basic 2-dimensional limit concept , we will express tabulation in terms of this requirement ; we note that existence of pullbacks in @xmath18 follows easily from tabulation . in the bicategory @xmath1 , the comonads @xmath22 are precisely the symmetric spans @xmath23 ; the map @xmath24 together with @xmath25 provides an eilenberg - moore coalgebra for @xmath23 . we will posit : [ emc]_eilenberg - moore for comonads:_each comonad @xmath26 in @xmath15 has an eilenberg - moore object . conversely , any map ( left adjoint ) @xmath24 in @xmath1 provides an eilenberg - moore object for the comonad @xmath27 . we further posit : [ mc]_maps are comonadic:_each left adjoint @xmath24 in @xmath15 is comonadic . from which , in our context , we can also deduce the frobenius and separable axioms . in fact we shall also give , in proposition [ easy ] below , a straightforward proof that @xmath18 is locally essentially discrete whenever axiom [ mc ] holds . but due to the importance of the frobenius and separability conditions in other contexts , we have chosen to analyze them in their own right . we recall from @xcite that a bicategory @xmath15 ( always , for convenience , assumed to be normal ) is said to be _ cartesian _ if the subbicategory of maps ( by which we mean left adjoint arrows ) , @xmath28 , has finite products @xmath29 and @xmath30 ; each hom - category @xmath31 has finite products @xmath32 and @xmath33 ; and a certain derived tensor product @xmath34 and @xmath35 on @xmath15 , extending the product structure of @xmath36 , is functorial . as in @xcite , we write @xmath37 and @xmath38 for the first and second projections at the global level , and similarly @xmath39 and @xmath40 for the projections at the local level . if @xmath41 is a map of @xmath15 an arrow of @xmath36 we will write @xmath42 for a chosen adjunction in @xmath15 that makes it so . it was shown that the derived tensor product of a cartesian bicategory underlies a symmetric monoidal bicategory structure . we recall too that in @xcite frobenius objects in a general cartesian bicategory were defined and studied . we will need the central results of that paper too . throughout this paper , @xmath15 is assumed to be a cartesian bicategory . as in @xcite we write @xmath43 \efig\ ] ] for the grothendieck span corresponding to @xmath44 where @xmath45 is the inclusion . a typical arrow of @xmath46 , @xmath47 can be depicted by a square @xmath48 \morphism(125,250)|m|<250,0>[`;\alpha ] \efig\ ] ] and such arrows are composed by pasting . a 2-cell @xmath49 in @xmath46 is a pair of 2-cells @xmath50 , @xmath51 in @xmath36 which satisfy the obvious equation . the ( strict ) pseudofunctors @xmath52 and @xmath53 should be regarded as _ domain _ and _ codomain _ respectively . thus , applied to ( [ square ] ) , @xmath52 gives @xmath41 and @xmath53 gives @xmath54 . the bicategory @xmath46 also has finite products , which are given on objects by @xmath34 and @xmath35 ; these are preserved by @xmath52 and @xmath53 . the grothendieck span can also be thought of as giving a double category ( of a suitably weak flavour ) , although we shall not emphasize that point of view . [ [ xredux ] ] the arrows of @xmath46 are particularly well suited to relating the various product structures in a cartesian bicategory . in 3.31 of @xcite it was shown that the local binary product , for @xmath55 , can be recovered to within isomorphism from the defined tensor product by @xmath56 a slightly more precise version of this is that the mate of the isomorphism above , with respect to the single adjunction @xmath57 , defines an arrow in @xmath46 @xmath58 \morphism(125,250)|m|<250,0 > [ ` ; ] \efig\ ] ] which when composed with the projections of @xmath46 , recovers the local projections as in @xmath59 \morphism(125,250)|m|<250,0 > [ ` ; ] \square(500,0)|amrb|[x{\otimes}x`x`a{\otimes}a`a;p_{x , x}`r{\otimes}s`r`p_{a , a } ] \morphism(625,250)|m|<250,0>[`;\tilde p_{r , s } ] \place(1250,250)[{\cong } ] \square(1550,0)|almb|[x`x`a`a;1_x`r\wedge s`r`1_a ] \morphism(1675,250)|m|<250,0>[`;\pi ] \efig\ ] ] for the first projection , and similarly for the second . the unspecified @xmath60 in @xmath46 is given by a pair of convenient isomorphisms @xmath61 and @xmath62 in @xmath36 . similarly , when @xmath63 is composed with @xmath64 the result is @xmath65 . [ [ bc ] ] quite generally , an arrow of @xmath46 as given by the square ( [ square ] ) will be called a _ commutative _ square if @xmath66 is invertible . an arrow of @xmath46 will be said to satisfy the _ beck condition _ if the mate of @xmath66 under the adjunctions @xmath67 and @xmath68 , as given in the square below ( no longer an arrow of @xmath46 ) , is invertible . @xmath69 \morphism(1125,250)|m|<250,0>[`;\alpha^ * ] \efig\ ] ] thus proposition 4.7 of @xcite says that projection squares of the form @xmath70 and @xmath71 are commutative while proposition 4.8 of @xcite says that these same squares satisfy the beck condition . if @xmath72 and @xmath73 are also maps and @xmath66 is invertible then @xmath74 gives rise to another arrow of @xmath46 , from @xmath41 to @xmath54 with reference to the square above , which may or may not satisfy the beck condition . the point here is that a commutative square of maps gives rise to two , generally distinct , beck conditions . it is well known that , for bicategories of the form @xmath1 and @xmath75 , all pullback squares of maps satisfy both beck conditions . a category with finite products has automatically a number of pullbacks which we might call _ product - absolute _ pullbacks because they are preserved by all functors which preserve products . in @xcite the beck conditions for the product - absolute pullback squares of the form @xmath76 \efig\ ] ] were investigated . ( in fact , in this case it was shown that either beck condition implies the other . ) the objects for which these conditions are met are called _ frobenius _ objects . [ mcifro ] for a cartesian bicategory , the axiom _ maps are comonadic _ implies the axiom _ frobenius_. it suffices to show that the 2-cell @xmath77 below is invertible : @xmath78 \morphism(750,500)|r|<0,-250>[a{\otimes}a`(a{\otimes}a){\otimes}a;d{\otimes}1 ] \morphism(750,250)|r|<0,-250>[(a{\otimes}a){\otimes}a`a{\otimes}(a{\otimes}a);a ] \morphism(175,250)|a|<150,0>[`;\delta_1 ] \square(0,-500)|blrb|/<-`->`->`<-/<750,500>[a{\otimes}a`a{\otimes}(a{\otimes}a)`a`a{\otimes}a;1{\otimes}d^*`r`r`d^ * ] \morphism(250,-250)|a|<150,0>[`;\tilde r_{1_a , d^ * } ] \efig\ ] ] the paste composite of the squares is invertible ( being essentially the identity 2-cell on @xmath79 ) . the lower 2-cell is invertible by proposition 4.7 of @xcite so that the whisker composite @xmath80 is invertible . since @xmath38 is a map it reflects isomorphisms , by maps are comonadic , and hence @xmath77 is invertible . [ frobclofin ] it was shown in @xcite that , in a cartesian bicategory , the frobenius objects are closed under finite products . it follows that the full subbicategory of a cartesian bicategory determined by the frobenius objects is a cartesian bicategory which satisfies the frobenius axiom . in this section we look at separability for objects of cartesian bicategories . since for an object @xmath19 which is both separable and frobenius , the hom - category @xmath21 is essentially discrete , for all @xmath20 , we shall then be able to show that @xmath18 is essentially discrete by showing that all objects in @xmath15 are separable and frobenius . but first we record the following direct argument : [ easy ] if @xmath15 is a bicategory in which all maps are comonadic and @xmath18 has a terminal object , then @xmath18 is locally essentially discrete . we must show that for all objects @xmath20 and @xmath19 , the hom - category @xmath21 is essentially discrete . as usual , we write @xmath30 for the terminal object of @xmath18 and @xmath81 for the essentially unique map , which by assumption is comonadic . let @xmath82 be maps from @xmath20 to @xmath19 . if @xmath83 is any 2-cell , then @xmath84 is invertible , since @xmath30 is terminal in @xmath18 . but since @xmath85 is comonadic , it reflects isomorphisms , and so @xmath66 is invertible . furthermore , if @xmath86 is another 2-cell , then @xmath87 by the universal property of @xmath30 once again , and now @xmath88 since @xmath85 is faithful . thus there is at most one 2-cell from @xmath41 to @xmath89 , and any such 2-cell is invertible . in any ( bi)category with finite products the diagonal arrows @xmath90 are ( split ) monomorphisms so that in the bicategory @xmath36 the following square is a product - absolute pullback @xmath91 \efig\ ] ] that gives rise to a single @xmath46 arrow . [ sep ] an object @xmath19 in a cartesian bicategory is said to be _ separable _ if the @xmath46 arrow above satisfies the beck condition . of course the invertible mate condition here says precisely that the unit @xmath92 for the adjunction @xmath93 is invertible . thus axiom [ sepax ] , as stated in the introduction , says that , for all @xmath19 in @xmath15 , @xmath94 is invertible . [ sepcat ] for a map @xmath41 it makes sense to define _ @xmath41 is fully faithful _ to mean that @xmath95 is invertible . for a _ category @xmath19 _ the diagonal @xmath96 is fully faithful if and only if @xmath19 is an ordered set . [ sepmeans ] for an object @xmath19 in a cartesian bicategory , the following are equivalent : 1 . @xmath19 is separable ; 2 . for all @xmath97 in @xmath36 , the diagram @xmath98 is a product in @xmath99 ; 3 . @xmath100 is a product in @xmath101 ; 4 . @xmath102 is a monomorphism in @xmath101 ; 5 . for all @xmath103 in @xmath101 , the diagram @xmath104 is a product in @xmath101 . @xmath105 @xmath106 $ ] a local product of maps is not generally a map but here we have : @xmath107 @xmath108 @xmath109 $ ] is trivial . @xmath110 @xmath111 $ ] note the use of pseudo - functoriality of @xmath112 : @xmath113 @xmath110 @xmath114 $ ] to say that @xmath100 is a product in @xmath101 is precisely to say that @xmath115 \efig\ ] ] is a pullback in @xmath101 which in turn is precisely to say that @xmath102 is a monomorphism in @xmath101 @xmath116 @xmath117 $ ] it is a generality that if an object @xmath73 in a category is subterminal then for any @xmath118 , necessarily unique , @xmath119 is a product diagram . @xmath120 @xmath109 $ ] is trivial . [ mcisep][of @xmath121 for a cartesian bicategory , the axiom _ maps are comonadic _ implies the axiom _ separable_. we have @xmath122 for the map @xmath123 . it follows that the unique @xmath124 is @xmath125 . since @xmath85 is comonadic , @xmath125 is the equalizer shown : @xmath126 and hence a monomorphism . [ copt][of @xmath121 for separable @xmath19 in cartesian @xmath15 , an arrow @xmath22 admits at most one copoint @xmath103 depending upon whether the unique arrow @xmath127 factors through @xmath128 . [ sepclofin ] in a cartesian bicategory , the separable objects are closed under finite products . if @xmath19 and @xmath129 are separable objects then applying the homomorphism @xmath130 we have an adjunction @xmath131 with unit @xmath132 which being an isomorph of the adjunction @xmath133 with unit @xmath134 ( via middle - four interchange ) shows that the separable objects are closed under binary products . on the other hand , @xmath135 is an equivalence so that @xmath35 is also separable . [ eos ] for a cartesian bicategory , the full subbicategory determined by the separable objects is a cartesian bicategory which satisfies the axiom _ separable_. [ ordhom ] if @xmath19 is a separable object in a cartesian bicategory @xmath15 , then , for all @xmath20 in @xmath15 , the hom - category @xmath136 is an ordered set , meaning that the category structure forms a reflexive , transitive relation . suppose that we have arrows @xmath137 in @xmath136 . in @xmath99 we have @xmath138 \morphism(500,500)|m|<0,-500>[g`f\wedge f;\gamma ] \morphism(500,0)|b|<-500,0>[f\wedge f`f;\pi ] \morphism(500,0)|b|<500,0>[f\wedge f`f;\rho ] \efig\ ] ] by proposition [ sepmeans ] we can take @xmath139 and @xmath140 so that we have @xmath141 . it follows that @xmath136 is an ordered set . [ discrete ] an object @xmath19 in a cartesian bicategory is said to be _ discrete _ if it is both frobenius and separable . we write @xmath142 for the full subbicategory of @xmath15 determined by the discrete objects . beware that this is quite different to the notion of discreteness in a bicategory . an object @xmath19 of a bicategory is discrete if each hom - category @xmath99 is discrete ; @xmath19 is essentially discrete if each @xmath99 is equivalent to a discrete category . the notion of discreteness for cartesian bicategories defined above turns out to mean that @xmath19 is essentially discrete in the bicategory @xmath18 . from proposition [ sepclofin ] above and proposition 3.4 of @xcite we immediately have [ eod ] for a cartesian bicategory @xmath15 , the full subbicategory @xmath142 of discrete objects is a cartesian bicategory in which every object is discrete . and from proposition [ ordhom ] above and theorem 3.13 of @xcite we have [ dishom ] if @xmath19 is a discrete object in a cartesian bicategory @xmath15 then , for all @xmath20 in @xmath15 , the hom category @xmath136 is an equivalence relation . if both the _ frobenius _ axiom of @xcite and the _ separable _ axiom of this paper hold for our cartesian bicategory @xmath15 , then every object of @xmath15 is discrete . in this case , because @xmath36 is a bicategory , the equivalence relations @xmath136 are stable under composition from both sides . thus writing @xmath143 for the set of objects of @xmath136 we have a mere category , @xmath0 whose objects are those of @xmath36 ( and hence also those of @xmath15 ) and whose hom sets are the quotients @xmath144 . if the @xmath145 are regarded as discrete categories , so that @xmath0 is a locally discrete bicategory then the functors @xmath146 constitute the effect on homs functors for an identity on objects biequivalence @xmath147 . to summarize [ odibld ] if a cartesian bicategory @xmath15 satisfies both the frobenius and separable axioms then the bicategory of maps @xmath36 is biequivalent to the locally discrete bicategory @xmath0 . in the following lemma we show that any copointed endomorphism of a discrete object can be made into a comonad ; later on , we shall see that this comonad structure is in fact unique . [ diag ] if @xmath19 is a discrete object in a cartesian bicategory @xmath15 then , for any copointed endomorphism arrow @xmath148 , there is a 2-cell @xmath149 satisfying @xmath150 \morphism(500,500)|m|<0,-500>[g`gg;\delta ] \morphism(500,0)|b|<-500,0>[gg`g;g\epsilon ] \morphism(500,0)|b|<500,0>[gg`g;\epsilon g ] \efig\ ] ] and if both @xmath151 are copointed , so that @xmath152 is also copointed , and @xmath153 is any 2-cell , then the @xmath154 s satisfy @xmath155 \place(1000,0)[\mbox{and } ] \atriangle(1500,0)/<-`->`->/[ghgh`gh`gh;\delta`(g\epsilon)(\epsilon h)`1 ] \efig\ ] ] we define @xmath156 to be the pasting composite @xmath157 \square(500,0)|amma|[aa`aa`aaa`aaa;g1`1d`1d`g11 ] \square(1000,0)|amma|[aa`a`aaa`aa;d^*`1d`d`d^*1 ] \qtriangle(500,-500)|abr|[aaa`aaa`aaa;g11`ggg`11 g ] \square(1000,-500)|arma|[aaa`aa`aaa`aa;d^*1`11g`1g`d^*1 ] \qtriangle(1000,-1000)|amm|[aaa`aa`a;d^*1`d_3^*`d^ * ] \morphism(0,500)|b|/{@{->}@/_4em/}/<1500,-1500>[a`a;g\wedge g\wedge g ] \morphism(0,500)|b|/{@{->}@/_8em/}/<1500,-1500>[a`a;g ] \morphism(0,500)|a|/{@{->}@/^3em/}/<1500,0>[a`a;g ] \morphism(1500,500)|r|/{@{->}@/^3em/}/<0,-1500>[a`a;g ] \morphism(800,-250)|m|<150,150>[`;g\epsilon g ] \morphism(300,-800)|m|<150,150>[`;\delta_3 ] \place(1250,250)[1 ] \place(750,675)[2 ] \place(1700,-250)[3 ] \place(750,250)[4 ] \place(1250,-250)[5 ] \place(600,-450)[6 ] \efig\ ] ] wherein @xmath112 has been abbreviated by juxtaposition and all subregions not explicitly inhabited by a 2-cell are deemed to be inhabited by the obvious invertible 2-cell . a reference number has been assigned to those invertible 2-cells which arise from the hypotheses . as in @xcite , @xmath158 s denote 3-fold diagonal maps and , similarly , we write @xmath159 for a local 3-fold diagonal . the invertible 2-cell labelled by ` 1 ' is that defining @xmath19 to be frobenius . the 3-fold composite of arrows in the region labelled by ` 2 ' is @xmath160 and , similarly , in that labelled by ` 3 ' we have @xmath161 . each of these is isomorphic to @xmath162 because @xmath19 is separable and @xmath162 is copointed . the isomorphisms in ` 4 ' and ` 5 ' express the pseudo - functoriality of @xmath112 in the cartesian bicategory @xmath15 . finally ` 6 ' expresses the ternary local product in terms of the ternary @xmath112 as in @xcite . demonstration of the equations is effected easily by pasting composition calculations . [ wedge=. ] if @xmath162 and @xmath163 are copointed endomorphisms on a discrete @xmath19 in a cartesian @xmath15 then @xmath164 is a product diagram in @xmath101 . if we are given @xmath165 and @xmath166 then @xmath167 is also copointed and we have @xmath168 as a candidate pairing . that this candidate satisfies the universal property follows from the equations of lemma [ diag ] which are precisely those in the equational description of binary products . we remark that the ` naturality ' equations for the projections follow immediately from uniqueness of copoints . [ comsim ] if @xmath19 is discrete in a cartesian @xmath15 , then an endo - arrow @xmath22 admits a comonad structure if and only if @xmath162 has the copointed property , and any such comonad structure is unique . the theorem shows that the arrow @xmath169 constructed in lemma [ diag ] is the product diagonal on @xmath162 in the category @xmath101 and , given @xmath170 , this is the only comonad comultiplication on @xmath162 . it is clear that @xmath171 is terminal with respect to the copointed objects in @xmath101 . [ subterm ] if an object @xmath129 in a bicategory @xmath15 has @xmath172 subterminal in @xmath173 then , for any map @xmath174 , @xmath41 is subterminal in @xmath175 and @xmath176 is subterminal in @xmath177 . in particular , in a cartesian bicategory in which every object is separable , every adjoint arrow is subterminal . precomposition with a map preserves terminal objects and monomorphisms , as does postcomposition with a right adjoint . the starting point of this section is the observation , made in the introduction , that a comonad in the bicategory @xmath1 is precisely a span of the form @xmath178 in which both legs are equal . we will write @xmath179 for the bicategory of comonads in @xmath15 , @xmath180 being one of the duals of street s construction @xmath181 in @xcite . thus @xmath182 has objects given by the comonads @xmath26 of @xmath15 . the structure 2-cells for comonads will be denoted @xmath183 , for counit and @xmath156 , for comultiplication . an arrow in @xmath182 from @xmath26 to @xmath184 is a pair @xmath185 as shown in @xmath186 \morphism(125,250)|a|<250,0>[`;\phi ] \efig\ ] ] satisfying @xmath187 \place(1000,250)[\mbox{and } ] \square(1500,0)/`->``->/[fg``fgg`hfg;`f\delta``\phi g ] \square(2000,0)/``->`->/[`hf`hfg`hhf;``\delta f`h\phi ] \morphism(1500,500)<1000,0>[fg`hf;\phi ] \efig\ ] ] ( where , as often , we have suppressed the associativity constraints of our normal , cartesian , bicategory @xmath15 ) . a 2-cell @xmath188 in @xmath182 is a 2-cell @xmath189 in @xmath15 satisfying @xmath190 \efig\ ] ] there is a pseudofunctor @xmath191 given by @xmath192 from @xcite it is well known that a bicategory @xmath15 has eilenberg - moore objects for comonads if and only if @xmath191 has a right biadjoint , which we will denote by @xmath193 . we write @xmath194 and the counit for @xmath195 is denoted by @xmath196 \morphism(125,250)|a|<250,0>[`;\gamma_g ] \ctriangle(2500,0)/<-`->`->/<500,250>[a`a_g`a;g_g`g`g_g ] \place(1500,250)[\mbox { or , using normality of $ { \mathbf{b}}$ , better by } ] \morphism(2700,250)|m|<200,0>[`;\gamma_g ] \efig\ ] ] with @xmath197 abbreviated to @xmath198 when there is no danger of confusion . it is standard that each @xmath199 is necessarily a map ( whence our lower case notation ) and the mate @xmath200 of @xmath201 is an isomorphism which identifies @xmath202 and @xmath203 . we will write @xmath204 for the locally full subbicategory of @xmath182 determined by all the objects and those arrows of the form @xmath205 , where @xmath41 is a map , and write @xmath206 for the inclusion . it is clear that the pseudofunctor @xmath191 restricts to give a pseudofunctor @xmath207 . we say that the bicategory @xmath15 _ has eilenberg - moore objects for comonads , as seen by @xmath36 _ , if @xmath207 has a right biadjoint . ( in general , this property does not follow from that of eilenberg - moore objects for comonads . ) [ rmk : d - for - dummies ] in the case @xmath208 , a comonad in @xmath15 can , as we have seen , be identified with a morphism in e. this can be made into the object part of a biequivalence between the bicategory @xmath204 and the category @xmath209 of arrows in @xmath0 . if we further identify @xmath36 with @xmath0 , then the inclusion @xmath210 becomes the diagonal @xmath211 ; of course this does have a right adjoint , given by the domain functor . [ simcom ] if @xmath15 is a cartesian bicategory in which every object is discrete , the bicategory @xmath212 admits the following simpler description : 1 . an object is a pair @xmath26 where @xmath19 is an object of @xmath15 and @xmath22 admits a copoint ; 2 . an arrow @xmath213 is a map @xmath174 and a 2-cell @xmath214 ; 3 . a 2-cell @xmath215 is a 2-cell satisfying @xmath216 satisfying equation ( [ comtrans ] ) . we have i ) by corollary [ comsim ] while iii ) is precisely the description of a 2-cell in @xmath204 , modulo the description of the domain and codomain arrows . so , we have only to show ii ) , which is to show that the equations ( [ comarrow ] ) hold automatically under the hypotheses . for the first equation of ( [ comarrow ] ) we have uniqueness of any 2-cell @xmath217 because @xmath41 is subterminal by proposition [ subterm ] . for the second , observe that the terminating vertex , @xmath218 , is the product @xmath219 in @xmath220 because @xmath221 is the product @xmath222 in @xmath223 by theorem [ wedge=. ] and precomposition with a map preserves all limits . for @xmath218 seen as a product , the projections are , again by theorem [ wedge=. ] , @xmath224 and @xmath225 . thus , it suffices to show that the diagram for the second equation commutes when composed with both @xmath224 and @xmath225 . we have @xmath226 \square(500,0)/``->`->/[`hf`hfg`hhf;``\delta f`h\phi ] \morphism(0,500)<1000,0>[fg`hf;\phi ] \qtriangle(500,-500)|blr|[hfg`hhf`hf;h\phi`hf\epsilon`h\epsilon f ] \morphism(0,-500)|b|<1000,0>[fg`hf;\phi ] \morphism(0,0)<0,-500>[fgg`fg;fg\epsilon ] \square(2000,0)/`->``->/[fg``fgg`hfg;`f\delta``\phi g ] \square(2500,0)/``->`->/[`hf`hfg`hhf;``\delta f`h\phi ] \morphism(2000,500)<1000,0>[fg`hf;\phi ] \ptriangle(2000,-500)|blr|[fgg`hfg`fg;\phi g`f\epsilon g`\epsilon fg ] \morphism(2000,-500)|b|<1000,0>[fg`hf;\phi ] \morphism(3000,0)|r|<0,-500>[hhf`hf;\epsilon hf ] \efig\ ] ] in which each of the lower triangles commutes by the first equation of ( [ comarrow ] ) already established . using comonad equations for @xmath162 and @xmath163 , it is obvious that each composite is @xmath227 . finally , let us note that @xmath204 is a subbicategory , neither full nor locally full , of the grothendieck bicategory @xmath46 and write @xmath228 for the inclusion . we also write @xmath229 for the composite pseudofunctor @xmath230 . summarizing , we have introduced the following commutative diagram of bicategories and pseudofunctors @xmath231 \morphism(500,500)<500,0>[{\mathbf{d}}`{\mathbf{g}};k ] \morphism(0,500)|a|/{@{->}@/^2em/}/<1000,0>[{\mathbf{m}}`{\mathbf{g}};\iota ] \efig ; \ ] ] note also that in our main case of interest @xmath208 , each of @xmath36 , @xmath204 , and @xmath46 is biequivalent to a mere category . ultimately , we are interested in having a right biadjoint , say @xmath232 , of @xmath233 . for such a biadjunction @xmath234 the counit at an object @xmath235 in @xmath46 will take the form @xmath236 \morphism(200,250)|m|<200,0>[`;\omega_r ] \efig\ ] ] ( where , as for a biadjunction @xmath237 , a triangle rather than a square can be taken as the boundary of the 2-cell by the normality of @xmath15 ) . in fact , we are interested in the case where we have @xmath234 and moreover the counit components @xmath238 enjoy the property that their mates @xmath239 with respect to the adjunction @xmath240 are invertible . in this way we represent a general arrow of @xmath15 in terms of a span of maps . since biadjunctions compose we will consider adjunctions @xmath241 and @xmath242 and we begin with the second of these . [ g(r ) ] for a cartesian bicategory @xmath15 in which every object is discrete , there is an adjunction @xmath243 where , for @xmath235 in @xmath46 , the comonad @xmath244 and its witnessing copoint @xmath245 are given by the left diagram below and the counit @xmath246 is given by the right diagram below , all in notation suppressing @xmath112 : @xmath247 \morphism(-1400,1350)|a|<150,0>[`;\simeq ] \morphism(-1500,1000)|l|<0,-500>[xxa`xaa;xra ] \morphism(-1000,1500)|r|<0,-1500>[xa`xa;1_{xa } ] \morphism(-1375,750)|m|<250,0>[`;\tilde p_{1,3 } ] \btriangle(-1500,0)[xaa`xa`xa;xd^*`p_{1,3}`1_{xa } ] \morphism(-1400,150)|a|<150,0 > [ ` ; ] \morphism(-1500,1500)|l|/{@{->}@/_3.5em/}/<0,-1500>[xa`xa;g(r ) ] \ptriangle(0,1000)/->`->`<-/[xa`x`xxa;p`da`p_2 ] \morphism(0,1000)|l|<0,-500>[xxa`xaa;xra ] \morphism(500,1500)|r|<0,-1500>[x`a;r ] \btriangle(0,0)[xaa`xa`a;xd^*`p_2`r ] \morphism(100,1350)|a|<150,0>[`;\simeq ] \morphism(125,750)|m|<250,0>[`;\tilde p_2 ] \morphism(100,150)|b|<150,0 > [ ` ; ] \morphism(0,1500)|l|/{@{->}@/_3.5em/}/<0,-1500>[xa`xa;g(r ) ] \efig\ ] ] moreover , the mate @xmath248 of the counit @xmath249 is invertible . in the left diagram , the @xmath250 collectively denote projection from the three - fold product in @xmath46 to the product of the first and third factors . in the right diagram , the @xmath251 collectively denote projection from the three - fold product in @xmath46 to the second factor . the upper triangles of the two diagrams are the canonical isomorphisms . the lower left triangle is the mate of the canonical isomorphism @xmath252 . the lower right triangle is the mate of the canonical isomorphism @xmath253 . given a comonad @xmath254 and an arrow @xmath255 \morphism(125,250)|a|<250,0>[`;\psi ] \efig\ ] ] in @xmath46 , we verify the adjunction claim by showing that there is a unique arrow @xmath256 \morphism(125,250)|a|<250,0>[`;\phi ] \efig\ ] ] in @xmath204 , whose composite with the putative counit @xmath249 is @xmath257 . it is immediately clear that the unique solution for @xmath41 is @xmath258 and to give @xmath259 is to give the mate @xmath260 which is @xmath261 and can be seen as a @xmath46 arrow : @xmath262 \morphism(125,250)|a|<250,0>[`;(\alpha,\beta,\gamma ) ] \efig\ ] ] where we exploit the description of products in @xmath46 . from this description it is clear , since @xmath263 as a composite in @xmath46 , that the unique solution for @xmath264 is @xmath265 . we have seen in theorem [ comsim ] that the conditions ( [ comarrow ] ) hold automatically in @xmath204 under the assumptions of the theorem . from the first of these we have : @xmath266 \morphism(125,250)|a|<250,0>[`;(\alpha,\beta,\gamma ) ] \square(500,0)|amrb|[xxa`xa`xaa`xa;p_{1,3}`xra`1_{xa}`p_{1,3 } ] \morphism(625,250)|a|<250,0>[`;p_{1,3 } ] \place(1375,250)[= ] \square(2000,0)|arrb|[t`xa`t`xa;(x , a)`1_t`1_{xa}`(x , a ) ] \morphism(2125,250)|a|<250,0>[`;\kappa_{(x , a ) } ] \morphism(2000,500)|l|/{@{->}@/_3em/}/<0,-500>[t`t;h ] \morphism(1750,250)|a|<200,0>[`;\epsilon_h ] \efig\ ] ] so , with a mild abuse of notation , we have @xmath267 , uniquely , and thus the unique solutions for @xmath66 and @xmath201 are @xmath268 and @xmath269 respectively . this shows that @xmath227 is necessarily the mate under the adjunctions considered of @xmath270 . since @xmath204 and @xmath46 are essentially locally discrete this suffices to complete the claim that @xmath242 . it only remains to show that the mate @xmath248 of the counit @xmath249 is invertible . in the three middle squares of the diagram @xmath271 \square(0,0)|alrm|/<-`->`->`<-/[xa`x`xxa`xx;p^*`da`d`p^ * ] \morphism(125,250)|a|<250,0>[`;\tilde p^*_{d,1_a } ] \square(0,-500)|mlmm|/<-`->`->`<-/[xxa`xx`xaa`xa;p^*`xra`xr`p^ * ] \morphism(125,-250)|a|<250,0>[`;\tilde p^*_{xr,1_a } ] \square(0,-1000)|mlrb|/<-`->`->`->/[xaa`xa`xa`a;p^*`xd^*`r`r ] \morphism(175,-750)|a|<150,0>[`;\simeq ] \square(500,-500)|amrb|[xx`x`xa`a;r`xr`r`r ] \morphism(625,-250)|a|<250,0>[`;r_{1_x , r } ] \morphism(500,500)|r|<500,-500>[x`x;1_x ] \morphism(1000,-500)|r|<-500,-500>[a`a;1_a ] \efig\ ] ] the top two are invertible 2-cells by proposition 4.18 of @xcite while the lower one is the obvious invertible 2-cell constructed from @xmath272 . the right square is an invertible 2-cell by proposition 4.17 of @xcite . this shows that the mate @xmath248 of @xmath249 is invertible . [ unit ] it now follows that the unit of the adjunction @xmath242 is given ( in notation suppressing @xmath112 ) by : @xmath273 \dtriangle(1500,0)/<-`->`->/[ttt`t`tt;d_3`td^*`d ] \morphism(1500,1500)|l|<0,-1500>[t`t;h ] \morphism(2000,1000)|l|<0,-500>[ttt`ttt;hhh ] \morphism(2000,1000)|r|/{@{->}@/^3em/}/<0,-500>[ttt`ttt;tht ] \morphism(1750,1350)|a|<150,0>[`;\simeq ] \morphism(1550,750)|m|<175,0>[`;\tilde d_3 ] \morphism(2050,750)|m|<225,0>[`;\epsilon h\epsilon ] \morphism(1750,150)|a|<150,0>[`;\simeq ] \efig\ ] ] where the @xmath158 collectively denote 3-fold diagonalization @xmath274 in @xmath46 . the top triangle is a canonical isomorphism while the lower triangle is the mate of the canonical isomorphism @xmath275 and is itself invertible , by separability of @xmath276 . before turning to the question of an adjunction @xmath241 , we note : [ maplikeanm ] in a cartesian bicategory in which maps are comonadic , if @xmath277 with @xmath89 and @xmath278 maps , then @xmath279 is also a map . by theorem 3.11 of @xcite it suffices to show that @xmath279 is a comonoid homomorphism , which is to show that the canonical 2-cells @xmath280 and @xmath281 are invertible . for the first we have : @xmath282 simple diagrams show that we do get the right isomorphism in this case and also for the next : @xmath283 which gives @xmath284 since the map @xmath285 reflects isomorphisms . [ emasbm ] if @xmath15 is a cartesian bicategory which has eilenberg - moore objects for comonads and for which maps are comonadic then @xmath15 has eilenberg - moore objects for comonads as seen by @xmath36 , which is to say that @xmath207 has a right adjoint . moreover , the counit for the adjunction , say @xmath286 , necessarily having components of the form @xmath287 with @xmath89 a map , has @xmath200 invertible . it suffices to show that the adjunction @xmath237 restricts to @xmath288 . for this it suffices to show that , given @xmath289 , the @xmath290 with @xmath277 which can be found using @xmath195 has @xmath279 a map . this follows from lemma [ maplikeanm ] . [ tabulation ] a cartesian bicategory which has eilenberg - moore objects for comonads and for which maps are comonadic has tabulation in the sense that the inclusion @xmath229 has a right adjoint @xmath232 and the counit components @xmath238 as in ( [ tabcounit ] ) have the property that the mates @xmath239 , with respect to the adjunctions @xmath240 , are invertible . using theorems [ g(r ) ] and [ emasbm ] we can construct the adjunction @xmath234 by composing @xmath241 with @xmath242 . moreover , the counit for @xmath234 is the pasting composite : @xmath291 \morphism(0,250)|l|/{@{->}@/^4.0em/}/<1000,250>[t`x;u ] \morphism(0,250)|l|/{@{->}@/_4.0em/}/<1000,-250>[t`a;v ] \morphism(300,160)|m|<0,180>[`;\gamma ] \square(500,0)|amrb|<500,500>[x{\otimes}a`x`x{\otimes}a`a;p`g(r)`r`r ] \morphism(650,250)|m|<200,0>[`;\mu ] \efig\ ] ] where the square is the counit for @xmath242 ; and the triangle , the counit for @xmath241 , is an eilenberg - moore coalgebra for the comonad @xmath244 . the arrow component of the eilenberg - moore coalgebra is necessarily of the form @xmath292 , where @xmath54 and @xmath293 are maps , and it also follows that we have @xmath294 . thus we have @xmath295 where the first two isomorphisms are trivial , the third arises from the invertibility of the mate of @xmath201 as an eilenberg - moore structure , and the fourth is invertibility of @xmath249 , as in theorem [ g(r ) ] . [ mapbhaspb ] for a cartesian bicategory @xmath15 with eilenberg - moore objects for comonads and for which maps are comonadic , @xmath18 has pullbacks satisfying the beck condition ( meaning that for a pullback square @xmath296 \morphism(200,250)<100,0>[`;\simeq ] \efig\ ] ] the mate @xmath297 of @xmath298 in @xmath15 , with respect to the adjunctions @xmath299 and @xmath300 , is invertible ) . given the cospan @xmath301 in @xmath18 , let @xmath302 together with @xmath303 be a tabulation for @xmath304 . then @xmath297 , the mate of @xmath305 with respect to @xmath299 , is invertible by theorem [ tabulation ] . we have also @xmath306 , the mate of @xmath305 with respect to @xmath300 . since @xmath19 is discrete , @xmath306 is also invertible and is the only 2-cell between the composite maps in question . if we have also @xmath307 , for maps @xmath54 and @xmath293 with @xmath308 , then the mate @xmath309 ensures that the span @xmath307 factors through @xmath302 by an essentially unique map @xmath310 with @xmath311 and @xmath312 . [ tabonadic ] in a cartesian bicategory with eilenberg - moore objects for comonads and for which maps are comonadic , every span of maps @xmath4 gives rise to the following tabulation diagram : @xmath313 \morphism(300,160)|m|<0,180>[`;a\eta_x ] \efig\ ] ] a general tabulation counit @xmath238 is given in terms of the eilenberg - moore coalgebra for the comonad @xmath314 and necessarily @xmath294 . it follows that for @xmath315 , it suffices to show that @xmath316 . consider the diagram ( with @xmath112 suppressed ) : @xmath317 \vtriangle(0,500)|mmm|/`->`->/[sa`xs`xsa;`(x , s)a`x(s , a ) ] \atriangle(0,1000)|lrm|/->`->`/[s`sa`xs;(s , a)`(x , s ) ` ] \ctriangle(-500,0)|lml|/->``->/[sa`xa`xxa;xa``da ] \dtriangle(1000,0)|mrr|/`->`->/[xs`xa`xaa;`xa`xd ] \morphism(500,1500)|l|/{@{->}@/_3em/}/<-1000,-1000>[s`xa;(x , a ) ] \morphism(500,1500)|r|/{@{->}@/^3em/}/<1000,-1000>[s`xa;(x , a ) ] \efig\ ] ] the comonoid @xmath318 can be read , from left to right , along the ` w ' shape of the lower edge as @xmath319 . but each of the squares in the diagram is a ( product - absolute ) pullback so that with proposition [ mapbhaspb ] at hand we can continue : @xmath320 as required . [ [ cfun ] ] if @xmath15 is a cartesian bicategory with @xmath18 essentially locally discrete then each slice @xmath321 is also essentially locally discrete and we can write @xmath322 for the categories obtained by taking the quotients of the equivalence relations comprising the hom categories of the @xmath321 . then we can construct functors @xmath323 , where for an arrow in @xmath322 as shown , @xmath324 \dtriangle(500,0)|mrr|/->`->`<-/[m`x`n;h`x`y ] \efig\ ] ] we define @xmath325 and @xmath326 . if @xmath18 is known to have pullbacks then the @xmath322 become the hom - categories for a bicategory @xmath327 and we can consider whether the @xmath328 provide the effects on homs for an identity - on - objects pseudofunctor @xmath329 . consider @xmath330 \vtriangle(500,0)/`->`->/[n`m`a ; `` ] \atriangle(1000,0)/->`->`/[m`a`x;a`x ` ] \atriangle(500,500)/->`->`/[p`n`m;p`r ` ] \efig\ ] ] where the square is a pullback . in somewhat abbreviated notation , what is needed further are coherent , invertible 2-cells @xmath331 , for each composable pair of spans @xmath332 , @xmath333 , and coherent , invertible 2-cells @xmath334 , for each object @xmath19 . since the identity span on @xmath19 is @xmath335 , and @xmath336 we take the inverse of this composite for @xmath337 . to give the @xmath338 though is to give 2-cells @xmath339 and since spans of the form @xmath340 and @xmath341 arise as special cases , it is easy to verify that to give the @xmath338 it is necessary and sufficient to give coherent , invertible 2-cells @xmath342 for each pullback square in @xmath18 . the inverse of such a 2-cell @xmath343 is the mate of a 2-cell @xmath344 . but by discreteness a 2-cell @xmath344 must be essentially an identity . thus , definability of @xmath338 is equivalent to the invertibility in @xmath15 of the mate @xmath343 of the identity @xmath345 , for each pullback square as displayed in ( [ beck ] ) . in short , if @xmath18 has pullbacks and these satisfy the beck condition as in proposition [ mapbhaspb ] then we have a canonical pseudofunctor @xmath329 . [ spanmain ] for a bicategory @xmath15 the following are equivalent : 1 . there is a biequivalence @xmath346 , for @xmath0 a category with finite limits ; 2 . the bicategory @xmath15 is cartesian , each comonad has an eilenberg - moore object , and every map is comonadic . 3 . the bicategory @xmath18 is an essentially locally discrete bicategory with finite limits , satisfying in @xmath15 the beck condition for pullbacks of maps , and the canonical @xmath347 is a biequivalence of bicategories . that @xmath348 implies @xmath349 follows from our discussion in the introduction . that @xmath350 implies @xmath348 is trivial so we show that @xmath349 implies @xmath350 . we have already observed in theorem [ odibld ] that , for @xmath15 cartesian with every object discrete , @xmath18 is essentially locally discrete and we have seen by propositions [ mcifro ] and [ mcisep ] that , in a cartesian bicategory in which maps are comonadic , every object is discrete . in theorem [ mapbhaspb ] we have seen that , for @xmath15 satisfying the conditions of @xmath349 , @xmath18 has pullbacks , and hence all finite limits and , in @xmath15 the beck condition holds for pullbacks . therefore we have the canonical pseudofunctor @xmath329 developed in [ cfun ] . to complete the proof it suffices to show that the @xmath323 are equivalences of categories . define functors @xmath351 by @xmath352 where @xmath353 \morphism(200,250)|m|<200,0>[`;\omega ] \efig\ ] ] is the @xmath72-component of the counit for @xmath354 . for a 2-cell @xmath355 we define @xmath356 to be the essentially unique map satisfying @xmath357 \ctriangle(0,0)|rrr|/<-`->`->/[x`\tau r'`a;u'`r'`v ' ] \morphism(175,400)|a|<150,0>[`;\omega ' ] \morphism(-500,500)|l|<1000,500>[\tau r`x;u ] \morphism(-500,500)|l|<1000,-500>[\tau r`a;v ] \place(750,500)[= ] \ctriangle(1000,0)/<-``->/[x`\tau r`a;u``v ] \morphism(1500,1000)|l|/{@{->}@/_1em/}/<0,-1000>[x`a;r ] \morphism(1500,1000)|r|/{@{->}@/^1.5em/}/<0,-1000>[x`a;r ' ] \morphism(1150,400)|a|<150,0>[`;\omega ] \morphism(1450,400)|a|<150,0>[`;\alpha ] \efig\ ] ] ( we remark that essential uniqueness here means that @xmath356 is determined to within unique invertible 2-cell . ) since @xmath358 has mate @xmath359 invertible , because @xmath360 is a tabulation of @xmath72 , it follows that we have a natural isomorphism @xmath361 . on the other hand , starting with a span @xmath362 from @xmath20 to @xmath19 we have as a consequence of theorem [ tabonadic ] that @xmath362 is part of a tabulation of @xmath363 . it follows that we have a natural isomorphism @xmath364 , which completes the demonstration that @xmath328 and @xmath365 are inverse equivalences . in the previous section we gave a characterization of those ( cartesian ) bicategories of the form @xmath1 for a category @xmath0 with finite limits . in this final section we give a refinement , showing that @xmath1 has direct sums if and only if the original category @xmath0 is lextensive @xcite . direct sums are of course understood in the bicategorical sense . a _ zero object _ in a bicategory is an object which is both initial and terminal . in a bicategory with finite products and finite coproducts in which the initial object is also terminal there is a canonical induced arrow @xmath366 , and we say that the bicategory has _ direct sums _ when this map is an equivalence . just as in the case of ordinary categories , the existence of direct sums gives rise to a calculus of matrices . a morphism @xmath367 can be represented by an @xmath368 matrix of morphisms between the summands , and composition can be represented by matrix multiplication . for every object @xmath20 there is a unique span @xmath373 . by uniqueness , any map into @xmath374 must be invertible , and any two such with the same domain must be equal . thus when we compose the span with its opposite , as in @xmath375 , the resulting span is just @xmath376 . now by the universal property of @xmath371 once again , this must just be @xmath377 , and so @xmath378 , and our unique span @xmath379 is a map . clearly coproducts of maps are coproducts , and so the coproduct injections @xmath380 and @xmath381 are also maps . thus the coproducts in @xmath15 will restrict to @xmath0 provided that the codiagonal @xmath382 is a map for all objects @xmath20 . now the fact that the codiagonal composed with the first injection @xmath383 is the identity tells us that we have a diagram as on the left below @xmath384_{i ' } \ar@{=}[dl ] \ar@/^2pc/[ddrr]^{1 } & & & & & x \ar[dr]_{i ' } \ar@{=}[dl ] \ar@/^2pc/[ddrr]^{i } \\ & x \ar[dr]^{i } \ar@{=}[dl ] & & e \ar[dl]_{u } \ar[dr]_{v } & & & x \ar[dr]^{i } \ar@{=}[dl ] & & e \ar[dl]_{u } \ar[dr]_{u } \\ x & & x+x & & x & x & & x+x & & x+x } \ ] ] in which the square is a pullback ; but then the diagram on the right shows that the composite of @xmath385 with the injection @xmath383 is just @xmath386 . similarly its composite with the other injection @xmath387 is @xmath388 , and so @xmath385 is the identity . this proves that the codiagonal is indeed a map , and so that @xmath0 has finite coproducts ; we have already assumed that it has finite limits . to see that @xmath0 is lextensive observe that we have equivalences @xmath389 @xmath116 @xmath111 $ ] suppose that @xmath0 is lextensive . then in particular , it is distributive , so that @xmath390 , and we have @xmath391 which shows that @xmath392 is the coproduct in @xmath15 ; but a similar argument shows that it is also the product . there is a corresponding result involving partial maps in lextensive categories , although the situation there is more complicated as one does not have direct sums but only a weakened relationship between products and coproducts , and a similarly weakened calculus of matrices . see ( * ? ? ? * section 2 ) . there is also a nullary version of the theorem . we simply recall that an initial object in a category @xmath0 is said to be _ strict _ , if any morphism into it is invertible , and then leave the proof to the reader . once again the equivalence @xmath394 is a special case of @xcite .

we present medium - resolution spectroscopy and multicolor photometry for the optical transient psn j09093496 + 3307204 ( sn 2015bh ) in the galaxy ngc 2770 , which has transferred into the supernova phase . the observations were carried out between february 2015 to may 2016 . both at the phase of the sn impostor ( 2015a ) and at the supernova phase ( 2015b ) , besides balmer emissions , the strong fe ii emissions are seen in the spectrum ; so , these spectra resemble those of williams fe ii type classical novae . the star is located near the edge of a dark nebula and notably absorbed ( a = 114 ) . taking into account this absorption , we determined maximum absolute magnitudes of = 150 at the 2015a phase and of = 1814 at the 2015b phase . the light curve at the 2015b phase is similar to those of sn iil . the supernova progenitor is a luminous blue variable ( lbv ) star with the powerful h emission . we considered several hypotheses of supernovae explosions following optical transients related with lbv . the hypothesis of core collapse of an evolved massive star interrupting the process of its merging with massive companion in a binary system ( a failed luminous red nova ) was chosen as the preferable one for this event . [ firstpage ] stars : supernovae , impostors , luminous red novae , supernovae , photometry , spectroscopy stars : individual : sn 2015bh , psn j09093496 + 3307204

1609.00731

the optical transient psn j09093496 + 3307204 was discovered in the galaxy ngc 2770 on february 7 , 2015 in the catalina real - time transient survey , and on february 8 , 2015 by s. hoverton in the process of observations under the program snhunt . the transient is also known as sn 2015bh and snhunt275 . in the spectrum obtained in asiago observatory on february 9 , a broad emission h@xmath3 ( fwhm @xmath4 6800 kms@xmath5 ) with a narrow line with p cygni profile on top was visible @xcite . the absorption component of the profile showed the expansion velocity @xmath4950 kms@xmath5 . in 2008 2009 hst images , a faint star was detected coinciding in coordinates with the transient and variable within 215 228 in the red filter f606w . its spectrum and chaotic variability resembled supernova impostors 2000ch and 2009ip , before the explosion of latter as a supernova in june 2012 @xcite . the definition of `` supernova impostor '' was introduced by @xcite to identify the explosions of lbv stars in which stars survived and being observed several years after outbursts in contrast to sn iin , which were destroyed at the core collapse . the eruption of @xmath6 car in 1844 1850 can be an analog of a sn impostor . so , the transient in ngc 2770 was classified as a sn impostor or an lbv explosion . photometry and spectroscopy with the 10.4 m gtc telescope at canarias within the time interval between march 27 and april 14 , 2015 revealed a considerable increase of star brightness by 12 , and the object reached the absolute magnitude 142 in the sdss @xmath7 band @xcite . in the spectrum taken on april 14 , the h@xmath3 emission , other balmer lines , the distinct emissions fe ii in the range @xmath8 49505400 and the na i / he i blend were predominating . the h@xmath3 line profile was asymmetric with a red - side wing extending up to @xmath45000 kms@xmath5 , and a steeper blue - side decline . but the p cyg type profile in h@xmath3 line was already not observable . on may 16 , 2016 , @xcite informed about a sharp increase of the transient brightness by 2@xmath9 . the absolute magnitude in the @xmath10 band reached in new spectra of may 16 from the keck i telescope , the star displayed a hot continuum with balmer , he ii and he i emission lines @xcite . h@xmath3 remained asymmetric with the half - width fwhm @xmath41200 kms@xmath5 . it became clear that it was a new explosion of the impostor , and the events were developing according to the scenario of sn 2009ip @xcite . the observations continued by @xcite and @xcite confirmed the assumptions that the transient passed to the sn ii phase . by analogy with other papers , we designate the first outburst or the impostor phase as 2015a and the second brightening or the sn phase as 2015b . the host galaxy of sn 2015bh , ngc 2770 is located at a distance of d = 29.70 @xmath13.4 ( distance module @xmath11 = 3235 @xmath10.24 ) ; its radial velocity is @xmath12 = 1947 kms@xmath5 ( redshift z = 0.006494 ) ; the galaxy extinction in this direction is @xmath13 = 0062 ( ned ) . before the 2015 event , three sne of the rare ib type exploded in the galaxy ngc 2770 during a short period of 10 years , that is why the galaxy was called `` a factory of type ib supernovae '' @xcite . these are sne 1999eh , 2007uy and 2008d . the latter one is interesting by the fact that it was related to an x - ray transient . @xcite show in their fig . 1 the galaxy image assembled from three vlt frames of march 16 , 2008 , including one with the filter centered on 6604 at @xmath14 = 0.007 and the width 64 , and another one in the filter h@xmath3 at @xmath14 = 0 with the center at 6563 and the width 61 . besides the two sne of 2007 and 2008 , one can see there a sn 2015bh progenitor , which is distinguished by its excess of emission in the h@xmath3 line . this is one of the brightest stars of the galaxy . photometry in the johnson @xmath15 and cousins @xmath16 bands was taken with the focal reducer scorpio at the bta @xcite in the process of spectral observations . additionally , ccd photometers were used at the 1-m telescope of sao ras , at the 70-cm telescope of sai msu in moscow , at the 60-cm reflector and the 50-cm maksutov meniscus telescope of the crimean observational station of msu , and at the 1-m and 60-cm telescopes of the simeiz observatory in crimea . our observations were performed in the period from february 24 , 2015 to may 30 , 2016 with a break from june 16 to september 10 , 2015 , when the object was not observed due to conjunction with the sun . the first images after the break suitable for measurement were obtained with the 1-m telescope of sao ras in the @xmath17 and @xmath18 filters on september 11 , 2015 , against the dawn background . the standard stars from @xcite near ngc 2770 in the johnson @xmath15 and cousins @xmath16 system were used as comparison stars . altogether we have 43 observations , both multicolor ones and those taken in individual filters . they are listed in table 1 . the full collection of observations of sn 2015bh including published observations of other authors and supplemented by new data is available in internetgoray / psn0909.ne3]@xmath19 . the light and color curves can be examined with a java - compatible browsergoray / psn0909.htm]@xmath19 . the accuracy of measurements is given in table 1 in brackets after each observation in units of the last significant digit . the table comment contains information on telescopes and detectors used in observations . there are small systematic differences between sets of our observations obtained with different devices . these differences were determined from simultaneous measurements and compensated for corrections . after the conjunction with the sun when the star brightness was faint , we were accumulating signal from the star during a long time , 1300 6000 s , to achieve a high signal - to - noise ratio . the main factor of the errors during long accumulation was an inhomogeneous surrounding background of the galaxy , because the star is located in a spiral arm near a dark dust nebula . [ figure1 ] presents fragments of images with the highest angular resolution taken with the hubble space telescope and with the ground - based telescope gemini ( a fragment of the color image by s. rankin in flickr.com ) . the progenitor , an lbv star is identified in the image . the hst images have been already analyzed by @xcite . we have measured the color image of gemini of march 6 , 2008 composed from components in the sdss @xmath20 filters and h@xmath3 filter by decomposing it into rgb components . the faintest stars from the sdss database in the vicinity of the galaxy ngc 2770 were used as comparison stars . at that moment the star brightness was @xmath21 and @xmath22 in the system ab95 . these values correspond to @xmath23 = 252 . in the filters @xmath24 the star was much brighter , 233 @xmath102 . the @xmath25 light curves are shown in fig . [ figure2 ] ( from bottom to top ) . besides our observations , here the data published by @xcite ( converted from sdss values to the @xmath26 system ) and by @xcite are collected . these observations and our ones are shown by points . observations from a collection of s. hoverton , a. cason and r. arbour are plotted on the @xmath23 band light curve by circles . these observations were made with different ccds and mainly without filters , but they are consistent with our observations in the @xmath23 filter rather well . triangles denote observations by zhijian xu and xing gao from the hoverton s collection . the first observation ( 2015 april 7.74 ut , mag = 17.8 , no filter ) is considerably ( by 08 ) brighter than the mean light curve and the observation by @xcite obtained at nearly the same time . it is possible that they registered a short flare . separate observations by g. locatelli ( flickr.com ) and g. masi ( cbat ) are also plotted with circles . the phase 2015a is covered by observations from december 21 , 2014 to april 25 , 2015 , or within jd 2457013 2457138 . at that time photometry shows a gradual rise of brightness with an average rate of 0015 per day ( 075 during 50 days ) . the observed brightness fluctuations in amplitude were of several tenths of a magnitude , there were a short brightness weakening by 09 near jd 2457066 and brightness decline by about 1@xmath9 before the rise to the main maximum . at the phase 2015a , the star achieved the level 184 ( ccd without filter ) . the decline of brightness before the rise to the main maximum ( jd 2457153 , 193 ) registered by hoverton is a very important observation for the understanding the nature of 2015bh , but it should be noted that this observation is single . the sn brightness rise ( the beginning of the phase 2015b ) has been registered starting from may 13 , 2015 ( jd 2457156 ) . the brightness maximum of the phase 2015b was on may 22 , 2015 , at the moment jd 2457165 @xmath27 . maximum magnitudes at the phase 2015b were as follows : @xmath28 = 1475 , @xmath29 = 1557 , @xmath23 = 1535 , @xmath17 = 15.21 , @xmath18 = 15.10 measured with the sao 1 m telescope , and @xmath29 = 1554 , @xmath23 = 1535 , @xmath17 = 1518 measured with the 60 cm telescope of simeiz observatory . in the sn light curve ( 2015b ) , we allocate the initial linear decay of brightness after maximum of may 22 october 17 , 2015 ( jd 24571652457313 ) and the secondary decay after october 17 , 2015 ( jd > 2457313 ) . an abrupt transition in the brightness decay rate from the initial one to the secondary one is seen in all bands except @xmath18 . in the initial decay in the first 50 days , the rates of brightness decline in the @xmath25 filters were , respectively , 20 , 15 , 13 and 10 . a sharp peak in the maximum and a linear brightness decay in the magnitude scale continued during @xmath4150 days after the maximum are typical for light curves of sn iil . a considerable part of this period , the star was inaccessible for observation . however , september observations in the @xmath17 and @xmath18 filters carried out in sao ras right after appearance of the object from behind the sun confirm the linear character of the initial decay . the mean light curve of sne iil shows that the brightness decline rate changes within the range of 22 30 during 50 days in the @xmath29 filter . sn 2015bh demonstrates a lower brightness decay rate ( 20 ) in comparison with this summary light curve . yet , when comparing the decay rate in the @xmath23 filter with three separate sne iil 1980k ( 19 ) , 1998s ( 15 ) and 2008if ( 14 ) @xcite the light curve of sn 2015bh looks typical for sne iil . then , in the secondary decay at the end of october 2015 , the rate of brightness decline reduced considerably , especially in the filter @xmath17 . the slowest brightness decay rate in @xmath17 is obviously related to a gradual increase of contribution of the h@xmath3 emission along with the weakening of continuum in the long - wave range . in fig . [ figure2 ] , the heavy gray line shows the secondary light decay rate due to radioactive decay of the @xmath30co isotope ( as in sne i and in several sne ii ) . the average slope in the 200-day secondary brightness decay in the @xmath23 filter ( where the contribution of emission lines is minimum ) was 00092 per day ( 046 during 50 days ) , what corresponds to the hypothesis of the @xmath30co decay rate and confirms the assumption that the event 2015b was a sn . the color - index @xmath31 , @xmath32 and @xmath33 curves are presented in fig . [ figure3 ] together with the @xmath23 light curve in the scale of absolute @xmath23 magnitudes ( right ) . when plotting the color curves , we used not only simultaneous observations obtained in separate nights , but also the observations in consecutive nights in the interval 12 days . when the star was faint , and the light accumulation time amounted to 2 hours , it was impossible to do that for all filters during one night . on the other hand , in the secondary decay , the changes of the object brightness during 2 days were much lower than measurement errors , so , the color indices are quite realistic . at the phase 2015a , @xmath31 and @xmath32 color indices were gradually reducing , what reflected increasing star s temperature . however , the @xmath33 trend was opposite . at the explosion of sn at the phase 2015b , the star was the hottest in the rising branch at the moment jd 2457159 , i.e. 6 days before maximum , with the color indices @xmath34 = 085 , @xmath31 = + 008 ( this is data of swift / uvot in the system @xmath35 ab @xcite , converted into the vega @xmath15 system ) . this fact is confirmed by spectral observations with the keck telescope @xcite , when the hot continuum and the he ii 4686 emission were observed . taking into consideration the fact that at that time @xmath15 bands were predominated by the hot continuum , and the contribution of emissions was small ( it is expected to be < 3% ) we determined the color excess @xmath36 = 037 @xmath1005 and extinction @xmath13 = 114 @xmath1015 . this estimate of interstellar extinction includes both the extinction in our galaxy and that in the galaxy ngc 2770 together with the circumstellar environment . with due regard to such extinction along the line of sight , the absolute visual magnitude of sn in the brightness maximum was @xmath2 = 1814 @xmath1030 , and in the impostor stage it achieved @xmath2 = 150@xmath103 ( or even @xmath2 = 157 , if relying on the observation by zhijian xu and xing gao discussed above ) . thus , it follows from observations that the event 2015b with its light curve shape and absolute magnitude in maximum can be interpreted as a real sn . from measurements of minimum brightness on march 6 , 2008 , with the telescope gemini @xmath23 = 252 @xmath103 with allowance made for extinction , the absolute magnitude of the sn 2015bh progenitor was m@xmath0 = 83 @xmath103 . this is close to the @xcite limit for the most massive stars and confirms classification of the progenitor as an lbv . stars of so high luminosity are not stable , they lose their mass and form the circumstellar medium rich with gas and dust . starting from the first moment of registration of the sn on the ascending branch of the 2015b outburst , a gradual reddening in all the color indices continued until the break in observations caused by conjunction with the sun . it is obvious that this reddening occurred due to expansion of the ejected optically thick gas shell and decreasing of surface temperature of this shell . however , after the break in observations , the behavior of color indices changed . @xmath31 and @xmath33 indices became bluer and revealed opposite trends in their changes , whereas the reddening of @xmath32 continued with the same rate . we explain that behavior by transition of the shell in the optically thin state , what leads to the fact that the radiation passes from hot inner layers of the shell . however , in the @xmath32 index , the contribution of the h@xmath3 emission increases relative to the hot continuum , and the increase of this index continues . the color index @xmath32 increased by @xmath37 during 9 months after the brightness maximum . similar changes of color indices occur in some classical novae , and they are related to analogous changes in the structure of ejected shells . moderate - resolution spectral observations were carried out with the russian 6 m telescope bta and the focal reducer scorpio @xcite during 8 nights in the period between february 23 , 2015 and may 30 , 2016 . the observations were obtained 88 and 87 days before the sn brightness maximum , in the initial brightness decay 22 days after this maximum , and in the secondary decay , in the time range between 176 to 374 days after the maximum . table 2 contains main information on the spectra : date , and julian date , time in days counted from the brightness maximum , full exposure in seconds , spectral range , spectral resolution , grism , heliocentric corrections to radial velocities , and signal - to - noise ratio in continuum in the middle of the spectral range . naturally , the signal - to - noise ratio in the profiles of spectral lines is much higher . the spectra were processed in os linux using the eso midas package and the long context ( for the long - slit spectra ) . in the red wavelength range of @xmath8 > 6800 , the spectra are distorted by interference pattern ( fringes ) . since the signal - to - noise ratio in spectra obtained in november 2015 and later is too small ( 24 ) , we smoothed them by the moving - average method with the averaging interval 3.5 15 , which is equal to actual spectral resolution of each spectrum . to convert the spectra into energy units , we used spectrophotometric standards by @xcite and simultaneous photometric observations . the spectra are available in digital form in internetbars / spectra / psn0909/ ] . the moments of our spectral observations are plotted in the light curve ( fig . [ figure3 ] , _ sp _ ) . spectral observations published by other authors are also marked in the figure with ( _ ps _ ) ; these spectra or their descriptions are available in internet . [ figure4 ] shows the spectra of sn 2015bh in the blue and green range at the phase 2015a ( top ) and at the phase 2015b immediately after the brightness maximum ( bottom ) . the total spectrum of the star at the phase 2015a composed of the spectra taken on february 23 and 24 , 2015 including the red range is shown in fig . [ figure5 ] ( top ) . approximately , these spectra correspond to descriptions given by @xcite and @xcite . in the spectrum of february 23 , 2015 at the phase 2015a , the h@xmath3 profile consists of a narrow emission line with a weak absorption component shifted to the blue side , both ones superimposed on a broad emission base . the emission peak is at a velocity of 1900 kms@xmath5 , or , taking into account the galaxy redshift , at 47 kms@xmath5 . the width of the narrow emission component at the half - maximum intensity level corrected for the instrument profile , @xmath38 is 640 kms@xmath5 . this component has an emission hump in the red wing at a velocity of + 870 kms@xmath5 . the narrow absorption component is located at a velocity of 900 kms@xmath5 , symmetrically relative to the emission maximum . these velocities are measured relative to the emission peak . the wide wings of h@xmath3 emission extend up to 10000 kms@xmath5 to each side from the peak . the h@xmath39 line profile has the identical structure , there is also a hump at a velocity of + 870 kms@xmath5 . in emission of this line , one can trace only the short - wavelength wing extending up to 8500 kms@xmath5 . the long - wavelength wing is distorted by strong fe ii emission lines . undoubtedly , the wide wings of balmer lines are not related to motions of large mass of material , but they result from thomson scattering of balmer photons by free electrons . the strongest fe ii lines show the p cyg type profiles . in the blue spectrum of june 13 , 2015 ( the phase 2015b ) , a wide and deep absorption component extending to the velocities 4000 kms@xmath5 appeared in the profiles of balmer lines . in the depth of this component , a narrow absorption detail at a velocity 900 kms@xmath5 observed previously at the phase 2015a , remained , and a new absorption detail at a velocity of 1900 kms@xmath5 appeared ( these details were marked in fig . [ figure4 ] ) . similar absorptions with double absorption details were observed in the strong fe ii lines at the phase 2015b . a considerable asymmetry of emission components in the p cyg profiles of h@xmath3 , h@xmath39 , strong emissions fe ii , their triangle shape are due to excess absorption in the blue wing of the profile within the velocity range between 3300 and 250 kms@xmath5 . ( this fact can be set if superimpose the profile with its mirror reflection , as was done , e.g. , in fig . 6 by @xcite for the spectrum of the luminous red nova ngc 4490-ot2011 ) . such a structure of profiles is formed by dense optically - thick ejected shell , its formation continued as early as at the phase 2015a . the observation of hot continuum with the balmer emissions , the lines he ii and he i on may 16 , 2015 by @xcite indicated the exit of shock wave at the brightness rise of the 2015b outburst . because of this , during a month , the high - velocity absorption components appeared in the profiles , but the component with a velocity of 900 kms@xmath5 formed before the 2015 events , remained in the line profiles . it is formed in a medium where the shock wave has not reached yet . at the phase 2015b a part of this medium got acceleration by the light pressure of the outburst , and the second narrow absorption component at a velocity of 1900 kms@xmath5 arose . the interesting observational fact is that the velocity of the narrow absorption component in the blue wing of the balmer line profiles is equal but oppositely directed to the velocity of the emission component ( hump ) in the red wing . this may mean that the initial medium was formed by a bipolar wind which was outflowing at a big space angle . a part of this conic outflow directed to the observer absorbs the light of the optically thick shell , and the opposite part is seen in emission . besides , these narrow components show that the primary circumstellar medium was expanding to considerably larger distances than photosphere of the ejected shell forming the p cyg type profiles . strong fe ii lines are a typical feature of the sn 2015bh spectra at both phases , 2015a and 2015b . with these lines , the spectra resemble ones of @xcite fe ii type classical novae more , than those of supernovae . classical novae are cataclysmic or symbiotic systems comprising a far - evolved star ( a white dwarf ) and a normal companion ( a red dwarf or a red giant ) which is an accretion donor . the outburst of a nova is a thermonuclear runaway of a hydrogen envelope accumulated on the surface of a white dwarf due to accretion . of course , the sn 2015bh explosion scale exceeds the scale of classical novae . its luminosity at the phase 2015a exceeds the maximum luminosity of brightest classical novae 100 times , and at the phase 2015b does 1600 times . the formal resemblance between spectra of novae and sn 2015bh demands a detailed study , because @xcite note presence of the he i and even he ii lines , which are not observed in fe ii type classical novae except rare cases of hybrid novae . classical novae of another type , the williams he / n , have the he i and he ii lines , and this is the main sign of the he / n type , but the iron lines are not found in their spectra . the strongest he i lines in the optical spectrum are expected at the wavelengths 4121 , 4143 , 4388 , 4471 , 4713 , 4922 , 5016 , 5876 and 6678 . resolution of our spectra is sufficient to determine that the he i components are absent in possible blends with strong emissions of fe ii 4922 and 5018 (the multiplet 42 ) , and in the blend with na i 5889/5896 at the phase 2015b . the emission peaks at 4922 and 5018 coincide with other fe ii lines in radial velocity . other he i lines were also not detected . there is a faint detail in the region of he i 4471 in the spectrum of june 13 , 2015 on the 22 day after the maximum at the 2015b phase , the identification of which with the helium line is doubtful . other he i lines including components of the fe ii blends mentioned above are not detectable . besides two mentioned in our spectra , fe ii emissions are reliably identified at the following wavelengths in angstroms ( the multiplet number is in brackets ) : 4179(28 ) , 4233(27 ) , 4303(27 ) , 4352(27 ) , 4417(27 ) , 4523(38 ) , 4556(37 ) , 5169(42 ) , 5198(49 ) , 5235(49 ) , the blend 5276(49)+5284(41 ) , 5317(48,49 ) , 5363(48 ) and 5425 ( 48,49 ) . all these lines are available in the list of emission lines of classical novae in optical range ( table 2 , @xcite ) . besides these lines , we have found two lines at 4590 and 4635 which are absent in this list . both emissions are strong in our spectra of the williams fe ii type classical nova v496 sct obtained on november 10 and 11 , 2009 . according to fig . 1.10 by @xcite , both lines are available in the @xmath6 car spectrum where they are identified as the blends fe ii and cr ii . at the phase 2015a , the absorption component of strongest lines fe ii 4924 , 5018 and 5169 ( multiplet 42 ) reached the velocity 1400 km / s . this is a typical velocity of an ejected shell in classical fe ii type novae . the half - width of the emission component was @xmath4940 kms@xmath5 . at the phase 2015b , the absorption component of the fe ii - line p cyg profiles became stronger as well as in the balmer lines , and the limiting velocities increased up to 4000 kms@xmath5 . such velocities of envelope ejection exceed velocities of fast high - luminosity fe ii type classical novae , but they are small for sne . at that , the half - width of the emission component remained identical to that at the phase 2015a , 940 kms@xmath5 . in june 13 , 2015 spectrum in the sn phase , we observed a very wide emission extending to the blue side of the h@xmath39 line to 18000 km / s with the maximum flux density at @xmath8 @xmath44640 ( it is marked as _ emis . _ in fig . [ figure4 ] below ) . the red part of this emission is overlaid by the absorption component of the h@xmath39 line . unfortunately , the emission is seen only in one spectrum , and we have no observations in the h@xmath3 line near the maximum brightness , so , it needs confirmation by independent observations . so far , the emission of so high - velocity outflow is the only available spectroscopic evidence that the event 2015b is a supernova . it is possible that in the spectrum of may 16 , 2015 by @xcite ( 6 days before maximum ) a fragment of this emission was identified with he ii 4686 . besides , in description of that spectrum , there are no mentions about the strong fe ii lines , which are seen in all our spectra . as a rule , the lines he ii and fe ii are not observed in spectra simultaneously because of difference between ionization potentials at which these lines form , and , naturally , one can see either ones or the others depending on temperature of the excitation source . in a case of a radially symmetric outflow of material at the velocity of 18000 kms@xmath5 , the absorbing and emitting envelope components moving with the velocities less than 4000 kms@xmath5 would be swept out by this material , or the high - velocity gas would underwent a fast deceleration . nevertheless , a so unusual symbiosis of properties is observed in the spectrum . this symbiosis can be explained only by the axial asymmetry of outflow at the sn explosion . that is to say , the high - velocity gas was ejected along the polar axis whereas the slower outflow was occurring in equatorial directions . it is not excluded that in the equatorial plane there already was a dense gas disk which was formed at an earlier stage of evolution or at the phase 2015a and it prevented from distribution of high - velocity gas in the equatorial plane . for example , spectropolarimetry of sn 2009ip by @xcite in the analogous phases 2012a and 2012b during its outburst in 2012 has shown that the outflows in these phases were strongly non - spherical , and the angles of polarization plane were orthogonal . in opinion by @xcite , in the case of 2009ip , the ejecta generated at the explosion of 2012a , in the phase 2012b were colliding in the equatorial plane with a flattened ( toroidal ) structure . evidences of bipolar ejections at lbv explosions are not rare ( this is seen in the balmer line profiles , e.g. , by @xcite , or in a bipolar structure of the `` homunculus '' nebula around the galactic lbv - type star @xmath6 car ) . at the case of sn 2015bh , probably , there was a `` shot '' of high - velocity gas in the observer direction related with the phase 2015b , whereas an opposite ejection was overlapped by a body of a thick equatorial disk or a toroidal structure at a relatively small inclination angle between this structure and the line of sight . the spectra of the object in the blue and green ranges obtained at the phase 2015b after it appeared from behind the sun , i.e. at the secondary decay of brightness ( 176374 days from the brightness maximum ) are shown in fig . [ figure6 ] , and the total spectrum in day 209 in fig . [ figure5 ] ( bottom ) . in blue spectra , strengthening of the short - wave emission component in h@xmath39 line profiles was seen , this component being located at the place of the absorption component early observed near the light maximum . the same emission components grew at the same part of fe ii profiles what led to appearance of an extended blend in the region @xmath84400 4700 and 5150 5450 . if there were no strong hydrogen lines , such a spectrum would be similar to that of sn ic 1994i , 2003jd , 2010gx , 2004aw ( see fig . 4 by @xcite ) . changes in the h@xmath39 profiles are shown in fig . [ figure7 ] . note that the december 2015 spectrum was obtained with the low spectral resolution . in june 13 , 2015 spectrum , the absorption component extended to the short - wave side from 400 to 4000 kms@xmath5 . this velocity range in the profile was gradually filled with emission , and on march 10 , 2016 we already see in the spectrum a strong emission with a peak at 1300 kms@xmath5 . in the h@xmath3 profiles shown in fig . [ figure8 ] , the similar effect is seen ( except a profile with the absorption component of june 13 , 2015 ; at that time the red - range spectrum was not obtained ) . in the may 30 , 2016 spectrum , the continuum was already lower than the noise level , but the h@xmath3 profile remained emissive and kept the double shape . the identical behavior was described at late stages of light curve of the luminous red nova ngc 4490-ot2011 @xcite . some difference from the event 2015bh was a smaller range of the absorption component velocities , from 0 to 650 kms@xmath5 , so , the blue emission wing of the h@xmath3 profile was covered with the absorption only partly , and an emission hump was observed at 650 kms@xmath5 . the maximum absorption depth in this profile , and then the maximum strength of emission were observed at the velocity of 280 kms@xmath5 . these velocities are 45 times lower than those ones of sn 2015bh . @xcite give an entirely plausible explanation of such evolution of the profile : `` a massive shell was ejected at these speeds and was initially seen in absorption , but as the underlying photosphere cooled ( as indicated by the spectral evolution during the decline from peak ) and as the optical depth dropped , the same dense shell is seen in h@xmath3 emission . this likely indicates ongoing shock heating of this ejected shell '' . to explain transition from absorption to emission in 2015bh one should add a factor of fast expansion of the ejected shell ( @xmath12 up to 4000 kms@xmath5 ) causing it to become optically thin quickly . @xcite interpreted the event ngc 4490-ot2011 as a luminous red nova , the merger in a massive system . as a rule , the expanding envelopes of luminous red novae do not transit to the optically thin phase at all , and if they do , it turns out to be cold unionized and even molecular gas , as in the case of v4332 sgr . note that the ngc 4490-ot2011 light curve had two maxima , and in the second outburst ( with a cold - star spectrum ) it achieved a record absolute magnitude 14 at merging of cores of massive stars , a shock wave forms , and when it comes out on the surface , the first brightness maximum is observed ( an example is lrn 2015/m 101 , also a massive system studied by @xcite ) . the relation between sne impostors and lbvs is established quite reliably . some sne iin are also obviously related with lbvs . specific evidences of these relations were considered in detail by @xcite and in studies cited there . however , only four impostors are known which eruptions led to appearance of sne iin . besides the above - mentioned event 2009ip , these are 2010mc @xcite , 2011ht @xcite , and lsq13zm @xcite . sn 2015bh demonstrates a certain similarity with these objects , but there are differences both in light curves and in spectra . the nature of lbv explosions , which lead to `` detonation '' of more large - scale explosions of sne still remains unclear , although there are several hypotheses explaining this phenomenon . \(1 ) ppi `` pulsation pair instability '' , related to massive formation of electron - positron pairs in cores of massive hot stars , which results in a reduction of radiation pressure , a partial collapse of a star and a thermonuclear explosion . the thermonuclear explosion stops the collapse and gives rise to ejection of the envelope . in this case , the repeated outburst can be explained by interaction between ejecta and dense circumstellar medium @xcite . the detailed review of final stages of evolution and mechanisms of sne in massive stars is given by @xcite . it was established that the thermal energy released at the collision of rapidly moving ejecta with the dense circumstellar medium may be higher than that of the core collapse . note that the ppi mechanism is valid for stars with initial mass within 80 280 m@xmath40 , and it can generate repeated eruptions of an impostor , similar to the case of sn 2009ip , and , eventually , leads either to the core collapse and formation of a black hole , or to a full destruction of the star ( at the photodisintegration of nickel for a mass within the range 130 280 m@xmath40 ) . \(2 ) core collapse of a faint supernova and interaction of ejecta with a medium @xcite . \(3 ) repeated outbursts related with approaching of components at elliptical orbits with the interaction in a binary system , and a final outburst at the merger or direct collision ( this hypothesis was considered for sn 2009ip , in which several outbursts of an impostor were observed before the sn explosion @xcite ) . \(4 ) last explosion of lbv ( a sn impostor ) followed by the core collapse ( sn iin ) @xcite . these hypotheses were considered in detail by @xcite . we think one more scenario is possible . \(5 ) merger of components of a massive system with formation of a common envelope broken by the core collapse of one of stars being at a later evolutionary stage because of an excess of the critical mass ( a failed luminous red nova in a massive binary system ) . in such a case , the process of the merger of stars does not need to end by merging of their cores , and then the sn explosion leads to a binary system containing an optical remnant of a companion ( the accretion donor ) + a compact component . the scenario determines if a sn is a final stage of evolution of a massive star finished by a compact object , or it is a single event in an lbv history . this question is to be solved by observations of remnants of similar double explosions . some events were discovered as sn iin ( e.g. , lsq13zm ) , but then their relation with an impostor might be detected from archive data . are all type iin sne related to previous lbv explosions ? this question also requires special studies . now let us consider how each hypothesis agrees with observations of sn 2015bh . \(1 ) the ppi hypothesis . can the spectral similarities of sn 2015bh with classical fe ii type novae we found in this outburst be interpreted in favor of this hypothesis ? classical novae can be attributed based on their spectra to one of two types , he / n and fe ii , or they can be `` hybrid '' with transition from fe ii to he / n at an outburst . @xcite explains this difference by the origin of gas ejected in the explosion . the he / n type novae have in their spectra very wide emission lines ( a high velocity of ejection ) of elements higher ionization degree with the rectangular profiles and almost always without absorption components . this means that the gas is ejected at a thermonuclear runaway just from the white dwarf . spectra of fe ii type novae at the initial decay of light curve are characterized by numerous fe ii lines of low ionization degree which were excited by atomic collisions , though in the far red range there are lines of cno elements which are excited at recombination and fluorescence scattering . profiles of these lines are rounded , more narrow and often have p cyg - type absorption features of optically thick expanding gas . the intensities of the fe ii emissions decrease relatively slowly as the brightness decreases , and occasionally increase again in the episodes of `` secondary maxima '' which happen during several months after the beginning of the outburst . in the brightness maximum and in secondary maxima at the decay of brightness the emitting gas is colder than in other phases . at the early decay of brightness , fe ii novae display narrow absorptions of heavy elements of the iron peak . this means that the spectrum forms in the gas shell with the element composition close to the solar one . @xcite assumes that this gas arises from the secondary component , a normal star rather than from a white dwarf . probably , it is not even mixed with the explosion ejecta , and saturated with heavy elements generated in the process of evolution of the secondary star . hydrodynamic calculations show that the radiation and the material thrown off the white dwarf stimulate the secondary component to lose the mass through the lagrange points l1 and l3 , and the shock from the orbital motion and accretion disk forms a circumstellar shell . at the same time , it was established that hybrid novae develop only from the type fe iib novae ( with broad lines in spectra ) , but not from fe iin ones ( with narrow lines ) . the arguments against the ppi hypothesis are as follows : \(a ) the ppi hypothesis does not suppose that there is a component in a binary system . however , from the observed spectrum of sn 2015bh it follows that the composition of elements in the ejected shell and stellar wind at the outburst is close to solar one or only slightly evolved . as in the classical fe ii type novae , this material can originate in accretion from the secondary companion . \(b ) within the context of this hypothesis , the ppi event and thermonuclear explosion explain the phase 2015a , and the transformation of kinetic energy of ejecta into emission at interaction with circumstellar medium explains the 2015b phase . photometric and spectral signs of sn at the 2015b phase noted by us previously , contradict the ppi hypothesis . \(c ) the time of wave propagation inside the envelope released at the 2015a phase can be estimated as an interval between the end of the 2015a phase ( the beginning of brightness decay at the moment jd 2457138 , mag = 186 according to observations by hoverton ) and the outburst peak at the 2015b phase jd 2457165 , @xmath23 = 1535 . this is equal to 27 days . such a term is too large for propagation of dynamic changes , but it is typical for transfer of heat and radiation in massive envelopes ( in the luminous red novae ) . \(2 ) core collapse of a low - mass star , a faint supernova and interaction between the ejected shell and circumstellar medium . at that , the 2015a phase is interpreted as an sn iip outburst , and the 2015b phase as interaction between the ejecta and environment . the examples of faint core - collapse sne iip reaching the absolute magnitude 14@xmath9 in the maximum brightness , 1999br and 2010id are adduced by @xcite . for sn 2015bh there are no evidences of a supernova outburst in the phase 2015a , although at the 2015b phase they are available : the rate of the secondary decay in the light curve is the same as the calculated @xmath30co decay rate , emission in the h@xmath39 line with the velocity up to 18000 km / s . these are arguments against the hypothesis about a faint sn . \(3 ) approaching and final merging of components , a massive evolved star of mass 60 - 100 m@xmath40 in a system with a star of mass 1250 m@xmath40 moving on an elliptic orbit . paper by @xcite gives weighty arguments in favor of this hypothesis for sn 2009ip . we have no such arguments for the event 2015bh repeated outbursts of the impostor or humps at the brightness decay at the burst of 2015b . although @xcite detected `` short - term variability at least during 21 years '' before the event of 2015 . the merger of components even for massive stars leads to the phenomenon of a red nova , a cold explosion , but we have no spectral evidence of a red nova . \(4 ) the last explosion of lbv ( the sn impostor ) and a core collapse ( sn ii ) . the ppi event which provoked collapse of a massive core of lbv at a late stage of its evolution may be considered as the cause of explosion . the late stage of lbv evolution can be testified by unusually strong lines of iron . iron could get to the star envelope due to the mixing . this is one of the promising hypotheses . in that case the explosion remnant will be a single black hole . \(5 ) collapse of a core of a massive evolved lbv star in the system in a merger with a massive companion caused by exceeding critical mass of the core ( a failed red nova in a massive binary system ) . e.g. , in the galactic system in the orbit around lbv there is an o - class star of early or middle subclass with a mass within the range which passed a less evolutionary track than the main more massive component @xcite . before the 2015b outburst , a gradual rise of the stellar brightness was observed which finished by a decline by 1@xmath9 . the 2015a phase is usually called a sn impostor , but impostor light curves do not show gradual increase of brightness during outbursts . on the contrary , they show a decline typical for sne ( maybe , except sn 1954j ) . such a brightness rise is typical for luminous red novae , and it is related with the formation of a common envelope at merger . the similar effect was observed in the luminous red novae v4332 sgr , v1309 sco , lrn 2015/m31 ( m31n 2015 - 01a ) and lrn 2015/m 101 ( but it was absent in v838 mon ) . the brightness decline by 1@xmath9 before the luminous red nova outburst was observed in the case of v1309 sco . @xcite explain this decline by the merger of star cores and a slow shock from the star center which leads to transfer of the envelope in the expansion mode in the regime close to the adiabatic one . the term `` slow shock '' was first used by @xcite to explain the phenomenon of luminous red nova for v4332 sgr . at the merger of star cores in the massive red novae ngc 4490-ot2011 @xcite and lrn 2015/m 101 @xcite , a shock wave formed , and , after it came out to the surface , the envelope passed to the mode of adiabatic expansion , which is accompanied by a deep weakening of brightness before the secondary maximum . up to now , no spectra were obtained for the stage of common envelope formation in a luminous red nova , and , most probably , they are not similar to those of classical type fe ii novae . such a spectrum of sn 2015bh at the 2015a phase can be explained by accretion of hydrogen - rich material from the companion onto the evolved star ( lbv ) at the phase of a common envelope formation , development of thermonuclear burning in a layer and release of additional energy by the accretion . at a high luminosity exceeding the eddington limit , the mass outflow occurs with the typical rate for classical novae in outburst . at the merger of stars , the angular momentum is carried off with the material through the lagrange point l1 , which results in formation of a disk surrounding the system . formation of a thick disk around the massive star inside a common envelope can also be expected . in the moment when the critical mass was exceeded , the core collapse of evolved star occurred , and under such conditions , the explosion was asymmetric , with high - velocity gas outflow in the direction of axis of the inner disk . in the specific case of sn 2015bh , we observe the luminosity decline at the end of the 2015a phase , i.e. the slow shock already occurred , and the adiabatic expansion of envelope started . however , the core collapse stopped the adiabatic expansion of the envelope and interrupted the scenario of a luminous red nova . that is why the interrupted scenario can be called `` a failed luminous red nova '' . propagation of the shock wave inside the common envelope and its coming out to the surface can explain the light curve at the 2015b phase with its high luminosity and a sharp peak in maximum , along with a high temperature at the 2015b outburst decay ( in contrast to luminous red novae ) , and transition of the envelope to the optically thin state . the abundance of hydrogen in the spectrum of 2015bh , the origin of which can be in the envelope of the less evolved companion , can argue for hypothesis ( 5 ) . this hypothesis predicts a remnant being a binary system with a relativistic component , if the collapse occurred earlier than the total merger of the cores . the latter hypothesis explains the events observed in the light curve and in the star spectrum in details . however , some components of the scenario ( the high - velocity ejection in the h@xmath39 profile , the luminosity decline at the transition of the envelope to the mode of adiabatic expansion ) are based on single observations and need a confirmation . the question of whether the already expanding envelope can transfer to the adiabatic mode due to a shock from within the star center remains open . the dynamical simulations are required . on april 7 , 2015 zhijang xu and xing gao observed a short burst which may be related with the exit of a primary , weaker shockwave . this observation should be verified . the question of whether the 2015bh event is a final episode in the evolution of a massive star is not yet solved , too . it will be answered by the study of the explosion remnant with large telescopes at a level of stellar magnitudes larger than 25 spectra of the optical transient 2015bh / ngc 2770 in the phases of sn impostor and real sn obtained with bta / scorpio are unusual , and they are similar to spectra of williams fe ii type classical novae . from spectral data it was established that in the impostor phase ( 2015a ) , an optically thick expanding shell formed which was passed by a shock wave at the sn phase ( 2015b ) . eventually , the shell was accelerated and transferred to the optically thin phase . besides , we have revealed the absorption components in spectral line profiles related to the extended circumstellar medium which partly accelerated by radiation pressure of sn explosion . near brightness maximum , we have detected the emission component of high - speed , 18000 kms@xmath5 ejecta in the h@xmath39 line profile along with slower moving , up to 4000 kms@xmath5 absorption components , what suggests the asymmetric eruption . the light curve resembles the sn iil ones with the rate of the secondary decline corresponding to the rates caused by the radioactive decay of @xmath30co isotope . in maximum , the sn reached the absolute magnitude @xmath2 = 1814 @xmath1030 . the sn 2015bh progenitor is a luminous blue variable star ( lbv ) with the strong emission h@xmath41 , one of the brightest stars of the galaxy ngc 2770 . we consider that the most probable hypothesis to explain the sn 2015bh event is a core collapse of a more massive far - evolved star in a binary system breaking its merging with a less massive companion ( a failed luminous red nova in a massive binary system ) . arguments for and against other hypotheses were also considered . the core collapse of a single massive star induced by the ppi event is also possible . in the paper we used the sloan digital sky survey database , the nasa / ipac extragalactic database ( ned ) , the hubble space telescope archive , the hosting site of images flickr.com , the databases of supernovae by rochester astronomy and sternberg astronomical institute of the moscow university . the spectral and photometric observations performed in the special astrophysical observatory of the russian academy of sciences , their processing and analysis were financed by the russian science foundation ( rsf ) by the grant no.145000043 . the operation of the russian 6 meter telescope bta is financially supported by the ministry of education and science of the russian federation . , e.a.b . and a.f.v . thank the russian foundation for basic research ( rfbr ) for the financial support of this work with the grant 140200759 . thanks rfbr for the support by the grant 16020758 . the research by d.yu.ts . was partially supported by the rsf grant 161210519 . the work by i.m.v . was partially supported by a grant of the slovak information agency saia . afanasiev v. l. , moiseev a. v. , 2005 , astron . lett . * 31 * , 194 anderson j. p. , gonzlez - gaitn s. , hamuy m. , et al . , 2014 , apj * 786 * , 67 barsukova e. a. , goranskij v. p. , valeev a. f. , kaisin s. s. , 2014 , variable stars * 34 * , no . 4 , p. 1 barsukova e. a. , goranskij v. p. , valeev a. f. , zharova a. v. , 2014 , astrophysical bulletin * 69 * , 67 campana s. , thne c. c. , leloudas g. , et al . , 2015 , astronomer s telegram no . 7517 corcoran m. f. , ishibashi k. , 2012 , in _ eta carinae and the supernova impostors_. ed . k. davidson & r. m. humphreys ( springer ) , apss library , * 384 * , p. 195 de ugarte postigo a. , leloudas g. , thne c. c. , et al . , 2015a , astronomer s telegram no . 7409 de ugarte postigo a. , thne c. c. , lelaudas g. , aceituno f. , 2015b , astronomer s telegram no . 7514 duggan g. , bellm e. , leloudas g. , et al . , 2015 , astronomer s telegram no . 7515 elias - rosa n. , benetti s. , tomasella t. , et al . , 2015 , astronomer s telegram no . 7042 fraser m. , magee m. , kotak r. , et al . , 2013 , apj * 779 * , l8 goranskij v. p. , barsukova e. a. , spiridonova o. i. , et al . , 2016 , astrophysical bulletin * 71 * , 82 graham m. l. , sand d. j. , valenti s. , et al . , 2014 , apj * 787 * , 163 heger a. , 2012 , in _ eta carinae and the supernovae impostors _ , ed . k. davidson & r. m. humphreys ( springer ) , apss library , * 384 * , p. 299 humphreys r. m. , davidson k. , 1979 , apj * 232 * , 409 humphreys r. m. , martin j. c. , 2012 , in _ eta carinae and the supernova impostors_. ed . k. davidson & r. m. humphreys ( springer ) , apss library , * 384 * , p. 1 margutti r. , milisavljevic d. , soderberg a. m. , et al . , 2014 , apj * 780 * , 21 martini p. , wagner r. m. , tomaney a. et al . , apj * 118 * , 1034 mauerhan j. c. , smith n. , filippenko a. v. , et al . , 2013 , mnras * 430 * , 1801 mauerhan j. , williams g. g. , smith n. , et al . , 2014 , mnras * 442 * , 1166 modjaz m. , li w. , butler n. , et al . , 2009 , apj * 702 * , 226 oke j. b. , 1990 , aj * 99 * , 1621 pastorello a. , smartt s. j. , boticella m. t. , et al . , 2010 , apj * 724 * , l16 pastorello a. , cappellaro e. , inserra c. , et al . , 2013 , apj * 767 * , 1 richardson n. , artigau e. , 2015 , astronomer s telegram no . 7543 smith n. , mauerhan j. c. , prieto j. l. , 2014 , mnras * 438 * , 1191 smith n. , andrews j. e. , van dyk s. d. , et al . , 2016 , mnras * 458 * , 950 soker n. , & kashi a. , 2013 , apj * 764 * , l6 tartaglia l. , a. pastorello a. , sullivan m. , et al . , 2016 , mnras * 459 * , 1039 thne c. c. , michalowski m. j. , leloudas g. , et al . , 2009 , apj * 698 * , 1307 thne c. , de ugarte postigo a. , leloudas g. , et al . , 2016 , in _ frontiers of massive - star evolution and core - collapse supernovae_. symposium s16 at ewass 2016 . 1133 , abstract van dyk s. d. , schuyler d. , peng c. y. , et al . , 2000 , pasp * 112 * , 1532 vinko j. , k. sarneczky k. , vida k. , 2015 , astronomer s telegram no . 7541 williams r. , 2012 , aj * 144 * , 98 woosley s. e. , blinnikov s. , heger a. , 2007 , nature * 450 * , 390 l|c|c|c|c|l jd hel.24 ... & @xmath29 & @xmath23 & @xmath17 & @xmath18 & remark + 57076.491 & 19.65 ( 5 ) & 19.26 ( 5 ) & 18.71 ( 4 ) & 18.60 ( 5 ) & 6 m + 57078.373 & 19.58 ( 8) & 19.19 ( 5 ) & 18.576 ( 40 ) & 18.44 ( 5 ) & 6 m + 57100.284 & 19.372 ( 50 ) & 18.841 ( 30 ) & 18.446 ( 20 ) & - & so + 57164.255 & 15.579 ( 40 ) & 15.376 ( 20 ) & 15.214 ( 20 ) & 15.167 ( 40 ) & so + 57164.266 & 15.581 ( 40 ) & 15.368 ( 20 ) & 15.214 ( 20 ) & - & so + 57165.248 & 15.566 ( 40 ) & 15.347 ( 20 ) & 15.205 ( 20 ) & 15.101 ( 40 ) & so@xmath42 + 57165.338 & 15.54 ( 5 ) & 15.35 ( 3 ) & 15.18 ( 4 ) & - & iw + 57166.320 & - & 15.272 ( 20 ) & 15.136 ( 20 ) & - & so + 57167.28 & 15.572 ( 60 ) & 15.362 ( 50 ) & 15.265 ( 40 ) & 14.978 ( 50 ) & iw + 57168.29 & 15.611 ( 70 ) & 15.387 ( 50 ) & 15.193 ( 50 ) & 15.007 ( 50 ) & iw + 57169.32 & - & - & 15.24 ( 6 ) & - & a2 + 57174.27 & 15.906 ( 70 ) & 15.558 ( 50 ) & 15.399 ( 50 ) & 15.232 ( 50 ) & iv + 57175.28 & 16.022 ( 80 ) & 15.618 ( 20 ) & 15.453 ( 50 ) & 15.224 ( 50 ) & iv + 57175.30 & 16.002 ( 70 ) & 15.645 ( 20 ) & 15.423 ( 50 ) & 15.213 ( 50 ) & iv + 57176.36 & 16.137 ( 70 ) & 15.714 ( 50 ) & 15.444 ( 50 ) & 15.249 ( 50 ) & iv + 57177.34 & 16.106 ( 50 ) & 15.665 ( 30 ) & 15.415 ( 50 ) & 15.09 ( 1 ) & iv + 57179.31 & 16.084 ( 70 ) & 15.768 ( 50 ) & 15.497 ( 50 ) & 15.232 ( 50 ) & iv + 57180.3 & 16.213 ( 60 ) & 15.766 ( 50 ) & 15.450 ( 40 ) & 15.07 ( 1 ) & a5 + 57181.3 & 16.310 ( 60 ) & 15.863 ( 50 ) & 15.547 ( 50 ) & 15.281 ( 50 ) & iv + 57182.3 & 16.252 ( 60 ) & 15.883 ( 50 ) & 15.587 ( 40 ) & 15.340 ( 40 ) & a5 + 57183.3 & 16.301 ( 60 ) & 15.868 ( 50 ) & 15.691 ( 40 ) & 15.366 ( 40 ) & a5 + 57184.257 & - & 15.912 ( 20 ) & 15.676 ( 40 ) & - & so + 57184.27 & 16.420 ( 70 ) & 15.923 ( 20 ) & 15.639 ( 30 ) & 15.330 ( 40 ) & iv + 57185.272 & 16.513 ( 30 ) & 15.931 ( 20 ) & 15.644 ( 20 ) & 15.374 ( 40 ) & so@xmath43 + 57186.29 & 16.541 ( 70 ) & 15.936 ( 20 ) & 15.690 ( 40 ) & 15.384 ( 40 ) & iv + 57187.285 & - & 16.005 ( 30 ) & - & - & 6 m + 57188.33 & 16.648 ( 70 ) & 16.090 ( 50 ) & 15.777 ( 40 ) & 15.487 ( 50 ) & iv + 57190.3 & 16.726 ( 60 ) & 16.073 ( 50 ) & 15.805 ( 50 ) & 15.402 ( 40 ) & a5 + 57276.55 & - & - & 18.31 ( 20 ) & 17.49 ( 15 ) & so + 57313.59 & - & - & 19.45 ( 11 ) & - & fl + 57330.58 & - & - & 19.64 ( 10 ) & - & a5 + 57331.58 & - & - & - & 18.71 ( 8) & a5 + 57332.58 & - & 20.33 ( 20 ) & - & - & a5 + 57340.45 & - & - & 19.77 ( 14 ) & - & a5 + 57340.54 & 21.37 ( 17 ) & 20.67 ( 12 ) & 19.57 ( 2 ) & - & 6 m + 57369.50 & - & 20.84 ( 22 ) & 19.69 ( 3 ) & - & so + 57373.53 & 21.48 ( 6 ) & 20.85 ( 08 ) & 19.71 ( 11 ) & - & 6 m + 57426.483 & - & 21.60 ( 12 ) & 20.33 ( 2 ) & 20.27 ( 20 ) & so + 57456.468 & 22.16 ( 9 ) & 21.95 ( 18 ) & 20.48 ( 3 ) & - & so + 57457.404 & - & - & - & 20.58 ( 11 ) & so + 57458.326 & 21.96 ( 12 ) & 21.56 ( 12 ) & 20.20 ( 5 ) & - & 6 m + 57512.311 & - & 21.96 ( 17 ) & 20.47 ( 5 ) & - & so + 57539.3 & - & 22.4 ( 9 ) & - & - & 6 m + + + + + + + + + + + + + c|c|r|c|c|c|c|c|c date ut & jd hel.24 ... & @xmath44 & @xmath45 , s & @xmath8 , & r , & grism & @xmath46 , kms@xmath5 & s / n + 2015.02.23 & 57076.5160 & -88 & 3600 & 5751 7501 & 5.5 & vphg1200r & -11.7 & 22 + 2015.02.24 & 57078.3933 & -87 & 3600 & 4053 5847 & 5.0 & vphg1200 g & -12.7 & 15 + 2015.06.13 & 57187.2982 & + 22 & 1800 & 4056 5850 & 5.0 & vphg1200 g & -21.2 & 30 + 2015.11.14 & 57340.5885 & + 176 & 3600 & 4048 5846 & 5.0 & vphg1200 g & + 28.4 & 3 + 2015.12.17 & 57373.5602 & + 209 & 4800 & 3740 7877 & 14.6 & vphg550 g & + 20.5 & 2 + 2016.03.10 & 57458.2293 & + 293 & 3600 & 4043 5846 & 5.5 & vphg1200 g & -18.3 & 2 + 2016.03.12 & 57460.3973 & + 295 & 2400 & 6024 7096 & 3.5 & vphg1800r & -19.4 & 4 + 2016.05.30 & 57539.3346 & + 374 & 2400 & 6025 7098 & 3.0 & vphg1800r & -24.9 & @xmath42 + +

we construct a new godunov type relativistic hydrodynamics code in milne coordinates , using a riemann solver based on the two - shock approximation which is stable under existence of large shock waves . we check the correctness of the numerical algorithm by comparing numerical calculations and analytical solutions in various problems , such as shock tubes , expansion of matter into the vacuum , landau - khalatnikov solution , propagation of fluctuations around bjorken flow and gubser flow . we investigate the energy and momentum conservation property of our code in a test problem of longitudinal hydrodynamic expansion with an initial condition for high - energy heavy - ion collisions . we also discuss numerical viscosity in the test problems of expansion of matter into the vacuum and conservation properties . furthermore , we discuss how the numerical stability is affected by the source terms of relativistic numerical hydrodynamics in milne coordinates .

1607.03630

relativistic hydrodynamics has been widely used for description of macroscopic dynamics in various fields ranging from nuclear physics to astrophysics . high - energy heavy - ion collision experiment is one of the active areas of relativistic hydrodynamics applications . in 2005 at relativistic heavy ion collider ( rhic ) , production of strongly - interacting quark - gluon plasma ( qgp ) was achieved , which was supported not only by the experimental data but also theoretical analyses @xcite . studies based on relativistic hydrodynamics have shown remarkable success in understanding various observables such as particle distributions , collective flows , particle correlations and so on @xcite . the strong elliptic flow at rhic is a highlight of the success of hydrodynamic models and is one of the evidences that the qgp is not a weakly interacting gas but a strongly interacting matter . since then the hydrodynamic model has been one of the promising phenomenological models for the description of dynamics of hot and dense matter produced in the heavy - ion collisions . construction of relativistic viscous hydrodynamic model has been of practical importance and has formed a basis for the analyses of the heavy - ion collisions @xcite . in the last decade , the hydrodynamic model itself has also been developed through the analyses of experimental data of heavy - ion collisions at rhic and the large hadron collider ( lhc ) . by comparing the hydrodynamic model calculations and the experimental observables such as particle distributions and collective flows , detailed bulk properties of qgp such as the qcd equation of state and its transport coefficients have been investigated . also , physical qcd equation of state is now available by lattice qcd simulations at vanishing chemical potential @xcite and is applied to the hydrodynamic model . this enables us to bridge the first principle lattice qcd simulations and the experimental data in the heavy - ion collisions . in spite of the success of hydrodynamic models in high - energy heavy - ion collisions , there are still several issues under discussion . currently the hydrodynamic models often adopt israel - stewart theory @xcite and a second - order viscous hydrodynamics from ads / cft correspondence @xcite as their basic equations . however , we have not reached the conclusion on which relativistic viscous hydrodynamic equation is suitable for the description of relativistic heavy - ion collisions . this is because the extension from relativistic ideal hydrodynamic equation to viscous hydrodynamic equation is not straightforward and several possible candidates exist . also viscous second - order anisotropic hydrodynamics is proposed which reproduces the exact solution of the boltzmann equation in the relaxation - time approximation @xcite . furthermore it remains as an enormous challenge to understand why hydrodynamics can be applied to the dynamics shortly after a heavy - ion collision takes place . conclusive understanding of the mechanism of thermalization and hydrodynamization in such a short time scale is still missing . here we emphasize that numerical algorithm for solving the relativistic hydrodynamic equation is one of the important ingredients in developing the hydrodynamic models . recent high statistical experimental data at rhic and the lhc impose more rigorous numerical treatment on the hydrodynamical models . for example , at rhic and the lhc , higher harmonic anisotropic flow , which is expressed by the higher fourier coefficient of particle yields as a function of azimuthal angle , is reported @xcite . the origin of the higher harmonics is considered to be event - by - event initial fluctuations in the particle distributions . when comparing with those high statistical data , reducing the numerical dissipation of the numerical algorithm for relativistic hydrodynamic equations should allow us an access to more precise value of transport coefficients of the qgp . usually each algorithm has advantages or disadvantages in terms of coding , computational time , numerical precision and stability . up to now , unfortunately , only little attention has been paid to the numerical aspects in the hydrodynamic models for high - energy heavy - ion collisions . recently we developed the state - of - the - art numerical algorithm for solving the relativistic hydrodynamic equation with the qgp equation of state @xcite . in the algorithm , we use a riemann solver based on the two - shock approximation @xcite which is stable under existence of large shock waves @xcite . the new numerical scheme is stable even with small numerical viscosity and can reduce the numerical uncertainly when extracting the physical viscosity of the qgp from experimental data . however , this algorithm in ref . @xcite is developed in cartesian coordinates . meanwhile , at the high - energy heavy - ion collisions such as rhic and the lhc , the expansion in longitudinal direction is rapid compared with that in transverse direction . for the description of space - time evolution of high - energy heavy - ion collisions , milne coordinates are more suitable than cartesian coordinates . therefore we extend our algorithm of relativistic ideal hydrodynamics in cartesian coordinates to that in milne coordinates so that we can efficiently apply it to the analyses of high - energy heavy - ion collisions . the present article is organized as follows . we begin in section ii by showing basic equations for the hydrodynamic models in milne coordinates . in section iii we explain the numerical algorithm ; riemann problem in milne coordinates and our numerical scheme . section iv is devoted to several numerical tests , such as relativistic shock tubes , comparison with analytic solutions which describe the dynamics of realistic high - energy heavy - ion collisions . in addition , we discuss the propagation of longitudinal fluctuations . in section v we investigate the conservation property of our code . we end in section vi with our conclusions . relativistic hydrodynamics is based on the conservation equations of net charge , energy and momentum , @xmath0 where @xmath1 is the baryon number current and @xmath2 is the energy - momentum tensor . for the ideal fluid , the energy - momentum tensor and the baryon number current are given by @xmath3 where @xmath4 is the baryon number density , @xmath5 is the energy density , @xmath6 is the pressure , @xmath7 is the normalized four - velocity of the fluid , @xmath8 and @xmath9 is the metric tensor . in high - energy heavy - ion collisions at rhic and the lhc , approximate invariance under the longitudinal lorentz boost is observed in particle rapidity distributions around mid - rapidity @xcite . in such situations , milne coordinates are suitable for the description of space - time evolution of the hot and dense matter after the collisions . milne coordinates @xmath10 and @xmath11 are described by the rapidity @xmath12 and the proper time @xmath13 with cartesian coordinates . the coordinate transformation of the four - velocity between milne coordinates and cartesian coordinates is given by @xmath14 where the transverse components of the four - velocity , @xmath15 and @xmath16 are the same in both coordinates . the four - velocity in milne coordinates is written by three - dimensional velocity as @xmath17 where @xmath18 is the three - dimensional velocity in milne coordinates and @xmath19 represents the lorentz factor , @xmath20 the coordinate transformation of the three - dimensional velocity vector between milne coordinates and cartesian coordinates is given by @xmath21 where @xmath22 is the three - dimensional velocity in cartesian coordinates . in contrast to @xmath15 and @xmath16 , @xmath23 and @xmath24 components of the three dimensional velocity in milne coordinates are different from those in cartesian coordinates . the metric tensor is given by @xmath25 and the nonzero components of the christoffel symbols are @xmath26 in milne coordinates , the charge conservation equation eq . and the equation of energy and momentum conservation eq . are written by @xmath27 there are the geometric source terms in the right - hand side of eqs.- , which contain the effect from the coordinate expansion with @xmath11 . one can rewrite eqs.- to @xmath28 here the effect from the coordinate expansion with @xmath11 is absorbed into the jacobian @xmath11 in the derivative terms . there are the source terms in the right - hand side of eqs . and , which indicates that @xmath29 and @xmath30 are not the conserved quantities . instead of them , the conserved quantities are @xmath31 and @xmath32 . @xmath31 and @xmath32 are related to @xmath29 and @xmath30 through the coordinate transformation between milne and cartesian coordinates , @xmath33 using the conserved quantities , @xmath31 and @xmath32 , one can express the hydrodynamic equations in the conservative forms , @xmath34 we construct a new algorithm for the relativistic hydrodynamic equations using eqs.- which do not have the source terms . in numerical tests , sec.[sec : numerical_test ] , we shall discuss effects of existence of the source terms from the point of view of stability and numerical viscosity . the riemann problem is an initial - value problem for the hydrodynamic equation . the initial condition is given by two arbitrary constant hydrodynamic states @xmath35 and @xmath36 separated by a discontinuity , @xmath37 where @xmath38 and @xmath39 stand for an initial time and a location of the discontinuity , respectively . the hydrodynamic state @xmath40 contains information on fluid variables , namely baryon number density , fluid velocities and pressure . the initial discontinuity at @xmath41 decays into three nonlinear waves @xcite . two of them are shock waves and/or rarefaction waves . the other is a contact discontinuity moving with hydrodynamic flow . these waves evolve between the constant hydrodynamic states @xmath35 and @xmath36 with a constant velocity . the hydrodynamic states @xmath35 and @xmath36 do not change until characteristic information from the discontinuity arrives . therefore , the hydrodynamic state outside the light cone of the discontinuity @xmath42 remains @xmath35 or @xmath36 . we can also define a riemann problem in milne coordinates . the initial condition of the riemann problem in milne coordinates is set to @xmath43 where @xmath44 and @xmath45 are the initial proper time and the location of the discontinuity and they represent the same point as @xmath42 in eq . in cartesian coordinates . note that the components of @xmath46 are not @xmath47 but the same as those in eq . , @xmath48 . the velocity fields in cartesian coordinates @xmath49 are constant in the rapidity direction . however , the velocity fields in milne coordinates @xmath50 depend on rapidity . now we show that the analytical solution for the riemann problem in milne coordinates is obtained from that in cartesian coordinates by proper coordinate transformations , in which the key issue is to represent the hydrodynamic states as variables independent of @xmath10 . ( a ) and @xmath51 ( b ) . the thick solid lines indicate the hypersurfaces on which the initial conditions of the riemann problems are defined . lc stands for the light cone that originates from the discontinuity . [ fig : initial],title="fig:",width=264 ] ( a ) and @xmath51 ( b ) . the thick solid lines indicate the hypersurfaces on which the initial conditions of the riemann problems are defined . lc stands for the light cone that originates from the discontinuity . [ fig : initial],title="fig:",width=264 ] now we compare the two initial value problems eqs . and . without loss of generality , we can assume that the initial discontinuity represented by the solid circle is located at @xmath52 as in figs.[fig : initial](a ) and ( b ) . in fig.[fig : initial ] ( a ) , the thick solid line stands for the time at which we define the initial condition of the riemann problem in cartesian coordinates , eq .. in milne coordinates , the initial condition of the riemann problem is set on a hyperbola @xmath53 as in fig.[fig : initial](b ) . by comparison between figs.[fig : initial](a ) and ( b ) , the hyperbola @xmath53 in milne coordinates is located inside the constant hydrodynamic state @xmath35 or @xmath36 in cartesian coordinates . this suggests that the initial condition of eq . is satisfied by the solution of the riemann problem in cartesian coordinates eq .. in brief , the riemann problems in both coordinates are identical and thus so are their solutions . detailed explanation about this proof is given in [ app - proof ] . assuming that the hydrodynamic variables in the @xmath23 and @xmath24 directions are constant , from eqs.- we obtain @xmath55 where @xmath56 or @xmath57 . if the transverse components of the four - velocity @xmath15 and @xmath16 are vanishing , eq . expresses the one - dimensional longitudinal expansion . in our numerical scheme , we utilize the lagrange step @xcite in which the grid - cell boundary itself moves together with the hydrodynamic flow during a time step from @xmath58 to @xmath59 . we discretize eq . by space - time integration in a grid cell ( fig.[fig : taueta - lat ] ) , @xmath60 where @xmath61 or @xmath10 . here @xmath45 is the location of the @xmath62-th grid - cell boundary at the proper time @xmath58 , @xmath63 expresses a moving distance of the grid - cell boundary from @xmath64 to @xmath65 @xmath66 and @xmath67 indicates the location of the @xmath62-th grid - cell boundary at the proper time @xmath11 . the center of the @xmath62-th grid cell is located at @xmath68 at @xmath69 , where @xmath70 . using gauss s theorem for integration of eq . , we find the value of @xmath71 of the @xmath62-th grid cell at the next time step @xmath72 , @xmath73,\label{eq : evolve}\end{aligned}\ ] ] where @xmath74 is the lagrange grid - cell size at the proper time @xmath72 , @xmath75 is the trajectory of the lagrange grid - cell boundary at @xmath45 and @xmath76 is the unit normal vector to @xmath75 ( fig.[fig : taueta - lat ] ) . the average values of the conserved quantities in the grid cell at the proper times @xmath58 and @xmath72 are defined respectively by @xmath77 the second term of eq . indicates the flux of the conserved quantities passing through the grid - cell boundary . moves with fluid . @xmath74 is the lagrange grid - cell size at the proper time @xmath72 . see text for detailed explanation . , width=245 ] using the analytical solution of the riemann problem in milne coordinates , we evaluate the @xmath78 and the flux term in eq .. the grid - cell boundary corresponds to the initial discontinuity in the riemann problem and moves together with the contact discontinuity in the solution of the riemann problem . this indicates that the physical quantities on the trajectory @xmath79 are given by those on the contact discontinuity . we obtain the moving distance @xmath80 of the grid - cell boundary @xmath81 using @xmath82 @xmath83 in eqs . and , @xmath84 is the velocity of the grid - cell boundary at @xmath45 seen from an observer sitting at @xmath85 in milne coordinates . in the construction of algorithm , we use the lorentz boost transformation which is explained in the next paragraph . here we show the explicit form of eq .. for detailed derivation of it please see [ app - flux ] . up to the third order in @xmath86 , the flux terms are given by @xmath87 where @xmath88 is the pressure on the cell boundary located at @xmath45 . from now we explain the numerical algorithm for solving the discretized equation eq .. we express the hydrodynamic variables in milne coordinates as @xmath89 . the first step is the interpolation procedure in which the left and right hydrodynamic states at the grid - cell boundary @xmath90 @xmath91 are determined from the reconstruction of the distribution of volume - averaged hydrodynamic variables @xmath92 in a grid cell . if a linear interpolation method is used for reconstruction of the distribution of hydrodynamic variables , the second - order accuracy is achieved ( the mc limiter@xcite ) . for the third - order accuracy , we need to utilize a quadratic curve ( the piecewise parabolic method ( ppm)@xcite ) in reconstruction of the distribution of hydrodynamic states . for the test calculation in the next section , we use the ppm . next , using the constructed left and right states @xmath93 in milne coordinates , we prepare the initial condition of the riemann problem in terms of @xmath94 in cartesian coordinates . to obtain @xmath94 , we move a grid cell to @xmath95 with the lorentz boost transformation and perform the coordinate transformation to @xmath46 . here @xmath92 is invariant under the lorentz boost transformation . using these relations , @xmath96 and @xmath97 at @xmath95 , we obtain @xmath98 in the second step , we solve the riemann problem eq . with the initial condition eq .. @xmath88 and @xmath99 in eqs.- are determined by the analytical solution of the riemann problem we use the values of the vacuum , @xmath100 , instead of solving the riemann problem . ] . for solving the riemann problem we employ the two - shock approximation @xcite . the numerical procedure of this step depends on the equation of state . the riemann solution with the qcd equation of state is described in ref . @xcite . after solving the riemann problem , we perform the inverse lorentz transformation to the original frame . here @xmath101 is lorentz invariant . the flux terms eqs.- are defined on the original frame but written in terms of @xmath99 . in the third step , solving the discretized hydrodynamic equation eqs . , - with the values of @xmath88 and @xmath102 , we obtain distribution of the conservative quantities @xmath103 at the next time step using the lagrange scheme . we remap the grids which move in the lagrange step on the eulerian coordinate @xcite . in the final step , we construct the primitive variables @xmath92 from the conserved quantities @xcite . we obtain @xmath104 from @xmath105 from the coordinate transformation . the numerical method for construction of @xmath92 from @xmath106 is the same as that in ref.@xcite . the one - dimensional code is easily extended to a multi - dimensional code by strang splitting method @xcite , that is to say , multi - dimensional hydrodynamic evolution is realized by successive one - dimensional hydrodynamic calculations . to avoid counting the expansion effect of coordinates more than once , we extract one - dimensional hydrodynamic equations from eqs.- which have the source terms in the right - hand side . then we rewrite each one - dimensional hydrodynamic equation to that without the source terms . to be explicit , we express eqs.- as @xmath107 where the source term @xmath108 is given by @xmath109 applying the dimensional splitting method to eq . , we obtain the one - dimensional hydrodynamic equations , @xmath110 solving the one - dimensional equations eqs.- successively , we carry out the ( 3 + 1 ) dimensional calculations . for expansion in the rapidity direction , transforming eq . to eq . , we use the algorithm mentioned in sec.[sec:1d ] . in the @xmath23 and @xmath24 directions , we use the same algorithm in cartesian coordinates in ref.@xcite . for example , for expansion in the @xmath23 direction , we discretize eq . as explained in sec.[sec:1d ] , @xmath111 where @xmath88 and @xmath112 are determined by the solution of the riemann problem whose initial condition is given by eq .. the grid - cell size after the lagrange step is given by @xmath113 using the operator @xmath114 which represents one - dimensional evolution in the @xmath62 direction during the proper time @xmath115 , two - dimensional expansion in @xmath116 coordinates is given by @xmath117 and @xmath118 , @xmath119 similarly the three - dimensional expansion in @xmath120 coordinates is written by @xmath121 the courant - friedrichs - lewy ( cfl ) condition helps us determine an appropriate time - step size in solving partial differential equations . the cfl condition is determined so that the numerical propagating speed on the grid is larger than the physical propagating speed , @xmath122 where @xmath123 and @xmath124 are the numerical signal velocity and the characteristic velocity in the rapidity direction , respectively . note that the proper time @xmath11 is multiplied for dimensionless expression . this condition is important for stability of numerical calculations . if the cfl condition eq . is written as a function of the characteristic velocity in @xmath57 direction @xmath125 , @xmath126 @xmath127 the cfl condition in milne coordinates has rapidity dependence . at the large rapidity the right - hand side of eqs . and approaches to 1 , which is the same condition as that in the case of @xmath128 therefore the cfl condition in milne coordinates is determined by @xmath129 equation indicates that the proper time of the denominator is larger , a possible time - step size @xmath86 is larger . with these considerations , we define the courant number as @xmath130 with an initial proper time @xmath44 . when we numerically solve partial differential equations and update the value of a cell , the values of its neighboring cells are necessary . for the cell on the boundary of hydrodynamic grid , we need to prepare additional cells which are called ghost cells and put appropriate information to them . in the next section we shall discuss several numerical tests using our code ; shock tube problems in milne coordinates , expansion of matter to the vacuum , landau - kahlatnikov solution , fluctuations in longitudinal expansion , gubser flow and conservation property . for the shock tube problem in milne coordinates , we input an analytical solution as a boundary condition at the ghost cells . for the numerical test of expansion of matter to the vacuum , we use the physical values of the vacuum at the cell on the boundary . we employ the periodic boundary condition for the investigation of fluctuations in longitudinal expansion . for the landau - khalatnikov solution and the gubser flow , we copy the values of the cells on the boundary onto those of the ghost cells . in addition , we point out that in some cases we need a careful procedure at the boundary . for example , for the shock tube problem in milne coordinates , we use the mc limiter at the boundary and ghost cells to reduce numerical errors which originate from inward flow at the boundary . in expansion of matter to the vacuum , we observe that a numerical instability occurs at the discontinuity between matter and the vacuum . to stabilize the difficulty we employ the minmod limiter which is dissipative and smears out discontinuities compared to the mc limiter . we employ several problems to check correctness of the numerical algorithm in milne coordinates . for one - dimensional tests , we analyze with our code the riemann problem and landau khalatnikov solution @xcite in order to verify our riemann solver . the landau - khalatnikov solution is used for understanding experimental data of particle rapidity distributions in the high - energy heavy - ion collisions . next we discuss propagation of fluctuations around bjorken flow . we derive analytical solutions from linearized hydrodynamics and compare them to numerical calculations with our code . for multi - dimensional tests , we use the gubser flow @xcite which gives us three dimensional hydrodynamic expansion of hot and dense matter created after the high - energy heavy - ion collisions . in the test problems we use the ideal gas equation of state , @xmath131 . as a function of @xmath10 at @xmath132 fm . , width=302 ] using the property that the initial value problem of eq . in milne coordinates is the same as the riemann problem in cartesian coordinates , we carry out the shock tube test in milne coordinates . we solve the initial value problem of eq . with our numerical scheme in milne coordinates and compare the numerical results with the analytical solution of the riemann problem eq .. the analytical solution of the riemann problem in milne coordinates is obtained by the coordinate transformation from that in cartesian coordinates @xcite ( see [ app - riemann - taueta ] ) . we perform test calculations in two cases . in the first case we set the discontinuity at @xmath95 and in the second case we put it at @xmath133 at the initial time @xmath134 fm . in the first test problem , the temperature in @xmath135 ( @xmath136 ) is set to @xmath137 ( @xmath138 ) and in both regions the @xmath57 component of the velocity @xmath139 is vanishing . in our numerical scheme , we calculate time evolution of the rapidity component of the velocity @xmath140 directly , instead of @xmath139 . the transformation between @xmath139 and @xmath140 is given by eq . which suggests that @xmath140 is not vanishing even if @xmath141 . the initial condition of @xmath140 is given by @xmath142 which has @xmath10 dependence . we perform the numerical calculations with grid size @xmath143 and time - step size @xmath144 . figure [ fig : shocktube1 ] shows the energy density distribution , the velocity @xmath140 and the velocity @xmath139 which is transformed by eq . from @xmath140 as a function of rapidity @xmath10 together with the analytical solutions ( solid lines ) . the rarefaction wave moves to the negative direction from @xmath95 and the shock wave moves to the positive direction from @xmath95 . our numerical results are consistent with the analytical solutions . for the second numerical test , we put the initial discontinuity at @xmath145 . other conditions , @xmath146 , @xmath147 and @xmath139 are the same as those in the first test calculation . in cartesian coordinates the first and second numerical tests are essentially identical . in milne coordinates , however , the second numerical problem is different from the first problem , because the @xmath140 depends on @xmath10 differently from the first problem . the energy density distribution @xmath5 , velocity distribution @xmath140 and velocity distribution @xmath139 are shown in fig.[fig : shocktube2 ] . again , our numerical calculations show good agreement with the analytical solutions in the second numerical test . as one of the specific problems of the riemann problem , we consider the one - dimensional expansion of matter into the vacuum ; a rarefaction wave appears at the discontinuity and expands between the matter and the vacuum @xcite . this problem is useful for a realistic description of expansion of the qgp and hadronic matter into the vacuum in the high - energy heavy - ion collisions . we set an initial condition to , @xmath148 where @xmath140 in @xmath149 corresponds to @xmath141 in cartesian coordinates and in @xmath150 lies the vacuum . setting the vacuum for the boundary condition as in eq . , one can avoid the matter from flowing into the system through the boundaries . in the shock tube problems in milne coordinates in sec.[sec : shoch - tube ] , the matter comes into through the boundaries , which is a possible source of numerical error . here we discuss the importance of description of the hydrodynamic equation in the conservative form in developing numerical algorithm . for the investigation , we discretize the hydrodynamic equation with the source terms eq . and construct a code based on the same procedure explained in sec.[sec:1d ] . in the code , @xmath29 and @xmath30 are updated in the lagrange step , in stead of @xmath31 and @xmath32 . then we compare the results of the code without the source terms eq . and those with the source terms . in the both numerical calculations the grid size @xmath151 and the time - step size are set to 0.02 and @xmath152 , respectively . the hydrodynamic expansion starts at @xmath134 fm . figure [ fig : vac_t11 ] shows the @xmath140 distribution as a function of @xmath10 at @xmath132 fm . in @xmath153 , the matter is at rest in cartesian coordinates , which corresponds to the negative flow in milne coordinates . in @xmath154 , the rarefaction wave starts to expand ; it propagates with the sound velocity inward the matter at rest in cartesian coordinates and expands with the speed of light to the vacuum . in the rarefaction wave the steep velocity gradient is produced , where both of the codes with and without the source terms can not reproduce the analytical result , though the code without the source terms is closer to the analytical solution . figure [ fig : vacuum ] shows the numerical results of the energy density , @xmath140 and @xmath139 distributions as a function of @xmath10 together with the analytical solution at later time @xmath155 fm . in @xmath156 the velocity @xmath139 is vanishing in cartesian coordinates , which indicates the negative @xmath140 in milne coordinates . in @xmath157 the rarefaction wave is spreading and at @xmath158 the boundary to the vacuum exits , which moves out to the vacuum at the speed of light . for the energy density distribution , both codes with and without the source terms reproduce the analytical solution , though near the boundary between the matter and the vacuum the small difference between them is observed . in both codes , the value of @xmath140 around the boundary is larger than that of the analytical solution near the boundary . the stronger flow near the boundary in the numerical solution causes the smaller energy density compared with the analytical solution . on the other hand , there is no difference between the behaviors of @xmath139 of both codes . the stability of numerical calculation is sensitive to the differences in @xmath140 of codes which are seen near the boundary between matter and the vacuum , because at the boundary pressure becomes zero and lorentz factor becomes infinity in the analytical solution . ( top panel ) and the velocity @xmath140 ( bottom panel ) as a function of number of cell for the codes with and without the source terms . , title="fig:",width=302 ] ( top panel ) and the velocity @xmath140 ( bottom panel ) as a function of number of cell for the codes with and without the source terms . , title="fig:",width=302 ] to investigate the numerical accuracy of the codes with and without the source terms , we calculate the l1 norm which is defined by @xmath159 where @xmath160 is the energy density @xmath161 or the rapidity component of the velocity @xmath140 , @xmath151 is a grid - cell size . using eq.([eq : l1-v ] ) , we evaluate the deviation of the numerical results @xmath162 from the exact solutions . at the same time we can know the convergence speed to the exact solution of numerical algorithm . in fig.[fig : vacl1 ] , l1 norms of energy density and @xmath140 are shown . in both energy density and velocity , the values of l1 norm of the code without the source terms are smaller than those with the source terms , which means that the code without the source term has smaller numerical viscosity than that with the source terms . as expected , the existence of the source terms produces more artificial viscosity . if the initial discontinuity is set at the larger rapidity , which makes the velocity slope at rarefaction wave larger and gives more severe problems to the codes in milne coordinates . we find that the code without the source terms is more stable than that with the source terms . for example , if we set the initial discontinuity at @xmath163 , we find that numerical instability occurs in the code with the source terms . we employ the landau - khalatnikov solution @xcite as a one - dimensional numerical test problem . the initial condition of it is expressed by a thin slab of hot and dense matter created after the collisions , which is the same as the problem discussed in sec.[subsub : vacuum ] . in the expansion of the slab of matter , two rarefaction waves travel into the slab from both sides and start to overlap at the center of the slab . the region where rarefaction waves overlap is described by the landau - khalatnikov solution . the asymptotic form of the landau - khalatnikov solution for enough later time @xmath164 is written by @xmath165 \right\ } \label{eq : landau}\ ] ] and @xmath166 , where @xmath167 is the thickness of the slab . the asymptotic solution eq . is used for investigation of rapidity distributions of produced particles at rhic@xcite . 0.1 and 5 fm . from the top , the energy density distributions at @xmath168 and 1000 fm.,width=302 ] in numerical calculation , we start the simulation at @xmath169 fm with the initial condition given by eq . , where @xmath170 and the thickness size are set to 10 gev/@xmath171 and @xmath172 fm , respectively . the numerical calculation is performed with the grid size @xmath173 , the time - step size @xmath174 fm and @xmath175 fm which is determined by the cfl condition in sec.[sec : cfl ] . figure [ fig : landau ] shows the energy density distributions at @xmath176 510 , 600 , 700 and 1000 fm together with the analytical solution . calculations with the time - step sizes @xmath177 fm and @xmath178 fm can explain the analytical solution , which suggests that the computational time can be saved if the time - step size is determined by the cfl condition in sec.[sec : cfl ] . there is small deviation between numerical calculations and the analytical solution at large @xmath179 , which implies that the asymptotic form of the landau - khalatnikov solution eq . can not be applicable at large rapidity @xcite . [ ht ! ] a longitudinal fluctuation in particle distributions and collective flows is one of the interesting topics in high - energy heavy - ion collisions at lhc@xcite . for instance , the propagation of fluctuations around bjorken flow in heavy - ion collisions are investigated from linear analyses @xcite . the small longitudinal fluctuations around bjorken flow propagate according to the following linearized equations , @xmath180 here @xmath181 is the energy density from bjorken s scaling solution @xcite for the equation of state @xmath182 . to 1/3 . ] since the background is rapidity independent , we can obtain solutions with a definite wave number @xmath183 . the solutions consist of two modes as eq . is essentially a second - order ordinary differential equation in the @xmath184-space . the nature of the modes depends on the sign of @xmath185 . for @xmath186 the fluctuations do not propagate but just attenuate while for @xmath187 the fluctuations propagate as well as attenuate . the condition for @xmath188 is satisfied by @xmath189 , where @xmath190 is the sound velocity . general solutions of the linearized equations including the case @xmath188 are given in the [ app - general - solution ] . here we compare the analytical solutions and the numerical calculation with our hydrodynamic code for two cases , @xmath186 and @xmath187 . in both cases , we choose initial conditions so that we can single out a particular mode of attenuation ( @xmath186 ) and propagation ( @xmath187 ) . to be specific , for @xmath186 we choose @xmath191 and for @xmath187 @xmath192,\end{aligned}\ ] ] where @xmath193 is defined by @xmath194 . the phase velocity of the fluctuation is @xmath195 . we set the initial time to @xmath134 fm and use the ideal gas equation of state , @xmath131 and @xmath196 . the initial energy density @xmath170 in bjorken flow is given by @xmath197 @xmath198 . in the case of @xmath186 , we choose @xmath199 , @xmath200 and @xmath201 @xmath198 . in the case of @xmath187 , we choose @xmath202 , @xmath203 and @xmath201 @xmath198 . figures [ fig : sounddamp ] and [ fig : sound ] show the analytical and numerical results of the fluctuations of the energy density and the velocity around bjorken flow for @xmath186 and @xmath187 . in the numerical calculation , we use the periodic boundary condition . when the value of @xmath204 is positive , the fluctuation does not propagate and its amplitude decreases with time ( fig.[fig : sounddamp ] ) . on the other hand , when the value of @xmath204 is negative , the fluctuation propagates and its amplitude decreases with time ( fig.[fig : sound ] ) . however , if we chose different initial conditions such that more than one mode were involved , the amplitude of the fluctuation would at first grow and then reduce for both @xmath186 and @xmath187 due to the interference of two modes . the similar amplification of the fluctuations around bjorken flow is reported in ref.@xcite . our numerical results show good agreement with the analytical solutions . . top : at @xmath205 and 3 fm , bottom : at @xmath206 and 7 fm . [ fig : gubser_e],title="fig:",width=302 ] . top : at @xmath205 and 3 fm , bottom : at @xmath206 and 7 fm . [ fig : gubser_e],title="fig:",width=302 ] at @xmath207 and 7 fm . , width=302 ] an analytic solution to the relativistic , conformally invariant navier - stokes equation is constructed based on symmetry considerations in @xmath208 coordinate system @xcite . the gubser flow and its related solutions are utilized for checking or improvement of hydrodynamic codes @xcite . the solution is generalization of bjorken flow where the medium expands both longitudinally and radially , which gives us a realistic description of the space - time evolution of high - energy heavy - ion collisions . according to the solution for inviscid fluid , the transverse or radial velocity @xmath209 and energy density @xmath161 are given by @xmath210 @xmath211^{4/3 } } , \label{eq : gubsere}\ ] ] where @xmath212 is a dimensionless integration constant , @xmath213 is an arbitrary dimensional constant with unit of inverse length of the system size @xcite . we compare our numerical calculations and the analytical solution in figs.[fig : gubser_e ] and [ fig : gubser_v ] . in our numerical calculation , the parameters are set to @xmath214 @xmath215 and @xmath216 . the hydrodynamic expansion starts at @xmath134 fm . the numerical simulation is performed with the grid size @xmath217 fm , @xmath173 and the time - step size @xmath218 . first the consistency between our numerical calculations and the analytical solution suggests the strange splitting method in sec . [ sec:3d ] works correctly . in the gubser flow , the existence of the initial transverse flow is an origin of the strong transverse flow at later time . to reproduce the strong transverse flow in the gubser flow , we find that careful choice of the interpolation method is required in the numerical calculation . for example , if the second order interpolation method is employed , the energy density around @xmath219 fm of the numerical calculations is larger than that of the analytical solutions . however if we use the ppm interpolation method , we can reproduce the analytical solutions numerically . it is discussed in [ app - gubser ] . [ cols="^,^,^,^,^,^ " , ] we constructed the new godunov type relativistic hydrodynamic code in milne coordinates based on the algorithm in cartesian coordinates @xcite . we evaluated the flux terms , using the numerical solution of the riemann problem with the initial condition at the constant proper time @xmath11 . we checked correctness of our algorithm from the comparison between numerical calculations and analytical solutions of shock tube , expansion of matter into the vacuum , the landau - khalatnikov solution , propagation of fluctuation around bjorken flow and the gubser flow . we investigated the energy and momentum conservation of our code from calculation of longitudinal hydrodynamic expansion with an initial condition for high - energy heavy - ion collisions . in particular , we focused on the effects of the source terms in relativistic numerical hydrodynamics in milne coordinates on stability and numerical viscosity . we analyzed those effects in the test problems of expansion into the vacuum and the conservation property . in expansion of matter into the vacuum , we showed that numerical results from the code without the source terms is closer to the analytical solution compared with that with source terms . besides , the code without the source terms is more stable and has less numerical viscosity than the code with the source terms . in addition , we observed that the code written in the conservative form keeps the conservation property with high accuracy in the expansion from the fluctuating initial longitudinal profile for high - energy heavy - ion collisions even on the coarse grid . our algorithm is easily extended to the code with the qcd equation of state and finite viscosities @xcite . after that , we shall employ our hydrodynamic code to investigate experimental results at rhic and lhc and understand the detailed qgp bulk property using a reliable 3d relativistic viscous hydrodynamic expansion with small numerical viscosity . the work of cn is supported by the jsps grant - in - aid for scientific research ( s ) no . 26220707 and us department of energy grant de - fg02 - 05er41367 . is partially supported by jsps postdoctoral fellowships for research abroad . we show two initial problems eqs . and have the same analytic solution . first , we prove that the hydrodynamic state which satisfies the condition @xmath220 at some proper time keeps @xmath221 . this state corresponds to the initial condition of the riemann problem in milne coordinates eq .. for the @xmath222 dimensional case , we rewrite the energy conservations , eqs . and as @xmath223 inserting the conditions , @xmath220 , namely @xmath224 , @xmath225 and @xmath226 , into eqs . and , we obtain @xmath227 which means @xmath228 . this result indicates that in the riemann problem in milne coordinates eq . , the hydrodynamic state outside the light cone of the discontinuity @xmath229 remains @xmath35 or @xmath36 . in fig.[fig : finite - etai](b ) , the initial condition of the riemann problem is set on the hyperbolic curve @xmath53 with the discontinuity at @xmath45 , the hydrodynamic state outside the light cone of the discontinuity is given by @xmath35 or @xmath36 . figure [ fig : finite - etai](a ) shows the initial condition of the riemann problem in cartesian coordinates with the same discontinuity point at @xmath230 , which corresponds to @xmath229 in fig.[fig : finite - etai](b ) . from comparison between figs.[fig : finite - etai](a ) and ( b ) , the analytical solutions for initial - condition problems which are given by eqs . and satisfy the same boundary conditions on the light cone at the discontinuity @xmath231 . we conclude that the analytical solution of the riemann problem in milne coordinates , eq . is the same as that in cartesian coordinates , eq .. we show the detailed calculation of numerical flux terms eqs.- . we represent the velocity and pressure of the grid - cell boundary at @xmath45 as @xmath232 and @xmath88 , respectively . first we shift the grid - cell boundary at @xmath45 to @xmath95 by the lorentz boost transformation to derive the moving distance @xmath79 of the grid - cell boundary during @xmath233 . under the lorentz boost transformation , @xmath79 is invariant . the velocity of the grid - cell boundary @xmath99 after the lorentz boost transformation is related to @xmath232 with @xmath234 the moving distance @xmath79 of the grid - cell boundary from @xmath95 during @xmath235 is given by eq . with eq .. equation is derived from following simultaneous equations @xmath236 and @xmath237 with @xmath238 . next we evaluate the numerical flux term of eq . on the original frame ( where the grid - cell boundary move from @xmath45 ) . in the integration of eq . , the unit vector @xmath76 is written by @xmath239 with @xmath240 . the integrands of the flux terms become @xmath241 the integral element @xmath242 on @xmath75 is given by @xmath243 expanding eq . in terms of @xmath235 with eq . , we perform the integration of @xmath235 in @xmath244 . then we can get the explicit form of the numerical flux terms as eq.- . we represent the analytical solution of the riemann problem in cartesian coordinates @xcite , as that in milne coordinates . the analytical solution is composed of the four uniform hydrodynamic states , @xmath245 and @xmath246 which are separated by three nonlinear waves ; two of them are shock waves and/or rarefaction waves , the other is a contact discontinuity which separates @xmath247 and @xmath248 . the four hydrodynamic states and the three nonlinear waves move with constant velocity . suppose that the initial discontinuity in milne coordinates is located at @xmath231 . at proper time @xmath249 , the location of the discontinuity surface of the shock front , rarefaction front and contact discontinuity which move with constant velocity @xmath250 from @xmath45 , @xmath251 is written by @xmath252 where @xmath253 is given by @xmath254 the state of the rarefaction wave depends on @xmath255 in cartesian coordinates , where @xmath38 is the initial time and @xmath39 is the location of the initial discontinuity . here @xmath38 and @xmath39 are related with @xmath44 and @xmath45 in milne coordinates through the relations , @xmath256 and @xmath257 . in milne coordinates , the state of rarefaction wave is described by @xmath258 in the case of the expansion of matter into the vacuum , the location of the boundary to the matter at rest in cartesian coordinates is obtained by eq . , where @xmath250 is replaced by negative sound velocity , @xmath259 . the boundary to the vacuum which moves with the speed of light is given by @xmath260 we derive analytical solutions for fluctuations around bjorken flow @xcite . we start from the @xmath261 dimensional relativistic ideal hydrodynamic equations in milne coordinates , @xmath262 which are obtained from eqs . , and . focusing on propagation of fluctuations around bjorken flow , we put a small perturbation on top of bjorken expansion in eqs . and , @xmath263 where @xmath264 is the energy density from bjorken s scaling solution @xcite , the equation of state is given by @xmath182 , the sound velocity is @xmath190 . assuming that the fluctuations are small , we neglect the second and higher order terms of fluctuations in eqs . and and derive linearized relativistic hydrodynamic equations , @xmath265 we input the following initial conditions for @xmath266 and @xmath267 at @xmath53 , @xmath268 using the fourier transform of @xmath269 and @xmath270 , @xmath271 we obtain the analytical solutions , @xmath272 e^{ik\eta } \label{eq : deltae } , \\ \delta w^\eta(\tau,\eta ) & = \frac{1}{\sqrt{d}}\left(\frac{\tau}{\tau_0}\right)^{(\lambda-3)/2}\nonumber\\ & \times \biggl [ \left(- a_2a_2 + i\frac{k\lambda a_1}{\tau_0e_0(1+\lambda ) } \right ) \left(\frac{\tau}{\tau_0}\right)^{-\sqrt{d}/2 } \nonumber \\ & + \left ( a_1a_2 - i\frac{k\lambda a_1}{\tau_0e_0(1+\lambda ) } \right ) \left(\frac{\tau}{\tau_0}\right)^{\sqrt{d}/2}\biggr ] e^{ik\eta } \label{eq : deltav } , \end{aligned}\ ] ] where @xmath273 , @xmath274 and @xmath275 are given by @xmath276 @xmath277 respectively . equation indicates that the value of @xmath204 is always less than @xmath278 and decreases with the wave number @xmath184 . as a result , the first and second terms of eqs . and have a negative power of @xmath11 , which indicates that the amplitude of the fluctuation decreases with @xmath11 . in the case of @xmath186 , the initial fluctuations are attenuated without propagation . in particular cases of eq . and , if @xmath279 and @xmath280 satisfy the following relation , @xmath281 then only the first - term mode in eq . and remains . the analytical solutions become @xmath282 where the fluctuations with the smaller wave number attenuate faster . in the case of @xmath187 , the fist and the second terms of eqs . and represent the progressive and regressive waves , respectively . if the coefficients @xmath279 and @xmath280 satisfy eq . , then the analytical solutions become , @xmath283 where @xmath193 is defined by @xmath194 . the propagating speed of fluctuations is @xmath195 . the condition of @xmath187 is given by @xmath284 where the fluctuation with the larger wave number propagates faster . in the case of @xmath188 , the analytical solutions are given by @xmath285 e^{ik\eta } \label{eq_e_d0 } , \\ & \delta w^\eta(\tau,\eta ) = \left(\frac{\tau}{\tau_0}\right)^{(\lambda-3)/2 } \nonumber \\ & \times \left [ a_2 - \left ( \frac{1-\lambda}{2}a_2 + i \frac{k\lambda a_1}{\tau_0 e_0 ( 1+\lambda)}\right){\rm log}\frac{\tau}{\tau_0 } \right ] e^{ik\eta } \label{eq_v_d0}. \end{aligned}\ ] ] if @xmath279 and @xmath280 satisfy @xmath286 then the analytical solutions , eqs . and become @xmath287 fm . bottom panel : at @xmath206 fm . [ fig : comparison_e],title="fig:",width=302 ] fm . bottom panel : at @xmath206 fm . [ fig : comparison_e],title="fig:",width=302 ] at @xmath205 and 5 fm . [ fig : comparison_vx ] , width=302 ] we discuss the importance of the interpolation procedure in the existence of strong radial expansion like gubser flow . figure [ fig : comparison_e ] shows the energy distributions at @xmath205 fm and @xmath206 fm which are obtained by the 2nd and 3rd order interpolation procedures . in the calculation , we use the same parameters as those in sec . [ sec : gubser ] . the numerical calculations with the 3rd order interpolation procedure reproduce the analytical solutions . on the other hand , the energy density with the second order interpolation procedure is slightly larger than that of the analytical solutions . in particular , we observe the deviation from the analytical solution of energy density in @xmath288 fm ( @xmath289 fm ) at @xmath205 ( @xmath206 fm ) . we find the same behavior in other 2nd order interpolation procedures , minmod and superbee limiters . however , in the case of the one - dimensional expansion , even if the strong expansion exists , the numerical calculation with the 2nd order interpolation procedures shows good agreement with the analytical solution . we observe the deviation between the analytical solution and the numeral results in multi - dimensional calculation , which suggests that the operator splitting method is also a possible key issue for the problem . in fig.[fig : comparison_vx ] , we show the transverse velocities at @xmath205 and @xmath206 fm . the gradient of the transverse velocity increases rapidly up to @xmath290 ( @xmath291 ) fm at @xmath205 ( @xmath206 ) fm where the value of transverse velocity is slightly smaller than that of the analytical solutions , which implies that the 2nd order interpolation schemes do not satisfy the description of such a rapid expansion . the inadequate velocity growth causes the delay of the decrease of the energy density which is observed in fig.[fig : comparison_e ] . in ref . @xcite they point out the importance of adjusting the flux limiter in their algorithm ( kt algorithm ) , using the relativistic viscous hydrodynamics . they show that a free parameter @xmath292 in the van leer minmod filter is fixed from comparison between the solutions of shear - stress tensor from gubser flow and the numerical calculations . , a750 * , 1 ( 2005 ) r. derradi de souza , t. koide , and t. kodama , prog . phys . * 86 * , 35 ( 2016 ) . arxiv:1506.03863 [ nucl - th ] c. gale , s. jeon and b. schenke , int . j. mod . a * 28 * , 1340011 ( 2013 ) . arxiv:1301.5893 [ nucl - th ] t. hirano , p. huovinen , k. murase , and y. nara , prog . part . * 70 * , 108 ( 2013 ) . arxiv:1204.5814 [ nucl - th ] c. nonaka and m. asakawa , ptep * 2012 * , 01a208 ( 2012 ) . doi:10.1093/ptep / pts014 . arxiv:1204.4795 [ nucl - th ] h. song and u. w. heinz , phys . rev . c * 77 * , 064901 ( 2008 ) . arxiv:0712.3715 [ nucl - th ] r. baier and p. romatschke , eur . phys . j. c * 51 * , 677 ( 2007 ) . doi:10.1140/epjc / s10052 - 007 - 0308 - 5 . arxiv : nucl - th/0610108 k. dusling and d. teaney , phys . c * 77 * , 034905 ( 2008 ) . arxiv:0710.5932 [ nucl - th ] g. s. denicol , t. kodama , t. koide , and p. mota , phys . rev . c * 80 * , 064901 ( 2009 ) . arxiv:0903.3595 [ hep - ph ] b. schenke , s. jeon , and c. gale , phys . rev . * 106 * , 042301 ( 2011 ) . arxiv:1009.3244 [ hep - ph ] p. bozek , phys . rev . c * 85 * , 034901 ( 2012 ) . arxiv:1110.6742 [ nucl - th ] . l. g. pang , y. hatta , x. n. wang , and b. w. xiao , phys . d * 91 * , 074027 ( 2015 ) . arxiv:1411.7767 [ hep - ph ] s. borsanyi , z. fodor , c. hoelbling , s. d. katz , s. krieg , and k. k. szabo , phys . lett . b * 730 * , 99 ( 2014 ) . arxiv:1309.5258 [ hep - lat ] . a. bazavov @xmath293 . ( hotqcd ) , phys . d * 90 * , 094503 ( 2014 ) . arxiv:1407.6387 [ hep - lat ] w. israel and j. m. stewart , ann . phys . ( n.y . ) * 118 * ( 1979 ) , 341 . doi:10.1016/0003 - 4916(79)90130 - 1 r. baier , p. romatschke , d. t. son , a. o. starinets , and m. a. stephanov , jhep * 04 * , 100 ( 2008 ) . doi:10.1088/1126 - 6708/2008/04/100 . arxiv:0712.2451 [ hep - th ] d. bazow , u. w. heinz , and m. strickland , phys . rev . c * 90 * , 054910 ( 2014 ) . arxiv:1311.6720 [ nucl - th ] a. adare @xmath293 . ( phenix collaboration ) , phys . . lett . * 107 * , 252301 ( 2011 ) . arxiv:1105.3928 [ nucl - ex ] p. sorensen ( star collaboration ) , j. phys . g * 38 * , 124029 ( 2011 ) . doi:10.1088/0954 - 3899/38/12/124029 . arxiv:1110.0737 [ nucl - ex ] . k. aamodt @xmath293 . ( alice collaboration ) , phys . 107 * , 032301 ( 2011 ) . doi:10.1103/physrevlett.107.032301 . arxiv:1105.3865 [ nucl - ex ] g. aad @xmath294 . ( atlas collaboration ) , phys . c * 86 * , 014907 ( 2012 ) . arxiv:1203.3087 [ hep - ex ] s. chatrchyan @xmath293 . ( cms collaboration ) , phys . c * 89 * , 044906 ( 2014 ) . arxiv:1310.8651 [ nucl - ex ] p. celella , siam j. sci . . comput . * 3 * , 76 ( 1982 ) . doi:10.1137/0903007 d. s. balsara , j. comput . phys . * 114 * , 284 ( 1994 ) . doi:10.1006/jcph.1994.1167 w. dai and p. r. woodward , siam j. sci . . comput . * 18 * , 982 ( 1997 ) . doi:10.1137/s1064827595282234 a. mignone , t. plewa , g. bodo , apjs * 160 * , 199 ( 2005 ) . arxiv : astro - ph/0505200 p. r. woodward and p. colella , j. comput . phys . * 54 * , 115 ( 1984 ) . doi:10.1016/0021 - 9991(84)90142 - 6 i. g. bearden @xmath295 ( brahms collaboration ) , phys . b * 523 * , 227 ( 2001 ) . doi:10.1016/s0370 - 2693(01)01333 - 8 . arxiv : nucl - ex/0108016 i. g. bearden @xmath295 ( brahms collaboration ) , phys . * 88 * , 202301 ( 2002 ) . arxiv : nucl - ex/0112001 b. b. back @xmath295 ( phobos collaboration ) , phys . . lett . * 91 * , 052303 ( 2003 ) . arxiv : nucl - ex/0210015 b. alver @xmath293 . ( phobos collaboration ) , phys . c * 83*. 024913 ( 2011 ) . arxiv:1011.1940 [ nucl - ex ] s. chatrchyan @xmath293 . ( cms collaboration ) , j. high energy phys . * 08 * , 141 ( 2011 ) . arxiv:1107.4800 [ nucl - ex ] g. aad @xmath293 , ( atlas collaboration ) , phys . lett . b * 710 * , 363 ( 2012 ) . arxiv:1108.6027 [ hep - ex ] e. abbas @xmath295 ( alice collaboration ) , phys b * 726 * , 610 ( 2013 ) . arxiv:1304.0347 [ nucl - ex ] j. m. marti and e. m@xmath296ller , j. fluid mech . * 258 * , 317 ( 1994 ) . doi:10.1017/s0022112094003344 j. a. pons , j. m. marti , and e. m@xmath296ller , j. fluid mech . * 422 * , 125 ( 2000 ) . arxiv : astro - ph/0005038 b. van leer , j. comput . phys . * 32 * , 101 ( 1979 ) . doi:10.1016/0021 - 9991(79)90145 - 1 . p. colella and p. r. woodward , j. comp * 54 * , 174 ( 1984 ) . doi:10.1016/0021 - 9991(84)90143 - 8 j. m. marti and e. m@xmath296ller , j. comp . phys . * 123 * , 1 ( 1996 ) . doi:10.1006/jcph.1996.0001 g. strang , siam j. numer . anal . * 5 * , 506 ( 1968 ) . doi:10.1137/0705041 l. d. landau , izv . nauk sssr * 17 * , 51 ( 1953 ) s. z. belenkij and l. d. landau , usp . fiz . nauk * 56 * , 309 ( 1955 ) ; nuovo cimento suppl . * 3 * , 15 ( 1956 ) . doi:10.1007/bf02745507 s. s. gubser , phys . d * 82 * , 085027 ( 2010 ) . arxiv:1006.0006 [ hep - th ] s. s. gubser and a. yarom , nucl . phys . * b846 * , 469 ( 2011 ) . arxiv:1012.1314 [ hep - th ] d. h. rischke , s. bernard , and j. a. marhn , nucl . * a595 * , 346 ( 1995 ) . doi:10.1016/0375 - 9474(95)00355 - 1 . arxiv : nucl - th/9504018 m. murray ( brahms collaboration ) , j. phys . g * 30 * , s667 ( 2004 ) . doi:10.1088/0954 - 3899/30/8/004 . arxiv : nucl - ex/0404007 i. g. bearden @xmath293 . ( brahms collaboration ) , phys . * 94 * , 162301 ( 2005 ) . arxiv : nucl - ex/0403050 p. steinberg , nucl . phys . * a752 * , 423 ( 2005 ) . arxiv : nucl - ex/0412009 m. murray ( brahms collaboration ) , j. phys . g * 35 * , 044015 ( 2008 ) . doi:10.1088/0954 - 3899/35/4/044015 . arxiv:0710.4576 [ nucl - ex ] c. y. wong , phys . rev . c * 78 * , 054902 ( 2008 ) . arxiv:0808.1294 [ hep - ph ] z. j. jiang , q. g. li , and h. l. zhang , phys . rev c * 87 * , 044902 ( 2013 ) . doi:10.1103/physrevc.87.044902 v. khachatryan @xmath295 ( cms collaboration ) , phys . rev . c * 92 * , 034911 ( 2015 ) . arxiv:1503.01692 [ nucl - ex ] s. floerchinger and u. a. wiedemann , j. high energy phys . * 11 * , 100 ( 2011 ) . arxiv:1108.5535 [ nucl - th ] j. noronha - hostler , g. s. denicol , j. noronha , r. p. g. andrade , and f. grassi , phys . c * 88 * , 044916 ( 2013 ) . arxiv:1305.1981 [ nucl - th ] l. del zanna , v. chandra , g. inghirami , v. rolando , a. beraudo , a. de pace , g. pagliara , a. drago , and f. becattini , eur . phys . j. c * 73 * , 2524 ( 2013 ) . doi:10.1140/epjc / s10052 - 013 - 2524 - 5 . arxiv:1305.7052 [ nucl - th ] i. karpenko , p. huovinen , and m. bleicher , comput . . commun . * 185 * , 3016 ( 2014 ) . arxiv:1312.4160 [ nucl - th ] h. marrochio , j. noronha , g. s. denicol , m. luzum , s. jeon , and c. gale , phys . c * 91 * , 014903 ( 2015 ) . arxiv:1307.6130 [ nucl - th ] c. shen , z. qiu , h. song , j. bernhard , s. bass , and u. heinz , comput . phys . * 199 * , 61 ( 2016 ) . arxiv:1409.8164 [ nucl - th ] f. becattini , g. inghirami , v. rolando , a. beraudo , l. del zanna , a. de pace , m. nardi , g. pagliara , and v. chandra , eur . phys . j. c * 75 * , 406 ( 2015 ) . doi : 10.1140/epjc / s10052 - 015 - 3624 - 1 . arxiv:1501.04468 [ nucl - th ] t. hirano , u. heinz , d. kharzeev , r. lacey , and y. nara , phys . b * 636 * , 299 ( 2006 ) . arxiv : nucl - th/0511046 a. k. chaudhuri and u. heinz , phys . rev . lett . * 97 * , 062301 ( 2006 ) . arxiv : nucl - th/0503028 b. betz , m. gyulassy , d. h. rischke , h. stocker , and g. torrieri , j. phys . g * 35 * , 104106 ( 2008 ) . doi:10.1088/0954 - 3899/35/10/104106 . arxiv:0804.4408 [ hep - ph ] y. tachibana and t. hirano , phys . rev . c * 90 * , 021902 ( 2014 ) . arxiv:1402.6469 [ nucl - th ] m. takamoto and s. inutsuka , j. comput . phys . * 11 * , 38 ( 2011 ) . doi:10.1016/j.jcp.2011.05.030 . arxiv:1106.1732 [ astro - ph ]

we propose a procedure for defining all single - photon observables in terms of positive - operator valued measures ( povms ) , in particular spin and position . we identify the suppression of-helicity photon states as a projection from an extended hilbert space onto the physical single - photon hilbert space . we show that all single - photon observables are in general described by povms , obtained by applying this projection to opportune projection - valued measures ( pvms ) , defined on the extended hilbert space . the povms associated to momentum and helicity reduce to pvms , unlike those associated to position and spin , this fact reflecting the intrinsic unsharpness of these observables . we finally extensively study the preparation uncertainty relations for position and momentum and the probability distribution of spin , exploring single photon gaussian states for several choices of spin and polarization .

1403.6405

the investigation of single - photon properties has experienced an increasing interest over the last years @xcite . one of the reasons has to be sought in the escalating request , in many quantum information and cryptography protocols @xcite , for highly accurate manipulations of spatial and polarization single - photon properties . the most appropriate description of single - photon observables is the subject of an ongoing debate . photon position , in particular , has been considered a controversial concept since t.d . newton and e.p . wigner first stated @xcite that no position operator can be defined in the usual sense for particles with mass @xmath1 . later , a.s . wightman @xcite extended the search for a notion of photon localization based on a projection - valued measure ( pvm ) , again obtaining a negative result . this evidence suggested k. kraus to define the single - photon position observable as a positive- operator valued measure ( povm ) @xcite . after the appearance of k. kraus seminal work , several authors proposed alternative definitions of the photon position povm relying on the theory of quantum estimate @xcite or on explicit models describing actual measurements performed in photocounting experiments @xcite . despite such remarkable achievements , the most appropriate single - photon description remains controversial , since the above approaches were conceived to solve the specific problem of photon localization , and do not appear amenable of an immediate generalization to all other single - photon observables . in particular , the photon spin is another notoriously delicate topic@xcite , often ignored in favor of the more familiar notions of helicity and polarization . a renewed interest , due to recent experimental developments especially concerning quantum cryptography protocols @xcite , for the spin of single photons nevertheless calls for an appropriate and manageable description of such observable , in a common picture with that of position . in the present work , we generalize k. kraus construction of the single - photon position observable , given within g. ludwig s axiomatic formulation of quantum mechanics @xcite , to a formalism in which _ all _ the fundamental single - photon observables are given in a unified way in terms of povms @xcite . following the construction of the single - photon hilbert space given by k. kraus and h. moses @xcite , based on the representation theory of the poincar group for mass @xmath1 and spin @xmath2 particles , given by e. p. wigner @xcite , we interpret the notorious suppression of @xmath0-helicity photon states as a projection from an extended hilbert space onto the single - photon hilbert space . extending k. kraus construction of the single - photon position observable , we show that all single - photon observables are described by povms obtained by applying this projection to pvms defined on the extended hilbert space and mutuated from the well - established quantum description of relativistic massive particles . we provide explicit expressions for the povms describing the joint measurement of spin and momentum , and of spin and position . the results show that momentum and helicity are described by pvms , while spin and position by povms . such difference naturally reflects the well - known@xcite circumstance that povms describe _ unsharp _ observables reflecting either practical limits in the precision of measurements ( in which case povms typically correspond to coarse - grained version of pvms ) or inherent difficulties in realizing a preparation in which the value of an observable is perfectly defined @xcite . in particular , the intrinsic unsharpness of position and spin results from the coupling between momentum and spin introduced by the suppression of @xmath0-helicity states , a specific consequence of the mass @xmath1 and spin @xmath2 of the photon . we finally apply this formalism to assess the increase of the statistical character of single - photon observables naturally brought along by the intrinsic _ unsharpness _ of povms @xcite . for this purpose we investigate preparation uncertainty relations for position and momentum , as well as the spin probability distribution . these quantities are analytically calculated for a broad class of physically meaningful single - photon states , namely gaussian states with definite polarization and projections of gaussian states with definite spin . the reasons behind the choice of gaussian states range from their great theoretical and experimental relevance to the fact that , in the non - relativistic context , they saturate the notorious inequality @xmath3 , identifying themselves as the most suitable candidate to investigate the increment of the statistical character of quantum theory brought into stage by the povms . our results show that the emergence of povms systematically increases the randomness of the outcomes@xcite . in particular , the inequality @xmath3 is saturated only in the limiting case of infinitely sharp states in the momentum space ; for any finite gaussian width , we observe instead an increase in the product @xmath4 , of which we give a fully analytic estimate . we observe a similar increase of randomness in the spin probability distribution . such increment appears to be a manifestation of the unsharpness of position and spin , and of the inherent impossibility of sharply localizing a single photon in a bounded space region @xcite , and of preparing it with definite spin along a spatial direction independent on its momentum @xcite . the paper is organized as follows : in section [ sec : formalism ] the quantum mechanical description of a single free photon is reviewed ; in section [ sec : uncertainty ] the procedure for constructing single - photon observables is delineated and the povms and probability densities of such observables are explicitly given . finally , in section [ sec : results ] , a detailed study of the preparation uncertainty relations for position and momentum and of the spin probability distributions of gaussian states is presented , and conclusions are drawn in section [ sec : conclusions ] . in the present section , a detailed review of the single - photon formalism given by k. kraus in @xcite will be given . particular attention will be devoted to the representation theory of the poincar group @xcite and on the introduction of an isomorphism based on the representation of the @xmath5 group for spin @xmath2 particles @xcite . all these elements will provide a framework for the discussion in section [ sec : uncertainty ] , where all the single - photon observables observables , including spin , will be defined as povms through a unified picture generalizing k. kraus treatment of the position observable @xcite . it is worth reminding that the term _ photon _ does not correspond to a unique notion in literature @xcite : photons are either treated as spin @xmath2 and mass @xmath1 irreducible representations of the poincar group @xcite , or as occupations of electromagnetic field modes . in the present work we will rely on the first approach , which naturally brings to the introduction of povms . wigner s seminal work @xcite on the representation of theory of the poincar group , revealed the existence of a deep connection between the symmetries underlying galilean or special relativity and the measurable quantities of an elementary particle . concretely , the mathematical description of a quantum particle existing in space - time should reflect its galilei or poincar invariance , and consequently the state space of such particle should carry an irreducible representation , characterized by its spin and mass , of either the galilei or the poincar group @xcite . with the purpose of introducing the single - photon formalism , let us start by considering the quantum mechanical description of a _ non - relativistic _ particle with spin @xmath2 , which takes place in the hilbert space : @xmath6 here @xmath7 is meant to be the momentum space , elements of @xmath8 are square - integrable @xmath9-component functions : @xmath10 and @xmath8 is equipped with the inner product : @xmath11 the roto - translation group admits the familiar @xcite unitary representation @xmath12 on @xmath8 : @xmath13 where @xmath14 is the matrix associated to the rotation of an angle @xmath15 around the axis @xmath16 , @xmath17 is the vector associated to a spatial translation , and @xmath18 denotes the vector of @xmath19 pauli matrices : @xmath20 it is a well - known fact that a projective irreducible representation of the galilei group can be obtained by introducing the non - relativistic position operator @xmath21 @xcite . the photon anyway is a purely relativistic particle and thus the formalism given above can not suit the purpose of describing it . in order to accomplish such task , let us observe that the matrices @xmath14 and @xmath18 are related to each other by @xcite @xmath22 where the unitary matrix @xmath23 reads @xcite @xmath24 the key relation is a peculiar prerogative of the spin @xmath2 case and suggests to introduce the following unitary transformation : @xmath25 on @xmath8 . it is immediately noticed that wavefunctions @xmath26 in @xmath27 transform as a vector field under roto - translations @xmath28 @xmath29 it is worth of notice that the difference between and consists in the replacements @xmath30 , @xmath31 and correspondingly @xmath32 . the simple change of basis operated by the matrix @xmath23 indicates to generalize the formalism outlined above by means of the following procedure : * endowing wavefunctions in @xmath27 with a fourth component , @xmath33 , this way promoting them to four - component functions : @xmath34 of the four - momentum @xmath35 satisfying the mass - shell condition @xmath36 where @xmath37 and einstein s convention is assumed . the functions lie in the vector space : @xmath38 * equipping @xmath39 with the following sesquilinear form : @xmath40 in which the poincar invariant measure @xmath41 replaces the non - relativistic roto - translationally invariant lebesgue measure @xmath42 . * introducing the following linear representation of the poincar group @xcite by isomorphisms on @xmath39 : @xmath43 which is a direct generalization of with @xmath44 is a four - vector and @xmath45 is a lorentz matrix ; in particular , wavefunctions @xmath46 transform as a four - vector under poincar transformations . the non - positivity of the scalar product in the _ pseudo - hilbert space _ @xmath39 , @xcite which is a direct consequence of the non - positivity of the minkowski inner product @xmath47 , prevents the possibility of giving a probabilistic interpretation to this formalism . nevertheless , such obstacle can be overcome in the special case of massless particles such as photons . in fact , the massless condition naturally identifies a subspace @xmath48 @xmath49 which is invariant under poincar transformations , since : @xmath50 whence @xmath51 for all poincar transformations . moreover , the restriction of results to be non - negative . to prove this result , it is convenient to expand the spatial part of a function @xmath52 on the _ intrinsic frame _ @xmath53 given by : @xmath54 where @xmath55 is an arbitrary unit vector @xcite independent on @xmath56 . it is worth of notice that the first elements of the intrinsic frame are related to the circular polarization vectors @xmath57 , solution of the eigenvalue equation @xmath58 , by the following relation : @xmath59 on this basis , functions in @xmath60 are expressed as : @xmath61 where : @xmath62 the condition @xmath52 translates into @xmath63 , and the restriction of the sesquilinear form onto @xmath60 reads : @xmath64\ ] ] equation is manifestly non - negative , and does not involve the components @xmath65 . following @xcite we interpret these components as irrelevant degrees of freedom , this fact being precursive of the gauge symmetry of the electromagnetic theory@xcite . this interpretation is confirmed if the action of poincar transformations on functions in @xmath60 is taken in consideration . in fact , it is immediate to show that that @xmath66 has components @xmath67 which depend only on @xmath68 through the relation : @xmath69 where @xmath70 indicates the spatial part of @xmath71 . these arguments lead immediately to the construction of the single - photon hilbert space : equation defines a seminorm on @xmath60 , and we can construct a normed space out of @xmath60 taking the quotient @xmath72 of @xmath73 by the equivalence relation : @xmath74 where @xmath75 denotes the seminorm induced by on @xmath60 @xcite . @xmath72 is equipped with the scalar product , which is now manifestly positive - definite , and carries an irreducible representation of the poincar group defined by . each wavefunction in @xmath76 is an equivalence class of functions ins @xmath60 , parametrized by a pair : @xmath77 of complex - valued square - integrable functions , thus making @xmath72 isomorphic to @xmath78 @xcite . each and every equivalence class in @xmath72 has a representative with the following _ transversal form _ : @xmath79 all other elements of the equivalence class are related to the transversal representative by the addition of an unphysical component @xmath80 . this fact shows that the single - photon hlbert space is also isomorphic to the space of square - integrable transverse wavefunctions : @xmath81 retrieving the result given by k. kraus in equations ( 2),(3 ) of @xcite . the construction outlined above has lead in a very natural way to the introduction of the single - photon state space @xmath76 by just considering the conditions of mass @xmath1 and spin @xmath82 , and requiring that @xmath76 carries an irreducible representation of the poincar group . the concrete realization of the single - photon hilbert space @xmath76 has the advantage of providing a one - to - one correspondence between states and transverse vector functions . the physically relevant photon states with linear and circular polarization are elements of @xmath76 of the form : @xmath83 and : @xmath84 finally , one might wish to relate such construction of the single - photon state space to the familiar one based on the requirement of helicity @xmath85 @xcite . the equivalence of the two approaches is readily proved by recalling the isomorphism @xmath23 , introduced to identify the spin as generator of rotations in . in fact , it can be explicitly shown that the images of the intrisic frame vectors under the action of the matrix @xmath23 are closely related to the eigenstates of the helicity operator : @xmath86 in particular @xmath87 is the eigenstate relative to the eigenvalue @xmath88 , while @xmath89 are the eigenstates relative to eigenvalues @xmath85 respectively . in the present section single - photon observables will be introduced as povms , generalizing the treatment of position given by k. kraus in @xcite . with such purpose , the single - photon hilbert space @xmath76 introduced in the previous section has to be regarded as a subspace of the extended hilbert space @xmath90 whose elements have the form : @xmath91 and thus differ from elements of @xmath76 by the addition of a longitudinal component @xmath92 . in equation , and in the remainder of the present work , we express states as functions of @xmath56 rather than @xmath93 with harmless abuse of notation . @xmath76 is obtained from @xmath94 by means of the projection operator : @xmath95 with @xmath96 the projector eliminates the longitudinal component of the triple @xmath97 and can thus be interpreted as an analogue of the helmholtz projection used for decomposing the electric and magnetic field of classical electrodynamics into a longitudinal and a transverse component . under roto - translations , states in @xmath98 must transform as vector fields : @xmath99 in order to retrieve . the introduction of the spin and helicity observables , nevertheless , requires a transformation law under roto - translations involving the vector of pauli matrices , as in , instead of the rotation matrix , as in . the comparison between equations and suggests that , to respond to this need , the isomorphism v must be used . throughout the remainder of the present work , we will denote using the symbol @xmath100 the isomorphic image of @xmath98 through the isomorphism @xmath23 and make use of the fact that @xmath76 corresponds to the image of @xmath101 through the action of the operator @xmath102 , see figure [ fig : spaces ] . the change of basis operated by the matrix @xmath23 not only elucidates the equivalence between the conditions of transversality and non - zero helicity but , as in the case of non - relativistic particles , leads to an irreducible representation of the roto - translation group that involves the vector @xmath18 of spin matrices . in fact , for a state @xmath103 in @xmath101 , equations and imply : @xmath104 elements of @xmath101 therefore transform under roto - translations according to a formula that closely resembles , holding in non - relativistic context . [ fig : spaces ] and projection @xmath105 . wavefunctions in @xmath106 are represented by couples of complex - valued functions . states @xmath107 in @xmath108 transform according to under roto - translations , while states @xmath109 in @xmath110 according to . ] having constructed the single - photon state space , and embedded it into a suitable extended hilbert space , we will now define single - photon observables . to begin , let us consider the case of massive relativistic particles with spin @xmath112 , where each physical observable @xmath113 taking values in a sample space @xmath114 equipped with a suitable sigma - algebra @xmath115 is described in terms of a self - adjoint operator @xmath116 on @xmath117 . by making use of the spectral theorem @xcite , the probability that @xmath113 takes values in a measurable set @xmath118 is given by the expectation value of a projector @xmath119 . the function @xmath120 is notoriously a pvm @xcite . this construction can be generalized to a set @xmath121 of @xmath122 compatible observables , each taking values in the sample space @xmath123 equipped with a suitable sigma - algebra @xmath124 . the compatible observables @xmath121 , in fact , take value in the sample space @xmath125 , equipped with the product sigma - algebra @xmath126 . they are described by a set @xmath127 of @xmath122 self - adjoint commuting operators on @xmath128 , and also in such situation , the spectral theorem allows to express the probability that @xmath121 take values in a measurable set @xmath129 as the expectation value of a projector @xmath130 . the map @xmath131 is referred to as the joint pvm of the compatible observables @xmath121 , and the individual pvms associated to the various observables @xmath132 are readily obtained as marginals of the joint pvm through the operation : @xmath133 for massive particles , the fundamental observables of spin , momentum and position are described by the self - adjoint operators : @xmath134 where @xmath135 is the well - known newton - wigner position operator @xcite . these observables are notoriously covariant under roto - translations . motivated by the seminal work by k. kraus @xcite , we observe that the theory outlined in section [ sec : formalism ] enables to define , for each single - photon observable @xmath113 , a self - adjoint operator @xmath116 on the hilbert space @xmath136 with the same structure as in the case of massive particles , e.g. @xmath137 the only difference with being that the massless condition implies @xmath138 . in the light of equation it is evident that the observables share the roto - translational covariance of . nevertheless , since the physical space of a single photon is @xmath76 , the image of @xmath136 through the operator @xmath139 , the probability that @xmath113 takes values in a borel set @xmath140 is the expectation value of the _ positive operator _ : @xmath141 where @xmath142 and the equality holds by virtue of the idempotence of the pvm @xmath143 . the pvm @xmath144 is turned into the povm @xmath145 on @xmath76 . in fact , the operators @xmath146 , referred to as _ effect operators _ in published literature , are still positive and bounded by the identity operator , but in general are not projectors . the idempotence property characterizing pvms is recovered if only if the single - photon hilbert space @xmath76 is invariant under the action of the projectors @xmath147 . in the remainder of the present section , we will show that relevant examples of such _ projective or sharp observables _ observables are momentum , polarization and helicity . on the other hand , we will show that the relevant position and spin observables are described by povms . in published literature @xcite , observables described by povms are referred to as _ unsharp _ , since the emergence of povms reflects either practical limits in the precision of measurements performed on the system ( in which case povms appear as coarse - grained versions of pvms ) @xcite or the inherent impossibility of realizing a preparation in which the value of an observable can be perfectly defined @xcite . in the case of photons , after explicitly constructing the pvms and povms associated to the fundamental single - photon observables , we will interpret the need of describing position and spin as povms rather than pvms as a consequence of the elimination of @xmath0-helicity states . in the context of open quantum systems theory , povms are obtained as projections of opportune pvms defined on larger hilbert spaces ; from a mathematical point of view , this situation is described by the so - called naimark s dilation theorem@xcite . within the formalism presented in the present work , all this construction emerges in a very natural way by the treatment of the single - photon hilbert space exposed in section [ sec : formalism ] . in the remainder of the present section , we apply the general procedure to explicitly show fundamental examples of povms . as first application , we consider the momentum and spin-@xmath150 observables , which admit the representation on @xmath136 , giving rise to the familiar joint pvm : @xmath151 where @xmath152 is the indicator function of the borel set @xmath153 and @xmath154 with @xmath155 . the corresponding povm on @xmath76 , obtained by applying to such pvm , reads : @xmath156_s\ ] ] and gives rise to the following joint probability distribution : @xmath157_s\bigr|^2 \\ \end{split}\ ] ] since the joint eigenfunctions of the newton - wigner position and spin-@xmath150 operators are the following elements of @xmath136 : @xmath159 and the associated pvm is : @xmath160 \ , \left [ { \mathbf{u}}_{{\mathbf{x}},s}({\mathbf{p } } ' ) \right]_{s'}\ ] ] the corresponding povm on @xmath76 is obtained applying to such pvm , and yields the following joint probability distribution : @xmath161_s\,\biggr|^2 \end{split}\ ] ] where the probability amplitude @xmath162_s$ ] reads : @xmath163_s = \int \frac{d^3p}{|{\mathbf{p}}| } \ , \sqrt{|{\mathbf{p}}| } \ , \frac{e^{\frac{i}{\hbar } { \mathbf{x}}\cdot { \mathbf{p } } } } { ( 2\pi\hbar)^{\frac{3}{2 } } } \ , \sum_{i=1}^2 \tilde{\psi}_v^i({\mathbf{p } } ) \left [ v\tilde{\mathbf{e}}_i({\mathbf{p } } ) \right]_s\ ] ] and can be therefore regarded to as the wave - function for the photon in the configuration space @xcite . it can be noticed that , by virtue of , this amplitude is covariant under roto - translations : @xmath164 reproducing the expected transformation laws for the probability density of spin and position and , consequently , for the momenta of such probability distributions . the position povm obtained as marginal of coincides with the photon position observable constructed by k. kraus in @xcite . on the single - photon hilbert space @xmath76 , states with definite circular polarization take the form . it is easy to verify that such states are eigenstates of the helicity operator @xmath165 with eigenvalues @xmath85 . consequently , the joint povm for momentum and helicity on @xmath76 takes the form : @xmath166 it is easy to verify that @xmath111 is invariant under the action of the operators , which makes it a pvm . clearly , the marginals of describing the momentum and helicity observables are also pvms . in the light of , the joint probability distribution of momentum and helicity reads : @xmath167 from which the probability distribution of helicity is obtained choosing @xmath168 . the probability distribution of the spin projection @xmath170 along an arbitrary spatial direction @xmath16 is readily found recalling that the matrix @xmath171 admits in @xmath172 the spectral decomposition @xmath173 . the joint pvm of @xmath169 and momentum on @xmath136 reads : @xmath174_{s ' } \,\ , { \mbox{\boldmath$\phi$}}^*_{{\mathbf{n}},m_s } \cdot { \mbox{\boldmath$\psi$}}({\mathbf{p}})\ ] ] the corresponding povm on @xmath76 , obtained by applying to such pvm , leads to the joint probability distribution : @xmath175 from which @xmath176 is readily obtained choosing @xmath177 . in the light of and , it becomes interesting to investigate the preparation uncertainty relations for position and momentum observables of a single photon . on @xmath136 , where both these observables are defined in terms of self - adjoint operators with usual commutator @xmath178 = i \hbar \ , \delta_{ij}$ ] , the familiar inequality : @xmath179 holds , and is saturated by gaussian states with definite spin along an arbitrary spatial direction . equation retains its validity on the physical space @xmath111 , but , due to the projection @xmath105 , it is no longer saturated by the minimum uncertainty states on @xmath110 . mean values and variances will be computed taking the marginals of the joint probability distributions and over the spin degrees of freedom . in the case of position , we are left with the following expression : @xmath180_s\,\biggr|^2\ ] ] where the configurational wave - function @xmath181 is known to exhibit polynomial@xcite or even exponential@xcite decay , ensuring the existence of the first momenta of . the latter are readily obtained inserting in : @xmath182_s}{\sqrt{|{\mathbf{p}}| } } \ , \frac{\partial}{\partial p_k } \left(\frac{\left[\tilde{{\mbox{\boldmath$\psi$}}}_v(p)\right]_s}{\sqrt{|{\mathbf{p}}|}}\right ) \\ \langle x^2_k \rangle & = - \hbar^2 \ , \sum_{s=1}^3 \int d^3 p \,\ , \frac{\left[\tilde{{\mbox{\boldmath$\psi$}}}_v(p)\right]_s}{\sqrt{|{\mathbf{p}}|}}\ , \frac{\partial^2}{\partial p_k^2 } \left(\frac{\left[\tilde{{\mbox{\boldmath$\psi$}}}_v(p)\right]_s}{\sqrt{|{\mathbf{p}}|}}\right ) . \\ \end{split}\ ] ] in the remainder of the work , the quantity @xmath183 will be referred to as uncertainty product . equation represents the main contribution of the present work . it generalizes k. kraus construction of the single - photon position observable to a formalism in which all the single - photon observables are defined in a unified way in terms of povms . in the present section , several relevant single - photon observables have been considered and the povms describing them have been derived , according to , as projections on @xmath76 of pvms defined on @xmath136 , whose expression is borrowed form the quantum theory of relativistic massive particles . it is worth of notice that position and spin are described by povms , while momentum and helicity by pvms , a circumstance which is amenable of a clear physical interpretations . observables @xmath113 represented as pvms are commonly understood to correspond to measurements with perfect accuracy , and are therefore called _ sharp observables_. in fact , the generic projector @xmath119 associated to the event @xmath184 identifies a subspace : @xmath185 of preparations for which the event @xmath184 has probability @xmath186 . this property is notoriously not shared by generic povms , which are therefore associated to _ unsharp observables _ @xcite . in this sense , we understand that the emergence of povms enhances the statistical character of quantum theory : concretely , the recursion to povms typically reflects either imperfect measurements or inherent difficulties in realizing a preparation in which the event @xmath184 has probability @xmath186 . in the case of photons , the need of treating position and spin as unsharp observables reflects the properties of single - photon preparations rather than those of measurement procedures . in particular , the suppression of @xmath0-helicity states determines a coupling between momentum and spin , expressed by the key role of the helicity , which limits the possibility to prepare a photon with definite spin along a fixed spatial direction independent on its momentum . similarly , since transverse wavefunctions are linear combinations of @xmath187 and @xmath188 and the latter are not analytic functions of @xmath56 @xcite . the paley - wiener theorem @xcite poses severe limitations on the possibility of localizing photons in the configuration space @xcite . these inherent difficulties find a direct correspondence in our construction , which highlights the need of treating spin and position as unsharp quantum observables . finally , it is worth of notice that equation can be further generalized to the case of a set of incompatible ( non - commuting ) observables @xmath189 on @xmath94 . in this case the probability that @xmath189 take values in a measurable set @xmath190 is given in terms of the expectation value of an effect operator @xmath191 . the family of such operators is referred to as the povm associated to the unsharp joint measurement of the incompatible observables @xmath189 @xcite . the projection onto the physical single - photon space @xmath76 leads therefore to another povm which is related to that defined on @xmath94 by means of the equation : @xmath192 this expression , which clearly extends , can be used for example to describe the joint measurement of position and momentum observables . in the present section we will consider two classes of physically relevant single - photon states , namely gaussian states with definite polarization and projections of gaussian states with definite spin , and extensively investigate : ( i ) the preparation uncertainty relations @xmath193 and ( ii ) the probability distribution of spin along a certain direction . the study of both these properties will highlight and quantitatively estimate the increase of the statistical character naturally brought along by the intrinsic _ unsharpness _ of povms @xcite . the results will show both an increment in the product @xmath183 for gaussian states , which in non - relativistic framework saturate the inequality @xmath194 , and the increase of randomness in the probability distribution of spin along a fixed spatial direction . let us consider gaussian states with definite polarization in @xmath195 , i.e. states of the form : @xmath196 where @xmath197 , @xmath198 , @xmath199 and : @xmath200 is a positive , dimensionless parameter which takes into account the wavefunction s spread in momentum space . in the following , without any loss of generality , the chioce @xmath201 will be made . the square norm of states in @xmath111 of the form is obtained from a straightforward calculation and reads : @xmath202 so that , if @xmath203 , the mean values and the variances of momentum components are given by : @xmath204 on the contrary , the calculation of the expectation value and variance of position components is complicated by the circumstance that the gradient appearing in acts non - trivially on the vectors @xmath205 . applying formula to the states , it is readily found that the mean values of @xmath206 and @xmath207 are bilinear functions of the coefficients @xmath208 , defined by the matrices : @xmath209_{kl } & = i \hbar \int d^3 p \ , g({\mathbf{p } } ) \ , \tilde{{\mathbf{e}}}_k({\mathbf{p } } ) \cdot \partial_{p_j } \tilde{{\mathbf{e}}}_l({\mathbf{p } } ) \\ \left [ ( x^j)^2 \right]_{kl } & = - \hbar^2 \ , \int d^3 p \ , g({\mathbf{p } } ) \ , \left [ \tilde{{\mathbf{e}}}_k({\mathbf{p } } ) \cdot \partial^2_{p_jp_j } \tilde{{\mathbf{e}}}_l({\mathbf{p } } ) -\frac{p_j - p_{0 j}}{4 a p_0 ^ 2 } \tilde{{\mathbf{e}}}_k({\mathbf{p } } ) \cdot \partial_{p_j } \tilde{{\mathbf{e}}}_l({\mathbf{p } } ) \right ] + \delta_{kl } \ , \frac{\hbar^2}{8 a p_0 ^ 2 } + ( x_0^j)^2 \ , \delta_{kl } \\ \end{split}\ ] ] with @xmath210 . performing the integrations using shifted spherical coordinates one finds : @xmath211_{kl } & = 0 \quad \quad \forall j = 1,2,3\\ \left[(x^3)^2\right]_{kl } & \equiv \left[(z)^2\right]_{kl}= \hbar^2 \delta_{kl } \ , \frac{1 - 4 a + 4 \sqrt{a } \ , ( 1 + 2 a ) \ , { \mathcal{d}\left ( \frac{1}{2\sqrt{a } } \right)}}{8 a p_0 ^ 2 } \\ \left[(x^1)^2\right]_{kl } & \equiv \left[(x)^2\right]_{kl}= \hbar^2 \delta_{kl } \ , \frac{1 + 8 a -16 a^{3/2 } { \mathcal{d}\left ( \frac{1}{2\sqrt{a } } \right)}}{8 a p_0 ^ 2 } \\ \left[(x^2)^2\right]_{kl } & \equiv \left[(y)^2\right]_{kl}= \hbar^2 \delta_{kl } \ , \frac{1 + 8 a -16 a^{3/2 } { \mathcal{d}\left ( \frac{1}{2\sqrt{a } } \right)}}{8 a p_0 ^ 2 } \\ \end{split}\ ] ] where @xmath212 denotes the dawson s function ( sometimes also referred to as dawson s integral ) of argument @xmath213 @xcite . the uncertainty products _ both _ for @xmath214 _ and _ @xmath215 , therefore read : @xmath216 and are illustrated in figure [ fig : e1ofp ] . , in units of @xmath217 , along directions parallel ( red solid line ) and perpendicular ( blue dashed line ) to @xmath218 , for states with definite polarization . ] the results are monotonically increasing functions of the parameter @xmath219 and remarkably , in the limit @xmath220 , the minimum value @xmath221 is obtained for every component thus indicating that only for an infinitely sharp wavefunction in the momentum space the familiar heisenberg limit is retrieved . in the opposite limit @xmath222 , the uncertainty product approach from below an asymptotic finite value which is equal to @xmath223 for every component . in table , the asymptotic expansions of the uncertainty products are given . @xmath224 c c c + + axis & small @xmath219 & large @xmath219 + + + z & @xmath225 & @xmath226 + + x ( y ) & @xmath227 & @xmath228 + + + [ table : limiting ] the increase in the uncertainty product witnessed for @xmath229 is a direct consequence of the _ unsharpness _ of the povm associated with the position observable , and gives a direct example of the circumstance , well known in the quantum theory of measurement @xcite , that povms enhance the randomness of measurement outcomes and the statistical character of the underlying quantum theory . the explicit expression of the probability distribution of @xmath148 for gaussian states of the form will now be given . it reads : @xmath230_{ij } \ , \gamma^j\ ] ] where @xmath154 and @xmath231_{ij } = \int d^3{\mathbf{p } } \ , \left [ \tilde{{\mathbf{e}}}_i({\mathbf{p } } ) \right]^*_s \left [ \tilde{{\mathbf{e}}}_j({\mathbf{p } } ) \right]_s \,\ , g({\mathbf{p}})\ ] ] equation is a quadratic function of @xmath232 and can be expressed in terms of three functions ( the first two of which will appear also in the forthcoming discussion relative to states with definite spin ) : @xmath233 the matrices @xmath234 are given by @xmath235 from which it can be immediately seen that : \a ) the eigenvectors of are the states @xmath236 with definite circular polarization \b ) @xmath237 for @xmath238 and @xmath237 for @xmath239 are equal to each other . \c ) @xmath240 for @xmath238 and @xmath241 for @xmath239 are equal to each other . figure shows the probability distribution of @xmath148 for a gaussian state with circular polarization @xmath242 . ( red dashed line for @xmath243 , blue dot - dashed line for @xmath244 , green solid line for @xmath245 ) on states with definite polarization @xmath242 . ] finally , the asymptotic behavior of the probability distribution of @xmath148 for gaussian states with definite polarization is listed in table . c c c c + @xmath246 & polarization & small @xmath219 & large @xmath219 + + + 1 & + 1 & @xmath247 & @xmath248 + 0 & + 1 & @xmath249 & @xmath250 + -1 & + 1 & @xmath251 & @xmath252 + 1 & -1 & @xmath251 & @xmath252 + 0 & -1 & @xmath249 & @xmath250 + -1 & -1 & @xmath247 & @xmath248 + [ table : szhel ] gaussian states with definite spin along an arbitrary spatial axis will now be considered , which are defined in @xmath253 as : @xmath254 the corresponding physical states are elements in @xmath111 given by : @xmath255 the quantity @xmath256 being a normalization constant . for states in @xmath110 of the form , denoting for simplicity @xmath257 we readily find that the square norm @xmath258 and the first momenta of the position and momentum probability distributions are quadratic functions of @xmath259 defined by the matrices : @xmath260_{kl } & = \int d^3 p \ , g({\mathbf{p } } ) [ \pi({\mathbf{p}})]_{kl } \\ [ p^j]_{kl } & = \int d^3 p \ , g({\mathbf{p } } ) \ , p^j \ , [ \pi({\mathbf{p}})]_{kl } \\ [ ( p^2)^j]_{kl } & = \int d^3 p \ , g({\mathbf{p } } ) \ , ( p^j)^2 \ , [ \pi({\mathbf{p}})]_{kl } \\ [ x^j]_{kl } & = i \hbar\ , \int d^3 p \ , g({\mathbf{p } } ) \ , \left [ \frac{\partial}{\partial p_j } \ , [ \pi({\mathbf{p}})]_{kl } - \frac{p^j - p_0^j}{4 a p_0 ^ 2 } \ , [ \pi({\mathbf{p}})]_{kl } \right ] \\ [ ( x^2)^j]_{kl } & = - \hbar^2 \int d^3 p \ , g({\mathbf{p } } ) \ , \left [ \frac{\partial^2}{\partial p_j^2 } \ , [ \pi({\mathbf{p}})]_{kl } - 2 \ , \frac{p^j - p_0^j}{4 a p_0 ^ 2 } \ , \frac{\partial}{\partial p_j}[\pi({\mathbf{p}})]_{kl}\right.\notag\\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad+\left.\left [ \left ( \frac{p^j - p_0^j}{4 a p_0 ^ 2 } \right)^2 - \frac{1}{4 a p_0 ^ 2 } \right ] [ \pi({\mathbf{p}})]_{kl } \right ] \end{split}\ ] ] of which we below the analytic expression , resulting from a calculations in shifted spherical coordinates , which remarkably can be expressed in terms of the functions @xmath261 and @xmath262 introduced above in . the normalization reads : @xmath263 the first and second momenta of the momentum probability distribution are given by : @xmath264 while the first and second momenta of the position probability distributions read : @xmath265 the matrices associated to the components of momentum and position observable along the @xmath266 and @xmath267 axes differ one from each other , but share the same spectrum . the uncertainty products on the @xmath150 and @xmath268 axis , which correspond parallel and perpendicular to @xmath218 , are given by : * if @xmath269 , or @xmath270 : @xmath271 \big [ \frac{1}{4a } + \left ( \frac{1}{2 } - \frac{1}{8a } \right ) u_2(a ) - u_1(a ) \big ] } } { 1 - 2 a \ , u_2(a ) } \\ \delta x \cdot \delta p_x & = \hbar \frac{\sqrt { 2a\left ( 1 - 4a\,u_2(a ) \right)\left(\frac{1}{2}u_1(a)- u_2(a)\left ( \frac{1}{2}+\frac{1}{4a } \right ) + \frac{3}{8a}\right)}}{1 - 2a\ , u_2(a)}\\\end{aligned}\ ] ] * if @xmath272 @xmath273\left(\frac{3}{4a } - \left ( 3 + \frac{3}{4a } \right ) u_2(a ) + 2 u_1(a)\right ) } } { 2 u_2(a ) } \\ \delta x \cdot \delta p_x & = \hbar \frac{\sqrt{u_1(a)\left(2u_2(a)-u_1 ( a)\right)}}{u_2(a)}\end{aligned}\ ] ] these results are presented in figures and respectively . in units of @xmath217 , along directions parallel ( red line ) and perpendicular ( blue line ) to @xmath218 , for eigenstates of @xmath274 with eigenvalue @xmath275 . ] in units of @xmath217 , along directions parallel ( red solid line ) and perpendicular ( blue dashed line ) to @xmath218 , for eigenstates of @xmath274 with eigenvalue @xmath276 . ] as highlighted in the case of states with definite polarization , the preparation uncertainty relations explicitly show a highly non - trivial dependence on the width parameter @xmath219 . for @xmath277 ( @xmath278 ) the uncertainty products are monotonically increasing functions of @xmath219 , converging to @xmath279 in the @xmath220 limit and to finite values in the opposite @xmath222 limit . the asymptotic expansions of and are listed , in units of @xmath280 , in table [ table : limiting2 ] . for @xmath281 , on the other hand , this behavior is exhibited only by the uncertainty product along the direction of @xmath218 : the other uncertainty product appears a monotonically decreasing function of @xmath219 , converging to @xmath186 in the @xmath220 limit . the apparently anomalous behaviour of the uncertainty product in figure [ 010 ] is a consequence of the fact that the state has , for decreasing @xmath44 , vanishing projection on the physically relevant space @xmath76 . @xmath224 c c c c + + @xmath259 & axis & small @xmath219 & large @xmath219 + + + @xmath282 & z & @xmath225 & @xmath283 + + @xmath282 & x & @xmath284 & @xmath285 + + @xmath286 & z & @xmath287 & @xmath288 + + @xmath286 & x & @xmath289 & @xmath290 + + + [ table : limiting2 ] finally , and depend quadratically on @xmath291 : in appendix , the minimum and maximum valued attained by the uncertainty product for different choices of @xmath259 have been calculated and plotted over a meaningful range of values for the parameter @xmath219 , and the results compared with those shown in figures , . the explicit expression of the probability distribution of @xmath148 for a spin state of the form will now be given . it reads : @xmath292_{ij } \ , \mathbf{h}_j}{k}\ ] ] where @xmath293 , @xmath231_{ij } = \int \frac{d^3{\mathbf{p}}}{p^0 } \ , [ \pi({\mathbf{p}})]^*_{i s } \ , [ \pi({\mathbf{p}})]_{s j } \ , g^2({\mathbf{p}})\ ] ] and @xmath294 is the normalization constant introduced above . equation is then the ratio between two quadratic functions of @xmath259 . the denominator is explicitly given by , while the matrix @xmath295 in the numerator is most conveniently expressed in terms of the function @xmath296 and reads : @xmath297 from which it is evident that : + a ) @xmath298 for the spin state relative to the choice @xmath299 + b ) @xmath300 for the spin state with @xmath301 and @xmath302 for the spin state with @xmath303 are equal to each other . figure [ fig : spinddpe1 ] shows the cases @xmath304 . * ( a ) * + ( red dashed line for the eigenvalue @xmath243 , green solid line for @xmath244 , blue dot - dashed line for @xmath245 ) on the spin states with @xmath305 ( a ) , @xmath306 ( b ) and @xmath307 ( c),title="fig : " ] + * ( b ) * + ( red dashed line for the eigenvalue @xmath243 , green solid line for @xmath244 , blue dot - dashed line for @xmath245 ) on the spin states with @xmath305 ( a ) , @xmath306 ( b ) and @xmath307 ( c),title="fig : " ] + * ( c ) * + ( red dashed line for the eigenvalue @xmath243 , green solid line for @xmath244 , blue dot - dashed line for @xmath245 ) on the spin states with @xmath305 ( a ) , @xmath306 ( b ) and @xmath307 ( c),title="fig : " ] the asymptotic expansions of the probability distributions for the spin @xmath274 are listed in table . c c c c + @xmath246 & @xmath259 & small @xmath219 & large @xmath219 + + + 1 & ( 1,0,0 ) & @xmath308 & @xmath309 + 0 & ( 1,0,0 ) & @xmath310 & @xmath311 + -1 & ( 1,0,0 ) & @xmath312 & @xmath313 + 1 & ( 0,1,0 ) & @xmath314 & @xmath315 + 0 & ( 0,1,0 ) & @xmath316 & @xmath317 + [ table : sz ] due to the projection onto the physical hilbert space , the spin of the photon therefore ceases to be a definite quantity and is characterized by a probability distribution resulting by integration over the momentum variable . this result is consequence of the fact that states with definite spin exist only in the extended hilbert space and puts under an even more clear evidence the enhancement of the statistical character of the quantum theory stemming from the unsharpness of povms . in the present work , we have constructed a unified procedure for treating all single - photon observables as povms , including the notoriously delicate position and spin observables , and for computing all joint probability distributions . starting from the representation theory of the poincar group for spin @xmath2 and mass @xmath1 particles , we have demonstrated how the suppression of the @xmath0-helicity component of the single - photon wavefunction , which corresponds through the isomorphism @xmath318 to the suppression of the longitudinal component , can be viewed as the action of a projection operator from an extended hilbert space onto the single - photon hilbert space . we have then shown how this theoretical construction naturally brings along the notion of povms , since _ any _ observable can be introduced as a pvm on the extended hilbert space , the form of which is very naturally suggested by the well - established quantum theory of relativistic particles with mass @xmath319 and spin @xmath2 , and then turned , by means of the projection , into a directly manageable povm on @xmath76 . the suppression of @xmath0-helicity states determines a coupling between momentum and spin , expressed by the sharpness of the helicity observable . the intrinsic unsharpness which characterizes the povms of spin and position , on the other hand , reflects the impossibility of preparing a photon with definite spin along a fixed arbitrary direction or localizing it in a bounded region of the configuration space , according to the paley - wiener theorem . we have explicitly derived the joint probability distributions for spin and momentum , and for spin and position , along with their marginals and with the probability distribution of helicity . finally , we have extensively investigated the consequences of the introduction of the povms , i.e. the effects of the projection @xmath320 , by studying the preparation uncertainty relations for position and momentum and the probability distribution of spin @xmath274 . we have considered different classes of gaussian states , which in the non - relativistic context saturate the inequality @xmath321 , namely those with definite polarization and with definite spin along a spatial axis . the results we have found show a clear enhancement of the statistical character of position and spin observables brought into stage by the appearance of povms , which ranges from the increment of the uncertainty products for any value of the gaussian width @xmath219 , to the impossibility of preparing a state with definite spin along an axis independent on momentum . only in the limiting case of infinitely sharp wavefunction in the momentum space ( which corresponds to @xmath322 ) , a single - photon state can have a defined value of the spin and the uncertainty products return to the value @xmath323 . such behaviour is observed both for gaussian states with definite polarization and spin , confirming the coherence of the underlying formalism , which can be immediately applied to all single - photon observables , and to all single - photon states and density matrices . we believe there exist favourable prospects for this formalism to be applied in a high - precision description of interference phenomena at the basis of experiments with single photons @xcite . we express our deep gratitude to dr . andrea smirne for support and fruitful discussions in merit , and to prof . luca molinari and dr . davide galli for their useful feedback . m.m . also acknowledges fundings from the dr . davide colosimo award , celebrating the memory of physicist davide colosimo . in this appendix we identify the quantities @xmath324 and @xmath325 which respectively maximize and minimize the heisenberg uncertainty relations and . for the purpose of simplifying the forthcoming calculations , let us parametrize @xmath326 as follows : @xmath327 it can be easily verified that @xmath328 if and only if @xmath329 . this choice leads to the following expressions : @xmath330 \\ \quad\quad\quad & \braket{(p_z)^2 } = p_0 ^ 2 \ , \left [ \frac{1 + 16 a^2 u_1(a)}{1 - 2 a u_2(a ) } ( 1-\rho ) + \frac{1 - 8 a u_1(a)}{u_2(a ) } \rho \right ] \\ \\ & \braket{z } = 0 \\ \quad\quad\quad & \braket{z^2 } = \frac{\hbar^2}{p_0 ^ 2 } \ , \ , \left [ \frac{\frac{1}{4a } + \frac{u_2(a)}{2 } + \frac{u_2(a)}{8a } - u_1(a)}{1 - 2 a u_2(a ) } ( 1-\rho ) + \frac{\frac{3}{4a } - \left ( 3 + \frac{3}{4a } \right ) u_2(a ) + 2 u_1(a)}{4 a u_2(a ) } \rho \right ] \end{split}\ ] ] where @xmath331 , @xmath332 . the square of the heisenberg product constructed from is a polynomial of third degree in the variable @xmath333 possessing , for each fixed value of @xmath219 , a global maximum at a value @xmath334 in the interval @xmath335 and a global minimum at @xmath336 , where @xmath337 . we do not detail the rather intricate analytic form of the minimum and maximum values of the heisenberg product , but limit ourselves to show , in figure [ fig : h1h2h3_guarmotta ] , their values against those in figures [ 100 ] and [ 010 ] relative to the choices @xmath301 and @xmath299 . * ( a ) * + , in comparison with @xmath301 ( red dotted line ) and @xmath299 ( green dot - dashed line).,title="fig : " ] + * ( b ) * + , in comparison with @xmath301 ( red dotted line ) and @xmath299 ( green dot - dashed line).,title="fig : " ] on the @xmath213 axis , it is convenient to parametrize also @xmath338 as follows : @xmath339 obtaining : @xmath340 \\ \notag\\ & \braket{\hat{x } } = 0 \\ \notag\\ & \braket{\hat{x}^2 } = \frac{\hbar^2}{(2 p_0)^2 } \ , \bigg [ \frac { u_1(a ) - \frac{u_2(a)}{a } - 4 u_2(a ) + \frac{3}{2a}}{1 - 2 a u_2(a ) } \ , \lambda + \notag\\ & + \frac { 3 \ , u_1(a ) - u_2(a ) + \frac{3}{2a}}{1 - 2 a u_2(a ) } \ , ( 1-\lambda ) ( 1-\xi ) + \frac{2u_2(a)-u_1(a)}{a u_2(a ) } \ , ( 1-\lambda ) \xi \bigg ] \\ \end{split}\ ] ] the square of the heisenberg uncertainty product is a linear and monotonically decreasing function of @xmath341 . the minimum is attained at @xmath342 , @xmath343 and @xmath344 , @xmath345 , and the maximum at @xmath346 , @xmath347 and @xmath348 . @xmath349 results from a straightforward but quite lengthy maximization procedure . the minimum and maximum values of the heisenberg uncertainty product are shown in fig . [ fig : h1h2h3_guarmotta ] against the corresponding values for @xmath301 and @xmath299 . c c c c + + @xmath259 & axis & small @xmath219 & large @xmath219 + + + @xmath350 & z & @xmath351 & @xmath288 + + @xmath352 & z & @xmath353 & @xmath354 + + @xmath350 & x & @xmath355 & @xmath356 + + @xmath352 & x & @xmath357 & @xmath358 + + + 99 m. tsang , _ phys . rev . lett . _ * 102 * , 253601 ( 2009 ) g. vallone , v. dambrosio , a. sponselli , s. slussarenko , l. marrucci , f. sciarrino and p. villoresi , _ phys . rev . lett . _ * 113 * , 060503 ( 2014 ) i. bialynicki - birula , _ phys . rev . lett . _ * 80 * , 5247 ( 1998 ) r. y. chiao and y .- s . wu , _ phys . _ * 57 * , 933 ( 1986 ) a. tomita and r. y. chiao , _ phys . lett . _ * 57 * , 937 ( 1986 ) g.s . paraoanu , _ phys . lett . _ * 97 * , 180406 ( 2006 ) y. liu et . al . , _ phys . lett . _ * 111 * , 130502 ( 2013 ) d. g. grier , nature * 424 * , 810 - 816 ( 2003 ) . t.d . newton and e.p . wigner , _ rev . * 21 * , 400 ( 1949 ) a.s . wightman , _ rev . * 34 * , 845 ( 1962 ) k. kraus `` position observable for the photon '' in `` uncertainty principles and the foundation of quantum mechanics '' w.c . price , s.s . chissich , ed . wiley ( 1977 ) a. holevo , `` statistical structure of quantum theory '' , lecture notes in physics monographs , springer ( 2001 ) a.s . holevo , `` probabilistic and statistical aspects of quantum theory '' , north - holland publ . cy . , amsterdam ( 1982 ) . a. s. holevo , _ rep . * 13 * , 3 ( 1978 ) a. s. holevo , _ russian microelectronics _ * 13 * , 21 ( 2006 ) l. mandel and e. wolf , optical coherence and quantum optics ( cambridge university press , cambridge , 1995 ) . m. hawton _ phys . rev . a _ * 82 * , 012117 ( 2010 ) a. i. akhiezer and v. b. berestetskii , quantum electrodynamics ( interscience publishers , new york , 1965 ) . s. j. van enk and g. nienhuis , _ europhys . _ * 25 * , 497 ( 1994 ) , _ j. mod . opt . _ * 41 * , 963 ( 1994 ) . k. y. bliokh , m. a. alonso , e. a. ostrovskaya and a. aiello , _ phys . rev . a _ * 82 * , 063825 ( 2010 ) g. ludwig , an axiomatic basis for quantum mechanics , berlin : springer(1985 ) h.e . moses , _ j. math . phys . _ * 8 * , 1134 ( 1967 ) ; h.e . moses , _ j. math . * 9 * , 16 ( 1968 ) e. p. wigner , ann . math . * 40 * , 149 ( 1939 ) p. busch , m. grabowski and p. lahti , operational quantum physics , lecture notes in physics monographs , springer ( 1995 ) h .- p . breuer and f. petruccione , _ the theory of open quantum systems _ ( oxford university press , oxford , 2002 ) . p. busch and g. jaeger , _ found . _ * 40 * , 1341 ( 2010 ) p. busch and a. shimony , _ stud . * 27 * , 397 ( 1996 ) p. busch and f.e . schroeck jr . _ found . * 19 * , 807 ( 1989 ) s. massar , phys . a * 76 * , 042114 ( 2007 ) r. e. a. c. paley and n. wiener , fourier transforms in the complex domain ( american mathematical society , new york , 1934 ) . amrein , helv . acta 42 ( 1969 ) m. hamermesh , group theory and its application to physical problems , dover publications ( 1989 ) . g. mccabe , the structure and interpretation of the standard model , elsevier ( 2007 ) g. w. mackey , mathematical foundation of quantum mechanics , dover ( 1963 ) j. m. lvy - leblond , _ commun . math . phys . _ * 6 * , 286 ( 1967 ) considering the particular @xmath82 unitary and irreducible representation of the spin operators @xmath359 on @xmath360 and the complexification of @xmath361 , one has that the operators @xmath359 and @xmath362 , with @xmath363 denoting the generators of the rotations in @xmath361 , satisfy the same commutation relations . then , the irreducibility of the @xmath82 representation implies that there exists a unitary transformation @xmath318 such that @xmath364 . the term pseudo - hilbert is inherited from the usual terminology which refers to the pseudo - euclidean character of the relativistic four - dimensional metric . i. bialynicki - birula and z. bialynicka - birula , j. phys . a , * 46 * 053001 ( 2013 ) s. gupta , _ proc . * a63 * , 681 ( 1950 ) k. bleuler , _ helv . * 5 * , 567 ( 1950 ) v. bach , j. frhlich and i. m. sigal , adv . math . 137 , 299 - 395 ( 1998 ) s. weinberg , the quantum theory of fields , cap . ( 2.5 ) and ( 5.9 ) cambridge university press ( 2000 ) e.g. beltrametti and g. cassinelli , the logic of quantum mechanics , addison - wesley ( 1981 ) v. moretti , spectral theory and quantum mechanics , springer unitext ( 2013 ) k. kraus , states , effects and operations : fundamental notions of quantum theory , springer , berlin ( 1983 ) . i. bialynicki - birula , _ acta physica polonica _ , * 86 * , 97 ( 1994 ) , i. bialynicki - birula , in _ progress in optics _ , edited by e. wolf ( elsevier , amsterdam , 1996 ) , vol . xxxvi . j. e. sipe , phys . a , * 52 * , 1875 ( 1995 ) c. adlard , e. r. pike , s. sarkar , _ phys . * 79 * , 1585 ( 1997 ) k. w. chan , c. k. law , and j. h. eberly , _ phys . rev . lett . _ * 88 * , 100402 ( 2002 ) p. busch , _ found . _ * 39 * , 712 ( 2009 ) m. abramowitz and i. stegun handbook of mathematical functions , dover ( 1964 ) m. larqu , a. beveratos , i. robert - philip , eur . j. d * 47 * , 119125 ( 2008 ) a. beveratos , r. brouri , t. gacoin , a. villing , j .- poizat , p. grangier , phys . * 89 * , 187901 ( 2002 )

we investigate the spectrum of a typical non - self - adjoint differential operator acting on , where is a constant matrix . we impose dirichlet and neumann boundary conditions in the first and second coordinate respectively at both ends of\subset { \mathbb{r}}$ ] . for we explore in detail the connection between the entries of and the spectrum of , we find necessary conditions to ensure similarity to a self - adjoint operator and give numerical evidence that suggests a non - trivial spectral evolution . -1 in _ _ -1.5em 1.5em lyonell s. boulton department of mathematics , king s college london the strand , london wc2r 2ls , u.k . email : [email protected]

math0102170

in this paper we investigate spectral properties of the linear operator @xmath6 acting on @xmath8 where @xmath2 is a @xmath9 constant matrix and @xmath10 denotes the ordinary differential operator @xmath11 the apparently simple combination of dirichlet and neumann boundary conditions allows self - adjointness if , an only if , @xmath2 is real and diagonal . if @xmath2 is non - diagonal and upper - triangular the numerical range of @xmath6 is a large sector of @xmath12 . otherwise it is the whole of @xmath12 preventing us from applying the theory of sectorial sesquilinear forms in a straightforward manner . our main goal is to explore the connection between the entries of the matrix @xmath2 and the location of the spectrum of @xmath6 in the complex plane . in @xcite r. f. streater considers the particular case @xmath13 in order to find necessary conditions for the stability of small perturbations about the stationary solution of certain non - linear system of parabolic equations . streater s system represents a thermodynamical model for hot fluid in one dimension and the localization of the spectrum is achieved by constructing a non - unitary transformation that makes @xmath6 similar to a non - negative self - adjoint operator , hence the spectrum of @xmath6 is real and non - negative . this similarity transformation does not work for other matrices and a slight modification of the entries of @xmath2 can destroy reality of the spectrum ( cf . sections [ s6]-[s7 ] ) so the general case should be attacked by other methods . although this paper mainly concerns @xmath5 , the results of sections [ s2]-[s5 ] refer to any complex @xmath3 matrix . the core results are to be found in section [ s6 ] where we present an exhaustive description of the spectrum of @xmath6 in terms of the entries of @xmath2 . among various other unexpected conclusions , the following three epitomize the complexity of the problem to be considered : * when @xmath2 is triangular and non - diagonalizable , @xmath6 is not similar to a self - adjoint operator but the spectrum of @xmath6 is real ( theorem [ t11 ] ) . * the spectrum of @xmath6 can be non - real even when both eigenvalues of @xmath2 are positive and equal ( theorem [ t32 ] ) . * there is a continuous family of matrices @xmath2 whose eigenvalues do not intersect the real line but such that the spectrum of @xmath6 is real ( theorem [ t31 ] ) . the last two assertions show that the spectra of @xmath2 , @xmath10 and @xmath6 are typically unrelated . the crucial idea in section [ s6 ] is to reduce the four - parameter problem of localizing the spectrum of @xmath6 in terms of the entries of @xmath2 , to five two - parameter cases and describe separately each of these cases . sections [ s2]-[s5 ] are devoted to describing the various properties of @xmath6 we will use in section [ s6 ] , whereas section [ s7 ] is devoted to numerical computations which illustrate some of the results reported . in section [ s2 ] we find the boundary conditions associated to the adjoint of @xmath6 and compute the numerical range of @xmath6 . in section [ s3 ] we show that the resolvent of @xmath6 is compact for all non - singular @xmath2 . in section [ s4 ] we explore the stability of the spectrum of @xmath6 in the sense of @xcite and @xcite , and provide estimates which allow us to enclose the spectrum of @xmath6 in angular regions when @xmath2 is subject to various constraints . in section [ s5 ] we use standard ode methods to compute the transcendental function of the spectral problem associated to @xmath6 . let @xmath14 be a linear operator whose domain is denoted by @xmath15 . throughout this paper @xmath16 stands for the spectrum of @xmath14 and the numerical range of @xmath14 is defined to be @xmath17 we recall that the numerical range of any linear operator is convex and that if @xmath18 , then @xmath19 if @xmath20 and @xmath21 , we will say that @xmath14 is positive and write @xmath22 . if @xmath20 and @xmath23 , we will say that @xmath14 is non - negative and write @xmath24 . below and elsewhere @xmath25 denotes the norm of a vector @xmath26 . the norm of any @xmath27 is the standard hilbert tensor product norm @xmath28 unless explicitly stated , we denote @xmath29 the complex numbers @xmath30 denote the eigenvalues of @xmath2 and the non - zero @xmath31 vectors @xmath32 denote the eigenvectors @xmath33 if @xmath34 and @xmath35 are real and different , we adopt the convention @xmath36 . notice that the @xmath37 are not necessarily orthogonal . let @xmath38 be the sobolev space of all @xmath39 , such that the generalized derivative @xmath40 . we define rigorously the domain of @xmath6 as @xmath41 if @xmath2 is invertible , it is standard to show that @xmath6 is always a closed densely defined linear operator acting on @xmath42 . [ t27 ] if @xmath2 is singular , then @xmath6 is not closed in the domain @xmath43 . * . * let @xmath44 be a non vanishing vector such that @xmath45 and let @xmath46 . clearly @xmath47 . let @xmath48 be a sequence of smooth functions whose support is compact in @xmath49 and such that @xmath50 in @xmath51 . then @xmath52 and @xmath53 . also @xmath54 so that @xmath55 is a convergent sequence in @xmath56 . we complete the proof by noticing that if @xmath6 was closed , then we would have @xmath57 . @xmath58 for the rest of this section and in sections [ s3]-[s5 ] we will assume without further mention that @xmath2 is non - singular . in section [ s6 ] we will consider again singular @xmath2 . in order to show that @xmath6 is in general non - self - adjoint , let us compute the adjoint @xmath59 . let @xmath60 then the boundary conditions for @xmath10 can be rewritten as @xmath61 [ t18 ] the adjoint of @xmath6 is @xmath62 for @xmath63 subject to the boundary conditions @xmath64 where @xmath65 is the rank one projection such that @xmath66 * . * for @xmath57 and @xmath67 , @xmath68 we ought to find a complex @xmath69 matrix @xmath70 and impose boundary conditions on @xmath71 , for @xmath72 and @xmath73 this must be true in particular for all @xmath74 and @xmath71 with compact support in @xmath49 , so clearly @xmath75 . let the boundary conditions for @xmath76 be given by ( [ e1 ] ) where @xmath65 is a non - necessarily orthogonal projection on @xmath31 , we show ( [ e25 ] ) . if @xmath77 are smooth functions supported in @xmath78 , then @xmath79 where @xmath80 , @xmath81 , @xmath82 and @xmath83 are arbitrary vectors in @xmath31 . if @xmath84 , the right hand side should vanish for all @xmath85 , so that @xmath86 if @xmath87 , the left hand side should vanish for all @xmath88 , so that @xmath89 since @xmath2 is non - singular these two spaces are one dimensional . [ t19 ] @xmath6 is self - adjoint , if and only if @xmath2 is real and diagonal . * . * using the notation of lemma [ t18 ] , @xmath6 is self - adjoint , if and only if @xmath90 the latter occurs , if and only if @xmath91 these conditions ensure @xmath2 real and diagonal . @xmath58 we now show that due to the boundary conditions we have chosen , @xmath92 for a large family of non - diagonal matrices @xmath2 . this prevents us from employing the theory of sectorial sesquilinear forms in order to find the spectrum . [ t17 ] let @xmath2 be a non - singular matrix . * if @xmath2 is an upper triangular matrix ( that is @xmath93 ) , then @xmath94 * if @xmath2 is not an upper triangular matrix ( that is @xmath95 ) , then @xmath96 * . * since @xmath97 is always an eigenvalue of @xmath6 ( cf . section [ s3 ] ) , then @xmath98 . for @xmath57 , @xmath99 : call @xmath100 then @xmath101 is a convex set and @xmath102 if @xmath103 , @xmath104 so that @xmath105 this and the fact that @xmath101 is closed and convex , yield @xmath106 in order to prove the reverse inclusion , let @xmath44 be such that @xmath107 and let @xmath108 for all @xmath109 , let @xmath110 then @xmath111 , @xmath112 let @xmath113 . by construction @xmath114 and @xmath115 thus by taking @xmath116 , from the fact that @xmath117 and since @xmath118 is convex , we gather @xmath119 : now @xmath95 . let @xmath120 be a fixed non - zero complex number . our aim is to find functions @xmath121 parameterized by @xmath122 , such that @xmath123 and @xmath124 is close to @xmath120 for small @xmath125 . for @xmath126 , let @xmath127 & \mathrm{if } & { \varepsilon}/2 \leq x \leq { \varepsilon}\\ 0 & \mathrm{if } & { \varepsilon}\leq x \leq 1 . \end{array}\right.\ ] ] then , straightforward computations show @xmath128 , @xmath129 , @xmath130 for all @xmath122 small enough , we define the required test function @xmath131 as @xmath132 where @xmath133 is independent of @xmath134 . by construction @xmath121 and @xmath135 according to ( [ e13 ] ) , @xmath136 since @xmath137 as @xmath138 , the above shows @xmath139 as @xmath138 , so that @xmath120 is an accumulation point of @xmath118 . by moving @xmath140 , any complex number is accumulation point of @xmath118 . since @xmath118 is convex , the only possibility for @xmath118 is to be the whole complex plane . @xmath58 in this section we show that the resolvent of @xmath6 is compact for all non - singular @xmath2 . in general it is false that the product of a bounded operator and an operator whose resolvent is compact has compact resolvent , however if we know in addition that the spectrum of the product is not the whole of @xmath12 , then the assertion is true . we first show that the resolvent of @xmath10 is compact by making use of its self - adjointness . since the constant function @xmath141 is in @xmath43 and @xmath142 vanishes , @xmath143 [ t13 ] if @xmath2 is a diagonal matrix , then @xmath144 the zero eigenvalue is always non - degenerate and all the remaining eigenvalues are of multiplicity no greater than 2 . * . * let @xmath145 be as above . for all @xmath146 , let @xmath147 then @xmath148 , @xmath149 and @xmath150 is a complete orthonormal set in @xmath151 . @xmath58 according to corollary [ t19 ] and the above proposition , @xmath152 and @xmath153 since the eigenfunctions @xmath150 form a complete orthonormal set , the resolvent of @xmath10 is compact . let us now rule out the possibility @xmath154 . [ t20 ] for any non - singular @xmath155 , @xmath156 * . * fix the matrix @xmath2 . since @xmath157 the complex number @xmath158 , if and only if @xmath159 let @xmath160 . then the family of operators @xmath161 with domain @xmath43 independent of @xmath162 is a holomorphic family of type ( a ) for all @xmath163 . since 0 is a non - degenerate isolated eigenvalue of @xmath164 and @xmath165 is bounded , there exist an open neighbourhood @xmath166 such that @xmath161 has a non - degenerate isolated eigenvalue , ( denoted by @xmath167 ) close to 0 for all @xmath168 and @xmath167 is a complex valued holomorphic function in @xmath169 ( cf . * th.xii.8 ) ) . if there exists some @xmath170 satisfying @xmath171 , then @xmath172 so that @xmath173 . hence , in order to show that @xmath174 , it is enough to show that @xmath175 . for this we find the first coefficients in the rayleigh - schrdinger series expansion of @xmath176 about @xmath97 . let @xmath177 since @xmath178 , @xmath179 . since @xmath180 and @xmath181 , we compute directly @xmath182 ( cf . * remark 2.2 , p.80 ) ) by @xmath183 if @xmath2 is such that @xmath184 , @xmath185 so that @xmath182 does not vanish and hence @xmath186 . let @xmath2 be such that @xmath187 . then @xmath188 so we compute @xmath189 . let @xmath190 be the eigenfunctions of @xmath10 as in ( [ e2 ] ) so that @xmath191 for all @xmath192 . let @xmath193 for all @xmath146 so that @xmath194 then ( cf . * remark 2.2 , p.80 ) ) @xmath195 we compute each term in the series . since @xmath187 and @xmath2 is invertible , then @xmath196 and @xmath197 do not vanish and @xmath198 hence @xmath199 so that @xmath200 and @xmath201 for @xmath202 integer and @xmath146 . on the other hand @xmath203 so that @xmath204 and @xmath205 for @xmath202 integer and @xmath146 . this yields @xmath206 for @xmath207 . thus @xmath208 so that @xmath209 as we required . @xmath58 [ t1 ] for all @xmath210 , the resolvent @xmath211 is compact . * . * since @xmath10 is non - negative and it has compact resolvent , @xmath212 has a compact inverse . let @xmath213 , then @xmath214 hence @xmath215 so that @xmath216 thus @xmath211 is compact as needed . @xmath58 theorem [ t1 ] shows that the spectrum of @xmath6 consists entirely of isolated eigenvalues of finite multiplicity . since the eigenvalue problem @xmath217 is a constant coefficient system of second order ordinary differential equations , due to the fact that we have a combination dirichlet and neumann boundary condition at both ends of the interval , the multiplicity of each eigenvalue is never greater than 2 . we now investigate the asymptotic behaviour of the resolvent norm of @xmath6 . the results we discuss in this section are connected with the stability of the heat semigroup @xmath218 . they are also relevant from the computational point of view and they are closely related to both local and global stability of the spectrum ( cf . @xcite , @xcite , @xcite and the reference therein ) . the present approach is motivated by analogous reports on non - self - adjoint schrdinger operators in @xcite , @xcite and @xcite . let @xmath219 below and elsewhere we will denote by @xmath220 . according to lemma [ t19 ] , @xmath221 . according to lemma [ t13 ] , @xmath222 each eigenvalue being of multiplicity 1 . we will employ part b ) of the following theorem in the proof of theorem [ t12]-b ) . [ t2 ] assume that there exists a non - singular diagonal matrix @xmath70 such that @xmath223 . then * @xmath6 is similar to a non - negative self - adjoint operator . * @xmath224 is similar to a self - adjoint operator whose numerical range is the whole real line . * . * let @xmath225 so that @xmath226 . since diagonal matrices commute with the boundary conditions , @xmath6 is similar to @xmath227 . for the same reason and since diagonal matrices also commute with @xmath228 , @xmath224 is similar to @xmath229 . by hypothesis , the square root @xmath230 . then @xmath231 where @xmath232 so that @xmath227 is similar to @xmath14 . since @xmath152 , then @xmath233 . analogously @xmath229 is similar to @xmath234 where @xmath235 . since @xmath236 , then @xmath237 furthermore , since @xmath238 and @xmath239 the numerical range of @xmath240 is the whole real line . @xmath58 let @xmath2 be as in the hypothesis . the similarity to a self - adjoint operator ensures the existence of a constant @xmath241 such that @xmath242 and @xmath243 these identities show that although the numerical range of @xmath6 and @xmath224 are in general the whole complex plane , the eigenvalues of these operators are stable in the sense of @xcite . if we assume the weaker condition @xmath244 , we show how to recover part of the above estimate . we start with a preliminary lemma . [ t22 ] let @xmath2 be such that @xmath245 then @xmath246 and there exists @xmath247 independent of @xmath120 , such that @xmath248 * . * let @xmath249 and let @xmath250 . then @xmath251 \\ = & a\left [ 1+(r -a^{-1 } ) z(d - r z)^{-1 } \right](d - r z).\end{aligned}\ ] ] therefore @xmath252 , whenever @xmath253 we show that there is always @xmath249 independent of @xmath120 , such that this holds for all @xmath254 . since @xmath255 and @xmath256 , @xmath257 thus @xmath258 the hypothesis we imposed on @xmath2 is equivalent to saying @xmath259 then @xmath260 for all @xmath44 , @xmath261 hence there exists a constant @xmath262 independent of @xmath263 ( and @xmath120 ) , such that @xmath264 when @xmath263 is large enough . for such an @xmath263 , identity ( [ e4 ] ) holds for any @xmath254 . this shows that @xmath265 must be enclosed in the right hand plane . furthermore @xmath266 so ( [ e26 ] ) is also proven . @xmath58 below and elsewhere we denote by @xmath267 the set of non - singular diagonal matrices and @xmath268 [ t21 ] if there exists @xmath269 such that @xmath270 then @xmath271 and for any small enough @xmath122 there exists @xmath272 independent of @xmath120 , such that @xmath273 * . * let @xmath225 , so that @xmath274 since @xmath70 commutes with the boundary conditions , @xmath6 is similar to @xmath227 and so it is enough to show the theorem for @xmath227 . now , for all @xmath275 @xmath276 so we just have to apply lemma [ t22 ] to @xmath277 . @xmath58 the constant @xmath278 of this theorem is in general strictly greater than @xmath279 , therefore this is weaker than the similar condition for m - sectorial operators in @xcite . if @xmath2 is triangular , the hypothesis of the above theorem does not necessarily hold . for instance if @xmath280 then @xmath281 and so for small @xmath282 the numerical range contains the origin . nonetheless by using a similarity transformation and an approximation argument , we can show positivity of the spectrum whenever both of the eigenvalues of @xmath2 are positive ( @xmath283 in our example ) . the conclusion about the spectrum of the following result will be improved in theorem [ t11 ] . [ t8 ] let @xmath2 be either upper or lower triangular . if @xmath284 , then @xmath285 and for all @xmath122 there exists @xmath272 independent of @xmath120 , such that @xmath286 for all @xmath287 . * . * if @xmath2 is upper triangular the proof is similar so let us assume that @xmath288 let @xmath289 then @xmath6 is similar to @xmath290 for all @xmath291 . put @xmath292 then @xmath293 . the eigenvalues of @xmath294 are @xmath295 thus for small @xmath249 , @xmath296 . the numerical range of @xmath297 is an ellipse with focus at @xmath298 and principal axis in the vertical direction of the order of @xmath263 . by taking @xmath299 , theorem [ t21 ] completes the proof . if @xmath2 is as in the hypothesis of corollary [ t8 ] , there does not exist @xmath269 such that @xmath300 or @xmath301 so theorem [ t2 ] is not applicable . we show that at least in one case @xmath6 fails to be similar to self - adjoint . [ t24 ] let @xmath302 let @xmath122 and @xmath303 . then there exists a constant @xmath304 independent of @xmath263 , such that @xmath305 * . * fix @xmath122 and let @xmath306 . without loss of generality we can assume @xmath307 . throughout the proof the constants @xmath308 are assumed to be positive , possibly depending upon @xmath125 but independent of @xmath263 . in order to show the desired conclusion , it is enough to find @xmath309 and @xmath310 , such that @xmath311 for all large enough @xmath263 . let @xmath312 then @xmath313 hence @xmath314 and @xmath315 we now define the appropriate @xmath316 satisfying ( [ e7 ] ) . let @xmath317 then @xmath318 . let @xmath319 be such that @xmath320 is smooth and * @xmath321 for all @xmath322 $ ] , * @xmath323 for all @xmath324 , * @xmath325 for all @xmath324 . then @xmath326 and @xmath327 hence @xmath328 and @xmath329 for all large enough @xmath263 . if @xmath330 , @xmath331 /(4\pi^2 r^2 ) \\ & = & -{\varepsilon}^2\cos(2\pi r x)/(4\pi^2r^2).\end{aligned}\ ] ] then , a ) and c ) yield @xmath332 } l_6 + l_7/r^4 { { \rm d}}x \nonumber\\ & \leq & l_6r^{-1 } + l_5r^{-4 } + l_7r^{-5 } \label{e11}.\end{aligned}\ ] ] also , @xmath333 then for @xmath330 , @xmath334 and thus c ) yields @xmath335 } |i{\varepsilon}\cos ( 2\pi rx ) - \phi''(x)|^2 { { \rm d}}x \nonumber \\ & \leq & \int_{x\not\in[1/r,1 - 1/r ] } l_8 { { \rm d}}x \nonumber\\ & \leq & l_8r^{-1}. \label{e12}\end{aligned}\ ] ] in order to complete the proof for @xmath336 , notice that ( [ e10 ] ) , ( [ e11 ] ) and ( [ e12 ] ) , show ( [ e7 ] ) . on the other hand , if @xmath337 it is enough to substitute @xmath320 by @xmath338 and repeat the above computations . @xmath58 this result is still valid for @xmath339 indeed , it is enough to put @xmath340 , @xmath341 and carry out similar calculations . since the resolvent norm of self - adjoint operators remains bounded in horizontal lines , the above @xmath6 can not be similar to any self - adjoint operator . let @xmath342 be the set of all non - degenerate real diagonal matrices . if @xmath2 does not satisfy the hypothesis of theorem [ t21 ] ( for instance the numerical range of @xmath2 is an ellipse centered at the origin ) , but @xmath2 is `` close '' in some sense to @xmath342 , an alternative to theorem [ t21 ] can be established . we will employ this result in the proof of theorem [ t11 ] . [ t25 ] let there exist @xmath343 such that @xmath344 let @xmath345 with @xmath346 . then @xmath347 and for any small enough @xmath122 there exist @xmath272 independent of @xmath120 , such that @xmath348 for all @xmath349 . * . * if @xmath350 , @xmath351 so the conclusion is consequence of corollary [ t19 ] . let @xmath352 , let @xmath353 and put @xmath354 . then @xmath355 and according to the hypothesis @xmath356 . let @xmath140 be such that @xmath357 . then @xmath358(cd - z).\end{aligned}\ ] ] since @xmath227 is self - adjoint and by definition @xmath359 , @xmath360 so that @xmath361\ ] ] is invertible . hence @xmath362 and @xmath363^{-1 } \label{e8}\ ] ] for all @xmath364 . this encloses @xmath265 . in order to show the second part , let @xmath365 for small @xmath122 . then there exist a constant @xmath366 independent of @xmath120 , such that @xmath367 also , there exist a constant @xmath368 independent of @xmath120 , such that @xmath369 these two estimates , ( [ e9 ] ) and ( [ e8 ] ) yield @xmath370 this shows that if @xmath371 is a sequence of non - singular matrices and there exists @xmath343 such that @xmath372 as @xmath373 , then @xmath374 [ t23 ] let @xmath2 be either upper or lower triangular . if @xmath375 and @xmath376 , then @xmath377 and for all @xmath122 there exists @xmath272 , such that @xmath286 for all @xmath378 . * . * it is similar to the proof of corollary [ t8 ] . assume without loss of generality that @xmath379 and let @xmath380 then @xmath6 is similar to @xmath290 for all @xmath249 . put @xmath381 then @xmath382 let @xmath383 . according to theorem [ t25 ] , for all @xmath384 @xmath385 by taking @xmath263 small enough , theorem [ t25 ] yields the desired estimate for the resolvent norm . by taking @xmath299 , _ a fortiori _ @xmath386 . in this section we find an entire function whose zeros coincide with @xmath265 . this is made by computing the transcendental function associated with the @xmath3 system of ordinary differential equations associated to @xmath6 via standard ode arguments . let the @xmath3 constant coefficients second order eigenvalue problem @xmath387 we will say that the complex number @xmath162 is an eigenvalue of the system ( [ e5])-([e15 ] ) , if there exist a non - vanishing @xmath388 satisfying ( [ e5 ] ) and the boundary conditions ( [ e15 ] ) . by regularity , @xmath389 is an eigenvalue of @xmath6 , if and only if @xmath162 is an eigenvalue of ( [ e5])-([e15 ] ) . our aim is to find a holomorphic function , denoted by @xmath390 below , whose zeros coincide with the eigenvalues of ( [ e5])-([e15 ] ) . we proceed in the classical manner . let the decomposition in jordan canonical form of @xmath2 be @xmath391 where the jordan matrix @xmath392 is either @xmath393 and @xmath394 then ( [ e5])-([e15 ] ) is equivalent to the @xmath3 system @xmath395 in order to solve ( [ e16])-([e17 ] ) , we reduce it to a first order @xmath396 system as follows . for all @xmath397 , let @xmath398 and let @xmath399 by regarding @xmath400 one sees that ( [ e16])-([e17 ] ) is equivalent to @xmath401 in order to solve ( [ e18])-([e19 ] ) we must find a fundamental system of solutions . let @xmath402 be the standard orthonormal basis of the euclidean space @xmath403 . a straightforward computation show that @xmath404 , \quad j=1,2,3,4\ ] ] is indeed a linearly independent fundamental system for ( [ e18])-([e19 ] ) . hence , @xmath162 is an eigenvalue of this system , if and only if there exist @xmath405 , such that @xmath406 is non - vanishing and satisfies the boundary conditions . we now proceed to compute @xmath390 . the exponential of @xmath407 is given by @xmath408 for @xmath322 $ ] . in theorems [ t3 ] and [ t26 ] below , we split our computation into two cases depending upon the jordan matrix @xmath392 . [ t3 ] when @xmath409 @xmath162 is an eigenvalue of the system ( [ e18])-([e19 ] ) , if and only if @xmath410 for @xmath411 + { } } \\ & & { } - \left [ v_1 ^ 2v_4 ^ 2\frac{\sqrt{a_+ } } { \sqrt{a_-}}+ v_2 ^ 2 v_3 ^ 2 \frac{\sqrt{a_-}}{\sqrt{a_+ } } \right ] \sin \left(\frac { x}{\sqrt{a_+}}\right)\ , \sin \left(\frac { x}{\sqrt{a_- } } \right).\end{aligned}\ ] ] * . * notice that @xmath412 . assume @xmath413 . according to the hypothesis , @xmath414 then @xmath415 let @xmath416 be a particular solution given as in ( [ e20 ] ) , where the complex parameters @xmath417 are to be determined . then @xmath418 and @xmath419 the solution @xmath101 satisfies the boundary conditions ( [ e19 ] ) , if and only if @xmath420 the determinant of this @xmath396 system of linear equations in @xmath417 is precisely @xmath421 . @xmath58 [ t26 ] when @xmath422 @xmath162 is an eigenvalue of the system ( [ e18])-([e19 ] ) , if and only if @xmath410 for @xmath423 * . * notice that @xmath412 . assume @xmath413 . one can verify directly that @xmath424 then the four @xmath425 blocks of the matrix @xmath426 are @xmath427 @xmath428 equal to @xmath429 & \frac{a_+^{1/2}}{{\lambda } } \sin \frac{{\lambda}x}{\sqrt{a_+ } } \end{matr2}\ ] ] and @xmath430 equal to @xmath431 & -\frac{{\lambda}}{a_+^{1/2 } } \sin \frac{{\lambda}x}{\sqrt{a_+ } } \end{matr2}.\ ] ] let @xmath416 be a particular solution given as in ( [ e20 ] ) , where the complex parameters @xmath417 are to be determined . then @xmath418 and @xmath432 where @xmath433 and @xmath434 the solution @xmath101 satisfies the boundary conditions ( [ e19 ] ) , if and only if @xmath435 a rather long but straightforward computation shows that the determinant of this @xmath396 system of linear equations in @xmath417 is @xmath421 . @xmath58 we show that @xmath6 can have non - real eigenvalues even when the spectrum of @xmath2 is positive . [ t4 ] put @xmath436 then the eigenvalues of @xmath2 are @xmath437 , @xmath438 , and the eigenvectors @xmath439 thus @xmath440 so that @xmath421=0 , if and only if @xmath441 hence @xmath442 where @xmath443 . in this section we explore some connections between the entries of the matrix @xmath2 and the global behaviour of @xmath265 when @xmath444 . alongside we discuss conditions to ensure similarity to a self - adjoint operator . for completeness of the picture , below and elsewhere we allow @xmath445 . our first task is to reduce to two parameters the four that are initially given as entries of @xmath2 . this leads us to five different types of matrices to deal with . for @xmath446 , let @xmath447 we show that the @xmath448 generate any @xmath449 via similarity transformations . [ t9 ] if @xmath5 , then @xmath6 is similar to @xmath450 for some @xmath451 and @xmath452 . * . * let @xmath453 if @xmath454 , the proof is trivial . let @xmath455 then , @xmath6 is similar to @xmath290 for all @xmath291 . if @xmath456 , @xmath457 for @xmath458 , @xmath459 and @xmath460 . if @xmath461 , @xmath462 for @xmath463 , @xmath464 and @xmath465 . @xmath58 the case @xmath466 was already described in corollary [ t19 ] . indeed if @xmath467 then @xmath468 is similar to a self - adjoint operator and @xmath469 [ ss6.1 ] since @xmath282 and @xmath471 are real , @xmath472 . let @xmath473 be the eigenvalues of @xmath470 . then @xmath474 so that * @xmath475 , if and only if @xmath476 and @xmath477 , * @xmath478 , if and only if @xmath476 and @xmath479 , * @xmath480 and @xmath481 have opposite signs , if and only if @xmath482 . [ t10 ] the following statements are true . * if @xmath483 then @xmath484 . * if @xmath476 and @xmath477 then @xmath485 is similar to a non - negative operator so that @xmath486 . * if @xmath476 and @xmath479 then @xmath487 is similar to a non - negative self - adjoint operator so that @xmath488 $ ] . * if @xmath482 then @xmath489 . * . * if @xmath483 , the matrix @xmath470 is singular so according to lemma [ t27 ] , @xmath485 is not a closed operator . this shows a ) . statement b ) is consequence of i ) and theorem [ t2 ] , and statement c ) is consequence of ii ) and theorem [ t2 ] . let us show d ) . for @xmath490 , let @xmath491 then @xmath492 and @xmath493 according to theorem [ t21 ] , @xmath494 and @xmath495 since @xmath496 is a holomorphic family of type ( a ) in a neighbourhood of @xmath497 and @xmath498 , _ a fortiori _ @xmath499 although @xmath472 , it is unclear to us whether @xmath485 is similar to self - adjoint in the latter case . [ ss6.2 ] since the results for the matrix @xmath501 are analogous and shown in a similar manner as for @xmath500 , we will only consider the latter . [ t11 ] the following statements are true . * if @xmath502 then @xmath503 . * if @xmath467 then @xmath504 . * if @xmath505 , for all @xmath122 there exists @xmath272 independent of @xmath120 , such that @xmath506 where the symbol @xmath507 is chosen according to the symbol of @xmath282 . * if @xmath376 , then for all @xmath122 there exists @xmath272 independent of @xmath120 , such that @xmath508 for all @xmath509 * if @xmath510 , let @xmath122 and @xmath511 . then there exists a constant @xmath272 independent of @xmath263 , such that @xmath512 for all @xmath513 . * . * if @xmath502 , the matrix @xmath500 is singular so according to lemma [ t27 ] , @xmath514 is not a closed operator . this shows a ) . let us show b ) . if @xmath515 , the matrix @xmath500 is diagonalizable and @xmath516 then , according to theorem [ t3 ] , @xmath517 where @xmath518 is constant in @xmath134 . if @xmath519 , @xmath500 is already in jordan form and according to theorem [ t26 ] , @xmath520 ^2.\ ] ] hence in both cases @xmath521 statements c ) is consequence of corollary [ t8 ] and statement d ) is consequence of corollary [ t23 ] . for statement e ) use theorem [ t24 ] and the fact that @xmath522 is of order @xmath523 . @xmath58 [ ss6.3 ] formally speaking , so far the spectrum of @xmath448 for @xmath525 reproduces the spectrum of @xmath526 in the following sense : if @xmath526 is non - degenerated and both eigenvalues of @xmath526 are positive ( negative ) then @xmath527 is non - negative ( non - positive ) , and if the eigenvalues are of opposite sign then @xmath448 possess both positive and negative spectrum . there is no reason to expect the same for @xmath528 , in fact this case is less simple due to the way the entries of @xmath524 interact with the boundary conditions . the eigenvalues of @xmath524 are given by @xmath529 then * @xmath530 , if and only if @xmath531 . in this case @xmath524 is not a diagonalizable matrix . * @xmath473 are real and have opposite signs , if and only if @xmath532 . * @xmath533 , if and only if @xmath534 , @xmath535 and @xmath536 . * @xmath537 , if and only if @xmath534 , @xmath535 and @xmath538 . * @xmath473 are non - real with @xmath539 , if and only if @xmath540 . * @xmath524 is singular , if and only if @xmath541 . motivated by this and for simplicity , we can divide the plane into @xmath542 disjoint regions @xmath543 , @xmath544 clearly @xmath545 . below we establish the spectral results for @xmath546 separately in each region @xmath543 . ( 300,300)(55,120 ) two cases are similar to what we have found so far . [ t12 ] the following statements are true . * if @xmath547 , then @xmath548 . * if @xmath549 , then @xmath550 and @xmath546 is similar to a self - adjoint operator whose numerical range is the whole real line . * . * if @xmath541 , the matrix @xmath524 is singular so according to lemma [ t27 ] , @xmath546 is not a closed operator . this shows a ) . let us show b ) . let @xmath228 be as in section [ s4 ] . then @xmath551 here @xmath552 and the eigenvalues of @xmath553 are @xmath554 since @xmath532 , @xmath555 are either both positive or both negative . if they are both positive , @xmath556 so that theorem [ t2]-b ) provides the desired conclusion . if they are both negative apply the above argument to @xmath557 . @xmath58 in order to find @xmath558 in @xmath543 for @xmath559 , we ought to rely on properties of the transcendental function @xmath390 . nonetheless , theorem [ t30 ] provides some indication of what we should expect , it bases on the observation that if both @xmath282 and @xmath471 are positive , @xmath560 so by virtue of lemma [ t22 ] , @xmath561 . [ t30 ] if both @xmath282 and @xmath471 are positive , then @xmath562 where @xmath563 for @xmath564 . * . * the numerical range of @xmath524 is an ellipse whose foci are @xmath473 and largest diameter is of length @xmath565 . it is easy to see that @xmath566 is the minimal sector that contains such an ellipse . use theorem [ t21 ] to complete the proof . @xmath58 since @xmath567 and because of diagonal matrices commute with the boundary conditions , @xmath568 where both @xmath282 and @xmath471 are negative . this also shows that the spectral results for @xmath546 are symmetric with respect to the transformation @xmath569 . below we will employ this symmetry often without mention . in order to describe @xmath558 in @xmath570 , we will make use of the following technical result . [ t16 ] let @xmath571 be such that @xmath572 , let @xmath573 and let @xmath574 then @xmath575 has an infinite number of zeros in the complex plane and * if @xmath576 , then @xmath577 , if and only if @xmath578 , * if @xmath579 , then @xmath577 , if and only if @xmath580 , * if @xmath581 , then @xmath575 only has a finite number of zeros lying on the real and imaginary axis . * . * let @xmath582 so that @xmath583 and let @xmath584 for @xmath585 . in order to show a ) , assume @xmath576 . then @xmath586 \cos[\overline{{\alpha } } ( x_1+ix_2 ) ] + \right.\\ & & \qquad \qquad \qquad + \left.\sin[{\alpha}(x_1+ix_2 ) ] \sin[\overline{{\alpha } } ( x_1+ix_2)]\right| ^2 \\ & = & 4 [ \cos^2(\rho x_1 ) - \cosh^2(\rho x_2)]^2.\end{aligned}\ ] ] hence @xmath587 if and only if @xmath588 and @xmath589 . this gives a ) . similarly for b ) , assume @xmath579 . then @xmath586 \cos[\overline{{\alpha } } ( x_1+ix_2 ) ] + \right.\\ & & \qquad \qquad \qquad - \left.\sin[{\alpha}(x_1+ix_2 ) ] \sin[\overline{{\alpha } } ( x_1+ix_2)]\right| ^2 \\ & = & 4 [ \cosh^2(\mu x_1 ) - \cos^2(\mu x_2)]^2.\end{aligned}\ ] ] hence @xmath587 if and only if @xmath590 and @xmath591 . let us show assertion c ) . if @xmath592 , then @xmath593 and @xmath594 hence , if @xmath581 , @xmath595 since @xmath575 is a smooth function , c ) follows . finally let us show that @xmath575 has a infinite number of zeros . suppose that @xmath596 only has a finite number of zeros @xmath597 where the @xmath598 repeats as many times as its order . then @xmath599 is an entire function with no zeros . by virtue of the weierstrass factorization theorem , there is an entire function @xmath600 such that @xmath601 . then @xmath602 { { \rm e}}^{g(x ) } = : p(x ) { { \rm e}}^{g(x ) } .\ ] ] since it is a combination of sines and cosines , the order ( cf . @xcite ) in the sense of entire functions of @xmath575 is @xmath603 . thus by virtue of hadamard s factorization theorem , @xmath600 is a polynomial of degree 1 in @xmath134 and so @xmath604 for suitable @xmath605 . since @xmath606 is a polynomial , this is clearly a contradiction , so @xmath575 must have an infinite number of zeros . @xmath58 [ t31 ] let @xmath607 . * if @xmath608 , then @xmath609 ^ 2 \right\}_{k\in { \mathbb{z } } } \subset ( -\infty,0].\ ] ] * if @xmath610 , then @xmath611 ^ 2 \right\}_{k\in { \mathbb{z } } } \subset [ 0,\infty).\ ] ] * if @xmath612 , then @xmath558 is infinite but it only intersects the real line in a finite number of points . * . * by virtue of v ) , @xmath524 is diagonalizable . we assume @xmath613 , so that @xmath614 let @xmath615 then @xmath616 let @xmath617 and @xmath618 so that @xmath619 . then @xmath620- \left(\frac{{\gamma}_+^2}{{\gamma}_-^2 } \sqrt{\frac{b_+}{b_- } } + \sqrt{\frac{b_-}{b_+}}\right ) \sin({\alpha}x ) \sin(\overline{{\alpha}}x ) \\ & = \frac{2{\gamma}_+}{\overline{{\gamma}_+ } } [ 1-\cos({\alpha}x)\cos(\overline{{\alpha}}x)]- \left(\frac{{\gamma}_+^2}{\overline{{\gamma}_+}^2 } \frac{\overline{{\alpha}}}{{\alpha } } + \frac{{\alpha}}{\overline{{\alpha}}}\right ) \sin({\alpha}x ) \sin(\overline{{\alpha}}x ) \\ & = 2{{\rm e}}^{i2{\vartheta}}[1-\cos({\alpha}x)\cos(\overline{{\alpha}}x)]- \left ( { { \rm e}}^{i4{\vartheta}}\frac{\overline{{\alpha}}}{{\alpha } } + \frac{{\alpha}}{\overline{{\alpha } } } \right ) \sin({\alpha}x ) \sin(\overline{{\alpha}}x ) \\ & = 2{{\rm e}}^{i2{\vartheta}}[1-\cos({\alpha}x)\cos(\overline{{\alpha}}x ) - c \sin({\alpha}x)\sin(\overline{{\alpha}}x ) ] \\ & = 2{{\rm e}}^{i2{\vartheta } } f(x ) , \end{aligned}\ ] ] where @xmath575 and @xmath621 are as in lemma [ t16 ] . let us show a ) . the hypothesis @xmath622 ensures @xmath623 . furthermore @xmath579 , if and only if @xmath624 the latter occurs , if and only if @xmath625 by simplifying this identity , we gather that @xmath579 for @xmath626 which is precisely our assumption . then , lemma [ t16]-b ) and ( [ e14 ] ) complete the proof of a ) . for b ) , notice that since @xmath627 , the constant @xmath197 is now such that @xmath628 and @xmath576 , if and only if @xmath624 therefore a similar argument as for a ) and lemma [ t16]-a ) show this case . in order to prove c ) use the fact that @xmath581 in @xmath629 lemma [ t16]-c ) and ( [ e14 ] ) . @xmath58 [ t29 ] in the regions @xmath630 and @xmath631 , @xmath558 is infinite , and @xmath632 \end{gathered } \qquad \qquad \begin{gathered } ( a , d ) \in r_3 \\ ( a , d ) \in r_4 , \end{gathered}\ ] ] where in both cases the constant @xmath633 only depends upon @xmath634 . * . * we show the result only for @xmath630 . according to iii ) , in this case @xmath635 and @xmath524 is diagonalizable . by expressing the trigonometric functions in exponential form , @xmath636 + \\ & \hspace{2 in } - \frac{k_2+k_1}{4 } \left [ { { \rm e}}^{i({\alpha}-{\beta})x } + { { \rm e}}^{-i({\alpha}-{\beta})x } \right]\end{aligned}\ ] ] where @xmath637 and @xmath638 are constants we do not need to specify here . a similar argument involving hadamard s theorem as in the proof of lemma [ t16 ] shows that @xmath558 is infinite . by putting @xmath639 where @xmath640 , @xmath641 and @xmath642 , @xmath643 + \\ & \hspace{1.5 in } - \frac{k_2+k_1}{4 } \left [ { { \rm e}}^{-{\delta}y}{{\rm e}}^{i { \delta}r } + { { \rm e}}^{{\delta}y}{{\rm e}}^{-i { \delta}r } \right].\end{aligned}\ ] ] since @xmath644 , if we chose @xmath645 , the term @xmath646 dominates the expression and so @xmath647 for a suitable @xmath197 independent of @xmath263 . if we chose @xmath648 , the term @xmath649 is the one that dominates and again @xmath650 is large this shows that all the zeros of @xmath390 must be contained in a band @xmath651 . @xmath58 the above theorem does not rule out the possibility of negative eigenvalues when @xmath376 . we will see in the numerical examples , evidence of points in this region such that @xmath546 has indeed negative spectrum . with regard to finding the minimal @xmath652 . we will see in section [ s7 ] an argument involving chebyshev polynomial that allows us to compute in closed form @xmath558 for a certain dense subset of @xmath630 . we will also illustrate this technique in various examples where the parabolic region is found explicitly . if @xmath653 , the matrix @xmath524 is not diagonalizable and so @xmath390 is given by theorem [ t26 ] instead of theorem [ t3 ] . nevertheless , similar techniques to the ones we have seen so far apply to this case . [ t33 ] let @xmath654 and let @xmath655 ^ 2 \qquad \qquad x\in { \mathbb{c}}.\ ] ] then @xmath575 has an infinite number of zeros in the complex plane but only a finite number of them lie on @xmath656 and on @xmath657 . * . * see the proofs of lemma [ t16 ] and theorem [ t29 ] . @xmath58 [ t32 ] let @xmath653 . if @xmath658 , then @xmath659 . otherwise @xmath558 is infinite but it only intersects the real line in a finite number of points . * . * if @xmath660 , @xmath661 and if @xmath662 , @xmath663 then @xmath664 ^ 2 , \qquad \qquad a - d=\pm 2.\ ] ] the first statement follows from the fact that if @xmath665 , then @xmath666 and so the trigonometric term disappear . the second follows from lemma [ t33 ] . @xmath58 notice that the curve @xmath626 meets the region @xmath667 at @xmath668 . these are the only points where @xmath558 is finite . since all self - adjoint operators with compact resolvent must have an infinite number of eigenvalues , @xmath546 is not similar to self - adjoint . all this suggests that for @xmath634 in a small neighbourhood of these points , @xmath558 must be highly unstable . in the next section we explore closely this idea . as mentioned previously , this section is devoted to investigating some aspects of the global spectral evolution of @xmath6 when we move the entries of the matrix @xmath2 . to be more precise , we consider @xmath669 ( see section [ s6 ] ) and compute @xmath558 as @xmath634 moves along various lines inside @xmath670 . we also introduce a technique that allows us to find explicitly @xmath558 when @xmath634 are in a certain dense subset of @xmath630 by computing the roots of certain polynomial @xmath671 . our first task is to decompose @xmath630 into a disjoint union of curves in order to find the dense subset . for @xmath672 , let @xmath673 and let @xmath674 then @xmath675 the motivation for this decomposition is found by observing that for @xmath676 @xmath677 , where @xmath678 are the eigenvalues of @xmath524 . that is , @xmath679 are level curves of @xmath680 in the @xmath634-plane . notice that @xmath681 the key idea behind finding @xmath671 is that for @xmath682 where @xmath683 , the zeros of the transcendental function are periodic in the horizontal direction . we show how to construct this polynomial . the transcendental function for @xmath546 is @xmath684 + \\ & \hspace{2 in } - k_2 \sin \left(\frac{x}{\sqrt{b_+}}\right ) \sin \left(\frac{x}{\sqrt{b_-}}\right)\end{aligned}\ ] ] where @xmath685 and @xmath686 are two real constants depending upon @xmath282 and @xmath471 which we do not need to specify here . since @xmath687 @xmath688 are rationally related and so the zeros of @xmath390 appear periodically in lines parallel to the real axis . by putting @xmath689 , @xmath690 - k_2\sin ( pz ) \sin ( qz ) \\ & = k_1 + \frac{k_2-k_1}{2 } \cos[(p+q)z ] - \frac{k_2+k_1}{2 } \cos[(p - q)z],\end{aligned}\ ] ] where @xmath691 . standard computations show that , @xmath692 where @xmath693 a polynomial of degree @xmath202 ( the @xmath694 chebyshev polynomial of first order ) . then by letting @xmath695 @xmath696 , if and only if @xmath697 . hence all the zeros of @xmath390 are of the form @xmath698 where @xmath699 is a root of @xmath671 . in this manner , @xmath558 is generated by translations of the roots of @xmath671 . although the above method computes @xmath558 explicitly for @xmath700 , its numerical implementation for large @xmath701 ( @xmath702 in a pc ) is highly unstable due to the well known instability of the roots of polynomials of high degree . nevertheless , no other procedure tried so far , has proven to be more efficient for estimating large eigenvalues in @xmath630 . figures [ f4 ] , [ f7 ] and [ f8 ] below were produced via this approach . [ [ spectral - behaviour - of - a_4-for - ad - close - to--1232 ] ] spectral behaviour of @xmath524 for @xmath634 close to @xmath703 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [ ss7.1 ] by virtue of theorem [ t32 ] , @xmath659 for @xmath704 . in any small neighbourhood of this point , the spectrum of @xmath546 is infinite so high instability is to be expected . since @xmath546 is holomorphic in @xmath282 and @xmath471 , every non - zero eigenvalue of @xmath546 either concentrates at zero or diverges to @xmath705 for @xmath706 . we explore this phenomenon in some detail . according to theorem [ t31]-a ) , if @xmath707 satisfy @xmath626 and @xmath708 , @xmath709 ^ 2\right\}_{k\in { \mathbb{z}}},\ ] ] where @xmath480 as in section [ ss6.3 ] . by taking @xmath710 and @xmath711 , @xmath712 so that @xmath713 . hence , all non - zero eigenvalues of @xmath546 remain negative and escape to @xmath714 as @xmath607 approach the critical point on the curve @xmath626 . in general , not every eigenvalue of @xmath546 need to be in the left hand plane when @xmath634 is close to @xmath703 . in figure [ f4 ] we consider the evolution of the first @xmath715 eigenvalues of @xmath546 for @xmath716 fixed and @xmath542 different values of @xmath471 from @xmath717 to @xmath718 . the awkward choice of @xmath471 correspond to the sensible values of @xmath719 ; each pair @xmath720 for @xmath721 . notice that for large @xmath722 the polynomial @xmath671 has @xmath701 roots and nonetheless all these roots but @xmath97 lie on the same curve . this curve moves away from the origin and there is always a negative eigenvalue . the positive eigenvalues also escape rapidly to @xmath723 and there are infinitely many of them . in figure [ f5 ] we isolate the negative eigenvalue for @xmath716 against 100 different values of @xmath471 close to @xmath724 . this provides indication of how rapidly it escapes to @xmath714 . in order to produce this picture , we made use of the algorithm that matlab provides to find the zero of @xmath390 for @xmath134 on the imaginary axis . comparing with the comment we made earlier in section [ ss6.3 ] , this provides points in @xmath630 such that @xmath546 has a negative eigenvalue of arbitrarily large modulus . [ ss7.2 ] we now explore the transition from real to non - real spectrum by considering the spectral evolution of @xmath524 on the line @xmath725 close to @xmath726 . in figure [ f6 ] we show the first @xmath727 eigenvalues of @xmath546 for @xmath187 and @xmath728 . we produced this graphic by reducing the equation @xmath696 to a single real variable and then making use of the algorithm that maple provides to find zeros of real functions . according to theorem [ t32 ] , we know that @xmath558 is infinite but there is only finite intersection with the real line . as the picture suggests , in this case the origin seems to be the only real eigenvalue . figure [ f7 ] shows the evolution of the first @xmath729 eigenvalues ( counting multiplicity ) of @xmath524 when @xmath187 for various different values of @xmath471 from @xmath730 to @xmath731 . each pair @xmath732 respectively for @xmath733 , @xmath734 , @xmath735 , @xmath736 , @xmath737 , @xmath738 , @xmath739 , @xmath740 . the numerical evidence suggests that for @xmath730 the spectrum is close to the real line and each eigenvalue is of multiplicity @xmath736 . each of these operators has infinitely many real eigenvalues . unfortunately the method we employed to find the roots of @xmath671 , is unable to deal with a finer partition of the @xmath471-interval . nonetheless , the global behaviour of the spectrum can be appreciated , as @xmath471 approaches to @xmath736 , each real eigenvalue eventually splits into two conjugate non - real single eigenvalues stabilizing close to the region in figure [ f6 ] ( see the step @xmath741 ) . notice that there is no spectrum in the left hand plane and compare with theorem [ t30 ] . [ ss7.3 ] another type of peculiar behaviour can be observed as @xmath743 approach the region @xmath742 , where the matrix @xmath524 is singular and @xmath548 . here we concentrate on the point @xmath744 . figure [ f8 ] shows the evolution of the first @xmath745 eigenvalues of @xmath546 ( represented by dots ) as @xmath746 approaches to @xmath747 . alongside we also picture the remaining eigenvalues ( represented by crosses ) that lie on the box @xmath748\times[-300,300]$ ] . a very similar behaviour occurs for @xmath749 as @xmath750 for other values of @xmath719 . it can not be appreciated in the graph provided but there are two conjugate eigenvalues whose real part is negative . these eigenvalues approach to the origin as @xmath751 . all the remaining spectrum concentrates on the real line suggesting that @xmath752 as @xmath751 this is in contrast with the fact that @xmath154 at @xmath753 . here we have chosen @xmath754 and @xmath755 . this means that @xmath671 is only of order @xmath756 and so the spectrum is always generated by @xmath756 points . it is not difficult to show analytically that all three roots converge to @xmath97 and then rigorously prove that @xmath757 . the author wish to thank prof . e. b. davies and prof . r. f. streater for valuable discussions of different aspects of this paper . he also would like to thank dr . a. aslanyan for many helpful comments and suggestions .

we investigate the evolution of string order in a spin-1 chain following a quantum quench . after initializing the chain in the affleck - kennedy - lieb - tasaki state , we analyze in detail how string order evolves as a function of time at different length scales . the hamiltonian after the quench is chosen either to preserve or to suddenly break the symmetry which ensures the presence of string order . depending on which of these two situations arises , string order is either preserved or lost even at infinitesimal times in the thermodynamic limit . the fact that non - local order may be abruptly destroyed , what we call _ string - order melting _ , makes it qualitatively different from typical order parameters in the manner of landau . this situation is thoroughly characterized by means of numerical simulations based on matrix product states algorithms and analytical studies based on a short - time expansion for several simplified models .

1604.02823

within landau theory , phases of matter are identified by their local order parameters , and transitions taking place between different phases are successfully understood by analyzing how order parameters change on crossing the transition points . in this framework , local quantities ( including two - point correlation functions ) fully characterize the phase . nonetheless , phases that can not be detected through local quantities do exist . they are usually referred to as topological phases , both protected and non - protected by symmetries @xcite . the simplest example of non - local order in one dimension is probably the string order ( so ) , first discussed in the context of the haldane phase @xcite , which characterizes the ground state of integer - spin heisenberg chains . so accounts for correlations between spins that are not detected by a simple two - point correlation function @xcite , and reveals the presence of a hidden order parameter , that is related to a non - local symmetry of the hamiltonian @xcite . whereas the resilience of so to local static perturbations has been thoroughly characterized @xcite , much less is known on its robustness to dynamical perturbations . building on the results reported in an earlier work @xcite , in this paper we continue the characterization of the time evolution of so in a specific dynamical protocol : a quantum quench @xcite . here , an initially - prepared many - body state evolves under the action of a time - independent hamiltonian . when the quenched hamiltonian is extensive , as it is in the case considered in this work , one refers to a global quench . the key result upon which this paper is built is the following : so can be abruptly destroyed after an infinitesimal amount of time . we call this phenomenon _ string - order melting _ @xcite . this is in stark contrast with landau local order , e.g. the magnetization of a spin lattice , where the order parameter of the system can not disappear instantaneously , but only after a finite amount of time , because of continuity of its time evolution . the goal of this paper is to elucidate the nature of this peculiar situation with numerical simulations and analytical discussions of specific limits . a number of theoretical works dealt with the problem of studying the dynamics in topological systems after a quantum quench @xcite . a general analysis of the behavior of all the indicators of a symmetry - protected topological phase in the presence of a quantum quench is still lacking . to the best of our knowledge , the sudden disappearance of so , one of the mentioned indicators , has not been observed yet , neither theoretically nor experimentally . our result opens the intriguing perspective that symmetry - protected topological phases might disappear abruptly in the presence of a time - dependent perturbation . this should be the object of further investigations . in this paper , we focus on the dynamics of so in a spin-1 chain after a global quantum quench . we consider the protocol already introduced in ref . initially , the system is prepared in the ground state of the affleck - kennedy - lieb - tasaki ( aklt ) hamiltonian @xcite , known as aklt state , which is the simplest example of a state in the haldane phase , featuring its typical properties . the persistence of so after the quench is related to the symmetries of the hamiltonian ruling the time evolution @xcite . the case in which the symmetries allow the presence of so at finite times has been characterized in ref . @xcite . here , we focus on cases in which the symmetries of the quenched hamiltonian are such that so is either preserved or suddenly destroyed after the global quantum quench , and characterize this situation in detail . our numerical results are based on simulations using matrix product states ( mps ) @xcite . a rich phenomenology can be observed at different length scales . we observe the clear presence of a short - length region , roughly identified as the thermalization region , where correlations induced by the quench have propagated , and of a long - length region , where the physics of melting is occurring . we corroborate these results with a simple analytical model which explicitly illustrates the mechanism of so melting . properly tuned symmetry properties simplify the analytic description , thus allowing a quantitative description of the dynamics of so after the quench . understanding the robustness of topological order to a dynamical perturbation is not the only motivation of this work . the non - equilibrium dynamics of quantum many - body systems has been theoretically studied since several decades @xcite . the impressive experimental progresses in the field of ultra - cold atomic gases ( see , e.g. , @xcite ) spurred a great deal of interest and a consequent renewed theoretical activity @xcite . a prominent portion of this activity on non - equilibrium dynamics has focused on quenches , this special attention being in part motivated by the fact that current experiments can easily implement this protocol . thanks to the amazing progresses in the quantum simulation of effective spin models through one - dimensional chains of trapped ions and cold atoms @xcite , these questions are not a mere academic curiosity , but retain an experimental relevance . in particular , since both string operators @xcite and the short - time dynamics of closed quantum systems are experimentally accessible , it is possible to envision an experimental verification of our results in the near term future . the paper is organized as follows . in sec . [ sec : quantumquenchesandstringorder ] , we review the main properties of the haldane phase and of so in a spin-1 chain . in sec . [ sec : nonlocalorderunderdynamicalperturbations ] , we summarize the recent results on out - of - equilibrium so and report our numerical data on the case of a quantum quench with sudden melting of so . in sec . [ sec : analyticalresults ] , we present an analytical model based on perturbation theory to describe numerical results , and in sec . [ sec : pureakltmodel ] , we discuss a simplified model to test the previous analytical results . conclusions are drawn in sec . [ sec : conclusions ] , while some technical details are provided in the appendixes . throughout the paper , we set @xmath0 , @xmath1 . let us review the basic notions of the haldane phase that will be useful in the analysis of our results @xcite . we consider a one - dimensional spin-1 chain governed by the aklt hamiltonian @xcite @xmath2 , \label{aklthamiltonian}\ ] ] where @xmath3 and @xmath4 are spin-1 operators . the ground state @xmath5 , usually known as aklt state , is exactly known and has a simple mps representation with bond link @xmath6 @xcite . it is the simplest example of a state which belongs to the so - called haldane phase . the presence of non - local order is identified by analyzing the properties of the string operator , defined as @xmath7 \hat s_{k+l}^{\alpha } \ , , \quad ( \alpha = x , y , z ) \ , . \label{definitionofstringorder}\ ] ] the system displays so when : @xmath8 when open boundary conditions ( obc ) are considered , the model defined by @xmath9 exhibits zero - energy spin-@xmath10 edge modes . the presence of these edge modes makes the aklt state four - fold degenerate on an open chain , whereas in the presence of periodic boundary conditions it is non - degenerate . this phenomenology is intimately related to the existence of so . indeed , both these features can be regarded as a consequence of a hidden symmetry breaking related to the @xmath11 group : @xmath12 note that even if @xmath9 is left invariant by the action of @xmath13 , there is no contradiction in the fact that symmetry - breaking might occur in a one - dimensional system at zero - temperature because the group is finite . the aklt open chain can be mapped onto a local ferromagnetic chain with four symmetry - broken magnetic states by means of a non - local transformation ( the kennedy - tasaki transformation ) @xcite . string operators in the aklt chain are mapped onto two - point spin correlators which reveal ferromagnetic order , and thus , in this dual picture , string operators reveal the presence of hidden , long - range antiferromagnetic order , as earlier proposed in ref . so can be considered as the nonlocal order parameter which identifies states in the haldane phase , and its relation to hidden long - range order has also been discussed in different contexts @xcite . because of the special role of the symmetry group @xmath14 , the haldane phase is better referred to as a symmetry - protected topological phase . indeed , it has been shown that a perturbation @xmath15 which is not left invariant by the group @xmath14 destroys so in the ground state . more rigorously , it has been demonstrated that the presence of so is equivalent to the existence of a local symmetry @xcite ; in the case of the aklt state , this means that so is related to the fact that @xmath5 is invariant with respect to the action of @xmath14 . in the absence of such symmetry , so will exponentially vanish on increasing the length of the string @xcite . in this section , we start recapitulating the theory of global quantum quenches applied to so developed in ref . @xcite . we then define the concept of so melting and present some general considerations on it . the quench protocol is defined as follows . consider a system at zero temperature described by the hamiltonian @xmath16 which depends on a parameter @xmath17 . let @xmath18 be the initial value of @xmath17 , so that at the beginning the system is in the ground state of @xmath19 . at time @xmath20 , @xmath17 is abruptly varied from @xmath18 to a new value @xmath21 such that the new hamiltonian is @xmath22 , and thus the system is suddenly driven out of equilibrium . if we denote with @xmath23 the ground state of @xmath24 , the state at time @xmath25 is @xmath26 . as mentioned in the introduction , we consider @xmath27 . the discussion of so at finite times is intimately related to the symmetry properties of @xmath5 and @xmath28 ; we thus recall some notions of group theory . let us begin by stressing that @xmath29 is an abelian group and admits four irreducible ( one - dimensional ) representations : @xmath30 ( the identity ) , @xmath31 , @xmath32 and @xmath33 . from now on , we dub _ even _ an operator @xmath34 which transforms under action of the group according to the representation @xmath30 , i.e. it is left invariant @xmath35 , @xmath36 . an operator @xmath37 is dubbed _ odd _ if , on the contrary , it is a linear combination of operators which transform under the action of the group according to the representations @xmath31 , @xmath32 and @xmath33 . the hamiltonian @xmath38 can be formally decomposed as @xmath39 . the necessary and sufficient condition for the simultaneous existence of so at infinitesimally short times along all the three directions @xmath40 is that @xmath41 it is very important to stress that this result depends on the fact that we explicitly consider a state which displays so at time @xmath20 . if eq . is not satisfied , so is not present for @xmath42 , and thus it disappears abruptly for at least one of the axes @xmath43 , @xmath44 , or @xmath45 . if eq . is fulfilled , the time evolution of so is continuous and so is present not only at infinitesimally short times , but up to a finite time @xmath46 . this result was first discussed in ref . @xcite , to which the interested reader is referred . for completeness , we report the proof of eq . in appendix [ appendixproofofequation ] . the previous relation highlights the role of symmetry : if @xmath47 , the evolved state @xmath48 has the same symmetry of @xmath49 at any time , and so is well defined at finite short times . if @xmath50 , two situations may arise . if the relation is satisfied , the state nevertheless keeps the original symmetry , and so is well defined . however , when eq . is not fulfilled , so is abruptly lost . this is possible because , although the dynamics of every string operator @xmath51 is continuous in time , the limit procedure contained in eq . introduces a singularity in the thermodynamic limit . as we will see , the expectation value of the string operator decays as the string length increases , approaching the zero value in the thermodynamic limit . this is the situation which , in this paper , is referred to as so melting . we now present numerical simulations of a quantum quench where so melting takes place . after initializing a spin-1 one - dimensional chain in the aklt state @xmath5 , we perform a quantum quench with the hamiltonian @xmath52 \,\ , . \label{xxzhamiltonian}\ ] ] when @xmath53 , we recover the xxz hamiltonian @xmath54 which identifies the even part @xmath55 . the spin-1 xxz model is known to be in the haldane phase for @xmath56 @xcite . since it is @xmath57-invariant , no so melting is induced ; this is the case which has been studied in ref . @xcite . in eq . , the odd part of the hamiltonian is given by @xmath58 . the numerical procedure that we employ here consists of two steps : * even if the aklt state has an exact mps representation , for technical reasons we find the mps representation of @xmath5 by means of a variational search in the mps space @xcite using the aklt hamiltonian . we start from a random mps state with initial bond link @xmath59 and iterate the variational procedure sweeping the chain until convergence is reached , i.e. the ground - state energy reaches a constant value . in our simulations we adopt obc . * after the ground state is found , we quench the system with the xxz hamiltonian ( [ xxzhamiltonian ] ) with @xmath60 and compute the time - evolved state @xmath48 by means of time - evolving block decimation ( tebd ) algorithm @xcite . the time evolution is decomposed by using a fourth - order trotter expansion @xcite . finite - temperature simulations are carried out by using the purification method of a mixed quantum state @xcite using a sixth - order trotter expansion . in order to observe so melting , @xmath61 is a necessary condition . the results are shown in fig . [ fig : oldsimulationsxxzodd ] , where we consider a chain of length @xmath62 . we plot the expectation value of the string operator along the @xmath44-axis @xmath63 for @xmath64 . because of the choice of obc , to compute the expectation value of the string operator we have to discard those sites close to the chain ends ; we choose to discard the @xmath65 of the chain from both ends . the results displayed in fig . [ fig : oldsimulationsxxzodd ] , top panels , show that @xmath63 depends exponentially on @xmath66 , the string length . however , there are two typical decay lengths , one which is valid for short length scales , and one which is valid for long length scales . we first address the physics of the system at short length scales . the time dependence of @xmath63 ( fig . [ fig : oldsimulationsxxzodd ] , bottom panels ) shows a steady behavior for short string lengths . we now provide evidence that this short - length behavior is related to the occurrence of thermalization . we show in fig . [ fig : thermalizationofstringorderxxz ] the expectation value of the string operator along the @xmath43 and @xmath44 axes as a function of string length at fixed time @xmath67 . the numerical data are compared with the thermal average @xmath68 $ ] for @xmath69 , where @xmath70 is the corresponding partition function , and @xmath71 is given by hamiltonian ( [ xxzhamiltonian ] ) with @xmath72 and @xmath73 . the inverse temperature @xmath74 is defined , as customary , by the relation @xmath75 $ ] . from our simulations , by matching the energy of the aklt state with that of a thermal state , we find @xmath76 . as shown in fig . [ fig : thermalizationofstringorderxxz ] , the properties of this thermal state describe with very good fidelity the long - time and short - length behavior of our system , thus pointing out the occurrence of thermalization in this model . this result complements the occurrence of thermalization discussed at @xmath77 in ref . @xcite . we now move to the discussion of the system at long length scales . from fig . [ fig : oldsimulationsxxzodd ] we observe that by increasing the strength of @xmath78 , the long - length decay becomes more pronounced , without significantly affecting the thermalized region . this effect is a clear indication of the radically different physical origin of the two distinct exponential decays . the long - length regime is the one associated to the melting of so . in order to better understand the melting process and its interplay with thermalization , we now proceed with a combined analytical and numerical analysis . the problem of describing the expectation value of the string operator as a function of time and of string length can be tackled by means of perturbation theory . by considering the strength of the magnetic field as a small expansion parameter , it is possible to explicitly see the relation between the symmetry properties of the post - quench hamiltonian and so . this is the subject of the following sections . . the system is quenched with hamiltonian ( [ xxzhamiltonian ] ) by using @xmath72 and @xmath73 ( see fig . [ fig : oldsimulationsxxzodd ] , leftmost panels ) . the thermal average of the expectation value of the string operator along the @xmath43 axis as a function of string length ( blue thick line ) with inverse temperature @xmath76 is overlapped to the expectation value of the string operator along the @xmath43 axis as a function of the string length , computed at fixed time @xmath67 ( red dots ) . in the inset we show the same analysis for the expectation value of the string operator along the @xmath44 axis . in both cases we numerically observe thermalization of so at short lengths.,width=325 ] before concluding , let us comment on the fact , displayed in fig . [ fig : thermalizationofstringorderxxz ] , that so along the @xmath43 axis survives the quench . this is not violating the results in sec . [ sec : stringorderandquantumquenches ] , which refer to the symmetry properties of the hamiltonian which are necessary and sufficient to ensure the existence of so along all the three axes @xmath40 . here , so melts along @xmath44 and @xmath45 axes ( not shown ) and thus there is no contradiction . in sec . [ sec : quenchwiththexxzhamiltonianinatransversefield ] we discuss the symmetry properties of a quench hamiltonian in order to let so survive only along some of the three cartesian axes . finally , we stress that when the system is quenched with a magnetic field along the @xmath45 axis , the presence of the magnetic field has no effect on the dynamics of so . this follows from the fact that @xmath79=0 $ ] and @xmath80 . by using @xmath81 and @xmath82 , it is straightforward to see that eq . is fulfilled . a curious question is whether , in the presence of so melting , there exists a _ rotating _ frame of reference in which so is restored . consider the case of a _ rotating string operator _ , defined by @xmath83 where the _ rotating operator _ is @xmath84 . we take an odd hamiltonian given by the sum of only on - site operators , i.e. , @xmath85 , where @xmath86 has support on the @xmath87th site only ; this requirement is crucial in order to preserve the local structure of @xmath88 when acting with @xmath89 . equation becomes @xmath90\hat s^{\alpha}_{k+l}(t ) \,\ , , \label{stringorderrotaingoperatorlocal}\ ] ] where @xmath91 . the rotating so parameter in eq . could be measured in experiments by applying a global spin - rotation pulse before the detection . the introduction of the rotating so operator allows us to show under which conditions the rotating operator @xmath89 restores so after the quench . let us assume that eq . does not hold , namely , that there is so melting . if we consider for instance the situation : ( i ) @xmath92 and ( ii ) @xmath93=0 $ ] , we can show that , at short times , the time - evolved system displays rotating so : @xmath94 in order to show this statement , we observe that eq . is equivalent to requiring that the state @xmath95 is @xmath57-invariant . indeed , an expansion at first order in time yields : @xmath96 |\psi_0\rangle . \label{eq : first_order}\ ] ] it follows that @xmath11 invariance is present only if @xmath97 . this is clearly true in the case @xmath98 corresponding to hypothesis ( i ) . moving to second order , we obtain : @xmath99\right)|\psi_0\rangle+\dots \,\ , .\nonumber\\ \label{movedtimeevolvedstate}\end{aligned}\ ] ] when @xmath93=0 $ ] , which is hypothesis ( ii ) , the state @xmath100 displays so , since @xmath57 symmetry is fully preserved at all orders in @xmath25 . conditions ( i ) and ( ii ) are thus sufficient to ensure rotating so , although they unfortunately select only a limited range of situations . a generalization of such conditions would give a more comprehensive understanding of the phenomenon of so in a time - evolved frame . in this paper , we do not provide a detailed analytical and numerical analysis of this fact , which remains a subject of further study . we conclude by showing that , in the case studied numerically in sec . [ sec : numericalresults ] , the melting of so along the @xmath44 axis is not associated with the existence of so with respect to a rotating spin given by @xmath101 . in this case , @xmath102 and thus the state @xmath103 is @xmath11-invariant at first order in @xmath25 . however , @xmath104\neq0 $ ] , and in particular it can be shown that @xmath104|\psi_{\rm aklt}\rangle\neq0 $ ] . using eq . , one observes that the state @xmath105 is not @xmath11-invariant . this is sufficient to ensure the absence of rotating so at finite times . a more detailed analysis of the onset of so melting can be achieved by considering the dynamics of the string at short times . here it is possible to construct a perturbative expansion that we are going to compare with the numerical simulations . in order to perform a short - time expansion , we use the interaction picture to describe the evolution of the state after the quench . the general description is the following : let us consider the even hamiltonian before the quench @xmath106 , and its even ground state @xmath23 , such that @xmath107 @xmath108 . the hamiltonian which rules the dynamics is decomposed as in sec [ sec : stringorderandquantumquenches ] . since we are going to develop a perturbative treatment , it is convenient to rename @xmath109 , where @xmath110 and @xmath111 is a dimensionless coupling constant . here @xmath112 denotes the operatorial norm of the operator @xmath15 defined by @xmath113 . the time - evolved state is @xmath114 . let @xmath115 be a generic observable , then its time evolution after the quench reads as @xmath116 time evolution can be recast in the interaction picture : [ interactionpicture ] @xmath117 where the subscripts @xmath118 and @xmath119 denote schrdinger and interaction pictures respectively . in the interaction picture , the wavefunction evolution can be recast in terms of the propagator @xmath120 , such that @xmath121 and defined as @xmath122 where @xmath123 is the perturbation in the interaction picture . after eqs . , the wave function in the schrdinger picture reads as @xmath124 . the subscripts @xmath118 and @xmath119 will be henceforth omitted , and since at time @xmath20 ( before quenching the system ) the chain is in the state @xmath23 , the time - evolved state becomes @xmath125 . the time evolution of the operator @xmath115 is then given by @xmath126 where @xmath127 . the time integrals in eq . are evaluated by expanding the operators as @xmath128 where we adopt the notation @xmath129 for the nested commutator of order @xmath130 between @xmath131 and @xmath15 , @xmath132 $ ] being the commutator between @xmath131 and @xmath15 . equation represents the exact time evolution and coupling constant dependence ( embedded in @xmath133 ) of the operator @xmath115 . we consider a perturbation as the sum of @xmath134 on - site perturbations ( which is here referred to as a _ global _ quantum quench ) , @xmath134 being the chain length @xmath135 where @xmath136 acts on the @xmath87th site only . using eq . we are interested in describing the evolution of the expectation value of string operators along the @xmath137 axis , highlighting the dependence on time , coupling constant @xmath111 , and string length : @xmath138 we apply perturbation theory to the model presented in sec . [ sec : numericalresults ] . in what follows we rename by @xmath111 the coupling constant @xmath78 in eq . and identify @xmath81 . first , it should be observed that eqs . and imply that no so melting arises when quenching the system with @xmath139 . since @xmath140=0 $ ] , we have that @xmath141 and @xmath142 . additionally , @xmath143 and thus @xmath144 , from which it follows that @xmath145 . the perturbation , in this case , does not have any effect . this is in agreement with what is discussed by means of generic arguments in ref . @xcite and recalled in sec [ sec : numericalresults ] . in order to gain a better insight for the case in which the condition is violated , we write explicitly the time evolution propagator up to second order in @xmath111 : @xmath146 where the single and double time integrals have been indicated with @xmath147 and @xmath148 , and are respectively given by @xmath149 [ singleanddoubleintegral ] the operator in the right - hand - side of eq . becomes @xmath150\nonumber\\ & & + \epsilon^2\left\{\hat \sigma(t)\hat{\mathcal{o}}^{(\alpha)}_l(t)\hat \sigma(t)\right.\nonumber\\ & & -\left.\hat{\mathcal{o}}^{(\alpha)}_l(t)\hat d(t)-\hat d^\dag(t)\hat{\mathcal{o}}^{(\alpha)}_l(t)\right\ } , \nonumber\\ \label{operatorforexpectationvalueexpansionxxz}\end{aligned}\ ] ] where we introduced @xmath151 . the key result is that the expectation value of the string operator is predicted to vary as @xmath152 the proof of eq . is postponed to the end of this section . the first term in eq . represents the unperturbed time evolution , which is the term responsible for thermalization at long time and short length that we observed in fig . [ fig : oldsimulationsxxzodd ] and in ref . the second term instead represents the leading term in the coupling constant describing the action of the odd perturbation @xmath15 : it is the term which introduces melting . since this is a perturbative expansion , it is not possible to observe the melting phenomenon as a clear scaling @xmath153 . it is , however , possible to distinguish the situations in which @xmath154 depends on @xmath66 ( so melting ) from those in which it does not . due to the nested - commutator structure arising from the expansion , a necessary condition for so to melt is that the odd perturbation must not commute with some element of @xmath29 @xcite , i.e. @xmath155\neq0\label{microscopicconditionstringordermelting } \,\ , , \ ] ] for some @xmath137 . in this case , the expectation value in eq . depends on @xmath66 and describes the onset of melting along the @xmath137 axis . the relation between so melting and the symmetry properties of the hamiltonian was already clarified by eq . . here , a compatible necessary condition for so melting is obtained from a different perspective . however , eq . provides a novel insight : in order to observe so melting , the quench has to extend over the whole chain as in eq . , i.e. the quench has to be global . this is necessary in order to ensure the mentioned @xmath66 dependence of @xmath156 . the necessary condition in eq . also clarifies the fact that if @xmath15 belongs to the @xmath157 representation of @xmath29 ( @xmath40 ) , then so will not melt along the @xmath137 axis ; indeed , in this case , the necessary condition is not valid . this result is consistent with the numerical results reported in sec . [ sec : numericalresults ] , where we observed that a magnetic field along @xmath43 did not lead to so melting along the @xmath43 axis . in eq . , we keep only the second - order term in @xmath111 and neglect higher orders since the quadratic scaling in the coupling constant has been numerically verified , which tells us that the term of order @xmath158 is indeed the leading term at small coupling constant . we show the @xmath111 dependence in fig . [ fig : quadraticscalingepsilonxxz ] ( left ) . the numerical simulations reported in fig . [ fig : quadraticscalingepsilonxxz ] ( right ) show that for small @xmath111 , the dependence of @xmath159 on @xmath66 is linear . this is consistent with the fact that the term of order @xmath158 in eq . actually catches the lowest - order dependence on @xmath66 of so melting . at this stage , however , the generic structure of the term of order @xmath158 in eq . makes the problem of extracting the explicit dependence on @xmath66 quite difficult to be tackled . we can overcome this problem by considering small times , since in this limit we can explicitly use the expansion as in eqs . however , since @xmath160 for all @xmath137 , we see from eqs . that we have to expand at least to third order in time , which makes the explicit computation of the expectation value in eq . tedious . the proof of eq . follows below . the reader uninterested in the technicalities can go directly to sec . [ sec : pureakltmodel ] . . since the time evolution is due to @xmath161 , which is even , any even operator evolved in time by @xmath161 remains in the @xmath30 representation of @xmath29 at any time . the nested commutators , which appear in eqs . , belong to the same @xmath29-representation of @xmath15 . the same is true for @xmath147 and @xmath162 $ ] ( the string operator is even ) . on the contrary , @xmath148 is even by construction . we conclude that , for any odd perturbation @xmath15 , the terms in eq . proportional to an odd power of @xmath111 do not belong to the @xmath30 representation . the previous discussion implies that their expectation value over the aklt state , which belongs to the @xmath30 representation of @xmath14 , vanishes . the application of an odd operator on @xmath23 yields a state which is not in the @xmath30 representation . this state is orthogonal to @xmath5 by symmetry . for the same reason , one can generalize this statement to any odd power of @xmath111 , since in the term of order @xmath163 , with @xmath164 , the perturbation @xmath15 appears @xmath165 times . @xmath166 in order to get a deeper understanding of the melting dynamics , we now present a simplified model for which we are able to describe the behavior of so in time , string length and coupling constant . as in sec . [ sec : stringorderandquantumquenches ] , we initialize the system in the aklt state @xmath49 , but now we perform a quench with the hamiltonian @xmath167 , where @xmath15 is an odd perturbation which violates eq . . we are interested in the short time and small coupling constant regime ; note that this does not prevent the possibility of observing the physics related to the melting phenomenon , because it is expected to occur at infinitesimal times . since the even part of the hamiltonian is not changed by the quench , the unperturbed evolution of @xmath168 is trivial , and to second order in time , following eq . , the expectation value of the string operator varies as @xmath169\right]\right\rangle \,\ , , \label{stringorderintimeaklt}\ ] ] where @xmath170 . with this form of perturbation , the dependence on @xmath66 , embedded in @xmath88 , can be explicitly extracted . by using the expression of the string operator and of the perturbation , one can expand the operator in the expectation value appearing in eq . and separate terms which give a dependence on the string length ( @xmath171 ) from those which do not ( @xmath172 ) : @xmath173\right]=\hat f_0+\hat f_l \,\ , . \label{operatorexpectationvalueexpansion}\ ] ] it is convenient to use perturbations @xmath174 , where @xmath175 , from the orthonormal set as in table [ tab : tablebasisoperators ] . by applying the scheme described before , the following analytical results have been found : if we quench the system with @xmath176 , then so melting occurs along the @xmath177 axes following the law @xmath178 \label{expressionstringordermelting}\,\ , , \ ] ] where @xmath170 , @xmath179 , and @xmath180 . conversely , so along the @xmath137 axis does not melt and it evolves according to @xmath181 where @xmath182 ( thus , it is independent of @xmath66 ) . the proof of eqs . and is postponed to the end of this section . away from the perturbative regime an exponential decay of the expectation value of the string operator is observed ( see fig . [ fig : datalengthandtimeakltgx_xy_long ] ) . the linear dependence on @xmath66 found in eq . , valid in the perturbative regime , is thus the lowest order in @xmath66 describing so melting . .elements of the basis @xmath183 . odd perturbations can be chosen to be @xmath184 ( @xmath185 ) . [ cols="<,^,^",options="header " , ] we show in fig . [ fig : datalengthandtimeakltgx_xy ] the results of a simulation with perturbation @xmath186 at @xmath187 . in the top panels , we display the expectation value of the string operator along @xmath43 , which is seen to evolve as a function of time and string length according to eq . , whereas the expectation value of the string operator along @xmath45 behaves as so along @xmath44 ( not shown ) . in the bottom panels , we plot the expectation value of the string operator as in the top panels but along the @xmath44 axis . for small time and string length , the expectation value of the string operator along @xmath44 is numerically seen to evolve following eq . . in fig . [ fig : datalengthandtimeakltgx_xy_long ] , the same simulations as in fig . [ fig : datalengthandtimeakltgx_xy ] , up to @xmath188 , are presented . even if the expectation value of the string operator along the @xmath43 axis is independent of @xmath66 in the bulk of the chain ( no so melting ) , a string - length dependence arises for longer times at short string length ( fig . [ fig : datalengthandtimeakltgx_xy_long ] , top panels ) as also observed in fig . [ fig : oldsimulationsxxzodd ] , for the model described in sec . [ sec : stringorderandquantumquenches ] . in the present case , our simulations are , however , still too short to reveal the occurrence of a thermalized region . this is due to the exponential increase of the von neumann entropy in time , which is in turn related to the bond link dimension of the mps representation @xcite . notice that these results also apply to the model described in sec . [ sec : nonlocalorderunderdynamicalperturbations ] and sec . [ sec : analyticalresults ] if one uses the @xmath189 as odd perturbations and expands for small times , since the evolution of the expectation value of the string operator will follow eq . also in this case , but with the only difference that the zero - order term will not be constant and equal to @xmath190 but will rather vary in time because of the change of the even part of the hamiltonian , as in eq . . this , of course , does not affect the dependence on @xmath66 and @xmath111 in the thermodynamic limit ; the predictions given by eqs . and also hold in this case for so melting described by the @xmath158 dependence . this fact has been confirmed by numerical simulations . a similar comparison can be done for the @xmath111 dependence . we show in fig . [ fig : dataepsilonakltgx ] numerical data of the expectation value of the string operator along the @xmath43 , @xmath44 , and @xmath45 axes when we quench with @xmath186 as a function of the coupling constant @xmath111 . we set the chain length @xmath62 and compute the expectation value of the string operator at fixed time @xmath191 and , to avoid border effects , we choose @xmath192 since we are interested in bulk values of the string operator . the quadratic behavior in @xmath111 predicted in eqs . and is numerically confirmed , as is evident from the picture . we also checked that the curvature , obtained by a quadratic fit of the data , perfectly agrees with the analytical estimates of eqs . and ( see the caption to fig . [ fig : dataepsilonakltgx ] ) . axis ( red crosses ) , along the @xmath44 axis ( blue circles ) , and along the @xmath45 axis ( magenta dots , over the blue circles ) as a function of the coupling constant @xmath111 for @xmath186 perturbation . in each case data are fitted by a parabola , confirming the quadratic scaling predicted in eqs . ( [ expressionstringordermelting ] ) and ( [ expressionstringordernomelting ] ) . data are taken at fixed time @xmath191 and string length @xmath192 with @xmath62 . numerical values of the curvature extracted by the quadratic fits are : @xmath193 ( @xmath43 axis ) , @xmath194 ( @xmath44 axis ) and @xmath195 ( @xmath45 axis ) . these values can be compared to those predicted by eqs . and which are : @xmath196 ( @xmath43-axis ) and @xmath197 ( @xmath44 and @xmath45 axes ) . analytical predictions are confirmed.,width=325 ] the same results predicted by eqs . ( [ expressionstringordermelting ] ) and ( [ expressionstringordernomelting ] ) are found if we quench with @xmath198 or @xmath199 , with the difference that so melting does not occur along the @xmath44 and @xmath45 axes , respectively . this is a consequence of the special choice of perturbations that we made . from our data , we see that we are always in the regime such that @xmath200\right]\right\rangle\lesssim\mathcal{o}_0 $ ] , which is consistent with the fact that we are developing a perturbative expansion . . the terms arising from the expansion of the operator in eq . which yield a space - independent contribution are @xmath201\right]\left(\prod_{n = k+1}^{k+l-1}e^{i\pi \hat s^{\alpha}_n}\right)\hat s^{\alpha}_{k+l}+\label{constantterm0 } \nonumber\\ & & + 2\left[\hat v_k,\hat s^{\alpha}_k\right]\left(\prod_{n = k+1}^{k+l-1}e^{i\pi \hat s^{\alpha}_n}\right)\left[\hat v_{k+l},\hat s^{\alpha}_{k+l}\right]+\label{constantterm1 } \nonumber\\ & & + \hat s^{\alpha}_k\left(\prod_{n = k+1}^{k+l-1}e^{i\pi \hat s^{\alpha}_n}\right)\left[\hat v_{k+l},\left[\hat v_{k+l},\hat s^{\alpha}_{k+l}\right]\right ] \,\ , .\label{constantterm2}\end{aligned}\ ] ] the terms yielding a space dependence are @xmath202\left(\sum_{i = k+1}^{k+l-1}\left[\hat v_i , e^{i\pi \hat s^{\alpha}_i}\right]\prod_{n\neq i}e^{i\pi \hat s^{(\alpha)}_n}\right)\hat s^{\alpha}_{k+l}+\label{stringterm0}\nonumber\\ & & + 2\hat s^{\alpha}_k\left(\sum_{i = k+1}^{k+l-1}\left[\hat v_i , e^{i\pi \hat s^{\alpha}_i}\right]\prod_{n\neq i}e^{i\pi \hat s^{\alpha}_n}\right)\left[\hat v_{k+l},\hat s^{\alpha}_{k+l}\right]+\label{stringterm1 } \nonumber\\ & & + \hat s^{\alpha}_k\left(\sum_{i = k+1}^{k+l-1}\left[\hat v_i,\left[\hat v_i , e^{i\pi \hat s^{\alpha}_i}\right]\right]\prod_{n\neq i}e^{i\pi \hat s^{\alpha}_n}\right)\hat s^{\alpha}_{k+l}+\nonumber\\ & & + \hat s^{\alpha}_k\left(\sum_{i , j = k+1\,:\,i\neq j}^{k+l-1}\left[\hat v_i , e^{i\pi \hat s^{\alpha}_i}\right]\times\right.\nonumber\\ & & \hspace{1.5cm}\times\left.\left[\hat v_j , e^{i\pi \hat s^{\alpha}_j}\right]\,\prod_{n\neq i , j}e^{i\pi \hat s^{\alpha}_n}\right)\hat s^{\alpha}_{k+l}\label{stringterm3 } \,\ , .\end{aligned}\ ] ] the space dependence arises only from terms where there is at least one sum : space dependence is a consequence of the _ nonzero _ commutation between the perturbation and the element of @xmath29 , as in eq . . if @xmath203=0 $ ] ( which is true if @xmath204 belongs to the @xmath205 representation of the @xmath29 ) , the expression in eq . is zero . no space dependence arises and so along the @xmath137 axis is preserved . we can exploit the fact the aklt state has a simple mps representation identified by the the set of @xmath206 real matrices @xmath207 to compute all expectation values of the operators @xmath208 using the formalism of the _ transfer matrices _ @xcite ( see appendix [ appendixcomputationofexpectationvalues ] ) . to do this , it is convenient to introduce a set of operators forming an orthonormal basis ( hilbert - schmidt metric ) in the space of @xmath209 complex matrices , and furthermore belong to a specific representation of @xmath29 . a possible orthonormal basis can be @xmath210 , where the subscript denotes the symmetry representation . the operators are shown in table [ tab : tablebasisoperators ] . any operator in the @xmath211 representation of @xmath29 can be expressed as a linear combination of operators in the same representation . even perturbations can be taken to be @xmath212 for @xmath213 , and odd perturbations can be @xmath214 for @xmath215 to ensure they are hermitian . equations and are found using the method described in appendix [ appendixcomputationofexpectationvalues ] and the scheme here explained with perturbations as in table [ tab : tablebasisoperators ] . in this paper , we have presented an extensive study of so melting , namely of the sudden disappearance of so after a quantum quench . we have highlighted the fact that non - local order can be particularly fragile in a dynamical context , in the presence of a perturbation which breaks the symmetry to which so is related . this is qualitatively different from what would happen in the presence of a local order parameter in the manner of landau , where the continuity of the evolution rules out the disappearance of order in an infinitesimal time we analyzed this phenomenon by considering a quantum quench in a spin-1 chain initialized in the aklt state . we focused on the coupling constant , time and string - length evolution of the expectation value of the string operator after the quench , and related the presence of so after the quench to the symmetries of the post - quench hamiltonian . in our simulations , we observed a first exponential decay of the expectation value of the string operator as a function of the string length , which we identify as the thermalization region . this region can be followed either by a plateau ( when the system displays so ) or by a second exponential decay . this second regime is what it is associated to so melting . a short - time expansion allowed to predict the behavior of so just after the quench at short times and distances . at longer times , where thermalization occurs , non - perturbative effects come into play and we resorted only on the numerical results . the features that we studied in our simulations ( e.g. , light - cone effects ) could also be observed when quenching from non - topological states and measuring string correlations ( see , e.g. , the non - topological mott insulating phase of the bose - hubbard model @xcite or the fermionic correlations in the ising chain @xcite ) . the melting phenomenon is related to the non - locality of the observable , more than the topology of the state . however , in the aklt chain , so is related to the presence of topological order , and thus this out - of - equilibrium analysis can be linked to the survival or disappearance of topological properties in the system . so is only one of the features which characterize the properties of one - dimensional symmetry - protected phases of matter @xcite . as long as the properties of the ground state are concerned , it is known that different indicators , such as so , the presence of edge modes or the existence of degeneracies in the entanglement spectrum , may behave differently @xcite . we leave as an open intriguing perspective the characterization of the behavior of such indicators in an out - of - equilibrium context , as well as the extension of this discussion to higher dimensions . generally speaking , it would be interesting to assess whether symmetry - protected topological phases can disappear abruptly after a quantum quench . r.f . was supported by eu ( ip - siqs and strep - quic ) , and by scuola normale superiore ( progetto interno non - equilibrium dynamics of one - dimensional quantum systems " ) . r.f . also acknowledges the oxford martin school for support . l.m . was supported by labex ens - icfp : anr-10-labx-0010/anr-10-idex-0001 - 02 psl*. l.m . and d.r . acknowledge support by the italian miur ( firb project rbfr12nlna ) . we consider the state @xmath216 and the hamiltonian @xmath217 . we now prove that the time - evolved state @xmath218 displays so as defined in eq . at infinitesimally short times if and only if eq . is verified . for better readability , we report eq . here below : @xmath219 _ proof that so at short time implies eq . . we assume that there is a time @xmath25 such that the state @xmath220 displays so . since a necessary condition for a state @xmath221 to possess so is to be @xmath57-invariant , i.e. @xmath222 , for all @xmath223 , the time - evolved state satisfies @xmath224 @xmath36 . we now investigate whether the state @xmath48 can be @xmath57 symmetric in the presence of an odd hamiltonian contribution ( @xmath225 ) . we consider a small - time expansion : at first order in @xmath25 one has @xmath226|\psi_0\rangle$ ] . thus , @xmath48 keeps @xmath57 symmetry at first order in @xmath25 if @xmath227 . one can generalize this by induction to any order in @xmath25 : a full series expansion of @xmath228 yields that the state @xmath48 keeps @xmath57 symmetry if and only if @xmath229 for all @xmath230 , which is indeed eq . . since so for @xmath48 implies @xmath57 invariance , it also implies eq . . note that this proof is valid also for @xmath25 infinitesimal . _ proof that eq . implies so at short time . _ we prove that eq . implies that @xmath231 is continuous as a function of @xmath25 . note that at @xmath20 this expectation value is different from zero . this implies that there is a finite time range where so is present . at first order in @xmath25 one has @xmath232|\psi_0\rangle \,\ , , \label{eq : firsttimeexpansionstringorderforproof}\ ] ] where @xmath233 is discarded because of eq . . we assume @xmath55 to be short range , i.e. @xmath234 , where @xmath235 acts on @xmath236 neighboring spins , from @xmath87 to @xmath237 . the commutator @xmath238 $ ] acts at most on @xmath239 neighboring spins centered around @xmath240 and @xmath241 . this follows by explicit inspection : recalling eq . , we get @xmath242=\sum_j\big[\hat h_{e , j},\hat s_k^{\alpha } \left(\ , \prod_{n = k+1}^{k+l-1 } e^{i\pi \hat{s}_n^{\alpha } } \right ) \hat s_{k+l}^{\alpha}\big ] \,\ , . \label{eq : commutator}\ ] ] the commutator in eq . can be expanded into a _ finite _ number of terms , around @xmath243 and @xmath244 , since @xmath245 is even and commutes with the string , i.e. @xmath246=0 \,\ , . \label{eq : evenhamioltoniancommutaror}\ ] ] because of eq . , the number of terms arising from the expansion of eq . is upper bounded by @xmath239 also in the limit @xmath247 . moreover , the operator norm of each of these operators is bounded . it follows that the expectation value of the operator in eq . is bounded in the limit @xmath247 ; the first - order term in eq . is thus finite in the thermodynamic limit . a higher - order expansion in time of @xmath248 yields a series of nested commutators between @xmath249 and @xmath88 @xmath250 where @xmath251 denotes the nested commutator of order @xmath252 between @xmath249 and @xmath88 , as in eq . . using eq . one can see that , in the expansion , only terms involving @xmath55 are nonzero , and by means of eq . one can prove that the term of order @xmath253 contributes with a finite number of terms in the expansion , for all @xmath254 , in the limit @xmath247 . thus , the term of order @xmath130 in eq . is finite in the limit @xmath247 , for all @xmath254 , proving that so is well defined , at least at short times . this proves that eq . implies that the state @xmath48 displays so at short times . note that this proof is valid also for @xmath25 infinitesimal . expectation values are computed thanks to the formalism of transfer matrices . we write the aklt state as @xcite @xmath255|i_1\dots i_l\rangle$ ] and define the transfer matrix for a generic on - site operator @xmath256 as @xmath257}_{\alpha_n\alpha_{n+1}}a^{[j_n]}_{\beta_n\beta_{n+1}}\theta_{j_ni_n } \,\ , , \ ] ] where @xmath258 . the expectation value on the aklt state of a generic operator @xmath259 of the form @xmath260 is readily computed by means of transfer matrices as @xmath261 \,\ , .\end{aligned}\ ] ] in the limit @xmath262 the expectation value becomes @xmath263 where the fact that @xmath264 is hermitian and with unitary spectral radius was used , @xmath265 being the eigenvector of @xmath264 with unitary eigenvalue . similarly , the matrices @xmath266 are seen to be hermitian and with unitary spectral radius for all @xmath137 . they can be then written in the diagonal form @xmath267 where @xmath268 and @xmath269 for @xmath270 . by defining the coefficient @xmath271 one can write @xmath272 in our case , @xmath273 is independent of @xmath274 and tends to @xmath275 when @xmath247 since @xmath276 and @xmath277 are finite . the same reasoning applies to the other nonlocal operators . by using this fact , one can see that the first three terms in eq . yield a result which scales linearly with @xmath66 . the only term which in principle can give a scaling @xmath278 is the last term in eq . ( [ stringterm0 ] ) . for a specific form of perturbation [ which means that the @xmath275 coefficient in ( [ generalqexpression ] ) has to be computable for any form of @xmath279 , @xmath280 and @xmath281 a full , quantitative prediction of the evolution of the expectation value of string operator can be given . c. nayak , s. h. simon , a. stern , m. freedman , and s. das sarma , rev . * 80 * , 1083 ( 2008 ) . wen , isrn cond . phys . * 2013 * , 198710 ( 2013 ) . f. d. m. haldane , phys . a * 93 * , 464 ( 1983 ) . f. d. m. haldane , phys . lett . * 50 * , 1153 ( 1983 ) . m. den nijs and k. rommelse , phys . b * 40 * , 4709 ( 1989 ) . f. pollmann , a. m. turner , e. berg , and m. oshikawa , phys . b * 81 * , 064439 ( 2010 ) . f. pollmann , e. berg , a. m. turner , and m. oshikawa , phys . b * 85 * , 075125 ( 2012 ) . t. kennedy and h. tasaki , phys . b * 45 * , 304 ( 1992 ) . l. mazza , d. rossini , m. endres , and r. fazio , phys . b * 90 * , 020301(r ) ( 2014 ) . a. polkovnikov , k. sengupta , a. silva , and m. vengalattore , rev . phys . * 83 * , 863 ( 2011 ) . d. i. tsomokos , a. hamma , w. zhang , s. haas , and r. fazio , phys . a * 80 * , 060302(r ) ( 2009 ) . g. b. halsz and a. hamma , phys . * 110 * , 170605 ( 2013 ) . y. zeng , a. hamma , h. fan , arxiv:1509.08613v4 ( 2015 ) . g. kells , d. sen , j. k. slingerland , and s. vishveshwara , phys . b * 89 * , 235130 ( 2014 ) . a. chandran , f. j. burnell , v. khemani , s. l. sondhi , j. phys . : condens . matter * 25 * , 404214 ( 2013 ) . m. d. caio , n. r. cooper , and m. j. bhaseen , phys . lett . * 115 * , 236403 ( 2015 ) . l. dalessio and m. rigol , nat . commun . * 6 * , 8336 ( 2015 ) . a. russomanno and e. g. dalla torre , arxiv:1510.08866 ( 2015 ) . l. privitera and g. e. santoro , phys . b * 93 * , 241406(r ) ( 2016 ) . i. affleck , t. kennedy , e. h. lieb , and h. tasaki , phys . lett . * 59 * , 799 ( 1987 ) . i. affleck , t. kennedy , e. h. lieb , and h. tasaki , commun . phys . * 115 * , 477 ( 1988 ) . f. verstraete , j. j. garca - ripoll , and j. i. cirac , phys . 93 * , 207204 ( 2004 ) . u. schollwck , ann . phys . * 326 * , 96 ( 2011 ) . t. niemeijer , physica * 36 * , 377 ( 1967 ) . e. barouch and m. dresden , phys . * 23 * , 114 ( 1969 ) ; e. barouch , b. m. mccoy , and m. dresden , phys . rev . a * 2 * , 1075 ( 1970 ) ; e. barouch and b. m. mccoy , phys . rev . a * 3 * , 2137 ( 1971 ) . m. greiner , o. mandel , t. esslinger , t. w. hnsch , and i. bloch , nature * 415 * , 39 ( 2002 ) . t. stferle , h. moritz , c. schori , m. khl , and t. esslinger , phys . lett . * 92 * , 130403 ( 2004 ) . c. d. fertig , k. m. ohara , j. h. huckans , s. l. rolston , w. d. phillips , and j. v. porto , phys . lett . * 94 * , 120403 ( 2005 ) . t. kinoshita , t. wenger , and d s. weiss , nature ( london ) * 440 * , 900 ( 2006 ) . m. gring , m. kuhnert , t. langen , t. kitagawa , b. rauer , m. schreitl , i. mazets , d. a. smith , e. demler , and j. schmiedmayer , science * 337 * , 1318 ( 2012 ) . j. eisert , m. friesdorf , and c. gogolin , nat . phys . * 11 * , 124 ( 2015 ) . s. trotzky , y - a . chen , a. flesch , i. p. mcculloch , u. schollwck , j. eisert , and i. bloch , nat . * 8 * , 325 ( 2012 ) . m. cheneau , p. barmettler , d. poletti , m. endres , p. schauss , t. fukuhara , c , gross , i. bloch , c. kollath , and s. kuhr , nature ( london ) * 481 * , 484 ( 2012 ) . c. senko , p. richerme , j. smith , a. lee , i. cohen , a. retzker , and c. monroe , phys . x * 5 * , 021026 ( 2015 ) . i. cohen , p. richerme , z .- x . gong , c. monroe , and a. retzker , phys . a * 92 * , 012334 ( 2015 ) . m. endres , m. cheneau , t. fukuhara , c. weitenberg , p. schauss , c. gross , l. mazza , m. c. bauls , l. pollet , i. bloch , and s. kuhr , science * 334 * , 200 ( 2011 ) . m. endres , m. cheneau , t. fukuhara , c. weitenberg , p. schau , c. gross , l. mazza , m. c. bauls , l. pollet , i. bloch , and s. kuhr , appl . phys . b * 113 * , 27 - 39 ( 2013 ) . a. klmper , a. schadschneider , and j. zittartz , europhys . * 24 * , 293 ( 1993 ) . e. g. dalla torre , e. berg , and e. altman , phys . * 97 * , 260401 ( 2006 ) . e. berg , e. g. dalla torre , t. giamarchi , and e. altman , phys . b * 77 * , 245119 ( 2008 ) . k. hasebe and k. totsuka , phys . b * 84 * , 104426 ( 2011 ) . r. w. chhajlany , p. r. grzybowski , j. stasiska , m. lewenstein , and o. dutta , phys . lett . * 116 * , 225303 ( 2016 ) . d. perz - garca , m. m. wolf , m. sanz , f. verstraete , and j. i. cirac , phys . lett . * 100 * , 167202 ( 2008 ) . c. d. e. boschi and f. ortolani , eur . . j. b * 41 * , 503 ( 2004 ) . h. ueda , h. nakano , and k. kusakabe , phys . b * 78 * , 224402 ( 2008 ) . g. vidal , phys . * 91 * , 147902 ( 2003 ) . g. vidal , phys . rev . lett . * 93 * , 040502 ( 2004 ) . m. suzuki , prog . phys . * 56 * , 1454 ( 1976 ) . m. suzuki , j. math . phys . * 32 * , 400 ( 1991 ) . recall that , by definition of so [ eq . ] and by means of eqs . , and , the dependence on @xmath66 is embedded in the string operator . p. calabrese , f. h. l. essler , and m. fagotti , phys . * 106 * , 227203 ( 2011 ) .

let be a connected , linear algebraic group defined over , acting regularly on a finite dimensional vector space over with-structure . assume that posseses a zariski - dense orbit , so that becomes a prehomogeneous vector space over . we consider the left regular representation of the group of-rational points on the banach space of continuous functions on vanishing at infinity , and study the convolution operators , where is a rapidly decreasing function on the identity component of . denote the complement of the dense orbit by , and put . it turns out that the restriction of to is a smooth operator . furthermore , if is reductive , and and are irreducible hypersurfaces , corresponds , on each connected component of , to a totally characteristic pseudodifferential operator . we then investigate the restriction of the schwartz kernel of to the diagonal . it defines a distribution on given by some power of a relative invariant of and , as a consequence of the fundamental theorem of prehomogeneous vector spaces , its extension to , and the complex-plane , satisfies functional equations . a trace of can then be defined by subtracting the singular contributions of the poles of the meromorphic extension .

math0402139

let @xmath0 be a connected , linear algebraic group defined over @xmath1 , @xmath2 a @xmath18-dimensional vector space over @xmath3 with @xmath1-structure @xmath4 , and @xmath19 a @xmath1-rational representation of @xmath0 on @xmath2 with a zariski - dense @xmath0-orbit . the triple @xmath20 then represents a prehomogeneous vector space , and we denote the dual prehomogeneous vector space by @xmath21 . in case that @xmath0 is reductive , and the singular set @xmath11 is an irreducible hypersurface , there exists an irreducible , homogeneous polynomial @xmath22 such that @xmath23 , and @xmath24 . the singular set @xmath25 of the dual prehomogeneous vector space is then also an irreducible hypersurface given as the set of zeros of a homogeneous , irreducible polynomial @xmath26 satisfying @xmath27 . the rational functions @xmath22 and @xmath26 are called relative invariants . set @xmath28 , @xmath29 , denote by @xmath7 the group of @xmath1-rational points of @xmath0 , and let @xmath30 be the connected component containing the unit element . @xmath13 and @xmath31 decompose into the same number of connected components denoted by @xmath32 , respectively @xmath33 , @xmath34 , each of them being a @xmath35-orbit . the fundamental theorem of prehomogeneous vector spaces @xcite , regarding the fourier transform of a complex power of a relative invariant , then states that for rapidly decreasing functions @xmath36 , @xmath37 on @xmath4 , respectively @xmath38 , the integrals @xmath39 which converge for @xmath40 , can be extended analytically to meromorphic functions on the whole complex @xmath17-plane , and satisfy the functional equations @xmath41 where @xmath42 is given by a product of @xmath43-functions , and the @xmath44 are entire functions . originally , the theory of prehomogeneous vector spaces developed from an attempt to construct dirichlet series satisfying functional equations in a systematic way . since every smooth affine algebraic variety @xmath45 which carries the action of a reductive algebraic group @xmath35 , and posesses an open orbit , is isomorphic to a homogeneous vector bundle @xmath46 , where @xmath47 is a reductive subgroup of @xmath35 , and @xmath48 is a locally homogeneous @xmath47-vector bundle , the investigation of such @xmath35-varieties can be reduced to the study of locally transitive linear actions of connected reductive groups . in classical representation theory , for every admissible irreducible representation @xmath49 of @xmath35 on some hilbert space @xmath47 , the continuous linear operators @xmath50 , defined by @xmath51 , where @xmath10 is some test function on @xmath35 , and @xmath52 denotes haar measure on @xmath35 , are of trace class , and the global character of @xmath53 is defined in terms of this trace as a distribution on @xmath35 . in this paper , we will consider the left regular representation @xmath6 of the group @xmath7 on the banach space @xmath8 of continuous functions on @xmath4 vanishing at infinity , and our main goal will consist in giving a description of the corresponding operators @xmath54 , where @xmath10 is a rapidly decreasing function on @xmath35 . since we will work in a @xmath55 framework , avoiding hilbert space theory entirely , we will rely on the theory of pseudodifferential operators in order to formulate our results . it turns out that , for arbitrary prehomogeneous vector spaces , the restriction of @xmath9 to @xmath56 is a pseudodifferential operator with smooth kernel . in case that @xmath0 is reductive , and both @xmath11 and @xmath14 are irreducible hypersurfaces , each of the components @xmath32 is a @xmath55 manifold with boundary @xmath57 , the latter being smooth outside its intersection with the set of non - regular points @xmath58 of @xmath14 . as we will show , the restriction of @xmath9 to @xmath59 is a totally characteristic pseudodifferential operator lying in the class @xmath60 . these operators are locally given as oscillatory integrals @xmath61 where @xmath62 , and @xmath63 is a lacunary symbol . here @xmath64 are the standard coordinates in @xmath65 . the kernels of such operators are no longer smooth , and become singulare along @xmath66 . in the case of a smooth operator acting on a @xmath55 manifold , a trace can be defined in a natural way by restricting its kernel to the diagonal , which yields a density on the underlying manifold . if the manifold is compact , it can be integrated , and the so defined trace coincides with the usual @xmath67-trace . in our situation , the restriction of the schwartz kernel of @xmath9 to the diagonal defines a distribution on @xmath13 given by expressions of the form for some critical exponent @xmath17 , and its extension to @xmath68 , and the complex @xmath17-plane , satisfies functional equations according to . a trace of @xmath9 can then be defined by subtracting the singular contributions of the poles of the meromorphic extension . as an application , we consider the holomorphic semigroup @xmath69 of bounded linear operators on @xmath8 generated by a strongly elliptic differential operator associated with the representation @xmath6 . the latter is a differential operator of euler type , and the corresponding semigroup can be characterized by a convolution semigroup of complex measures , which are absolutely continuous with respect to haar measure . denoting the corresponding radon - nikodym derivative by @xmath70 , one has @xmath71 , and since @xmath72 is analytic in @xmath73 and @xmath74 , as well as rapidly decreasing , we can apply the above considerations . in particular , we get explicit expressions for the schwartz kernel of the operators @xmath75 , and their restrictions to the diagonal . let @xmath76 be a connected , linear algebraic group defined over @xmath1 , and @xmath77 the group of @xmath1-rational points of @xmath0 . we will assume that @xmath0 is symmetric , so that @xmath7 becomes a real reductive algebraic group in the sense of @xcite , page 42 . let @xmath78 be the lie algebra of @xmath7 , @xmath79 a maximal compact subgroup of @xmath7 with lie algebra @xmath80 , and @xmath81 the corresponding cartan decomposition of @xmath78 . the mapping @xmath82 represents an analytic diffeomorphism of @xmath83 onto @xmath7 , while @xmath84 is a surjection . let @xmath35 be the connected component of @xmath7 containing the unit element @xmath85 . clearly , @xmath35 is a lie group , and its lie algebra coincides with @xmath78 . choose a left invariant riemannian metric on @xmath7 , and denote the distance of two points @xmath86 by @xmath87 . we set @xmath88 . then @xmath89 in the following , we will say that a function @xmath10 on @xmath35 is _ at most of exponential growth _ , if there exists a @xmath90 such that @xmath91 for some constant @xmath92 . if @xmath93 is the cartan decomposition of an arbitrary element in @xmath35 , one computes @xmath94 where @xmath95 . putting @xmath96 , we obtain @xmath97 let @xmath98 be the dimension of @xmath7 , @xmath99 be a basis of @xmath80 and @xmath100 a basis of @xmath101 . since @xmath102 is an analytic diffeomorphism , there exists a @xmath103 , such that @xmath104 where @xmath105 is the length of @xmath106 . realizing @xmath107 as power series for matrices , relations and then imply that the matrix coefficients of @xmath108 and , consequently , of @xmath93 , are at most of exponential growth . let @xmath109 respectively @xmath110 be the left respectively right regular representation of @xmath35 on the frchet space @xmath111 of smooth , complex valued functions on @xmath35 , equipped with the usual topology of uniform convergence on compact subsets of a function and each of its derivatives , see @xcite , page 220 . let @xmath112 be the universal enveloping algebra of the complexification of @xmath78 , and denote the representations of @xmath112 on the space @xmath113 of differentiable elements by @xmath114 , respectively @xmath115 . let @xmath116 denote left invariant haar measure on @xmath35 . we introduce now the space of rapidly decreasing @xmath55 functions on @xmath35 . [ def : a ] the _ space of rapidly decreasing @xmath55 functions on @xmath35 _ , in the following denoted by @xmath117 , is given by all functions @xmath118 satisfying the following conditions : \(i ) for every @xmath119 , and @xmath120 , there exists a constant @xmath121 such that @xmath122 \(ii ) for every @xmath119 , and @xmath120 , one has @xmath123 . note that @xmath117 is @xmath124- and @xmath125-invariant . [ rem : a ] let @xmath126 . by the relation @xmath127 , @xmath128 , and with respect to the basis of @xmath78 introduced above , one calculates @xmath129=-dl({\mathrm{ad}\,}(g ) x_i)f(g)\\&=-\sum \limits _ { j=1}^d a_{ij}(g)dl(x_j)f(g),\end{aligned}\ ] ] where the matrix coefficients @xmath130 of the adjoint representation , and their derivatives , are at most of exponential growth . thus , for arbitrary @xmath120 and @xmath119 , @xmath131 satisfies the conditions @xmath132 and @xmath133 of the preceeding definition . in the sequel , the following partial integration formulas will be needed . let @xmath134 be the modular function of @xmath35 . [ prop : a ] let @xmath135 , and assume that @xmath136 , together with all its derivatives , is at most of exponential growth . then , for arbitrary multiindices @xmath137 , @xmath138 @xmath139 where @xmath140 , @xmath141 , and @xmath142 . first , note that @xmath143 is a differentiable function with respect to @xmath144 , and integrable on @xmath35 for all @xmath145 , since @xmath146 for some @xmath147 , and @xmath148 . furthermore , @xmath149 so that , for @xmath150 , @xmath151 where @xmath152 . thus , for @xmath153 , @xmath154 can be estimated from above by @xmath155 , @xmath156 being some constant . by the bounded convergence theorem of lebesgue , and the left invariance of haar measure , we therefore obtain @xmath157 d_g(g)\\ = \lim_{h \to 0 } h^{-1 } \int _ { g } f_1(g ) [ f_2({\,{\mathrm e}^{-hx_i}\,}g)-f_2(g ) ] d_{g}(g)=\lim_{h \to 0 } h^{-1 } \int _ { g } [ f_1({\,{\mathrm e}^{hx_i}\,}g)-f_1(g)]f_2(g ) d_{g}(g),\end{gathered}\ ] ] compare @xcite , pages 146 and 364 . since @xmath158 and @xmath159 , @xmath153 can be estimated in a similar way , a repeated application of lebesgue s theorem finally yields for @xmath160 . similarly , by taking into account remark [ rem : a ] , we deduce @xmath161 d_{g}(g)\\ = \lim_{h \to 0 } h^{-1 } \int _ { g } [ f_1(g{\,{\mathrm e}^{-hx_i}\,})\delta({\,{\mathrm e}^{-hx_i}\,})^{-1}-f_1(g)]f_2(g ) d_{g}(g),\end{gathered}\ ] ] obtaining for @xmath160 . the general formulas then follow by induction . in what follows , we will denote the coordinate functions in @xmath162 by @xmath163 . with the identification @xmath164 , and with respect to a basis @xmath165 of @xmath78 , the canonical coordinates of second type of a point @xmath166 are given by @xmath167 where @xmath168 denotes a sufficiently small neighbourhood of @xmath169 in @xmath170 , and @xmath171 . we will write for @xmath172 simply @xmath173 . as a real analytic submanifold of @xmath174 , @xmath7 is , in particular , embedded in @xmath175 , so that the matrix @xmath176 has rang @xmath98 for all @xmath177 and @xmath178 . for each tuple @xmath179 of indices we then define the sets @xmath180 in this way getting a finite covering of @xmath7 by open , although not necessarily connected , sets . let @xmath2 be a @xmath18-dimensional vector space over @xmath3 with @xmath1-structure @xmath4 , @xmath181 a @xmath1-rational representation of @xmath0 on @xmath2 , and suppose that @xmath182 has a dense @xmath0-orbit in the zariski topology . then @xmath5 is called a _ prehomogeneous vector space over @xmath1_. the complement set of the unique dense orbit , the _ singular set _ , is a zariski closed set , and will be denoted by @xmath11 . since @xmath183 , the restriction of @xmath182 to @xmath7 induces a regular @xmath7-action on @xmath4 . assume now that @xmath0 is reductive , and that the singular set @xmath11 is an irreducible hypersurface . then , there is an irreducible , homogeneous polynomial @xmath22 such that @xmath184 . it is called a _ relative invariant _ , and there exists a rational character @xmath185 , such that @xmath186 in this case , @xmath5 is a regular prehomogeneous vector space , and one has @xmath187 , @xmath188 , where @xmath189 . by multiplying @xmath22 with a scalar , we can assume that @xmath190 , and @xmath191 . let @xmath192 . as before , denote by @xmath35 the connected component of @xmath7 containing the unit element . then , by a theorem of whitney , @xmath13 decomposes into a finite number of connected components , each of them being a @xmath35-orbit , and we write @xmath193 . in what follows , we will identify @xmath2 with @xmath194 , and @xmath4 with @xmath195 , by choosing a basis in @xmath4 , and assume that @xmath196 without loss of generality . instead of @xmath197 we will then simply write @xmath198 . in particular , @xmath11 and @xmath14 become irreducible affine algebraic varieties in @xmath194 , respectively @xmath195 . their sets of regular points , @xmath199 , respectively @xmath200 can be provided with differentiable structures , the underlying topology being the induced one . let @xmath201 denote the dual prehomogeneous vector space of @xmath5 , and @xmath26 its relative invariant satisfying @xmath202 . let @xmath203 , respectively @xmath204 , denote the schwartz space of functions on @xmath4 , respectively @xmath38 , and let @xmath205 be the fourier transform of @xmath206 . let @xmath207 be lebesgue measure on @xmath4 . the following result is known as the fundamental theorem of prehomogeneous vector spaces , and was proved by sato in 1961 @xcite . [ thm : a ] let @xmath206 and @xmath208 . then the integrals @xmath209 converge for @xmath40 , and can be extended analytically to holomorphic functions on the whole @xmath17-plane , satisfying the functional equations @xmath210 where @xmath42 is given by a product of @xmath43-functions , and the @xmath211 are entire functions which do not depend on @xmath36 . the functions @xmath212 , @xmath213 are called _ this section is devoted to the exposition of some of the basic facts about pseudodifferential operators , needed to formulate our main results in the sequel . our main references for the theory will be @xcite and @xcite . consider first an open set @xmath215 in @xmath195 , and let @xmath216 be the standard coordinates . for any real number @xmath217 , we denote by @xmath218 the class of all functions @xmath219 such that , for any multiindices @xmath220 , and any compact set @xmath221 , there exist constants @xmath222 for which @xmath223 stands for @xmath224 , and @xmath225 . we further put @xmath226 . note that , in general , the constants @xmath222 also depend on @xmath227 . for any such @xmath227 one then defines the continuous linear operator @xmath228 by the formula @xmath229 where @xmath230 denotes the fourier transform of @xmath231 , and @xmath232 . an operator @xmath233 of this form is called a _ pseudodifferential operator _ , and we denote the class of all such operators for which @xmath234 by @xmath235 . the set @xmath236 consists of all operators with smooth kernel . such operators are called _ smooth operators_. by inserting in the definition of @xmath230 , we obtain for @xmath237 the expression @xmath238 which exists as an oscillatory integral , see @xcite . the schwartz kernel @xmath239 of @xmath233 is therefore given by the oscillatory integral @xmath240 and is a smooth function outside the diagonal in @xmath241 . consider next an @xmath18-dimensional @xmath55 manifold @xmath242 , and let @xmath243 be an atlas for @xmath242 . then a linear operator @xmath244 is called a _ pseudodifferential operator on @xmath242 of order @xmath217 _ if , for every chart diffeomorphism @xmath245 , the operator @xmath246 \circ \kappa_{\gamma}^{-1}$ ] given by the diagramm c^_c(u^ ) & ^a_|u^ & & c^(u^ ) + _ ^ & & & ( 2,1 ) _ ^ + c^_c(u^ ) & ^a^ & & c^(u^ ) + is a pseudodifferential operator on @xmath247 of order @xmath217 , and we write @xmath248 . note that , since the @xmath249 are not necessarily connected , we can choose them in such a way that @xmath250 is covered by the open sets @xmath251 . now , in general , if @xmath242 and @xmath252 are two smooth manifolds , and @xmath253 is a continuous linear operator , where @xmath254 and @xmath255 is the density bundle on @xmath252 , its schwartz kernel is given by the distribution section @xmath256 , where @xmath257 . observe that @xmath258 . in case that @xmath259 and @xmath260 , @xmath233 is given locally by the operators @xmath261 , which can be written in the form @xmath262 where @xmath263 , @xmath264 , and @xmath265 . the kernel of @xmath233 is then determined by the kernels @xmath266 . for @xmath267 , they are continuous and given by absolutely convergent integrals . in this case , their restrictions to the respective diagonals in @xmath268 define continuous functions @xmath269 which , for @xmath270 , satisfy the relations @xmath271 , thus defining a density @xmath272 on @xmath273 . we introduce now a special class of pseudodifferential operators associated in a natural way to a @xmath55 manifold @xmath242 with boundary @xmath274 . our main reference will be @xcite in this case . let @xmath275 be the space of functions on @xmath242 which are @xmath55 up to the boundary , and @xmath276 the subspace of functions vanishing to all orders on @xmath274 . the standard spaces of distributions over @xmath242 are @xmath277 the first being the space of extendible distributions , whereas the second is the space of distributions supported by @xmath242 . consider now the translated partial fourier transform of a symbol @xmath278 , @xmath279 which is @xmath55 away from @xmath280 . here we will say that @xmath227 is _ lacunary _ if it satisfies the _ lacunary condition _ @xmath282 the subspace of lacunary symbols will be denoted by @xmath283 . let @xmath284 be the standard manifold with boundary with the natural coordinates @xmath64 . in order to define on @xmath285 operators written formally as oscillatory integrals @xmath286 where @xmath287 is a more general amplitude and @xmath288 is lacunary , one rewrites the formal adjoint of @xmath233 by making a singular coordinate change . thus , for @xmath289 , one considers @xmath290 putting @xmath291 , @xmath292 one obtains @xmath293 for @xmath294 , the succesive integrals in converge absolutely and uniformly , thus defining a continuous bilinear form @xmath295 which by @xcite , propositions 3.6 and 3.9 . extends to a separately continuous form @xmath296 if @xmath297 , one then defines the operator @xmath298 written formally as , as the adjoint of @xmath299 . in this way , the oscillatory integral is identified with a separately continuous bilinear mapping @xmath300 the space @xmath301 of _ totally characteristic pseudodifferential operators on @xmath285 of order @xmath217 _ consists of those continuous linear maps such that @xmath302 is of the form with @xmath303 whenever @xmath304 . similarly , a continuous linear map on a smooth manifold @xmath242 with boundary @xmath274 is an element of the space @xmath305 of totally characteristic pseudodifferential operators on @xmath242 if , for a given atlas @xmath306 , the operators @xmath307 \circ \kappa^{-1}_\gamma$ ] are elements of @xmath301 , where @xmath249 are coordinate patches isomorphic to subsets in @xmath285 . let @xmath5 be a prehomogeneous vector space defined over @xmath1 , and identify @xmath2 with @xmath194 , and @xmath4 with @xmath195 , by choosing a basis in @xmath4 . consider the banach space @xmath8 of continuous , complex valued functions on @xmath4 vanishing at infinity , equipped with the supremum norm . let @xmath308 be the corresponding , continuous , left regular representation of @xmath7 , which was already introduced in @xcite . the representation of @xmath112 on the space of differentiable vectors @xmath309 will be denoted by @xmath310 . we will also consider the left regular representation of @xmath7 on @xmath311 which , equipped with the topology of uniform convergence on compact subsets , becomes a frchet space . this representation will be denoted by @xmath6 as well . as before , we write @xmath35 for the connected component of @xmath7 containing the unit element , and @xmath52 for haar measure on @xmath35 . let @xmath126 , and denote by @xmath9 the continuous linear operator @xmath312 in @xmath8 . its restriction to @xmath313 induces a continuous linear operator @xmath314 to see this , let @xmath315 be a sequence which converges to zero in @xmath313 , that is , assume that there exists a compact set @xmath316 such that @xmath317 , and @xmath318 for all @xmath319 and arbitrary multiindices @xmath320 . then , for @xmath321 , one has @xmath322 since @xmath323 , and @xmath324 by assumption . as a consequence , and according to schwartz , there exists a distribution @xmath325 such that @xmath326 for all @xmath327 . the properties of the schwartz kernel @xmath328 will depend on the analytic properties of @xmath10 , and the orbit structure of the underlying @xmath35-action , and our main effort in the following sections will be directed towards the elucidation of the structure of @xmath328 . denote the coordinates of a point in @xmath4 , respectively @xmath38 , by @xmath329 , respectively @xmath330 , and put @xmath331 clearly , @xmath332 . since @xmath333 , we can define the function @xmath334 @xmath335 is continuous in @xmath336 and @xmath337 , and one verifies @xmath338 . since @xmath339 is , together with all its derivatives , at most of exponential growth in @xmath74 , we can interchange the order of integration and differentiation in , yielding @xmath340 . the main issue of this section will consist in proving the following structure theorem . [ thm:1 ] let @xmath341 be a prehomogeneous vector space over @xmath1 , and @xmath342 the left regular representation of @xmath7 on @xmath4 . then , for @xmath126 , @xmath343 is a fourier integral operator given by @xmath344 with symbol @xmath345 on @xmath346 , @xmath9 is a pseudodifferential operator of class @xmath347 . for the proof , we will need some lemmas . first , note that on @xmath13 , one can express the canonical vector fields @xmath348 of @xmath4 locally as linear combinations in the fundamental vector fields of the underlying @xmath7-action . [ lem:1 ] let @xmath349 . then there exists a neighbourhood @xmath350 of @xmath351 , and rational functions @xmath352 in @xmath336 on @xmath215 , such that , for arbitrary @xmath353 , @xmath354 for some indices @xmath355 . for @xmath356 , one has @xmath357 because of the local transitivity of the @xmath7-action on @xmath358 , the @xmath359-matrix @xmath360 has maximal rang @xmath18 for @xmath361 , where , by assumption , @xmath362 . therefore , there exists a neighbourhood @xmath215 of @xmath351 , and indices @xmath355 , such that for all @xmath363 @xmath364 denoting the matrix coefficients of the corresponding inverse matrix by @xmath352 , the assertion follows with . consider the covering of @xmath7 by the sets @xmath365 introduced in , and put @xmath366 . [ prop:1 ] fix @xmath349 . then there exists a neighbourhood @xmath350 of @xmath351 , such that , for arbitrary multiindices @xmath367 , @xmath353 , and @xmath368 , @xmath369 the coefficient functions @xmath370 are rational expressions in the coordinates of @xmath336 and satisfy the estimates @xmath371 . by the previous lemma , there exists a neighbourhood @xmath215 of @xmath351 such that , for @xmath363 and all @xmath368 , @xmath372 now , @xmath373 where the functions @xmath374 are given by the equations @xmath375 , and are real analytic in @xmath144 and @xmath74 , compare @xcite . differentiating these equations with respect to @xmath144 one obtains @xmath376 let @xmath377 be a finite partition of unity subordinate to the covering @xmath378 of @xmath35 , that is , assume that @xmath379 , @xmath380 , and @xmath381 for each @xmath382 . note that we are allowing the @xmath383 to have non - compact support . for @xmath384 , @xmath385 becomes invertible . since @xmath386 , we therefore obtain , as a consequence of the last equation , the relation @xmath387 the coefficients of @xmath74 and the functions @xmath383 are , together with their derivatives , at most of exponential growth , so that equation implies that the derivates of the functions @xmath388 with respect to @xmath144 satisfy the estimates @xmath389 for some @xmath92 and @xmath390 . thus we obtain the assertion of the proposition for @xmath391 , and @xmath392 . assume that the assertion holds for @xmath393 . differentiating with respect to @xmath394 then yields the assertion for @xmath395 . assume @xmath349 , and let @xmath215 be a neighbourhood of @xmath351 as in the preceeding proposition . then , for arbitrary multiindices @xmath320 , @xmath396 , and @xmath368 , the relations @xmath397 hold , where the coefficients @xmath398 are rational functions in the coordinates of @xmath336 satisfying the estimates @xmath399 . the key step in proving the corollary will be to express @xmath400 as a linear combination of derivatives of @xmath401 with respect to @xmath336 . since @xmath402 , one computes @xmath403 and repeated differentiation leads to @xmath404 therefore , by first taking into account that implies @xmath405 and then applying , we obtain for @xmath400 the expression @xmath406 where the @xmath407 are rational functions in the matrix coefficients of @xmath74 . the corollary now follows , since by the previous proposition , one computes @xmath408 where the functions @xmath409 are rational expressions in the cordinate functions of @xmath336 satisfying the bounds . we are now in position to prove the structure theorem . fix @xmath410 , and assume that @xmath350 is a neighbourhood of @xmath351 as in proposition [ prop:1 ] . then , by proposition [ prop:1 ] and its corollary , one has @xmath411 d_g(g)\\ & = \frac 1 { ( 1+|\xi|^2)^n } \int_g f(g)\sum \limits_{|\gamma|\leq 2n}\sum _ { \delta_1+\delta_2=\beta } \frac{\beta!}{\delta_1!\delta_2!}\operatorname{\partial}_m^{\delta_1}d^\gamma_{\alpha , n}(m , g)\\ & \cdot \sum_{|{\varepsilon}|\leq|\delta_2| } b^{\varepsilon}_{\delta_2}(m , g ) \pi(g)d\pi(x^{\varepsilon}x^\gamma){\varphi}_\xi(m ) d_g(g ) , \qquad m \in u.\end{aligned}\ ] ] next , for arbitrary @xmath353 and @xmath120 , @xmath412 where we set @xmath413 . indeed , @xmath414=\lim _ { h \to 0 } h^{-1 } [ { \varphi}_m(g{\,{\mathrm e}^{hx_i}\ , } ) - { \varphi}_m(g)],\end{aligned}\ ] ] so is correct for @xmath415 . assuming that the assertion holds for @xmath416 with @xmath417 , we obtain for arbitrary @xmath418 @xmath419\\&=\lim _ { h \to 0 } h^{-1 } [ dr(x^\gamma){\varphi}_m(g{\,{\mathrm e}^{hx_i}\ , } ) - dr(x^\gamma ) { \varphi}_m(g)]=dr(x_ix^\gamma){\varphi}_m(g),\end{aligned}\ ] ] so that , by induction , we get for arbitrary @xmath120 . now , integrating by parts according to yields @xmath420 where we set @xmath421 and @xmath422 . note that , for fixed @xmath423 , @xmath424 , as a consequence of the estimates and , so that integration by parts is legitimate . let @xmath425 be a compact set in @xmath4 . for each point @xmath426 , let @xmath427 be a neighbourhood of @xmath351 as in proposition [ prop:1 ] . by heine - borel , @xmath425 can be covered by finitely many neighbourhoods @xmath428 , so that @xmath429 has a finite subcovering . fix @xmath430 , and let @xmath431 be an integer @xmath432 . assume that @xmath433 are arbitrary multiindices . setting @xmath434 , and taking into account the above expression for @xmath435 , one computes @xmath436 where @xmath437 . thus , @xmath438 for all @xmath439 , yielding @xmath440 . now , for @xmath441 , fourier transformation gives @xmath442 where the occuring integrals are absolutely convergent , so that we can interchange the order of integration . this proves theorem [ thm:1 ] . let @xmath5 be a prehomogeneous vector space with singular set @xmath11 . in this section , we restrict our attention to the case where @xmath12 coincides with the set of fixed points of the @xmath7-action . if @xmath7 is reductive , zero constitutes the only closed orbit contained in @xmath14 , and hence the only fixed point . this will be the case we will be mainly concerned with . if @xmath443 is a fixed point , the symbol of @xmath9 at @xmath336 is simply given by @xmath444 . thus , if @xmath14 coincides with the set of fixed points @xmath445 of the underlying @xmath7-action , one has @xmath446 in this situation , theorem [ thm:1 ] can be restated as follows . [ thm:2 ] let @xmath447 be a prehomogeneous vector space defined over @xmath1 , and assume that @xmath14 is equal to the set @xmath445 of fixed points of the underlying @xmath7-action . then @xmath448 is given by the family of oscillatory integrals @xmath449 where @xmath450 for @xmath451 , one has @xmath452 . its schwartz kernel @xmath453 is given by the continuous family of oscillatory integrals @xmath454 furthermore , one has @xmath455 , so that @xmath9 , as a mapping from @xmath313 to @xmath456 , represents a fourier integral operator . in particular , @xmath457 . [ ex:1 ] consider the simplest prehomogeneous vector space , @xmath458 , where @xmath459 acts by multiplication on @xmath3 . then @xmath460 , and @xmath461 is a relative invariant corresponding to @xmath462 , @xmath463 . one has @xmath464 with @xmath465 , @xmath466 . the fundamental theorem yields in this case the relations @xmath467 see @xcite , proposition 4.21 . let @xmath468 be a positive definite real symmetric matrix of degree @xmath469 , and consider the positive definite quadratic forms @xmath470 , @xmath471 on @xmath194 . we define @xmath472 . setting @xmath473 , @xmath474 , where @xmath475 , @xmath476 , one obtains actions of @xmath477 on @xmath478 . again , @xmath479 , and by the fundamental theorem one obtains @xmath480 where @xmath481 , see @xcite , proposition 4.22 . returning to the situation of the previous theorem , note that on @xmath482 , the schwartz kernel @xmath328 of @xmath9 is given by the function @xmath483 while , as distributions , @xmath484 as @xmath485 . in this way , the orbit structure of the underlying @xmath7-action is reflected in the singular behaviour of the schwartz kernel of @xmath486 at @xmath445 . in what follows , we will assume that @xmath479 . in order to get a better understanding of the schwartz kernel @xmath328 of @xmath9 , and , in particular , of its restriction to the diagonal , we define the auxiliary symbol @xmath487 by introducing the inverse fourier transform @xmath488 of @xmath489 , and since @xmath490 , @xmath491 , we obtain the relation @xmath492 [ lem:2 ] @xmath493 . by the structure theorem , @xmath494 , which implies that @xmath495 vanishes to all orders as @xmath496 . since @xmath495 only depends on the direction of @xmath336 , @xmath497 goes to zero to infinite order as @xmath498 , provided @xmath499 . by setting @xmath500 for @xmath501 , we therefore obtain @xmath502 . let @xmath109 be a compact set in @xmath503 . then , for arbitrary @xmath504 , @xmath505 for @xmath506 , the last integral reads @xmath507 where we introduced in @xmath508 the polar coordinates @xmath509 and @xmath510 , and assumed that @xmath109 is contained in a sphere of radius @xmath511 . since @xmath512 is a real valued , continuous function on the @xmath513-dimensional sphere which is bounded from below , the infimum is adopted . hence , there is a strictly positive number @xmath514 such that @xmath515 . this proves the lemma . the last lemma implies that @xmath328 is given by the locally integrable function @xmath516 . let us examine its restriction to the diagonal . by the structure theorem , @xmath517 as a consequence of theorem [ thm : a ] , we then have the following proposition . let @xmath5 be a reductive prehomogeneous vector space , and @xmath479 . set @xmath518 . then , for @xmath519 , the integrals @xmath520 can be continued to meromorphic functions on the whole @xmath17-plane . for @xmath521 , they satisfy the functional equations @xmath522 put @xmath523 , and assume @xmath524 . one then computes @xmath525 everything in sight being absolutely convergent . consider now the @xmath18-dimensional mellin transform @xmath526 which can be continued meromorphically in @xmath17 with simple poles at @xmath527 . since , by the above computation , @xmath528 we obtain the first assertion . the second one follows directly from the relations with respect to the quadratic form @xmath16 . now , although the integrals @xmath529 become singular for general @xmath519 , it is nevertheless possible to extend @xmath530 , as a distribution , to @xmath195 . first note that , since @xmath531 on @xmath532 , where @xmath533 , @xmath534 defines a homogeneous distribution on @xmath532 of degree @xmath535 according to @xmath536 by @xcite , theorem 3.2.4 , an extension of @xmath530 to @xmath537 can then be constructed as follows . let @xmath17 be a complex number with @xmath538 , and consider on @xmath1 the locally integrable function @xmath539 it is homogeneous of degree @xmath17 , and , as a distribution , can be extended to arbitrary complex values of @xmath17 by analytic continuation , except when @xmath527 . in that case one defines for @xmath540 @xmath541 for @xmath542 , @xmath543 is then a homogeneous distribution on @xmath1 . now , for @xmath544 , and @xmath545 , set @xmath546 , and let @xmath547 satisfy @xmath548 then , the extension of @xmath530 is defined as @xmath549 summing up , we have shown the following proposition . [ prop:2 ] let @xmath5 be a prehomogeneous vector space , and @xmath479 . then , the restriction of the kernel of @xmath550 to the diagonal is given by the homogeneous distribution @xmath551 of degree @xmath535 , which has an extension to @xmath4 given by . the extension @xmath552 is unique up to a linear combination of @xmath553-distributions at zero , and is no longer homogeneous . one could also have defined @xmath552 as the integral over the unit sphere of @xmath554 according to @xmath555 each of the extensions @xmath556 could then be regarded as a trace of @xmath9 . let @xmath5 be a reductive prehomogeneous vector space defined over @xmath1 , and assume that the singular set @xmath11 is an irreducible hypersurface . denote by @xmath22 the corresponding relative invariant with character @xmath557 such that is satisfied , and by @xmath308 the left regular representation of @xmath7 on the banach space @xmath8 . as already explained in section [ sec : iv ] , for @xmath126 , the restriction of the operator @xmath9 to @xmath313 defines a continuous linear operator @xmath558 now , if @xmath559 denotes the decomposition of the open @xmath7-orbit into its connected components , on each @xmath32 , @xmath560 only depends on the restriction of @xmath561 to @xmath32 , so that one naturally obtains the continuous linear operators @xmath562 which are elements in @xmath563 , by the structure theorem . throughout this section , we will make the following assumption . @xmath12 is an irreducible hypersurface . denote the set of non - regular points of @xmath14 by @xmath58 . we will then show that the restrictions of @xmath9 to @xmath564 are totally characteristic pseudodifferential operators in the class @xmath565 . let us introduce local coordinates in @xmath68 . since the fibers of the categorical quotient @xmath566 are , apart from the exceptional divisor @xmath14 , smooth affine varieties , we have @xmath567 on @xmath68 . defining the open subsets @xmath568 and the coordinates @xmath569 on @xmath570 , where @xmath571 , one gets @xmath572 by the inverse function theorem , there exists for every @xmath573 an open neighbourhood @xmath574 of @xmath336 such that @xmath575 becomes a diffeomorphism , and we obtain a covering of @xmath68 by charts . we define now the following closed subgroups of @xmath0 . let @xmath576 clearly , @xmath577 acts transitively on each generic fiber of the categorical quotient @xmath566 , and its lie algebra is given by @xmath578 , where @xmath579 denotes the infinitesimal representation corresponding to @xmath557 . there always exists a @xmath580 such that @xmath581 ; otherwise , @xmath13 would decompose into an infinite number of @xmath35-orbits . for the following considerations , we will make the basic assumption that @xmath577 acts transitively on @xmath200 . let us formulate this condition as follows . for all @xmath582 , there exists an open neighbourhood @xmath583 of @xmath336 and elements @xmath584 such that @xmath585 defines a section in @xmath586 for each @xmath587 . here we wrote @xmath588 for the fundamental vector field of the underlying @xmath7-action on @xmath4 induced by @xmath415 . fix @xmath573 , and let @xmath589 and @xmath285 be neighbourhoods of @xmath336 as specified above . let @xmath590 be open sets containing @xmath336 such that @xmath591 , and assume that @xmath215 is chosen in such a way that the set @xmath592 acts transitively on the @xmath7-orbits of @xmath215 . the so defined sets @xmath215 constitute an open covering of @xmath68 . by choosing a locally finite subcovering , we therefore obtain an atlas @xmath306 of @xmath68 with the following properties : \(i ) for @xmath593 , one has @xmath594 , the @xmath595 being given by the prescription @xmath596 for some @xmath597 ; \(ii ) for each @xmath249 , there exist open sets @xmath598 , stable under inverse , acting transitively on the @xmath7-orbits of @xmath249 ; \(iii ) for @xmath599 , one has @xmath600 ; \(iv ) there exist @xmath601 , such that @xmath585 defines a section in @xmath602 for each @xmath587 . [ ex:3 ] as the generic case in the theory of prehomogeneous vector spaces , consider an indefinite quadratic form @xmath22 with signature @xmath603 , @xmath604 , on @xmath1 . then , by sylvester s law , @xmath605 , for some @xmath606 . define the orthogonal , resp . special orthogonal , group of @xmath22 by @xmath607 , resp . @xmath608 , and introduce on @xmath609 an action of @xmath610 by setting @xmath611 . then @xmath22 is an irreducible relative invariant of @xmath5 , and @xmath612 , where @xmath613 . by the theorem of witt , @xmath614 is @xmath615-homogeneous , and @xmath616-homogeneous for @xmath469 . in this case , the fundamental theorem yields the functional equations @xmath617 where @xmath206 , and @xmath618 see @xcite , proposition 4.27 . as an example of a non - regular prehomogeneous vector space , consider the subgroup @xmath0 of upper triangular @xmath619-matrices in @xmath620 acting regularly on @xmath621 . in this case , @xmath622 . according to the structure theorem of the foregoing section , @xmath9 is a pseudodifferential operator on @xmath13 with smooth kernel . in the present section , we will be concerned with the action of @xmath9 as an operator on @xmath68 . with respect to the atlas @xmath243 , we define for @xmath623 the operators @xmath624 \circ \kappa_{\gamma}^{-1}$ ] , see section [ sec : iii ] . explicitely , one has @xmath625 let @xmath626 be a smooth function on @xmath7 with support in @xmath627 satisfying @xmath628 on @xmath629 . then , @xmath630 can be written as @xmath631 where we put @xmath632 . we define @xmath633 by lebesgue s bounded convergence theorem , we can differentiate under the integral sign , yielding @xmath634 , @xmath635 . set @xmath636 and define the symbol @xmath637 note that , by equation , @xmath638 , so that @xmath639 implying that @xmath640 can be expressed explicitely by @xmath641 yielding @xmath642 . put @xmath643 the main result of this section will consist in proving the following theorem . here @xmath306 denotes the atlas of @xmath68 constructed above . [ thm:3 ] let @xmath644 be a reductive prehomogeneous vector space whose singular set @xmath11 is an irreducible hypersurface , and let assumptions @xmath645 and @xmath646 be satisfied . then , for @xmath126 , @xmath9 is locally given by the operators @xmath647 where @xmath648 , and @xmath649 is given by . in particular , @xmath650 . note that the symbols @xmath651 are defined even in the case when the coordinates @xmath652 become singular . the expressions are then still absolutely convergent , so that @xmath9 is indeed given globally by the operators @xmath630 . we will divide the proof of theorem [ thm:3 ] in several parts . to begin with , let us first state a technical lemma . let @xmath653 . then , for @xmath128 , @xmath177 , and @xmath654 one has @xmath655.\ ] ] first , one computes @xmath656 . \end{aligned}\ ] ] by noting that @xmath657 we obtain @xmath658 . similarly , one verifies the relation @xmath659 , thus obtaining the assertion . for @xmath128 , one has @xmath660 as well as @xmath661 thus , @xmath662 is a measure for the transversality of the fundamental vector field @xmath663 with respect to the foliation @xmath664 as a first step towards theorem [ thm:3 ] , we prove the following [ prop:3 ] @xmath665 . as already noted , @xmath666 . while differentiation with respect to @xmath337 does not alter the growth properties of @xmath640 , differentiation with respect to @xmath667 yields additional powers in @xmath337 . let @xmath668 and @xmath669 be given in such a way that @xmath670 , and condition ( iv ) is satisfied . by the previous lemma , @xmath671 writing @xmath672 for the matrix appearing on the right hand side of the last equation , we obtain @xmath673 hence , @xmath674 vanishes if , and only if , the fundamental vector fields @xmath675 , @xmath676 , do span the tangent spaces @xmath677 at each point @xmath678 , and if their span is not perpendicular to the hypersurface @xmath679 . the latter condition is equivalent to @xmath680 , which in turn is equivalent to the condition @xmath681 for @xmath682 . by construction , these conditions are fulfilled for @xmath264 and @xmath683 , as a consequence of the properties ( iii ) and ( iv ) of the atlas @xmath684 . thus , for arbitrary @xmath137 , @xmath685 we now consider the extension of @xmath672 , as an endomorphism in @xmath686 $ ] , to the symmetric algebra @xmath687)\simeq { { \mathbb c}}[{{\mathbb r}}^n_\xi]$ ] . regarding the polynomials @xmath330 as a basis in @xmath686 $ ] , we obtain @xmath688 where @xmath672 denotes the image of the basis vector @xmath689 under the endomorphism @xmath672 . since , for @xmath264 , @xmath690 , @xmath672 is invertible , its extension to @xmath691)$ ] is also an automorphism , so that every polynomial @xmath692 can be written as a linear combination @xmath693\dots,\ ] ] where the @xmath694 are @xmath55 functions on @xmath695 which , in addition , are rational in @xmath74 . now , we need the following lemma . [ lem:3 ] under the identification @xmath696 one has @xmath697 where the coefficients @xmath698 are at most of exponential growth in @xmath74 , and independent of @xmath337 . for @xmath699 one has @xmath700 , and differentiating the latter equation with respect to @xmath701 we obtain , by taking into account , @xmath702 hence , the assertion of the lemma is correct for @xmath703 . now , assume that it holds for @xmath704 . setting @xmath705 in equation , and differentiating with respect to @xmath706 , yields for the left hand side @xmath707 by assumption , we can apply to the products @xmath708 , and to all other products of fewer terms . this proves the lemma . returning to the proof of proposition [ prop:3 ] , as an immediate consequence of equations and , one computes @xmath709 where the coefficients @xmath710 are rational expressions in the matrix coefficients of @xmath74 . now , @xmath711 is a finite sum of terms of the form @xmath712 the functions @xmath713 being at most of exponential growth in @xmath74 . making use of equation , and integrating by parts , we finally obtain for arbitrary @xmath714 the estimate @xmath715 where @xmath425 denotes an arbitrary compact set in @xmath247 and @xmath716 . this completes the proof of proposition [ prop:3 ] . in order to prove theorem [ thm:3 ] , we need to examine the symbols @xmath717 and the kernels of the operators @xmath718 more closely . the kernel of @xmath630 is given by the oscillatory integral @xmath719 where @xmath720 . it is a continuous function of @xmath721 with values in @xmath722 . one computes @xmath723 . by proposition [ prop:3 ] , we therefore obtain the estimates @xmath724 hence , @xmath720 . in particular , @xmath725 . let @xmath726 with @xmath727 in a neighbourhood of @xmath169 , and consider , for @xmath728 , the absolutely convergent integral @xmath729 according to @xcite , page 5 , it can be regularized using integration by parts , so that , by the bounded convergence theorem , one may pass to the limit as @xmath730 , in this way getting a distribution @xmath731 as a consequence , exists as oscillatory integral and is equal to @xmath732 , its kernel being given by . we are now in position to prove theorem [ thm:3 ] . according to proposition [ prop:3 ] , @xmath733 , and it remains to show that @xmath640 satisfies the lacunary condition . since , on @xmath734 resp . @xmath735 , @xmath736 only depends on the restriction of @xmath231 to @xmath737 resp . @xmath735 , we necessarily have @xmath738 by the previous lemma , the kernel of @xmath718 is given by the oscillatory integral . consequently , @xmath739 which is a @xmath55 function for @xmath740 , resp . @xmath741 , must vanish then . by making the substitution @xmath742 in case that @xmath743 , we arrive at the condition @xmath744 meaning that @xmath640 satisifies the lacunary condition . this completes the proof of theorem [ thm:3 ] . as a consequence of theorem [ thm:3 ] we can write the kernel of @xmath630 , for @xmath745 , in the form @xmath746 where @xmath747 denotes the inverse fourier transform of @xmath640 , @xmath748 since , for @xmath264 , @xmath640 is rapidly falling in @xmath337 , it follows that @xmath749 , the fourier transform being an isomorphism on the schwartz space . consider the restriction of the atlas @xmath306 to @xmath13 . by , the restriction of the kernel of the operator @xmath750 to the diagonal is given by @xmath751 and we denote the corresponding density on @xmath13 by @xmath530 . [ prop:4 ] let @xmath752 be a reductive prehomogeneous vector space whose singular set is an irreducible hypersurface , satisfying assumptions @xmath645 and @xmath646 . let @xmath753 be the decomposition of @xmath13 into its connected components , and set @xmath754 denote the corresponding density on @xmath13 by @xmath755 . then , for @xmath756 , the integrals @xmath757 converge for @xmath40 , and can be continued analytically to meromorphic functions in @xmath758 , satisfying functional equations of the form . let @xmath759 be a partition of unity subordinated to the covering @xmath760 . let @xmath207 denote lebesgue measure in @xmath4 . identifying the functions @xmath761 with the densities @xmath762 , we obtain @xmath763 since @xmath764 , the assertion follows with theorem [ thm : a ] . note that there exists a finite covering of @xmath765 by the open sets @xmath249 , so that , in particular , the sum over @xmath137 appearing in the last equation is finite . let us illustrate this in the situation of example [ ex:3 ] . by , one has , in this case , the functional equations @xmath766 where we set @xmath767 , and @xmath768 . we close this section by noting that a trace of @xmath9 can be defined in the situation of theorem [ thm:3 ] and proposition [ prop:4 ] by a regularization procedure analogous to the one in proposition [ prop:2 ] . this is closely related to the definition of the @xmath769-trace in the theory of totally characteristic pseudodifferential operators . as an application of the results of the previous sections , we study the holomorphic semigroup generated by a strongly elliptic differential operator associated with the left regular representation @xmath308 of a prehomogeneous vector space @xmath5 . let @xmath6 be a continuous representation of a lie group @xmath35 on a banach space @xmath770 , @xmath78 the lie - algebra of @xmath35 , and @xmath771 a basis of @xmath78 . a differential operator @xmath772 associated with the representation @xmath6 is called _ strongly elliptic _ , if , for all @xmath773 , the relation @xmath774 is satisfied , where @xmath90 is some positive number . by langlands @xcite , the closure of a strongly elliptic operator generates a strongly continuous holomorphic semigroup of bounded operators on @xmath770 given by @xmath775 where @xmath43 is a appropriate path in @xmath3 coming from infinity and going to infinity , such that @xmath776 for @xmath777 , and the integral converges uniformly with respect to the operator norm . here @xmath778 for an appropriate @xmath779 $ ] . the proof of this essentially relies on the verification of a criterium of hille @xcite within the theory of strongly continuous semigroups . the subgroup @xmath69 can be characterized by a convolution semigroup of complex measures @xmath780 on @xmath35 according to @xmath781 @xmath6 being measurable with respect to the measures @xmath780 . the measures @xmath780 are absolutely continuous with respect to haar measure @xmath52 on @xmath35 , and denoting by @xmath782 the corresponding radon - nikodym derivative , one has @xmath783 the function @xmath784 is analytic in @xmath73 and @xmath74 , and given universally for all banach representations . in the following , it will be called the _ langlands kernel _ of the holomorphic semigroup @xmath69 . though , in general , the langlands kernel is not explicitely known , it satisfies the following @xmath785- and @xmath786-bounds . a detailed exposition of these facts can be found in @xcite . let @xmath787 be the left regular representation of @xmath35 . then , for each @xmath788 , there exist constants @xmath789 , and @xmath790 such that @xmath791 for all @xmath533 , @xmath792 and multiindices @xmath320 . furthermore , @xmath793 for all @xmath368 , where @xmath794 , and @xmath217 denotes the order of @xmath795 . in particular , @xmath796 . let @xmath5 be a reductive prehomogeneous vector space defined over @xmath1 , and @xmath308 the regular representation of @xmath7 on @xmath4 . assume that @xmath795 is a strongly elliptic operator associated with @xmath6 , and consider the corresponding holomorphic semigroup @xmath71 with langlands kernel @xmath797 . it induces a continuous linear mapping @xmath798 and since @xmath799 , the result of the previous sections concerning the schwartz kernel of the operator are applicable . we will illustrate this for the classical heat kernel and the simplest prehomogeneous vector space , see example [ ex:1 ] . thus , let @xmath800 , @xmath801 . in this case , @xmath802 acts multiplicatively on @xmath803 . let @xmath804 be a basis of @xmath805 . the action of @xmath806 on the grding subspace @xmath807 corresponds on @xmath1 to the vector field @xmath808 the casimir operator @xmath809 of the considered @xmath35-representation is therefore given by @xmath810 and constitutes a differential operator of euler type . let @xmath811 be haar measure on @xmath812 . denoting by @xmath69 the holomorphic semigroup generated by @xmath813 , and by @xmath814 the corresponding langlands kernel , we get @xmath815 in order to compute @xmath814 explicitely , we consider the banach representation @xmath816 , introducing on @xmath817 the new coordinate @xmath818 . the casimir operator is then given by @xmath819 , and the langlands kernel coincides with the classical heat kernel @xmath820 . by transforming back , we get for @xmath814 the universal expression @xmath821 now , @xmath822 if we therefore define @xmath823 we get for @xmath69 the representation @xmath824 one has @xmath479 , and theorem [ thm:2 ] reads as follows . [ prop:6 ] the bounded linear operator @xmath825 is given by the family of oscillatory integrals @xmath826 where @xmath827 for @xmath828 , one even has @xmath829 . its schwartz kernel @xmath830 is given by the continuous family of oscillatory integrals @xmath831 furthermore , @xmath832 , so that @xmath69 , as a map from @xmath833 to @xmath834 , represents a fourier integral operator . on @xmath835 , @xmath836 . since the langlands kernel is explicitely known , it is illustrative to give a direct proof in this case . since @xmath35 is unimodular , one has @xmath837 where we defined the function @xmath838 , for @xmath839 , @xmath840 for @xmath841 . clearly , @xmath842 . indeed , @xmath843 for all @xmath844 , and by the substitution @xmath845 , we obtain @xmath846 for @xmath847 and @xmath848 , @xmath849 goes to zero together with all its derivatives , and it follows @xmath850 for arbitrary multiindices @xmath220 . hence we deduce @xmath851 for arbitrary @xmath852 , and @xmath853 for @xmath828 . since @xmath854 we finally obtain @xmath855 . the continuity of @xmath856 is a consequence of the fact that @xmath857 is continuous . this proves propostion [ prop:6 ] . now , by lemma [ lem:2 ] , on @xmath858 , the schwartz kernel @xmath859 of @xmath860 is given by the function @xmath861 introduced above . indeed , for @xmath828 , @xmath862 while , as oscillatory integral , @xmath863 where @xmath864 denotes the @xmath553-distribution mit support at zero . furthermore , @xmath865 . following the discussion in section [ sec : v ] , we define the auxiliary symbol @xmath866 so that @xmath867 . the inverse fourier transform @xmath868 is a schwartz function , which vanishes for @xmath869 together with all its derivatives . implying that @xmath870 satisfies the lacunary condition . for @xmath828 one computes @xmath871 furthermore , by the general theory of conormal distributions , @xmath872 , see @xcite , theorem @xmath873 . in concordance with proposition [ prop:2 ] , the restriction @xmath874 of @xmath69 to the diagonal defines a homogeneous distribution of degree @xmath875 on @xmath835 , which has an extension to @xmath1 given by . note that @xmath876 . setting @xmath877 , then yields in our situation the functional equations @xmath878

inhe superfluids quantized vorticity can take many different forms : it can appear as distributed periodic textures , as sheets , or as lines . in the anisotropiche - a phase in most cases the amplitude of the order parameter remains constant throughout the vortex structure and only its orientation changes in space . in the quasi - isotropiche - b phase vortex lines have a hard core where the order parameter has reduced , but finite amplitude . the different structures have been firmly identified , based on both measurement and calculation . what parallels can be drawn from this information to the new unconventional superconductors or bose - einstein condensates ?

cond-mat0012350

soon after the discovery of the @xmath0he superfluids in 1972 it was understood that they represented the first example of unconventional cooper pairing among fermi systems , a p - wave state with total spin @xmath1 and orbital momentum @xmath2 @xcite . this lead to a wide variety of new phenomena , of which one of the most important is the discovery of new vortex structures @xcite . these can be studied with nmr spectroscopy @xcite , when this is combined with a calculation of the order parameter texture @xcite . in recent years other unconventional macroscopic quantum systems have been found and have taken the centre stage . intermetallic alloys such as the heavy fermion metals , the high - temperature superconductors , and the most recent addition , the layered superconductors of sr@xmath3ruo@xmath4 type , do not fit in the conventional picture of s - wave pairing . is it possible that unconventional vortex structures , similar perhaps to some of those in the @xmath0he superfluids , might also be present in these new systems ? and where the order - parameter amplitude vanishes in the center . ( ii ) the _ non - singular _ vortex has a _ hard _ core in which the order parameter has a finite amplitude everywhere within the core . in principle , this amplitude can even be larger within the core than far outside . ( iii ) the _ continuous _ vortex has an almost constant order - parameter amplitude throughout the _ soft _ core , whose size is much larger than @xmath5 . within the soft core primarily the orientation of the order parameter changes . examples of _ continuous _ vortices are the doubly quantized vortex in @xmath0he - a and the vortex in a two - component bose - einstein condensate , which has a wide inflated core . the latter vortex is formed from the condensate fraction @xmath6 , while the soft core is filled by the superfluid component @xmath7 . , scaledwidth=85.0% ] current belief holds that the superconducting state in the tetragonal sr@xmath3ruo@xmath4 material is described by an order parameter of the same symmetry class as that in @xmath0he - a @xcite , an anisotropic superfluid with uniaxial symmetry ( where both time reversal symmetry and reflection symmetry are spontaneously broken ) . recent advances in optical trapping and cooling of alkali atom clouds to bose - einstein condensates have produced bose systems which also are described by a multi - component order parameter : the spinor representation of the hyperfine spin manifold @xmath8 , for instance , would allow the presence of similar vorticity as in @xmath0he - a . thus the existence of unconventional vorticity has moved in the centre of interdisciplinary debate : to what extent will reduced symmetry influence the structure of quantized vorticity ? vortex lines are defects of the order parameter field , which carry phase winding and circulation of the respective supercurrent . the conventional structure is built around a narrow _ singular hard vortex core _ : the order parameter vanishes in the center of the core . by now it has been thoroughly verified that quantized vorticity can take many other forms . some of the different core structures are schematically listed in fig . [ vorcoreschematics ] . there exist differences in nomenclature in the @xmath0he literature and the current theoretical discussion of vortex line structure in unconventional superconductors , which we try to unify here . \(i ) if the vortex has a _ hard _ core , whose radius is comparable to the temperature- and pressure - dependent superfluid coherence length @xmath5 and where the order - parameter amplitude vanishes in the center , then we call it _ singular_. conventional vortices in the @xmath9he - ii superfluid and in s - wave superconductors are singular . here : @xmath10 . in some approximation such a core can be pictured to be a tube with a diameter comparable to the coherence length and filled with normal - state material . \(ii ) if the vortex has a hard core but the order parameter has a finite amplitude everywhere within the core , then we call this a _ non - singular _ vortex , as is generally done in the literature on unconventional superconductivity . such a core can be viewed as consisting from a different broken symmetry state than the bulk phase outside the core . vortices in @xmath0he - b are non - singular : the core of the vortex is made up of some non - b - phase components of the order parameter , either the axial ( ie . @xmath0he - a ) or axiplanar state ( sec . [ brokensymmetry ] ) . the length scale , which determines the core radius , is @xmath11100 nm . \(iii ) the _ continuous _ vortex has an almost constant order - parameter amplitude throughout the _ soft _ vortex core . by the soft core we mean a core whose size is much larger than @xmath5 . within the soft core primarily the orientation of the order parameter changes . the larger the soft core diameter , the smoother is the distribution of the order - parameter amplitude . examples of continuous vortices are the _ doubly quantized vortex _ in @xmath0he - a ( sec . [ continuousvortex ] ) and the vortex in a two - component bose - einstein condensate ( sec . [ singulartocontinuous ] ) . in @xmath0he - a the length scale , which determines the soft - core radius , is the healing length of the dipolar spin - orbit interaction : @xmath12 40@xmath13 m . \(iv ) a vortex with a _ composite _ core has onion structure : it has a hard core ( with radius @xmath14 ) , which can be either singular or non - singular , but is embedded within a soft core ( with radius @xmath15 ) . ( a ) the _ singly quantized vortex _ in @xmath0he - a is the prime example ( sec . [ compositecore ] ) . it has a hard non - singular core within a large soft core . the superfluid circulation is generated by the soft core , but the hard core is needed to satisfy the boundary conditions on the orbital part of the order - parameter field , such that it becomes continuous with respect to the bulk fluid . ( b ) at high enough magnetic field , the singly quantized vortices in @xmath0he - b have composite cores . here the vorticity is concentrated in the hard core , while the soft core supports an inhomogeneous order - parameter distribution where the spin - orbit interaction is not minimized . this soft core is a deformation in the order - parameter texture which occurs primarily in the spin part . ( c ) the _ spin - mass vortex _ in @xmath0he - b is a composite defect with a narrow hard core , around which the superfluid circulation is trapped . this core is embedded within a planar domain - wall - like soliton defect where the spin - orbit interaction is not minimized and which supports a spin current . the distinction between non - singular and continuous vortices , and also the existence of a vortex with a composite core rely upon the presence of two length scales , which in turn are determined by two energy scales . in superfluid @xmath0he the two relevant energy scales are the weak spin - orbit interaction and the two orders of magnitude larger superfluid condensation energy . these fix the soft and the hard core sizes , respectively , and as a result the soft core is typically two orders of magnitude larger in diameter than the hard core . if the two scales become comparable in magnitude ( as typically occurs in superconductors ) , then the difference between the continuous and non - singular vortices is washed out . an understanding of the various structures , in which quantized vorticity may appear , has led to new insight in the physics of macroscopic quantum systems . this new understanding now promises to bring surprising rewards . the discovery of gap nodes in the spectrum of quasiparticle excitations is generally taken to be a signal for unconventional pairing in fermi systems . an important observation from recent years is the fact that in the vicinity of these gap nodes the energy spectrum is linear and the system acquires all the attributes of relativistic quantum field theory : the analogues of lorentz invariance , gauge invariance , general covariance , etc . all are present . therefore fermion superfluids and superconductors on one hand and quantum field theory on the other hand show surprising conceptual similarity . this makes it possible to treat the condensed - matter quantum systems as laboratory models to study physical principles which might also be effective in high energy physics or cosmology . the first examples of such work have been seen in cosmological " laboratory experiments . for instance , it was recently demonstrated that quantized vortex lines , or linear topological defects as they are known in field theory , are produced in quench - cooled transitions from the normal to the superfluid / superconducting state @xcite . this process has been suggested to mimic the production of cosmic strings in big - bang expansion . a second effort was related to the dynamics of vortex lines and was exploited to explain the matter - antimatter asymmetry in the early universe @xcite , if it is assumed to result from the axial anomaly of relativistic field theory . relativistic quantum field theory may just have found itself an unexpected accomplice ! superfluid @xmath0he has been blessed with the most ideal properties among the dense coherent quantum systems , approaching those of the alkali atom clouds in bose - einstein condensed states : ( i ) there are no bulk impurities since all alien particles are expelled to the container walls . ( ii ) the superfluid coherence length @xmath16 , in addition to being a function of temperature @xmath17 , also depends on the applied pressure @xmath18 . by choosing the pressure , the density and interactions can be varied and @xmath19 decreases from 65 nm at zero pressure to 12 nm at the liquid - solid transition ( @xmath20 bar ) . ( iii ) for the best wall materials surface roughness can be reduced close to the level of @xmath16 . experimentally this has important consequences , when the container walls approach ideal solid boundaries . ( iv ) being an isotropic liquid and a fermi system , liquid @xmath0he is theoretically more tractable than either superconductors or liquid @xmath9he . all complexity and anisotropy is simply derived from the order parameter , with cooper pairs in p - wave states @xcite . in practice this means that experimental observation can be confirmed by detailed theoretical calculation , although often only retrospectively , once the true behaviour is already known . one of the most striking results emerging from @xmath0he research is the existence of vortices with perfectly continuous singularity - free structure . in such a vortex the order - parameter amplitude remains constant throughout the whole structure while its orientation changes in a continuous manner . in the @xmath0he - a phase with axial anisotropy ( and cooper pairs in @xmath21 states ) , the orbital order - parameter orientation is denoted by the symmetry axis @xmath22 which points in the direction of the nodes of the superfluid energy gap . at very low magnetic fields vorticity is distributed over the entire primitive cell of the vortex array and thereby takes the form of a periodic order - parameter texture . at higher fields a region of concentrated vorticity is formed , the soft vortex core . an integral number of phase windings is reached along a closed path which encircles the boundary of the unit cell or the soft core . he - a with continuous structure in the soft vortex core : _ ( top right ) _ rotating container with quantized vortex lines . the pillars depict the soft vortex cores , with a diameter of roughly @xmath23 m @xmath24 . each soft core is encircled by a persistent superfluid circulation of two quanta @xmath25mm@xmath26/s . _ ( bottom right ) _ nmr spectroscopy of topological defects in @xmath0he - a . in an external magnetic field , which exceeds the equivalent of the spin - orbit interaction , @xmath27mt , the spin @xmath28 and orbital @xmath29 axes are not aligned parallel in the soft core . spin - orbit interaction exerts an extra torque on spin precession in nuclear magnetic resonance which shifts the nmr frequency @xcite . this torque has different value within and outside the soft core , and thus gives rise to a satellite absorption peak . both the frequency shift and the absorption intensity of the satellite are characteristic of the order - parameter texture in the soft core @xcite . _ ( left ) _ orientational distribution of the orbital quantization axis @xmath29 in the soft core , depicted in terms of the projection of @xmath29 on the plane perpendicular to the vortex axis . the @xmath29 orientations cover a solid angle of @xmath30 and the distribution is everywhere continuous . this gives rise to a superfluid circulation of two quanta around the soft core . , scaledwidth=100.0% ] the simplest possible vortex structure with continuous vorticity in @xmath0he - a is the doubly quantized vortex line . in fig . [ softvorcore ] the orientational distribution of @xmath22 within the soft vortex core is depicted with arrows . by following their flow , it is noted that all possible @xmath30 directions of the radius vector of the unit sphere are present here . this topology of order - parameter orientations ensures two quanta of circulation @xmath31 . the @xmath30 topology of @xmath22 orientations within the soft core is known as a _ skyrmion_. it can be divided into a pair of _ merons _ @xcite ( @xmath32 means fraction @xcite ) , which in the @xmath0he literature are called mermin - ho vortices . in the complete skyrmion the @xmath22 vector sweeps the whole unit sphere while each meron , or mermin - ho vortex , covers only the orientations in one hemisphere and therefore carries one quantum of vorticity @xmath33 . the meron covering the northern hemisphere forms a circular @xmath34 mermin - ho vortex , while the meron covering the southern hemisphere is the hyperbolic @xmath34 mermin - ho vortex . the centers of the merons correspond to minima in the potential of trapped spin - wave states , which generate the satellite peak in the nmr absorption spectrum and make the soft cores observable in nmr measurement . the satellite from the doubly quantized vortex was first detected in rotating @xmath0he - a in 1982 @xcite , but it is only recently that single - vortex sensitivity was reached and the quantization number @xmath35 was verified directly @xcite . skyrmions and merons are popular structures in physics : for instance , the double - quantum vortex , in the form of a pair of merons similar to that in @xmath0he - a , is also discussed in the quantum hall effect where it is formed by pseudo - spin orientations in the magnetic structure @xcite . in qcd merons are suggested to produce the color confinement @xcite . in superconductors continuous vortices have been discussed in ref . @xcite within a model where the spin - orbit coupling between the electronic spins and the crystal lattice is small and the spin rotation group @xmath36 is almost exact . the vortex has essentially the same topology as in @xmath0he - a except that instead of the orbital momentum @xmath22 it is the spin orientations @xmath37 of the cooper pairs which cover a solid angle of @xmath30 in the soft core . from the topological point of view , all vortices with the same winding number @xmath38 can be transformed to each other without changing the asymptotic behaviour of the order parameter , simply by reconstruction of the vortex core . to obtain the continuous vortex in @xmath0he - a ( fig . [ softvorcore ] ) we can start with a @xmath35 _ singular phase vortex _ , which has a hard core of the size of the coherence length and a uniformly oriented orbital momentum axis @xmath22 along @xmath39 . this is a pure phase vortex for the orbital component @xmath40 of the order parameter , which vanishes only on the vortex axis while other components are zero everywhere . if we now allow for the presence of the other components @xmath41 and @xmath42 , then we might fill the hard core with these components , such that the core becomes non - singular . the non - singular core can expand further to form a continuous soft core , which is much larger in radius than the coherence length . its size is limited by some other weaker energy scales , for instance in @xmath0he - a by the tiny spin - orbit interaction . within the core @xmath22 sweeps through all possible orientations and the topology becomes that of a skyrmion . this transformation between continuous and singular vorticity has direct application to bose - einstein condensates . suppose a mixture of two sub states can be rotated in a laser manipulated trap . here one starts with a single bose - condensate which we denote as the @xmath43 component . within this component a pure @xmath44 phase vortex with singular core is created . next the hard core of the vortex is filled with the second component in the @xmath45 state . as a result the core expands and a vortex with continuous structure is obtained . such a skyrmion has recently been observed with @xmath46rb atoms @xcite . it can be represented in terms of the @xmath22 vector which is constructed from the components of the order parameter as @xmath47 here @xmath48 and @xmath49 are the cylindrical coordinates with the axis @xmath50 along the vortex axis . the polar angle @xmath51 of the @xmath22 vector changes from 0 at infinity , where @xmath52 and only the @xmath43 component is present , to @xmath53 at the axis , where @xmath54 and only the @xmath45 component is present ( fig . 1 ) . thus the vector @xmath22 sweeps the whole unit sphere . as distinct from @xmath0he - a , where the orbital part of the order parameter has three components and the vortex has @xmath35 circulation , this mixture of bose - condensates has two components and the continuous vortex has @xmath44 phase winding . various schemes have recently been discussed by which a meron with phase winding can be created in a bose - condensate formed within a three - component @xmath8 manifold @xcite . the singly quantized vortex in @xmath0he - a has a composite core : it is a @xmath44 vortex with a non - singular hard core , but where the superfluid circulation is generated by the soft core . thus this @xmath44 vortex is not a simple @xmath55 phase vortex , which would have circulation trapped around a hard core in an otherwise homogeneous @xmath22 texture . in fact , the function of the hard core is only to provide the topological stability of the soft core , which with its @xmath34 orientational distribution of the @xmath22 field produces all the vorticity . this vortex structure has the lowest energy in a magnetic field ( @xmath56mt ) at low rotation velocity @xmath57rad / s ) @xcite . thus it can be created by cooling slowly through @xmath58 in rotation at low @xmath59 , which secures the equilibrium vortex state in an adiabatically slow transition . in contrast , if one starts to rotate the fluid when it is already in the superfluid state below @xmath58 , then the @xmath44 vortex is generally not formed , because the @xmath35 vortex has lower critical velocity and is created first . the larger critical velocity of the @xmath44 vortex reflects the much larger energy barrier involved in the creation of the hard vortex core ( sec . [ critvel - coresize ] ) . experimentally the singly and doubly quantized vortex lines can be distinguished by their very different nmr absorption satellites . in @xmath0he - a the double - quantum vortex line is not the only unconventional vortex structure with perfect continuity in the order parameter amplitude . the most unusual continuous structure is the _ vortex sheet _ @xcite , with planar topology ( fig . [ vorsheet ] ) . it consists of a folded domain - wall - like structure , a soliton sheet , which separates regions with opposite orientations of the @xmath22 vector . within this meandering sheet the continuous mermin - ho vortex lines with @xmath44 are confined . these vortices form a chain of alternating circular and hyperbolic merons , which are similar to the two constituents of the soft core in fig . [ softvorcore ] . each meron represents a kink in the soliton structure and is thus topologically trapped within the soliton : it can not exist as an independent object in the bulk liquid outside the soliton . an analogous example are bloch lines within a bloch wall in ferromagnetic materials . the sheet is attached along two connection lines to the lateral boundaries . it is through these connection lines that merons with @xmath44 vortex quanta can enter or leave the sheet . he - a : _ ( bottom left ) _ the macroscopic configuration of the equilibrium vortex sheet consists of a single continuously folded sheet which fills the rotating container evenly . in an axially oriented external magnetic field the meander forms a double spiral which mimics the original onsager suggestion of concentric coaxial cylindrical surfaces . _ orbital order - parameter texture in the vortex sheet . the arrows depict the projection of @xmath29 in the @xmath60-plane ( perpendicular to the rotation axis @xmath61 ) . this texture consists of an alternating chain of circular ( center marked with black dot ) and hyperbolic ( center marked with cross ) merons . if one follows a path along the centers of merons , the @xmath29 orientation winds continuously : thus the vorticity is continuously distributed along the sheet . outside the sheet both @xmath29 and the counterflow velocity @xmath62 are oriented parallel to the sheet , but in opposite directions on the two sides of the sheet . _ ( bottom right ) _ the integrated nmr absorption in the satellite peak of continuous vortex lines in fig . [ softvorcore ] follows a linear dependence on the rotation velocity @xmath59 or , equivalently , the number of vortex lines in the equilibrium state of rotation . in contrast , if a soliton sheet preexists in the container and rotation is slowly increased , then a satellite peak with a nonlinear dependence on @xmath59 is grown . this is the signature of the vortex sheet as a rotating state of superfluid . , scaledwidth=90.0% ] the vortex sheet is well known from classical turbulence as a thin interface across which the tangential component of the flow velocity is discontinuous . within this sheet vorticity approaches infinity while the width of the sheet approaches zero @xcite . historically in superfluids , vorticity was first suggested to be confined to sheets , when onsager @xcite and london @xcite described the superfluid state of @xmath9he - ii under rotation . it soon turned out , however , that in @xmath9he - ii a vortex sheet is unstable towards break - up into separated quantized vortex lines . nevertheless , a calculation on the sheet spacing by landau and lifshitz @xcite , who did not impose a quantization requirement , happens to be exactly to the point for the vortex sheet in @xmath0he - a . here the vortex sheet proved to be stable owing to the topological confinement of the vorticity within the topologically stable soliton sheet @xcite . the equilibrium state of the vortex sheet in a rotating vessel is constructed by considering the kinetic energy from the flow between the folds and the surface tension @xmath63 from the soliton sheet . it is then concluded that the distance between the parallel folds has to be @xmath64 . this is somewhat larger than the inter - vortex distance in a cluster of vortex lines . the areal density of circulation quanta has approximately the solid - body value @xmath65 . this means that the length of the vortex sheet per two circulation quanta is @xmath66 , which is the periodicity of the order - parameter structure in fig . [ vorsheet ] ( @xmath67 m at @xmath68rad / s ) . the nmr absorption in the vortex - sheet satellite is proportional to the total volume of the sheet which in turn is proportional to @xmath69 . this nonlinear dependence of the satellite absorption on rotation velocity is one of the experimental signatures . locally the equilibrium vortex sheet corresponds to a configuration with accurately equidistant layers . this is manifested by bragg reflections of spin waves between the folds of the sheet , which produces a characteristic frequency shifted absorption in the observed nmr line shape @xcite . the vortex sheet has also been discussed in unconventional superconductors @xcite . similar to superfluid @xmath0he - a the vorticity is trapped in a domain wall which separates two domains with opposite orientations of the @xmath22-vector @xcite . but , unlike the case of @xmath0he - a , the trapped kink is not a meron but a singular vortex with the fractional winding number @xmath70 ( in contrast to isolated vortex lines which are singly quantized ) . if there are many trapped fractional vortices , then they form a vortex sheet , which as suggested in ref . @xcite , can be responsible for the flux flow dynamics in the low - temperature phase of the heavy - fermion superconductor upt@xmath71 . in superconductors a vortex - sheet - like structure could also appear dynamically , when it is topologically not stable . this state would correspond to a smectic phase of the flowing vortex matter , as was argued in the case of eg . nbse@xmath3 @xcite . an unusual object is the _ half - quantum vortex_. here the order - parameter phase changes by @xmath72 on circling once around this line . the change of sign of the order parameter can be compensated by some extra degree of freedom , which usually is the spin . such a linear structure was originally predicted to appear in @xmath0he - a @xcite , but has not yet been observed there experimentally . later it was also predicted to appear in unconventional superconductors @xcite . some years ago it was discovered as the intersection line of three grain boundary planes in a thin film of yba@xmath3cu@xmath71o@xmath73 @xcite . the half - quantum vortex has also been suggested to exist in bose - einstein condensates with a hyperfine spin @xmath8 @xcite . based on the @xmath0he - a example more possible structures of the fractional vortices in unconvetional superconductors can be predicted . in @xmath0he - a the discrete symmetry , which supports the half - quantum vortex , arises from the changes in sign when the spin axis @xmath74 and the orbital axis @xmath75 are taken once around the line and rotate into @xmath76 and @xmath77 . when both of these axes change sign , then the order parameter returns to its initial value @xmath78 here @xmath49 is the azimuthal angle of the cylindrical coordinate system and the magnetic field is applied along @xmath79 to keep @xmath80 in the @xmath60 plane . the spin axis @xmath80 rotates by @xmath72 on circling around the half - quantum vortex . thus a `` spectator '' in @xmath0he - a , who travels around the vortex , would find its spin reversed with respect to the spin of a `` spectator '' who was at rest . this situation is the analog of the alice string in particle physics @xcite where a particle , which moves around the string , in a continuous manner flips its charge or parity or enters the `` shadow '' world @xcite . in superconductors the crystalline structure must be taken into account . in the simplest representation which preserves tetragonal symmetry , the @xmath81-wave order parameter in sr@xmath3ruo@xmath4 has the form @xmath82 where @xmath83 is momentum , @xmath84 is the phase of the order parameter , @xmath85 and @xmath86 are the elementary vectors in the basal plane of the crystal lattice . vortices with fractional quantization @xmath38 can now be constructed in two ways . if the @xmath28-field is sufficiently flexible , the analog of the vortex with @xmath70 in eq . ( [ halfquantumvortex ] ) becomes possible , where @xmath87 and @xmath88 on circling once around the vortex @xcite . another possibility is the mbius - strip geometry . here the crystal axes @xmath85 and @xmath89 are twisted continuously by the angle @xmath90 on traversing around the closed wire loop @xcite . this closed loop traps fractional flux , since the local orientation of the crystal lattice continuously changes by @xmath90 around the loop , @xmath91 and @xmath92 , which means that the order parameter becomes multiplied by @xmath93 . the single - valuedness of the order parameter requires that this change must be compensated by a change in the phase @xmath84 by @xmath90 . as a result the phase winding around the loop is @xmath90 and @xmath94 . this , however , does not mean that such a mbius loop in a chiral @xmath81-wave superconductor traps 1/4 of the magnetic flux @xmath95 of a conventional abrikosov vortex . because of the breaking of time reversal symmetry in chiral crystalline superconductors , persistent electric currents arise not only due to phase coherence but also due to deformations of the crystal @xcite : @xmath96 the magnetic flux trapped in the loop is obtained from the condition of zero current , @xmath97 in eq . ( [ superfluidcurrent ] ) . thus , the trapped flux depends on the parameter @xmath98 in the deformation current . in the limiting case of @xmath99 the flux is @xmath100 ( or @xmath101 if the underlying crystal lattice has hexagonal symmetry ) . in a nonchiral @xmath102-wave superconductor of layered cuprate - oxide structure the order parameter can be represented by : @xmath103 the same twisted wire loop which transforms @xmath91 and @xmath92 , produces a change of sign of the order parameter , which must be compensated by a change of the phase @xmath84 by @xmath72 . this corresponds to a circulation of half a quantum @xmath70 , i.e. the fractional flux trapped by this loop is @xmath104 , since the parameter @xmath98 in eq . ( [ superfluidcurrent ] ) is exactly zero in nonchiral superconductors . the same reasoning gives rise to the @xmath104 flux associated with the intersection line of three grain boundary planes , the tri - crystal line @xcite : around this line @xmath105 and @xmath92 . only the observation of a fractional flux different from @xmath104 would indicate the breaking of time reversal symmetry @xcite . in the quasi - isotropic @xmath0he - b phase ( with cooper pairs in the total angular momentum state @xmath106 ) , the simplest possible vortex is a singular vortex with @xmath44 , where all the order - parameter components are zero on the axis of the vortex core . however , this vortex is never realized . all vortices ( and other linear defects ) observed in @xmath0he - b have a non - singular hard core . this is a typical situation for superfluids / superconductors with a multi - component order parameter . as a rule the superfluid / superconductor does not tolerate a full suppression of the superfluid fraction in the core , if there is a possibility to escape this by filling the core with other components of the order parameter . this rule applies also to high - temperature superconductors with @xmath102-wave pairing , where the @xmath44 vortex supports a nonzero @xmath107-wave component in the vortex core @xcite . he - b : _ ( top ) _ pressure vs. temperature phase diagram of @xmath0he liquids , with the phase transition line for the b - phase vortex core structure . the exact intercept of the transition line at @xmath58 is unknown . _ ( bottom ) _ both measurement @xcite and calculation @xcite confirm that the axi - symmetric core at high temperatures spontaneously reduces its symmetry and transforms to a double - well structure at low temperatures . the magnitude of the square of the order - parameter amplitude has here been plotted in the @xmath60-plane ( perpendicular to the rotation axis ) for the two core structures . this quantity remains finite throughout the cross section of the core . , scaledwidth=100.0% ] the first nmr measurement on rotating @xmath0he - b revealed as a function of temperature and pressure a first order phase transition in the vortex core structure @xcite . this phenomenon in fig . [ vorcoretransition ] was the first example of a phase transition in the structure of any topological defect . other transitions have been identified in @xmath0he - a since then , but the b - phase transition remains the most prominent one . this transition separates two vortices with the same topology ( @xmath44 ) , but with a different structure of the hard core . the existence of the transition illustrates that vortices in @xmath0he - b have to be non - singular and have a complex structure of the hard core . later these two stable vortex core structures were theoretically identified as two different stationary minimum energy solutions as a function of pressure in the ginzburg - landau temperature regime . both structures have broken parity in the core . the high - pressure @xmath0he - b vortex has an axi - symmetric core in which the a - phase order - parameter components dominate . these components produce increase in the order parameter amplitude close to vortex axis which is shown in fig . [ vorcoretransition ] , bottom right . this non - singular vortex with a superfluid core lies lower in energy , than the simplest most symmetric solution with a normal core , and displays ferromagnetic spin polarization @xcite , which is observed in the rotating nmr experiments as a gyromagnetism @xcite . similar , but antiferromagnetic cores have been discussed for high-@xmath58 vortices within the popular @xmath108 model for superconductivity and antiferromagnetism @xcite . here it has been established that in the ginzburg - landau regime , in certain regions of the parameter values , a solution with normal vortex cores is unstable with respect to non - singular antiferromagnetic cores @xcite . also non - singular vortex cores in the heavy fermion superconductor upt@xmath71 could possibly explain why a three times larger flux - flow resistivity is observed parallel to the @xmath109 axis compared to the perpendicular directions @xcite . the low - temperature b - phase vortex has a non - axisymmetric core , i.e. the axial @xmath55 symmetry of the ferromagnetic core is spontaneously broken , to create a dumbbell - like double core ( fig . [ vorcoretransition ] , bottom left ) . it can be considered as a pair of half - quantum vortices , connected by a non - topological wall @xcite . the separation of the half - quantum vortices increases with decreasing pressure and thus the two - core structure is most pronounced at zero pressure @xcite . the asymmetry of the core has been verified by direct observation of a goldstone mode which is a direct consequence from the loss of axial symmetry : the deformed vortex core can become twisted in the presence of a special type of b - phase nmr mode , which is then detected as a reduction in rf absorption @xcite . related phenomena are also possible in superconductors . in ref . @xcite the splitting of the vortex core into a pair of half - quantum vortices has been discussed in heavy - fermion superconductors . in fact a vortex - core splitting may have been observed in high-@xmath58 superconductors @xcite . these observations were interpreted as tunneling of a vortex between two neighboring sites in the potential wells created by impurities . however , the phenomenon can also be explained in terms of vortex - core splitting . in addition to vortex structure , the second most important question becomes the creation of the different forms of vorticity . vortex formation by intrinsic mechanisms is a topic which has been discussed for decades in superfluidity , but which has been notoriously difficult . the problems are caused by interference from extrinsic effects , primarily from remanent vorticity trapped on rough surfaces . one of the major developments from the last ten years has been the emergence of reliable measurements on critical velocities in @xmath9he - ii . these have been performed by monitoring the superflow through sufficiently small sub - micron - size orifices @xcite . intrinsic vortex formation has thereby become an important phenomenon in superfluids quite unlike in superconductors , where typically vortices appear due to different types of extrinsic effects at the lower critical field @xmath110 . however , it is useful even in the case of superconductors to keep in mind the more ideal properties of vortex formation . also the process of unpinning of vortices which is important for a problem of vortex creep in superconductors can be discussed in terms of vortex nucleation : the motion of vortex from the pinning site is equivalent to creation of a vortex loop which annihilates the pinned part of the vortex line . many vortex phenomena , not only nucleation , but also the unpinning of remanent vorticity , involve energy barriers which are usually overcome by thermal activation . at the lowest temperatures quantum tunneling has been suggested as a possible mechanism , where a macroscopic amount of matter is assumed to participate coherently in a tunneling process . vortex nucleation in orifice - flow of @xmath9he - ii displays a characteristic low - temperature plateau in the temperature dependence of the critical velocity which has been claimed to support the quantum tunneling concept @xcite . this question is also discussed in superconductors in the context of unpinning and creep . however , firm proof for such interpretation is still missing . in @xmath0he - b the critical velocity was measured in rotating experiments with single - vortex resolution in the early 1990 s . this proved to be a more straightforward measurement than in @xmath9he - ii . the measured temperature dependence of the critical velocity resembles that of the superfluid energy gap @xmath111 @xcite , which at the lowest temperatures also approaches a temperature - independent plateau . the explanation here , however , does not involve quantum tunneling , but the superflow instability . this phenomenon resembles a second order transition where the energy barrier goes to zero as a function of the scanned variable , in this case the superflow velocity . when a cylinder with superfluid @xmath0he - b is slowly accelerated to rotation , the state with one single vortex line becomes energetically favorable when the superflow at the circumference exceeds the feynman critical velocity @xmath112 @xcite . with a container radius @xmath113 of a few millimeters and a circulation quantum @xmath114 mm@xmath26/s , this velocity is only @xmath115 mm / s . above this velocity remanent vorticity , which has been trapped on the cylinder wall , may start to unpin and then to expand to rectilinear vortex lines . it is the equivalent of @xmath110 in superconductors . however , if we exclude extrinsic mechanisms of vortex formation , then the vortex - free state will persist metastably to much higher velocities because of the nucleation barrier . in @xmath0he - b the nucleation barrier is practically impenetrable . the argument is the following : the vortex is nucleated on the wall as a segment of a vortex ring . the radius of a ring sustained by superflow at the velocity @xmath116 can be written as @xmath117 , where the vortex - core radius @xmath118 is of the order of the superfluid coherence length @xmath119 100 nm . the energy of a ring is @xmath120 , where @xmath121 is the superfluid density , and @xmath122 the interatomic distance . this energy constitutes the nucleation barrier . on dimensional grounds we may write @xmath123 , where @xmath124k is the degeneracy temperature of the @xmath0he quantum fluid . assuming that @xmath125 , we find @xmath126 . such a barrier height in @xmath0he - b is inaccessible at all temperatures below @xmath58 . in contrast in @xmath9he - ii , @xmath127 , and the barrier is low , @xmath128 . it can be thermally overcome , except at the lowest temperatures below 0.2k . the huge barrier in @xmath0he - b means that the vortex formation mechanism can not be thermal activation . when the superflow velocity is increased ( by increasing the rotation ) a threshold @xmath129 is finally reached above which homogeneous flow loses local stability . this occurs when the energy density of the superflow , @xmath130 , exceeds the energy responsible for the topological stability of a vortex , which is of order @xmath131 . an order of magnitude estimate of the maximum velocity is thus @xmath132 . at this velocity , the radius @xmath133 of the nascent ring becomes comparable to @xmath118 and the nucleation barrier goes to zero . the instability inevitably leads to the creation of a vortex when no other mechanism is available at lower @xmath116 . , for @xmath0he - b in a cylindrical container . _ top : _ the vertical axis denotes the height of a nmr absorption peak , where the intensity increases per each newly added vortex line by a constant amount . vortex formation starts with a first step increase at @xmath134rad / s , but the critical threshold is identified from the third step ( dashed vertical line ) where the maximum flow velocity @xmath135 reaches a stable value ( dashed horizontal line in the plot at bottom ) . _ bottom : _ the corresponding superflow velocity @xmath136 at the cylinder wall . each step increase in the upper plot corresponds to a drop by @xmath137 in the velocity @xmath136 , the equivalent of one circulation quantum @xmath138mm@xmath26/s . the average of the maximum velocities @xmath116 defines the critical velocity @xmath135 . the scatter from the average is probably of experimental origin . the rotation is here increased at a constant slow rate ( @xmath139 rad / s@xmath26 ) , the sample is contained in an epoxy - resin - walled cylinder with radius @xmath140 mm at a pressure @xmath141 bar , in a magnetic field @xmath142mt , and temperature @xmath143 . ( from ref . @xcite . ) , scaledwidth=85.0% ] in @xmath0he - b _ vs. _ temperature for pressures @xmath144 bar . the solid curves have been fitted to the data with @xmath145 , where the fitting parameters @xmath102 and @xmath146 characterize the surface roughness on the cylinder wall , @xmath147 is the calculated bulk liquid superflow instability @xcite and @xmath148 $ ] the superfluid coherence length . the roughness is modeled by the protuberance with height @xmath102 and apex angle @xmath149 which acts as the nucleation center . the fit gives @xmath150 m and @xmath151 ( apex angle @xmath152 ) . the measurements have been performed with an epoxy - resin - walled container with radius @xmath153 mm . . @xcite . ) , scaledwidth=80.0% ] in @xmath9he - ii well below @xmath154 , the estimate of @xmath129 in the form @xmath155 agrees in order of magnitude with the landau velocity , defined by the roton gap @xmath156 and momentum @xmath157 as @xmath158 m / s @xcite . however , this limiting velocity is not observed directly in the measurements of orifice flow , by extrapolating the thermally activated process to @xmath159 . the reason is that the measured quantity is the average flow velocity through the aperture and not the local critical velocity at the nucleation site . on the circumference of the orifice , the local velocity will be enhanced from the average value by surface roughness , in particular when the superflow is deflected around any sharp protuberances which match the length scale @xmath133 of the evolving vortex half ring . the most effective of such excrescences on the circumference will then selectively become the nucleation center @xcite . for this reason the measured critical velocity is expected to be roughly a factor of @xmath160 smaller than the ideal limiting value ( fig . [ critvelmeasurement ] ) . in @xmath0he - b , the estimate of @xmath161 is smaller by 3 orders of magnitude than in @xmath9he - ii , but again comparable to the appropriate landau limit , defined by the energy gap and the fermi momentum as @xmath162 . in this case the velocity @xmath129 is also known from direct calculations of the stability limit of homogeneous superflow @xcite . as in orifice flow , the measured average velocity at vortex formation is smaller than the calculated bulk @xmath129 and depends on surface roughness . the principal difference from @xmath9he - ii is the much larger length scale @xmath163 , which means that experimentally the influence of surface roughness is less prominent , remanent vorticity can be avoided in the presence of sufficiently smooth surfaces , and stable periodic vortex formation can be investigated with bulk liquid flowing past a flat wall . this is in stark contrast to @xmath9he - ii , where the coherence length is of atomic size , surface roughness generally provides an unlimited source of trapped remanent vorticity , and intrinsic nucleation can only be observed in flow through a sufficiently small orifice . and maximum experimental @xmath164 plotted for 3 superfluids as a function of their core size @xmath118 . for @xmath118 we use 0.1 nm in @xmath9he - ii , in @xmath0he - b it is the superfluid coherence length @xmath5 , and for continuous vortices in @xmath0he - a in a magnetic field the spin - orbit healing length @xmath165 . in @xmath9he - ii we use the landau limit @xmath166 for @xmath129 , in @xmath0he - b @xmath129 is the calculated maximum superflow velocity @xcite , and in @xmath0he - a it corresponds to the helical textural instability @xcite . for @xmath167 in @xmath9he - ii the measured value in ref . @xcite is used , for @xmath0he - b data from ref . @xcite , and for @xmath0he - a from ref . the line is a guide for the eye , but it obeys the relation @xmath168 . ( from ref . @xcite ) , scaledwidth=85.0% ] owing to the excessively high nucleation barrier , in @xmath0he - b it must be the velocity @xmath129 of superflow instability which becomes the appropriate velocity of vortex formation . the process then corresponds to a classical instability , which occurs at the pair - breaking velocity . thus in the case of @xmath0he - b , the reason for a plateau in @xmath169 in the @xmath170 limit is that the characteristic physical quantities , such as the gap amplitude @xmath171 , which determine the instability velocity , become temperature independent . consequently not only quantum tunneling , but also the intrinsic instability provides an explanation for the low - temperature plateaus which are observed in many different systems , including the case of @xmath9he - ii at the lowest temperatures . the order - of - magnitude expectation for the bulk superflow instability is @xmath172 , where @xmath118 is the core size of the emerging vortex . for @xmath9he - ii and @xmath0he - b it is the size of the hard core , which is on the order of the superfluid coherence length . for the continuous vortex in @xmath0he - a the length scale of the soft core structure is much larger : it is the healing length @xmath173 m of the spin - orbit coupling ( fig . [ softvorcore ] ) . the same length scale also applies to the structure of the soliton and vortex sheets . because of this long length scale the measured critical velocities in the a phase are two orders of magnitude lower than in the b phase , as shown in fig . [ critvellengthscaledependence ] . this explains why the continuous @xmath35 vortex is formed in the a liquid when it is accelerated into rotation , rather than the composite @xmath44 vortex with a hard core , although the latter might be energetically preferred . he - a as a function of the applied rotation @xmath59 . the texture transition occurs at @xmath174 , where a large number of vortex lines is simultaneously formed . after the transition in the transformed texture the regular periodic vortex formation process sets in , where the flow is limited by a constant critical velocity @xmath175 this process gives a linear dependence for the number of vortex lines @xmath176 as a function of @xmath59 , with the slope @xmath177 . , scaledwidth=85.0% ] at the container wall the boundary condition on the order parameter requires that the orbital quantization axis @xmath22 has to be oriented along the surface normal . therefore the center of vortex formation must evolve outside a surface layer with a width @xmath15 , out of the influence of surface roughness . this has been recently experimentally confirmed @xcite . here the only mechanism available is an instability of the order - parameter texture . an example of a critical process , which exemplifies the fact that the order - parameter texture is involved , is shown in fig . [ texturetransition ] . here the vertical axis is the peak height of the satellite absorption in fig . [ softvorcore ] , which is proportional to the number of vortex lines , and the horizontal axis is the external drive . initially on increasing @xmath59 , no vortex lines are formed . when the critical velocity @xmath174 is reached , a large number of vortex lines is suddenly simultaneously formed . during further increase of @xmath59 the system recovers and a characteristic linear slope is retrieved . the linear dependence represents a reproducible periodic process where one vortex at a time is formed at a constant critical velocity , similar to that for b - phase vortices in fig . [ critvelmeasurement ] . the extrapolation of the linear dependence back to zero vortex number gives for the onset a lower value @xmath178 than the actual initially measured @xmath174 at the sudden jump . this is an example where the global order parameter texture becomes unstable in the increasing superflow and finally a first order transition occurs in the texture to a new configuration with a lower critical velocity . it is also clear evidence for the fact that the value of the critical velocity depends on the global order parameter texture in the rotating container . measurements @xcite and calculations @xcite of the critical velocity in @xmath0he - a show that the maximum limit for the critical velocity is reached with an ordered texture which mimics one where @xmath22 is homogeneously oriented along the superflow @xmath179 . the minimum velocity , in turn , is close to an order of magnitude smaller and is obtained within a soliton sheet where the spin - orbit coupling is broken and the @xmath22 texture is inhomogeneous . the dependence of the critical velocity on the core size explains why vorticity with continuous singularity - free structure is formed , when @xmath0he - a is accelerated to rotation . but is it created in the form of lines or sheets ? this has turned out to be an interesting question of general validity . the vortex sheet is formed whenever a vertical dipole - unlocked soliton sheet is present in the container , while rotation is started . here the critical velocity is the lowest possible , ie . the energy barrier for adding more merons into the sheet vanishes at lower velocity than for any other type of vortex structure . the critical velocity of the vortex sheet is made up of several contributions . first of all there is the low critical velocity at the connection lines between the sheet and the cylinder wall , where the spin - orbit coupling is broken . some low superflow velocity is required even here , since at least the attractive interaction of the emerging meron with its image within the wall has to be overcome . a second contribution is a small , but nonzero resistance of the texture in the soliton connection line to reach the instability limit . these small contributions at the connection line are the only ones which are effective initially when rotation is started and the first circulation quanta enter the sheet . of the adiabatically grown vortex sheet as a function of applied rotation @xmath59 : this state consists of a single sheet , which in an axially oriented magnetic field is folded into a double spiral , as shown in fig . [ vorsheet ] . the solid curves are fits to the measured data with the dependence @xmath180 . an epoxy container with surface roughness of order @xmath181 m displays a larger magnitude of @xmath182 than a container with fused quartz walls and an order of magnitude smaller roughness ( @xmath183 m ) . the difference could arise from slight pinning of the two connection lines of the sheet along the cylindrical wall which would resist readjustments when new vorticity is added into the sheet during acceleration . the different symbols of data points ( quartz cylinder ) illustrate the reproducibility of the results from one adiabatically grown vortex sheet to another . , scaledwidth=85.0% ] as the sheet grows , a second contribution becomes effective . when a new meron is added , it experiences repulsion from the meron which already resides in the sheet close to the connection line . the repulsion depends on the distance between the merons in the sheet and the sheet s resistance to change its folding and the distribution of circulation in the container . thus the critical velocity becomes @xmath59-dependent . at low @xmath59 , when the merons are rare , this contribution is much below the critical velocity for the formation of isolated vortex lines or skyrmions . it was suggested and later experimentally confirmed that the critical velocity follows the qualitative dependence @xmath184 as shown in fig . [ critvel - vorsheet ] . ( the star as a superscript marks the fact that @xmath185 is obtained from the measured value of @xmath178 through the relation @xmath186 which is strictly valid only for an axially symmetric distribution of vorticity . ) the vortex sheet has unusual dynamic properties which make it to the preferred state if the rotation drive is rapidly changing in time @xcite . it is expected that analogous features can evolve in anisotropic p - wave superconductors with periodic vortex flow in the unpinned regime @xcite . the dynamic response of @xmath0he - a as a function of the frequency and amplitude of the external drive is currently under study @xcite . the time proven method to create the vortex sheet is to apply sinusoidally modulated rotation in the form @xmath187 , where the period of modulation is of order @xmath188s and the amplitude @xmath189rad / s exceeds that required for breaking the spin - orbit coupling . to grow the equilibrium vortex sheet , one applies the oscillating rotation until the signal from the vortex sheet is observed @xcite . the final step is to increase @xmath59 from zero to the desired value , using slow acceleration @xmath190 rad / s@xmath26 ) to ensure adiabatic growth of a single folded sheet . in the metastable regime of superflow at @xmath191 , vortex formation can be triggered by irradiation with ionizing radiation . experiments with superfluid @xmath0he - b @xcite have shown that quantized vortex lines are formed in the aftermath of a neutron absorption event . according to current belief , vortex formation occurs via the kibble - zurek ( kz ) mechanism @xcite , which was originally developed to explain the phase transitions of the early universe . in this scenario a network of cosmic strings is formed during a rapid non - equilibrium second order phase transition , owing to thermal fluctuations . in @xmath0he the absorption of a thermal neutron causes heating which drives the temperature in a small volume of @xmath192 m size above the superfluid transition . subsequently the heated bubble cools back below @xmath58 with a thermal relaxation time of order 1@xmath13s . this process forms the necessary conditions for the kibble - zurek mechanism within the cooling bubble . it is interesting to note that so far this is the only case of vortex formation in the @xmath0he superfluids which in principle is not confined to the vicinity of the bounding solid walls . in practice the mean free path of thermal neutrons in liquid @xmath0he is only @xmath193 m and therefore even this process is localized close to the wall . the real experimental conditions in the neutron irradiation experiment of @xmath0he - b @xcite ( and also probably in the early universe ) do not coincide , with a perfectly homogeneous transition , as is assumed in the kz scenario : the temperature distribution within the cooling `` neutron bubble '' is nonuniform , the transition propagates as a phase front between the normal and superfluid phases , and the phase is fixed outside the bubble . these considerations require modifications to the original kz scenario @xcite and even raise concerns whether the kz mechanism is responsible for the defects which are extracted from the neutron bubble and observed in the experiment @xcite . new measurements demonstrate that a joint defect the combination of a conventional ( mass ) vortex and spin vortex is also formed and is directly observed in the neutron irradiation experiment @xcite . this strengthens the importance of the kz mechanism and places further constraints on the interplay between it and other competing effects . in superconductors localized heating can cause the unpinning of vortices from defects in the crystal lattice which can be viewed as a creation of an intermediate vortex ring . as a liquid free of alien impurities , he superfluids do not display bulk pinning . surface pinning at protuberances on the container wall remains an issue which has been studied in @xmath9he - ii @xcite . in @xmath0he superfluids vortex core diameters are at least two orders of magnitude larger and pinning correspondingly weaker . measurements of vortex dynamics have so far not resulted in reliable estimates for pinning parameters . in fact , it has not been settled whether surface pinning plays any observable role in the presence of smooth walls . all indications point in the direction that , even in @xmath0he - b with its smaller vortex core sizes , vortex motion occurs in the limit of weak pinning @xcite , where collective effects are expected to dominate in pinning . for vortex lines with continuous structure in @xmath0he - a pinning is expected to be unimportant . the virtual absence of pinning has a number of important consequences : \(i ) it allows investigation of vortex dynamics without pinning in the whole temperature range from @xmath194 to @xmath195 . as a result , in @xmath0he - b three topologically different contributions to vortex dynamics have been distinguished from one another , owing to their different temperature behavior @xcite . these are 1 ) the magnus force with which the flowing superfluid component acts on a vortex line , 2 ) the kopnin force which is caused by the spectral flow phenomenon @xcite and which acts on the vortex line when it moves with respect to the normal component , and 3 ) the iordanskii force which is the analog of the gravitational aharonov - bohm effect @xcite . \(ii ) it becomes possible to prepare surfaces with specially prepared pinning sites by micro - fabrication techniques . one can then study vortex dynamics when an array of vortex lines becomes commensurate with a pre - fabricated lattice of surface pinning sites . \(iii ) one can study the trapping and unpinning of circulation from a columnar defect in the order - parameter field namely a thin wire stretched across the superfluid bath parallel to the rotation axis . this is the vinen vibrating wire configuration where one quantum of circulation can be trapped around the wire by rotating the container @xcite . when rotation is stopped , the trapped circulation can be observed to peel off from the wire as a precessing vortex @xcite . with each revolution of the precessing vortex , the phase difference between the ends of the container slips by @xmath34 . this process can be interpreted as a macroscopic manifestation of the ac josephson effect @xcite . the corresponding josephson frequency is remarkably low , approximately 4mhz . \(iv ) unimpeded by pinning on solid walls , it becomes possible to study interactions of vortex lines with different types of interfaces which can be prepared in the he systems . these interfaces include the free surface of the superfluid bath ( i.e. the gas liquid interface ) @xcite , the superfluid solid @xmath0he interface @xcite , the interfaces between @xmath0he and @xmath9he superfluids @xcite , and interfaces between the a and b phases in superfluid @xmath0he @xcite . although superflow and quantized vortex lines have been the essence of superfluid investigations since the start in the late nineteen forties , nevertheless uniform rotation has not become an important tool for generating vorticity in @xmath9he - ii , owing to the uncontrolled release of remanent vorticity . when the construction work of the first rotating nuclear demagnetization cryostat was started in 1978 it was feared that rotation might not prove a useful concept in @xmath0he superfluidity either . however , today we know that the @xmath59 axis is as important in the study of the @xmath0he superfluids as the other experimental parameters @xmath196 , @xmath18 , or @xmath17 , which together control the various order - parameter structures : measurements on @xmath0he superfluids without access to rotation would be as limited as studies on superconductors without the possibility to turn on the magnetic field ! the central areas so far in superfluid @xmath0he work have been the identification of different topologically stable defect structures in the order - parameter field , the conditions and critical values which control their formation , and phase transitions between different structures as function of the external variables . for these questions the @xmath0he superfluids have been the most ideal system , owing to the large variability in its order - parameter response . today the largest collection of different theoretically characterized and experimentally identified quantized vortex structures are found in the @xmath0he superfluids . in the coming years @xmath0he work will focus more and more on making use of its ideal text - book - like properties , to employ the @xmath0he liquids as a model system in quantum field theory for the study of such varied questions as the physical vacuum , black hole , or inhomogeneity in the accelerating universe . for many questions of this category working theoretical @xmath0he analogues have already been constructed @xcite . ruutu , v.b . eltsov , a.j . gill , t.w.b . kibble , m. krusius , yu.g . makhlin , b. plaais , g.e . volovik , wen xu : nature * 382 * , 334 ( 1996 ) ; c. buerle , yu.m . bunkov , s.n . fisher , h. godfrin , g.r . pickett : nature * 382 * , 332 ( 1996 ) . parts , e.v . thuneberg , g.e . volovik , j.h . koivuniemi , v.h . ruutu , m. heinil , j.m . karimki , m. krusius : phys . lett * 72 * , 3839 ( 1994 ) ; e.v . thuneberg : physica b * 210 * , 287 ( 1995 ) ; m.t . heinil , g.e . volovik , physica b * 210 * , 300 ( 1995 ) ; . parts , v.m . ruutu , j.h . koivuniemi , m. krusius , e.v . thuneberg , g.e . volovik : physica b * 210 * , 311 ( 1995 ) t.a . tokuyasu , d.w . hess , j.a . sauls : phys . b * 41 * , 8891 ( 1990 ) ; j.a . sauls : adv . phys . * 43 * , 113 ( 1994 ) ; n. ltke - entrup , r. blaauwgeers , a. huxley , s. kambe , m. krusius , p. mathieu , b. plaais , y. simon : to be published j.c . davis , j. steinhauer , k. schwab , yu.m . mukharsky , a. amar , y. sasaki , r.e . packard : phys . lett . * 69 * , 323 ( 1992 ) ; g.g . ihas , o. avenel , r. aarts , r. salmelin , e. varoquaux , phys . 69 * , 327 ( 1992 ) ; _ ibid . _ * 70 * , 2114 ( 1993 ) . parts , v.m . ruutu , j.h . koivuniemi , yu.m . bunkov , v.v . dmitriev , m. fogelstrm , m. huebner , y. kondo , n.b . kopnin , j.s . korhonen , m. krusius , o.v . lounasmaa , p.i . soininen , g.e . volovik : europhys . * 31 * , 449 ( 1995 ) ; j. low temp . phys . * 107 * , 93 ( 1997 )

we describe a general framework to study covariant symmetric broadcasting maps for mixed qubit states . we explicitly derive the optimal superbroadcasting maps , achieving optimal purification of the single - site output copy , in both the universal and the phase covariant cases . we also study the bipartite entanglement properties of the superbroadcast states .

quant-ph0602125

`` information '' is by its nature _ broadcastable_. what happens when information is quantum and we need to distribute it among many users ? indeed , this may be useful in all situations where quantum information is required in sharable form , e. g. in distributed quantum computation @xcite , for quantum shared secrecy @xcite , and , generally , in quantum game - theoretical contexts @xcite . however , contrarily to the case of classical information , which can be distributed at will , broadcasting quantum information can be done only in a limited fashion . indeed , for pure states ideal broadcasting is equivalent to the so - called `` quantum cloning '' , which is impossible due to the well - known `` no - cloning '' theorem @xcite ( see also @xcite ) . the situation is more involved when the input states are mixed , since broadcasting can be achieved with an output joint state which is indistinguishable from the tensor product of local mixed states from the point of view of individual receivers . therefore , the no cloning theorem can not logically exclude the possibility of ideal broadcasting for sufficiently mixed states . in ref . @xcite it was proved that perfect broadcasting is impossible from a single input copy to two output copies for an input set of non mutually commuting density operators . this result was then considered ( see refs . @xcite and @xcite ) as an evidence of the general impossibility of broadcasting mixed states drawn from a non - commuting set in a more general scenario , where @xmath1 equally prepared input copies are broadcast to @xmath2 users . however , for sufficiently many input copies @xmath3 and sufficiently mixed input states the no - broadcasting theorem does not generally hold @xcite , and for input mixed states drawn from a noncommuting set it is possible to generate @xmath2 output local mixed states which are identical to the input ones , by a joint correlated state actually , as proved in ref . @xcite , it is even possible to partially purify the local state in the broadcasting process , for sufficiently mixed input states . such a process of simultaneous purification and broadcasting was named _ superbroadcasting_. for qubits , the fully covariant superbroadcasting channel that maximizes the output purity ( i. e. the length of the output bloch vectors of local states ) when applied to input pure states coincides with the optimal cloning map @xcite . the possibility of superbroadcasting does not increase the available information about the original input state , due to unavoidable detrimental correlations among the local broadcast copies , which do not allow to exploit their statistics ( a similar phenomenon was already noticed in ref . @xcite ) . essentially , superbroadcasting transfers noise from local states to correlations . from the point of view of single users , however , the protocol is a purification in all respects , and this opens new interesting perspectives in the ability of distributing quantum information in a noisy environment , and deserves to be analyzed in depth . for qubits , it has been shown that for universal superbroadcasting is possible with at least @xmath4 input copies @xcite . is this the absolute minimum number for superbroadcasting , or does it hold only for this particular set of input states ? in this paper we will show that , indeed , for equatorial mixed states of qubits the minimum number is @xmath5 . however , for smaller non - commuting sets of qubit states the possibility of superbroadcasting with only @xmath6 input copies is still an open problem ( for larger dimension @xmath7 it is possible to superbroadcast also for @xmath6 , see e.g. ref . @xcite ) . we want to point put that clearly there are limitations to superbroadcasting . the input state must be indeed sufficiently mixed , since pure states can not be broadcast by the no cloning theorem . however , states with a pretty high purity can sill be superbroadcast , e. g. for universally covariant superbroadcasting @xcite from @xmath4 to @xmath8 it is possible to superbroadcast states with a bloch vector length up to 0.787 ( 0.935 for the phase covariant case ) . one can achieve superbroadcasting with even higher input purity for increasing @xmath3 , approaching unit input bloch vector length in the limit of infinitely many input copies . there are also some limitations in the absolute number of output copies for which one can achieve superbroadcasting . e. g. in the universal case , for @xmath4 input copies one can superbroadcast up to @xmath9 output copies , for @xmath10 up to @xmath11 , and for @xmath12 up to infinitely many . the output purity is clearly decreasing versus @xmath13 . in this paper we will further analyze all limitations to superbroadcasting also for the case of equatorial input qubits . regarding the possibility of achieving superbroadcasting experimentally , the first route to explore is to use the same techniques as for purification @xcite , since the superbroadcasting map for @xmath2 generalizes the purification map using the same protocol @xcite . this transformation involves a measurement of the total angular momentum of the qubits , then an optimal werner cloning @xcite in the universal case ( or an optimal phase covariant cloning @xcite in the phase covariant case ) . another possibility is to use the methods of ref . @xcite in order to classify all possible unitary realizations , and then seek for experimentally achievable ones using current technology . in this paper we present phase covariant superbroadcasting , and also give a complete derivation of the universal superbroadcasting map , presented in ref . the two maps are derived in a unified theoretical framework . the paper is organized as follows . in section [ s : symm - broad ] we introduce some preliminary notions regarding symmetric covariant maps . in sects . [ s : univ ] and [ s : pc ] we give a complete derivation of the optimal broadcasting maps in the universal case and in the phase covariant case respectively . in sect . [ sec : concurrence ] we study the entanglement properties of the states of two copies at the output of the universal and the phase covariant broadcasting maps . finally , in sect . [ s : conc ] we summarize and comment the main results of this paper . at the end of the paper we report the details of the calculations needed to derive the results presented in three appendices . in this section we introduce in a unified theoretical framework some preliminary concepts that will be employed to describe covariant symmetric qubit broadcasting maps . these concepts will be then specified to the universal and the phase covariant cases in the subsequent sections . a main tool we will extensively use in deriving the optimal maps is the formalism of the choi - jamiokowski isomorphism @xcite between completely positive ( cp ) maps @xmath14 from states on the hilbert space @xmath15 to states on the hilbert space @xmath16 , and positive bipartite operators @xmath17 on @xmath18 . such an isomorphism can be specified as follows @xmath19,\ ] ] where @xmath20 is the non normalized maximally entangled state @xmath21 in @xmath22 , @xmath23 gives the identity transformation and @xmath24 denotes the transposition of the operator @xmath25 on the same basis @xmath26 used in the definition of @xmath20 . in terms of @xmath17 , the trace - preservation condition for the map @xmath14 reads @xmath27=\openone_\sh\,.\ ] ] suppose that the map @xmath14 is covariant under the action of a group @xmath28 . in this case the covariance property is reflected to the form of the operator @xmath17 by the following correspondence @xmath29=0.\ ] ] in the above expression @xmath30 and @xmath31 are the unitary representations of @xmath32 on the input and output spaces respectively , while @xmath33 denotes complex conjugation on the fixed basis @xmath26 . in this framework it is also possible to study group - invariance properties of the map @xmath14 in terms of the operator @xmath17 . in this case we have the following equivalences @xmath34=0,\ ] ] and @xmath35=0.\ ] ] the above expressions refer to invariance properties on the input and output spaces respectively . in the following we will consider maps @xmath36 from states of @xmath3 qubits to states of @xmath13 qubits , namely cp maps from states on @xmath37 to states on @xmath38 . we will consider in particular symmetric broadcasting maps , namely transformations where all receivers get the same reduced state . the figures of merit which are commonly used are invariant under permutations of the output copies , and this allows to assume that the output state of a broadcasting map is permutation invariant without loss of generality . moreover , since the input consists in @xmath3 copies of the same state , there is no loss of generality in requiring that the map is also invariant under permutations of the input copies . these two properties , according to eqs . ( [ invariance1 ] , [ invariance2 ] ) , can be recast as follows @xmath39=0,\qquad\forall\sigma,\tau , \label{perminv}\ ] ] where @xmath40 and @xmath41 are representations of the output and input copies permutations , respectively . notice that permutations representations are all real , hence @xmath42 . a useful tool to deal with unitary group representations @xmath30 of a group @xmath28 on a hilbert space @xmath15 is the wedderburn decomposition of @xmath15 @xmath43 where the index @xmath44 labels the equivalence classes of irreducible representations which appear in the decomposition of @xmath30 . the spaces @xmath45 support the irreducible representations and @xmath46 are the multiplicity spaces , with dimension @xmath47 equal to the degeneracy of the @xmath44-th irreducible representation . correspondingly the representation @xmath30 decomposes as @xmath48 where @xmath49 is shorthand for @xmath50 . by schur s lemma , every operator @xmath25 commuting with the representation @xmath30 in turn decomposes as @xmath51 in the case of permutation invariance , the so - called schur - weyl duality @xcite holds , namely the spaces @xmath46 for permutations of @xmath13 qubits coincide with the spaces @xmath45 for the representation @xmath52 of @xmath53 , where @xmath30 is the defining representation . in other words , a permutation invariant operator @xmath54 can act non trivially only on the spaces @xmath45 , namely it can be decomposed as @xmath55 the clebsch - gordan series for the defining representation of @xmath53 is well - known in the literature ( see for example @xcite ) , and its wedderburn decomposition is given by @xmath56 where @xmath57 , @xmath58 equals 0 for @xmath13 even , 1/2 for @xmath13 odd , and @xmath59 in the case of the broadcasting maps the hilbert space @xmath18 on which the operator @xmath17 acts , supports the two permutation representations corresponding to the output and input qubits permutations . therefore it can be decomposed as @xmath60 by rearranging the factors in the above expression , we can recast the decomposition in a more suitable way , namely @xmath61 the operator @xmath17 , in order to satisfy the permutation invariance property , according to eq . ( [ eq : schur - weyl - dual - form ] ) , can be written in the following form @xmath62 where the operators @xmath63 act on @xmath64 . moreover , in order to fulfill the requirements of trace preservation and complete positivity , the operators @xmath63 must satisfy the constraints @xmath65=\frac{\openone_{2l+1}}{d_j } , \label{cpt}\ ] ] where @xmath66 denotes the partial trace performed over the space @xmath67 in the @xmath68-th term of the decomposition in eq . , and @xmath69 is shorthand for @xmath70 . we have now all the tools to study symmetric qubits broadcasting devices . in this work we are interested in the case of covariant broadcasting maps , which in general have to fulfill also the following covariance condition under the representations @xmath71 and @xmath72 of a group @xmath28 ( see eq . ( [ eq3 ] ) ) @xmath73=0.\ ] ] the above condition gives a further constraint on the form of the operators @xmath63 in eq . ( [ symmbro ] ) . actually , the group @xmath31 is in general just a subgroup of the defining representation @xmath30 of @xmath53 , and therefore the representation @xmath74 acts non trivially only on the subspaces @xmath64 , which are the ones supporting the operators @xmath63 . in the next sections we will consider two interesting cases , namely @xmath75 and @xmath76 , corresponding to universal and phase covariant broadcasting respectively , and we will see how the form of the operators @xmath63 depends on the particular choice of the considered covariance group . in addition to the wedderburn decomposition and the related schur - weyl duality reviewed above , another useful tool we will extensively use in the following is a convenient decomposition of an @xmath3-partite state of the form @xmath77 , representing @xmath3 qubits all prepared in the same generic state @xmath78 . in appendix [ appendixaa ] we report the complete derivation of the following identity , which was originally presented in @xcite , @xmath79 for the sake of simplicity , in the above expression we considered density operators of the form @xmath80 , @xmath81 , namely qubit states whose bloch vector ( of length @xmath82 ) is aligned along the @xmath83 axis , and consequently the states @xmath84 are eigenstates of the operator @xmath85 in the @xmath68 representation , namely @xmath86 . notice actually that the _ total _ angular momentum component @xmath85 of @xmath3 qubits is clearly permutation invariant and therefore it can be written as @xmath87 we want to point out that the decomposition ( [ eq : rho - decomposition ] ) holds for any direction of the bloch vector , provided that the eigenstates of @xmath85 in eq . ( [ eq : rho - decomposition ] ) are replaced by the eigenvectors of the angular momentum component along the direction of the bloch vector in the single qubit state @xmath78 . we will prove in appendix c that the single - site output copy @xmath88 of a covariant broadcasting map commutes with the input density operator @xmath78 . in order to quantify the performance of the broadcasting map @xmath36 and to judge the quality of the single - site output density operator @xmath89 $ ] we will then evaluate the length @xmath90 of its bloch vector , namely @xmath91=r^\prime\ , . \label{singsitra}\ ] ] notice , moreover , that the length of a bloch vector @xmath82 is simply related to the purity @xmath92 $ ] of the density operator @xmath78 as @xmath92=(1+r^2)/2 $ ] . therefore , maximizing the output bloch vector length @xmath93 is equivalent to maximizing the output single - site purity . notice also that so far we can not exclude that the input and output bloch vectors @xmath82 and @xmath93 are antiparallel , and this just implies that @xmath90 can range from @xmath94 to @xmath95 . we will now show how to evaluate @xmath93 according to eq . ( [ singsitra ] ) , which is the main quantity that we will consider in the next sections in the particular cases of universal and phase covariant broadcasting . the trace in eq . can be computed by considering that the global output state @xmath96 of the @xmath13 copies is by construction invariant under permutations , hence @xmath97=\tr\left[\left(\sigma_z\otimes\openone^{\otimes m-1}\right)\sigma\right]=\frac1{m!}\tr\left[\sum_\sigma\pi_\sigma\left(\sigma_z\otimes\openone^{\otimes m-1}\right)\pi_\sigma \sigma\right]\,.\ ] ] the last term on the r.h.s . of eq . ( [ eq : rprime - valutaz ] ) contains a sum over the @xmath98 possible permutations of the @xmath13 output qubits . notice that @xmath99 where the operator @xmath100 acts as @xmath101 on the @xmath102-th qubit and identically on the remaining qubits . now , by exploiting the permutation invariance of @xmath103 , we can write @xmath104 and clearly @xmath105\,.\ ] ] the explicit expression of @xmath93 in the universal case will be derived in sect . [ s : univ ] . in the phase covariant case , we will see that it is more convenient to take @xmath78 diagonal on the @xmath106 eigenstates . the above formula in this case is just substituted by @xmath107\,\ ] ] and will be explicitly calculated in sect . [ s : pc ] . we want to stress that maximization of the figure of merit @xmath93 allows to optimize the fidelity criterion as well . in fact , for @xmath108 the two criteria coincide , whereas for @xmath109 one can always achieve unit fidelity by suitably mixing the output state with optimal @xmath93 and the maximally mixed one . on the other hand , direct maximization of fidelity is not analytically feasible , since fidelity is a concave function over the convex set of covariant maps , whence it is not maximized by extremal maps . finally , we want to mention that in the next sections we explicitly maximize the scaling factor for @xmath3 inputs and @xmath13 outputs @xmath110 , which can be referred to as _ shrinking factor _ or _ stretching factor _ , depending whether it is smaller or greater than 1 , respectively . it is obvious that this maximization is equivalent to maximizing @xmath93 . superbroadcasting corresponds to the cases where @xmath111 . in this section we will give the explicit derivation of the optimal universal broadcasting maps . starting from the general broadcasting map described in the previous section we have to impose in this case the additional constraint @xmath112=0\,,\ ] ] where @xmath30 is the defining representation of the group @xmath53 . for the defining representation @xmath30 the following property holds @xmath113 by exploiting such a property , the commutation relation ( [ univ1 ] ) can be written more conveniently as follows @xmath114=0\ , , \label{universales}\ ] ] where @xmath115 . the complete positivity and trace - preservation constraints in terms of the operator @xmath116 are then equivalent to @xmath117=\openone_\sh\,.\ ] ] upon defining @xmath118 , the constraints for complete positivity and trace preservation are now given by the following conditions on the operators @xmath119 @xmath120=\frac{\openone_{2l+1}}{d_j}.\ ] ] by exploiting the fact that the clebsch - gordan series for @xmath64 is just @xmath121 , we can write @xmath122 notice that this is not the wedderburn decomposition , since not all the subspaces @xmath123 support inequivalent representations . however , the wedderburn decomposition can be recovered by a suitable rearrangement that takes into account the repetitions of the same representation @xmath124 . using the decomposition we can formulate the constraint in terms of the operators @xmath119 as follows @xmath125 where , by complete positivity , the coefficients @xmath126 are real and positive , and @xmath127 is the projection of the space @xmath64 onto the @xmath124 representation , satisfying @xmath128=\frac{2j+1}{2l+1}\openone_{2l+1},\qquad\tr_l\left[p^j_{j , l}\right]=\frac{2j+1}{2j+1}\openone_{2j+1}.\ ] ] the set of projectors @xmath127 is clearly orthogonal . the trace - preservation constraint ( [ trace - pres - abovo ] ) can now be written as @xmath129 which is equivalent to the conditions @xmath130 along with the complete positivity constraint @xmath131 , eq . ( [ eq : polytope ] ) defines a convex polyhedron whose extremal points are classified by functions @xmath132 and @xmath133 @xmath134 the classification of symmetric universally covariant maps is then completely determined in terms of the vectors @xmath135 and @xmath136 , whose elements can range from @xmath58 to @xmath137 and from @xmath138 to @xmath139 , respectively . extremal maps then correspond to the following form for the operators @xmath116 @xmath140 the optimization of the figure of merit @xmath90 or , equivalently , of the scaling factor @xmath141 can be obtained by explicit calculation from eq . ( [ eq : rprime ] ) . the output state @xmath103 of the broadcasting map applied to an input state @xmath77 can be represented as @xmath142=\tr_\sh[\openone^{\otimes m}\otimes{\tilde\rho}^{\otimes n}\ s]\ , , \label{maponinp}\ ] ] where @xmath143 denotes the orthogonal complement of @xmath78 , which just corresponds to the change @xmath144 ( or , equivalently , @xmath145 ) . using the decomposition in eq . ( [ eq : rho - decomposition ] ) for @xmath146 and the form ( [ s ] ) for the operator @xmath116 , we can express eq . as follows @xmath147\otimes\openone_{d_{j_l}}.\ ] ] we can now use eq . ( [ eq : rprime ] ) to evaluate the scaling factor , namely @xmath148.\ ] ] in appendix [ appendixb ] we report the explicit calculation of @xmath149 and we show that it can be written in the following form @xmath150 where @xmath151 since @xmath152 , the sum @xmath153 in eq . ( [ eq : univ - pr - to - opt ] ) is always negative . therefore , the function @xmath149 is maximized by the choice of @xmath136 and @xmath135 minimizing @xmath154 , which clearly implies @xmath155 . the form of the coefficient @xmath156 for @xmath157 is given by @xmath158 whereas for @xmath159 we have @xmath160 in both cases @xmath154 is minimized by choosing the maximum value of @xmath135 , and therefore the maximum scaling factor is achieved by @xmath161 . for @xmath2 the optimal value of the figure of merit is then univocally determined by the value of the function @xmath154 @xmath162 while the optimal scaling factor is given by @xmath163 the corresponding output state takes the form @xmath164 where @xmath165 denote the clebsch - gordan coefficients . as mentioned in the previous section , in appendix [ appendixa ] we prove that the single - site reduced output state @xmath166 $ ] commutes with @xmath101 , hence @xmath149 is definitely a scaling factor . two interesting cases we will consider in the following are the ones with @xmath167 and @xmath168 , for which the scaling factor takes the explicit forms @xmath169 the function @xmath170 is plotted in fig . [ prplotz ] for @xmath3 ranging from 10 to 100 in steps of 10 . we can see that for a suitable range of values of @xmath82 the scaling factor is larger than one . this corresponds to a broadcasting process with an increased single - site purity at the output with respect to the input . this phenomenon occurs for @xmath171 . in this case @xmath141 is actually a stretching factor , and we call such a phenomenon superbroadcasting . the maximum value of @xmath82 such that it is possible to achieve superbroadcasting will be referred to as @xmath172 and it is a solution of the equation @xmath173 versus @xmath82 for the universal broadcasting , for @xmath167 and @xmath3 ranging from 10 to 100 in steps of 10 . notice that there is a wide range of values of @xmath82 such that @xmath174 , corresponding to superbroadcasting.,width=340 ] it is clear that the optimal scaling factor for fixed @xmath3 is a non increasing function of @xmath13 . actually , by contradiction , suppose that the map with @xmath175 output copies has a higher purity than the optimal map with @xmath13 copies . then one could trace over @xmath176 copies from the former map , and he would obtain a map with @xmath13 output copies with purity higher than the optimal , which is obviously absurd . this implies that in general @xmath177 , and for large values of @xmath176 superbroadcasting may not be possible anymore . the maximum @xmath13 such that superbroadcasting can be achieved for @xmath3 input copies will be referred to as @xmath178 . it turns out that , apart from the values @xmath179 for which we have @xmath180 and @xmath181 , for @xmath182 one has @xmath183 , namely superbroadcasting is possible for any number of output copies . in fig . [ rstar ] we report the values of @xmath184 and @xmath185 for @xmath186 . by a numerical analysis we have evaluated the power laws for the two curves , which turn out to be in good agreement with @xmath187 and @xmath188 , respectively . we want to point out that for input pure states ( @xmath189 ) only the term with @xmath190 in the expression ( [ s ] ) is significant . the optimal map then corresponds to the optimal universal cloning for pure states derived in @xcite . and @xmath185 for @xmath186 . the upper line , corresponding to @xmath191 , has a power law @xmath188 . the lower line , corresponding to @xmath167 , has a power law @xmath187.,width=340 ] in this section we study the case of symmetric phase covariant broadcasting , where we restrict our attention to input states lying on an equator of the bloch sphere , say the @xmath192-plane . the equatorial qubit density operator in this case has the explicit form @xmath193 . the starting point , as in the case of universal broadcasting , is the requirement of permutation invariance ( [ perminv ] ) for input and output copies , that leads to the form ( [ symmbro ] ) of the operator @xmath17 . moreover , in this case we demand covariance under the action of the group of rotations along the @xmath83-axis @xmath194 . we proceed analogously to the case of universal broadcasting . by imposing the covariance condition for the map we require invariance of the @xmath17 operator , namely @xmath195=0 $ ] . by exploiting the wedderburn decomposition ( [ eq : wedder - for - u ] ) for the operator @xmath196 , namely @xmath197 the phase covariance requirement corresponds to the following additional condition for the operators @xmath63 @xmath198=0,\qquad\forall j , l,\ ] ] where @xmath199 is defined according to eq . ( [ jz ] ) . a convenient form for the operators @xmath63 satisfying eq . ( [ eq : rjl - covariance ] ) is the following @xmath200 when @xmath201 , and @xmath202 when @xmath203 . notice that there are two more running indices with respect to the universal case ( [ onlyonerunning ] ) . the index @xmath204 in eq . ( [ eq : rjl1 ] ) simply allows for off - diagonal contributions in the operator @xmath63 , while we will see that the index @xmath205 , which labels equivalence classes , is related to the direction of the reduced output state bloch vector . in particular we will show that , in order to get an equatorial output , the operators @xmath63 have to be symmetric in @xmath205 , in the sense that @xmath206 . notice also that @xmath205 takes integer values when @xmath207 is even and half integer values when @xmath207 is odd . the trace - preservation condition ( [ cpt ] ) now reads @xmath208 and , analogously to the universal case , the fact that the operators @xmath63 are diagonal with respect to the indices @xmath68 s and @xmath205 s implies that the extremal points are classified by functions @xmath209 and satisfy @xmath210 we will now compute the output density operator and the scaling factor for @xmath0 phase covariant broadcasting maps . without loss of generality , let us now consider an input state @xmath78 oriented along the @xmath211-axis , namely @xmath212 . the density operator @xmath77 can then be decomposed , analogously to eq . ( [ eq : rho - decomposition ] ) , as @xmath213 where @xmath214 is the eigenvector of @xmath215 corresponding to the eigenvalue @xmath216 . in the following , eigenvectors without explicit specification of the superscript axis , such as @xmath84 , are intended to be along the @xmath83-axis , namely @xmath217 . according to eq . ( [ eq : choi - jam ] ) , the density operator @xmath103 on @xmath218 , describing the output state of the @xmath13 copies , can be written as @xmath219=\tr_\sh\left[\left(\openone^{\otimes m}\otimes\rho^{\otimes n}\right)r\right]\\ & = ( r_+r_-)^{n/2}\bigoplus_{j = j_0}^{m/2}\sum_{l , k}d_l\sum_{n , n'}\sum_{n''}r^{j , l}_{n , n',k}\left(\frac{r_+}{r_-}\right)^{n''}(w_l^\dag)_{n'',n}(w_l)_{n',n''}|j , n+k\>\<j , n'+k|\otimes\openone_{d_j } , \end{split}\ ] ] where @xmath220 are the entries of the wigner rotation matrix in the @xmath205 representation which rotates the @xmath83-components into the @xmath211-components in the usual notation ( the one that is found , for example , in @xcite ) such entries are denoted as @xmath221 . as discussed previously in sect . [ s : symm - broad ] , the projection @xmath93 along the @xmath211 axis ( [ eq : rprime - phase ] ) of the bloch vector of the single - site output state is a convex ( linear ) function on the convex set of phase covariant broadcasting maps , and therefore it achieves its maximum on extremal broadcasting maps . let the functions @xmath132 and @xmath222 denote an extremal map . hence , starting from eq . ( [ eq : rprime - phase ] ) and specializing eq . ( [ eq : phase - form - pr ] ) , derived in appendix b , to the extremal case @xmath132 and @xmath222 , we can express the scaling factor in the following form @xmath223_{n , n+1}\left[j_x^{(j_l)}\right]_{n+k_l , n+k_l+1},\ ] ] where @xmath224_{n , m}$ ] denotes the matrix element of the operator @xmath225 evaluated with respect to the eigenstates of @xmath199 , i. e. @xmath84 . the final form in eq . ( [ rprime - to - opt ] ) is now suitable to be optimized . first of all , since the matrix elements @xmath226_{n , m}$ ] are non - negative , the maximum purity is reached by maximizing the off - diagonal elements of @xmath63 , namely for rank - one @xmath63 with all the matrix elements equal to @xmath227 ( see eq . ( [ eq - mancante ] ) ) . we now want to identify the values of @xmath135 and @xmath228 corresponding to the optimal scaling factor of the map . the matrix elements of @xmath229 take the explicit form @xmath230_{n+k_l , n+k_l+1}=\frac{1}{2}\sqrt{j_l(j_l+1)-(n+k_l ) ( n+k_l+1)}\;.\ ] ] since the above matrix elements are maximized in the central block of the matrix , the optimal map is achieved by choosing @xmath228 as close as possible to zero , for all the values of @xmath231 . when @xmath207 is even , the optimal choice corresponds to @xmath232 for all @xmath231 . when @xmath207 is odd there are two equivalent possible choices for @xmath228 , namely @xmath233 , for each value of @xmath231 . moreover , we set @xmath161 for all @xmath231 , namely as large as possible . for @xmath207 even , the global output @xmath103 and the scaling factor are given by @xmath234_{n , n'}\left|\frac{m}{2},n\right\>\left\<\frac{m}{2},n'\right|,\\ & p^{n , m}_e(r)=\frac{4}{mr}(r_+r_-)^{n/2}\sum_{l = l_0}^{n/2}d_l\sum_{n =- l}^{l-1}\left[\exp\left(j_x^{(l)}\log\frac{1+r}{1-r}\right)\right]_{n , n+1}\left[j_x^{(m/2)}\right]_{n , n+1}. \end{split}\ ] ] for @xmath207 odd we have many more solutions , corresponding to all the possible combinations of @xmath235 for all values of @xmath231 . as will be clear in the following discussion , we will examine the two cases of @xmath236 and @xmath237 for all @xmath231 . in the former case we can write @xmath238_{n , n ' } \left|\frac m2,n+\frac 12\right\>\left\<\frac m2,n'+\frac 12\right|,\\ & p^{n , m}_o(r)=\frac{4}{mr}(r_+r_-)^{n/2}\sum_{l = l_0}^{n/2}d_l\sum_{n =- l}^{l-1}\left[\exp\left(j_x^{(l)}\log\frac{1+r}{1-r}\right)\right]_{n , n+1}\left[j_x^{(m/2)}\right]_{n+1/2,n+3/2 } , \end{split}\ ] ] while for @xmath237 we have @xmath239_{n , n ' } \left|\frac m2,n-\frac 12\right\>\left\<\frac m2,n'-\frac 12\right|,\\ & p^{n , m}_o(r)=\frac{4}{mr}(r_+r_-)^{n/2}\sum_{l = l_0}^{n/2}d_l \sum_{n =- l}^{l-1}\left[\exp\left(j_x^{(l)}\log\frac{1+r}{1-r}\right ) \right]_{n , n+1}\left[j_x^{(m/2)}\right]_{n-1/2,n+1/2}. \end{split}\ ] ] notice that , since @xmath226_{m , m+1}= \left[j_x^{(j)}\right]_{-(m+1),-m}$ ] and the same property holds for the matrix elements of any power of @xmath240 , the scaling factors corresponding to the extremal maps with @xmath236 and @xmath237 are exactly the same . this means that the bloch vector components in the @xmath192-plane are scaled in the same way by the two maps . we want to point out that an extremal map with @xmath241 generates output density operators with a non vanishing component of the bloch vector along the @xmath83 direction . actually , for the input state ( [ decpc ] ) the output single - site density operator is given by @xmath242 where @xmath243 optimal broadcasting maps , where the bloch vector is just scaled along its input direction , can then be obtained for odd values of @xmath207 by equally mixing the two maps considered above , corresponding to @xmath236 and @xmath237 . as mentioned earlier , since the two maps give the same scaling factor , their mixture does not compromise optimality . notice that the optimal broadcasting maps we have derived in this way are independent of the input state . in the limit of pure input states , the above maps coincide with the optimal phase covariant cloning for pure equatorial states presented in ref . @xcite . we will now discuss more quantitatively the results derived above . the optimal scaling factors , reported in eqs . ( [ eq : phase - pr1 ] ) and ( [ eq : phase - pr2 ] ) , contain only known terms and can be studied numerically . versus @xmath82 for the phase covariant broadcasting , for @xmath167 and @xmath3 ranging from 4 to @xmath244 in steps of 8.,width=340 ] it turns out that phase covariant superbroadcasting is possible even for @xmath5 , with @xmath245 . moreover , it is possible to superbroadcast an infinite number of output copies starting from @xmath4 ( @xmath183 for @xmath246 ) . as for the universal case , we can easily compute the function @xmath141 for @xmath167 and @xmath168 , which is monotone decreasing in @xmath13 . in fig . [ pr - n+1 ] we report the plots of @xmath170 for values of @xmath3 such that @xmath247 in steps of 8 , and for @xmath167 . in fig . [ rstarphase ] we report the plots of the values of @xmath184 and @xmath185 , as defined in the universal case . and @xmath185 in the phase covariant broadcasting , for @xmath248 . the upper line , corresponding to @xmath191 , has a power law @xmath249 . the lower line , corresponding to @xmath167 , has a power law @xmath250.,width=340 ] the upper line refers to the case @xmath251 and shows a behaviour like @xmath249 . the lower line is for @xmath252 and scales like @xmath253 . as before , in the limit of pure input states ( @xmath189 ) , the optimal phase covariant superbroadcasting map coincides with the optimal phase covariant cloner for qubits of ref . in this section we analyze the entanglement properties of the output state of the optimal @xmath0 broadcasting maps . notice that , since broadcasting maps are always optimized with @xmath161 , the output state is supported on @xmath254 ( which has multiplicity @xmath255 ) , namely the completely symmetric subspace of @xmath256 . therefore , also the reduced state of two qubits @xmath257 $ ] is symmetric . we will analyze in particular bipartite entanglement in the output state , which is conveniently described in terms of the concurrence @xcite @xmath258 where @xmath259 are the decreasingly - ordered eigenvalues of the operator @xmath260 , and @xmath261 . we will first consider the universal case , where the output state @xmath103 is diagonal on the @xmath85 basis . as shown in appendix [ appendixa ] , the state @xmath262 commutes with @xmath263 , and therefore it can be written as a linear combination of independent powers of @xmath264 , namely @xmath265 in the above expression the positivity and unit trace constraints are given by @xmath266 the eigenvalue @xmath267 of @xmath268 is always 0 , corresponding to the null component of @xmath262 on the singlet . by the unit trace condition we can express @xmath269 as a function of @xmath270 and @xmath154 as @xmath271 , and the positivity condition in terms of the two independent parameters @xmath272 is just @xmath273 the above inequality defines a triangle with basis @xmath274 $ ] and height @xmath275 $ ] , as shown in fig . [ concurr ] ( left ) . a state in @xmath276 is then completely determined by the couple @xmath272 . notice that the only pure states of the form ( [ eq : rho2 ] ) are @xmath277 , @xmath278 and @xmath279 , which correspond to the vertices @xmath280 , @xmath281 and @xmath282 respectively of the triangle in fig . [ concurr ] . we will now express the concurrence in terms of @xmath272 . since @xmath262 is real it follows that @xmath283 , and therefore we can write @xmath284 it is easy to verify from eq . ( [ eq : rho2 ] ) that @xmath285 corresponds to the couple @xmath286 . moreover , since @xmath262 and @xmath285 commute , the operator @xmath268 can be simply written as @xmath287 . by exploiting some algebra , and taking into account the identities @xmath288 and @xmath289 , we get the following expression @xmath290 > from the above expression , by using the unit trace constraint , we can compute the eigenvalues of @xmath268 @xmath291 notice that the first eigenvalue is doubly degenerate . the concurrence can then be written as follows @xmath292 the above equation defines a parabola inside the triangle ( [ eq : triangle ] ) of states . such a parabola separates the region of separable states from that of entangled states , shown in light and dark gray in fig . [ concurr ] respectively . in order to analyze the amount of bipartite entanglement in the broadcast states , we have then to evaluate the couple @xmath272 for the reduced state of two output copies and then determine in which region of the triangle it lies . using eq . ( [ alfabeto ] ) derived in appendix c , we can numerically evaluate @xmath272 for the universal double - site reduced output density operator @xmath257 $ ] . in fig . [ concurr ] we report the parametric plot for the case of 4 input and 5 output copies . as we can see , the black line moves towards positive @xmath154 as the bloch vector length @xmath82 of the input state goes from 0 to 1 . it is possible to see in the magnified plot on the right that , as @xmath82 gets close to 1 , i. e. the input state gets pure , the output exhibits bipartite entanglement , since it crosses the parabola . in the limit of pure input states , these results agree with the ones derived in ref . @xcite . in the phase covariant case it is not possible to carry on the same analysis , since , as we notice in appendix [ appendixa ] , the global output state does not commute with @xmath240 . however , using the partial traces in eqs . ( [ alfabeto ] ) , ( [ alfabeto2 ] ) , and ( [ alfabeto3 ] ) , it is still possible to evaluate the concurrence numerically . in fig . [ concuphase ] we report the plots of the entanglement @xmath293 defined in ref . @xcite as follows @xmath294 for @xmath295 and @xmath167 as a function of the input bloch vector length @xmath82 , both in the universal and phase covariant case . notice that , contrarily to what happens in the universal case , in the phase covariant case bipartite entanglement vanishes in the limit of pure input states . the absolute value of @xmath296 goes to zero for increasing number @xmath3 of input copies . in this paper we studied symmetric broadcasting maps , where @xmath3 input qubits initially prepared in the same mixed state are transformed into an output state of @xmath13 qubits , all described by the same density operator . we considered covariant maps and we investigated in particular the universally covariant case and the phase covariant case . we have shown that for sufficiently mixed initial states and for @xmath297 it is also possible to partially purify the single qubit output density operator in the broadcasting operation . such a new process was named superbroadcasting . the new superbroadcasting channels open numerous interesting theoretical problems . the first problem is to extend the map to any dimension @xmath7 and to different covariance groups . indeed , for special cases it is easy to see that increasing the dimensionality and/or reducing the set of states to be broadcast makes superbroadcasting possible with smaller @xmath3 , and even with @xmath6 input states . as a matter of fact , this is the case of universally covariant superbroadcasting from @xmath4 to @xmath298 , which can also be regarded as superbroadcasting for @xmath299 for special states of the form @xmath300 , and for the covariance group @xmath301 . the case of dimension @xmath299 is most interesting , since it can be exploited to improve entanglement for bipartite states of qubits . also the infinite dimensional case ( the so - called `` continuous variables '' ) turns out to be interesting , and easily feasible experimentally @xcite . it should be emphasized that for dimension @xmath302 there are many ways of increasing purity , and certainly the most interesting case is the purification along the mixing direction of a noisy channel ( notice that most channels do not correlate , whence the produced state is the tensor product of identical mixed states ) . another major problem is the analysis of the detrimental correlations between two outputs , e. g. to establish whether they are quantum or classical . these correlations are exotic , in the sense that instead of increasing the local mixing as usual , they reduce it . such mechanism is new , and deserves a more thorough analysis . an interesting issue , for example , is that they can not be erased leaving the local state unchanged ( the de - correlating map which sends a state to the tensor product of its partial traces is non linear ) , and this raises the problem of the optimal de - correlating channel , which optimizes the fidelity between the input and the output local state . such optimal channel can be derived using the same technique for optimal covariant maps used in the present paper . finally , distributing quantum information and in particular superbroadcasting raises the new problem of the trade - off between broadcasting and cryptographic security . indeed , on one side , the presence of many identical uses seems to open more possibilities of eavesdropping , however the detrimental correlations may drastically reduce such possibility , and the opportunity of detecting the eavesdropping on the joint output state may be exploited to increase the security . this work has been co - founded by the ec under the program secoqc ( contract no . ist-2003 - 506813 ) and by the italian miur through firb ( bando 2001 ) and prin 2005 . the global density operator @xmath77 is clearly invariant under permutations of the @xmath3 qubits , and , according to the schur - weyl duality , it can be represented by the wedderburn decomposition ( [ eq : schur - weyl - dual - form ] ) @xmath303 where @xmath304 is a state on @xmath67 . in order to evaluate @xmath304 it is sufficient to evaluate the matrix elements , where @xmath84 are eigenstates of @xmath85 in the @xmath68 representation , and @xmath305 is an arbitrary state in @xmath306 . for the sake of simplicity we can suppose that the state @xmath78 has the bloch form @xmath307 . the problem is now to choose @xmath305 in a suitable way . it turns out that a clever choice is given by @xmath308 where the subscript @xmath309 means that @xmath310 is a vector in the symmetric subspace of the first @xmath309 qubits , while @xmath311 is a singlet , supporting an invariant representation on a couple of qubit spaces . notice that in eq . the tensor product on the l.h.s . refers to the `` abstract '' subspace @xmath312 in the wedderburn decomposition , whereas the one on the r.h.s . refers to the decomposition @xmath313 grouping separately the first @xmath309 qubits and the remaining @xmath314 . moreover , since the chosen density operator @xmath77 commutes with the total @xmath85 , given by the following expression @xmath315 then @xmath304 commutes with @xmath199 , which implies that @xmath304 is diagonal on the eigenstates @xmath84 of @xmath199 . therefore we can write @xmath316 since @xmath310 is symmetric , it is a linear combination of factorized vectors with @xmath317 qubits in the @xmath318 state and @xmath319 in the state @xmath320 . as a consequence we can also write @xmath321 where @xmath322 . by analogous arguments it follows that @xmath323 the matrix element @xmath324 has the following expression @xmath325 and the decomposition of @xmath77 is finally given by @xcite @xmath326 notice that this expression exhibits a singularity for @xmath189 due to the rearrangement of terms ( [ rearr ] ) . however , a finite limit for @xmath327 exists , as it can be seen from the equivalent expression @xmath328 which exhibits no singularities . in this appendix we will derive the explicit form of the scaling factor for the universal and phase covariant cases . in order to calculate @xmath149 we start rewriting eq . ( [ eq : to - calculate ] ) as @xmath329=\tr\left[|ln\>\<ln|\tr_j\left[\left(j_z^{(j)}\otimes\openone_{2l+1}\right)\ p^{j}_{j , l}\right]\right].\ ] ] let us define @xmath330,\ ] ] where @xmath331 , and @xmath332 . since @xmath333 the set @xmath334 transforms according to @xmath335\\ & = \tr_{j}\left[\left(\sum_{k=-1}^1(u^{(1)}_g)_{ik}j_k^{(j)}\otimes\openone_{2l+1}\right)\ p^{j}_{j , l}\right]\\ & = \sum_{k=-1}^1(u^{(1)}_g)_{ik}x_k^{(l)},\\ \end{split}\ ] ] and we conclude that @xmath336 is an irreducible tensor set . it can then be proved by the wigner - eckart theorem that @xmath337 , and in particular @xmath338 . from the last relation and from the identity @xmath339 where @xmath340 and @xmath341 , and @xmath342 , we have @xmath343\\ & = \sum_{i=-1}^1a_i\tr\left[\left(j_i^{(j)}\otimes j_{-i}^{(l)}\right)p^j_{j , l}\right ] , \end{split}\ ] ] by using the well known identity @xmath344 we can write the explicit form of the coefficient @xmath270 @xmath345 by using the above expression we finally have @xmath346=n\frac{2j+1}{2l+1}\beta(j , j , l).\ ] ] substituting the global output state @xmath103 given by eq . ( [ eq : form - for - sigma ] ) into eq . ( [ eq : rprime - phase ] ) , it is possible to compute the scaling factor @xmath149 as @xmath347_{n'+k , n+k}\times\\ & \sum_{n''=-l}^l(w_l)_{n',n''}\left(\frac{r_+}{r_-}\right)^{n''}(w_l^\dag)_{n'',n } , \end{split } \label{app1}\ ] ] where @xmath220 are entries of the wigner rotation matrix in the @xmath205 representation which rotates the @xmath83-components to the @xmath211-components ( in the usual notation , the one that is found , for example , in @xcite , such entries are denoted as @xmath221 ) . in eq . ( [ app1 ] ) the sum over @xmath348 gives the @xmath349-th matrix element of @xmath350 . therefore we can write @xmath351_{n',n}\left[j_x^{(j)}\right]_{n'+k , n+k}\\ & = \frac{4}{mr}(r_+r_-)^{n/2}\sum_{j , l , k}d_jd_l\sum_{n =- l}^lr^{j , l}_{n , n+1,k}\left[\exp\left(j_x^{(l)}\log\frac{1+r}{1-r}\right)\right]_{n , n+1}\left[j_x^{(j)}\right]_{n+k , n+k+1 } , \end{split}\ ] ] where in the last equality we used the fact that @xmath352 has non - null matrix elements only on the second - diagonals , and we multiplied the second line by a factor 2 considering in the sum only one of the two second - diagonals . notice that all matrix elements in the previous equations are calculated with respect to the @xmath83-oriented basis @xmath353 and @xmath354 . in this appendix we want to calculate the following partial trace @xmath355,\ ] ] where the operator to be partially traced acts on @xmath356 , and @xmath84 are eigenstates of @xmath199 , as usual . in order to do that , we first decompose the vector @xmath84 into its components onto @xmath357 using the clebsch - gordan coefficients @xmath358 and then trace the operator @xmath359 over @xmath360 . in this way we get @xmath361=\frac{\openone}{2}+\frac{m}{2j}\sigma_z.\ ] ] we now recall a fact related to the already mentioned schur - weyl duality , by which multiplicity spaces @xmath306 in the wedderburn decomposition ( [ eq : wedderburn - for - su2 ] ) support irreducible representations of the permutation group @xmath362 of @xmath13 qubits . hence , for any operator @xmath363 on @xmath312 one has @xmath364\otimes\openone_{d_j}.\ ] ] for convenience , let us write @xmath365 as we already did in eq . ( [ eq : clever - choice ] ) . with this choice , we get : @xmath366&=\tr_{m-1}\left[\frac{d_j}{m!}\sum_\sigma\pi_\sigma^m\left(|j , m\>\<j , m|\otimes|\psi^-\>\<\psi^-|^{\otimes\frac{m-2j}{2}}\right)\pi_\sigma^m\right]\\ & = \frac{d_j}{m!}\left[(m-1)!(m-2j)\frac{\openone}{2}+(m-1)!2j\left(\frac{\openone}{2}+\frac{m}{2j}\sigma_z\right)\right]\\ \end{split}\ ] ] the first term in the sum comes from excluding from trace one of the @xmath367 qubits in singlet state . the second term in the sum comes from excluding from trace one of the @xmath309 qubits in @xmath84 state , and from eq . ( [ partial - trace ] ) . rearranging the above equation , we get the final expression @xmath368=d_j\left(\frac{\openone}{2}+\frac{m}{m}\sigma_z\right).\ ] ] consider the global output state @xmath369 . if it commutes with the total angular momentum component along the direction @xmath83 , for example , then it is simple to prove that also @xmath370 $ ] commutes with @xmath101 . in fact , from eq . ( [ eq : final - partial - trace ] ) , it is simple to see that @xmath371\propto\sigma_z,\ ] ] for all @xmath68 , and consequently also @xmath372\propto\sigma_z$ ] . in the universal case , the global output @xmath103 is diagonal on eigenstates of @xmath373 , hence @xmath370 $ ] commutes with @xmath101 , according to previous arguments . in the phase covariant case , it is more difficult to prove on general grounds that @xmath374=0 $ ] , since in the phase covariant case @xmath375\neq 0 $ ] . the simplest thing we can do is to compute the partial trace of eqs . ( [ eq : phase - pr1 ] ) and ( [ eq : phase - pr2 ] ) using again clebsch - gordan coefficients ( [ eq : cg - coeff ] ) . first of all let us notice that , tracing over @xmath376 qubits , only the terms with @xmath377 contribute . among these , the terms with @xmath378 give one factor proportional to @xmath379 and one factor proportional to @xmath101 , whereas terms with @xmath380 contribute with factors proportional to @xmath106 , since the matrix of coefficients @xmath381 is symmetric , see eq . ( [ eq_che_manca ] ) . then , posing @xmath382 without loss of optimality , @xmath166 $ ] commutes with @xmath106 . we will show here how to compute the reduced output state of two copies , which is used in section [ sec : concurrence ] to compute the concurrence between two of the @xmath13 clones . clearly it does not matter which two clones we are considering , since the global output state is permutation invariant . using clebsch - gordan coefficients , it is possible to decompose a vector in @xmath67 into its components onto @xmath383 @xmath384 and then to compute the partial trace of @xmath385 over @xmath386 . for @xmath387 we have @xmath388=\begin{pmatrix } \frac{(m-2m)(m-2m-2)}{4m(m-1)}&0&0\\ 0&\frac{m^2 - 4m^2}{2m(m-1)}&0\\ 0&0&\frac{(m+2m)(m+2m-2)}{4m(m-1 ) } \end{pmatrix},\ ] ] when @xmath389 @xmath390=\\ & \begin{pmatrix } 0&\frac{(m-2m-2)\sqrt{(m-2m)(m+2m+2)}}{2\sqrt{2}m(m-1)}&0\\ \frac{(m-2m-2)\sqrt{(m-2m)(m+2m+2)}}{2\sqrt{2}m(m-1)}&0&\frac{(m+2m)\sqrt{(m-2m)(m+2m+2)}}{2\sqrt{2}m(m-1)}\\ 0&\frac{(m+2m)\sqrt{(m-2m)(m+2m+2)}}{2\sqrt{2}m(m-1)}&0 \end{pmatrix } , \end{split}\ ] ] and @xmath391 , @xmath392=\\ & \begin{pmatrix } 0&0&\frac{\sqrt{(m-2m)(m+2m+4)(m+2m+2)(m-2m-2)}}{4m(m-1)}\\ 0&0&0\\ \frac{\sqrt{(m-2m)(m+2m+4)(m+2m+2)(m-2m-2)}}{4m(m-1)}&0&0 \end{pmatrix}. \end{split}\ ] ] for @xmath393 , partial trace over @xmath394 copies gives null contribution . 99 l. grover , quant - ph/9704012 ; j i cirac , a ekert , s f huelga , and c. macchiavello , phys . a * 59 * , 4249 ( 1999 ) . c crepeau , d gottesman , and a smith , in proceedings of the 34th annual acm symposium on theory of computing ( new york , acm press , 2002 ) . s c benjamin and p m hayden , phys . a * 64 * , 030301 ( 2001 ) . w k wootters , w h zurek , nature * 299 * , 802 ( 1982 ) . d dieks , phys . a , * 92 * , 271 ( 1982 ) . h p yuen , phys . a*113 * 405 ( 1986 ) . in ref . @xcite it was shown that the cloning machine violates the superposition principle , which applies to a minimum total number of _ three _ states , and hence does not rule out the possibility of cloning _ two _ nonorthogonal states . it is violation of unitarity that makes cloning any _ two _ nonorthogonal states impossible , as proved in ref . @xcite . in reference @xcite it was shown that the proposal for superluminal communication does not work , proving the impossibility of cloning in this particular context due to linearity of evolution . according to ref . @xcite , previous to refs . @xcite the anonymous referee s report of g. ghirardi to ref . @xcite contained an argument which was a special case of the no - cloning theorem of refs . @xcite . more recently , after several attempts of determining the wave function of a single system appeared in the literature , ref . @xcite showed how it is impossible to determine the wave function from a single copy of the system , and connected such impossibility to the no - cloning theorem . g m dariano and h p yuen , phys . lett . * 76 * 2832 ( 1996 ) . n herbert , found . phys . * 12 * , 1171 ( 1982 ) . a peres , fortsch . phys . * 51 * , 458 ( 2003 ) . h barnum , c m caves , c a fuchs , r jozsa , and b schumacher , phys . 76 * 2818 ( 1996 ) . r clifton , j bub , and h halvorson , found . of phys . * 33 * 1561 ( 2003 ) . g m dariano , c macchiavello , and p perinotti , phys . lett . * 95 * , 060503 ( 2005 ) . r f werner , phys . rev . a * 58 * , 1827 ( 1998 ) . m. keyl and r. f. werner , ann . h. poincar , * 2 * , 1 ( 2001 ) . g m dariano , p perinotti , and m f sacchi , quant - ph/0602037 ; g m dariano , p perinotti , and m f sacchi , quant - ph/0601114 . j i cirac , a k ekert , and c macchiavello , phys . * 82 * , 4344 - 4347 ( 1999 ) . f buscemi , g m dariano , c macchiavello , and p perinotti , in proceedings of the 13th quantum information technology symposium ( qit13 ) ( sendai , japan , 2005 ) . f buscemi , g m dariano , and m f sacchi , phys . a * 68 * , 042113 ( 2003 ) . a jamiokowski , rep . phys . * 3 * , 275 ( 1972 ) ; m - d choi , lin . alg . appl . * 10 * , 285 ( 1975 ) . w fulton and j harris , _ representation theory : a first course _ ( springer - verlag , berlin , 1991 ) . a r edmonds , _ angular momentum in quantum mechanics _ ( princeton university press , princeton , 1960 ) . v buek , m hillery , and r f werner , phys . a * 60 * , r2626 ( 1999 ) . g m dariano and c macchiavello , phys . a * 67 * , 042306 ( 2003 ) . f buscemi , g m dariano , and c macchiavello , phys . rev a * 72 * , 062311 ( 2005 ) . w k wootters , phys . rev . * 80 * , 2245 ( 1998 ) . d bru and c macchiavello , found . phys . * 33 * ( 11 ) , 1617 ( 2003 ) .

qcd is constructed as a lattice gauge theory in which the elements of the link matrices are represented by non - commuting operators acting in a hilbert space . the resulting quantum link model for qcd is formulated with a fifth euclidean dimension , whose extent resembles the inverse gauge coupling of the resulting four - dimensional theory after dimensional reduction . the inclusion of quarks is natural in shamir s variant of kaplan s fermion method , which does not require fine - tuning to approach the chiral limit . a rishon representation in terms of fermionic constituents of the gluons is derived and the quantum link hamiltonian for qcd with a gauge symmetry is expressed in terms of glueball , meson and constituent quark operators . the new formulation of qcd is promising both from an analytic and from a computational point of view .

hep-th9704106

solving qcd is among the most challenging problems in theoretical physics . a non - perturbative formulation of qcd is provided by wilson s lattice gauge theory @xcite , which maps the problem to one of classical statistical mechanics . over the past twenty years a variety of tools have been developed to solve lattice field theories . at present the most powerful tool is the monte carlo simulation of the partition function of the corresponding classical statistical mechanics system . however , the most efficient numerical algorithms for solving lattice qcd suffer from critical slowing down when the continuum limit is approached and thus exhaust even the biggest supercomputers . the use of improved or even `` perfect '' actions may alleviate this problem @xcite , because it may allow one to extract continuum physics from rather coarse lattices . on the other hand , to obtain high - precision numerical data using the standard numerical techniques will still require large resources of computer power . hence , it is certainly reasonable to look for new formulations of the qcd problem . here we propose an alternative non - perturbative approach to qcd in the framework of quantum link models . such models were first discussed by horn @xcite , and studied in more detail by orland and rohrlich @xcite . in these models the classical statistical mechanics problem of standard lattice gauge theory is replaced by a problem of quantum statistical mechanics . in particular , the classical euclidean action is replaced by a hamilton operator . as a consequence , the elements of the link matrices that are ordinary c - numbers in the standard formulation of lattice gauge theory , turn into non - commuting operators acting in a hilbert space . recently , quantum link models have been formulated with a fifth euclidean direction and it has been shown how they are related to ordinary 4-d yang - mills theories via dimensional reduction @xcite . the previous work on quantum link models was limited to @xmath0 and @xmath1 gauge groups and it was not clear how to generalize the construction to @xmath2 . the quantum link models constructed in this paper have an @xmath2 gauge symmetry and provide a new non - perturbative formulation of qcd . we hope that this formulation will lead to alternative ways for attacking this long - standing problem , both with analytic and with numerical methods . quantum link qcd is quite different from the standard formulation of lattice qcd . the main novel feature of quantum link models is that the gluonic hilbert space on each link is finite . while the hilbert space in the hamiltonian formulation of standard lattice qcd contains all representations of @xmath3 on a link , the corresponding hilbert space of an @xmath2 quantum link model consists of a single representation of @xmath4 . this is achieved by formulating the theory in @xmath5 dimensions . in order to represent full qcd , of course , quarks must also be included . even in the standard formulation of lattice gauge theory , the hilbert space of the quarks is finite and indeed the representation of quarks in quantum link qcd is not very different . however , the boundary conditions of the quarks in the fifth euclidean direction must be carefully adjusted in order to ensure their proper dimensional reduction . antiperiodic boundary conditions do not work , because they do not lead to a long enough fermionic correlation length in four dimensions . periodic boundary conditions , on the other hand , allow an arbitrarily long 4-d fermionic correlation length . the resulting 4-d fermion resembles in many respects the standard lattice wilson fermion . in particular , with periodic boundary conditions for the quarks in the fifth direction , quantum link qcd suffers from a fine - tuning problem , when one wants to approach the chiral limit . fortunately , we can do better than that . kaplan proposed to represent 4-d chiral fermions by working with a domain wall in a five - dimensional world @xcite . in his construction the fifth dimension arises for completely different reasons than in quantum link models . however , since we already have five dimensions , it is natural to use them also to protect the chiral symmetries of the fermions . the scheme that fits most naturally with quantum link models is shamir s variant @xcite of kaplan s proposal , which works with a 5-d slab ( which is finite in the fifth direction ) with open boundary conditions for the fermions at the two boundaries . due to the boundary conditions , a left - handed fermion is bound to one side of the slab , while a right - handed fermion appears on the opposite side . remarkably , the correlation length of these fermions is exponentially large in the depth of the slab . this is exactly what one needs for them to survive the dimensional reduction in quantum link models . the corresponding geometry is shown in fig.1 . when shamir s boundary conditions are used for the quarks in quantum link qcd , no fine - tuning is necessary to approach the chiral limit of the theory . this is essential , because recovering the chiral properties of the quarks is one of the main problems in numerical simulations of wilson fermions . recently , using the standard formulation of lattice gauge theory , it was demonstrated convincingly that shamir s proposal is indeed practical from a computational point of view @xcite and that it leads to a great improvement of the chiral properties of quarks on the lattice . it is reassuring that the formulation of quantum link models in five dimensions fits naturally with this elegant solution of the fermion doubling problem in vector - like theories including qcd . should similar ideas ultimately lead to a lattice construction of chiral gauge theories , one may expect that the theory can also be formulated as a quantum link model . once the lattice chiral fermion problem is solved , the door may be open to a non - perturbative lattice formulation of supersymmetry . in that case quantum link models may offer the most natural framework , because they treat the hilbert spaces of bosons and fermions on an equal footing . in particular , in quantum link models the bosonic hilbert space is also finite ( on a finite lattice ) . due to the fact that the gluonic hilbert space of quantum link qcd is finite , one can reformulate the theory in terms of fermionic constituents of the gluons , which we call rishons following jewish tradition @xcite . the rishons carry color and transform in the fundamental representation of @xmath2 . each quantum link variable can be expressed as a rishon - anti - rishon pair . the rishon dynamics is particularly simple . the hamilton operator describes the hopping of rishons from one end of a link to the other , as illustrated in fig.2 . the rishon formulation of quantum link qcd resembles the schwinger boson and constraint fermion constructions of quantum spin systems @xcite and provides theoretical insight , that may lead to new analytic approaches to qcd , for example , in the large @xmath6 limit . the hamilton operator of quantum link qcd with a @xmath0 gauge group can also be expressed in terms of glueball , meson and constituent quark operators . these objects consist of two rishons , two quarks and a quark - rishon pair , respectively . it is conceivable that the theory can be bosonized in terms of these degrees of freedom . if large @xmath6 techniques can be applied successfully , major progress in the non - perturbative solution of qcd and other interesting field theories is to be expected . it is intriguing that @xmath0 quantum link qcd has two representations one exclusively in terms of color - triplet quark and rishon fermions , the other in terms of color - singlet bosons . in contrast to standard lattice gauge theories , the configuration space of quantum link models is discrete and their path integral representation resembles that of quantum spin models . hence , it seems plausible that highly efficient cluster algorithms similar to those for quantum spins @xcite become available for quantum link models . for classical spin models the development of cluster algorithms @xcite gave rise to a tremendous improvement of numerical data and has led to a satisfactory high - precision numerical solution of various models . despite many attempts , it has so far not been possible to construct efficient cluster algorithms in the standard formulation of lattice qcd . the 2-d classical o(3 ) model is in many respects similar to 4-d yang - mills theories . the most efficient way to simulate the @xmath7 model is provided by the wolff cluster algorithm @xcite . on the other hand , the same model can also be obtained from the 2-d antiferromagnetic quantum heisenberg model via dimensional reduction @xcite . for the heisenberg model a very efficient loop cluster algorithm is available @xcite , which has recently been modified to work directly in the euclidean time continuum @xcite . this algorithm has been used to simulate the correlation length of the 2-d classical o(3 ) model that results from dimensional reduction @xcite . the quality of the numerical data is compatible with those obtained from the wolff algorithm applied directly to the 2-d classical model . for example , using a finite size scaling technique , correlation lengths of up to about 37000 lattice spacings could be handled successfully . this suggests that , if a cluster algorithm can be constructed for quantum link qcd , its numerical simulation should be more efficient than that of standard lattice qcd , for which no cluster algorithm has been found . at present , it is unclear if superior numerical methods can be developed in the quantum link formulation of lattice qcd . however , this seems possible , because , for example , for a @xmath8 quantum link model a cluster algorithm operating in the continuum of the fifth euclidean direction has already been constructed @xcite . quantum link models with quarks resemble standard models of quantum statistical mechanics , as , for example , the hubbard model . this may eventually lead to problems with their numerical simulation , because such models are known to suffer from the notorious fermion sign problem . however , based on a fermion cluster algorithm @xcite , a recently proposed method @xcite may circumvent this difficulty . however , we postpone further considerations of numerical simulations of quantum link qcd with dynamical quarks . the main issue of this paper is to present the quantum link formulation of qcd . the paper is organized as follows . in section 2 quantum link models with @xmath0 and @xmath2 gauge symmetries are constructed and in section 3 it is discussed how these models are related to ordinary 4-d yang - mills theories via dimensional reduction . in section 4 quantum link models are reformulated in terms of rishons the fermionic constituents of the gluons . the inclusion of quarks is discussed in section 5 within the framework of shamir s variant of kaplan s fermion method . the representation of the @xmath0 gauge invariant quantum link hamiltonian , in terms of glueball , meson and constituent quark operators , is worked out in section 6 . finally , section 7 contains our conclusions . quantum link models were first discussed by horn @xcite . he constructed models with @xmath0 and @xmath1 gauge groups , but he pointed out that his construction can not be generalized to @xmath2 . further , the recent application of quantum link models to yang - mills theory via dimensional reduction was discussed only for @xmath1 @xcite . here we explicitly construct quantum link models with an @xmath2 gauge symmetry . obviously , this is essential for the quantum link formulation of qcd with an @xmath3 gauge group . let us first recall the standard formulation of lattice gauge theory . in that case there is an @xmath2 matrix @xmath9 associated with each link @xmath10 on a 4-d hypercubic lattice . the action is given by @xmath11 = - \sum_{x,\mu \neq \nu } \mbox{tr } [ u_{x,\mu } u_{x+\hat\mu,\nu } u_{x+\hat\nu,\mu}{^\dagger}u_{x,\nu}{^\dagger}],\ ] ] where the dagger denotes hermitean conjugation . the action is invariant under @xmath2 gauge transformations @xmath12 where @xmath13 is the vector of hermitean generators of @xmath2 with the usual commutation relations @xmath14 = 2 i f_{abc } \lambda^c , \ \mbox{tr } \lambda^a \lambda^b = 2 \delta^{ab}.\ ] ] the path integral is given by @xmath15),\ ] ] where @xmath16 is the non - abelian gauge coupling . formally , the above system can be viewed as one of classical statistical mechanics . then the action plays the role of the classical hamilton function and @xmath17 plays the role of the temperature . the 4-d @xmath2 lattice gauge theory is believed to have only one phase , in which the gluons are confined . due to asymptotic freedom , the continuum limit of the lattice model corresponds to @xmath18 . let us now construct a quantum link representation of @xmath2 lattice gauge theory . we replace the classical action ( or hamilton function in the language of statistical mechanics ) by a quantum hamilton operator @xmath19.\ ] ] here the elements of the @xmath20 link matrices @xmath21 are operators acting in a hilbert space not just c - numbers like in the standard formulation . naturally , the dagger now represents hermitean conjugation in both the hilbert space and the @xmath20 matrix space . note , however , that the trace in the above expression is only over the @xmath20 matrix space and not over the hilbert space . gauge invariance of the quantum link model requires that the above hamilton operator commutes with the generators @xmath22 of infinitesimal gauge transformations at each lattice site @xmath23 , which obey the standard local algebra @xmath24 = 2 i \delta_{xy } f_{abc } g^c_x.\ ] ] gauge covariance of a quantum link variable follows by construction if @xmath25 where @xmath26 is the unitary operator that represents a general gauge transformation in hilbert space . the above equation implies the following commutation relations @xmath27 = \delta_{x , y+\hat\mu } u_{y,\mu } \vec \lambda - \delta_{x , y } \vec \lambda u_{y,\mu}.\ ] ] in order to satisfy these relations we introduce @xmath28 where @xmath29 and @xmath30 are generators of right and left gauge transformations of the link variable @xmath21 . suppressing the link index @xmath10 , their commutation relations take the form @xmath31 = 2 i f_{abc } r^c , \ [ l^a , l^b ] = 2 i f_{abc } l^c , \ [ r^a , l^b ] = 0,\ ] ] i.e. @xmath32 and @xmath33 generate an @xmath34 algebra on each link . the @xmath32 and @xmath33 operators associated with different links commute with each other . the commutation relations of eq.([gucommutator ] ) imply @xmath35 = u \vec\lambda , \ [ \vec l , u ] = - \vec\lambda u.\ ] ] the above commutation relations are identical to those of the standard hamiltonian formulation of lattice gauge theories . there they are the canonical commutation relations between link variables , which play the role of coordinates , and electric field variables @xmath32 and @xmath33 , which play the role of canonical conjugate momenta . in the standard formulation the commutation relations are realized in an infinite dimensional hilbert space of square integrable wave functionals of the link variables . this hilbert space is necessarily infinite dimensional , because one insists that the elements of a link matrix commute with each other . quantum link models , on the other hand , have a finite dimensional hilbert space for each link , by allowing the elements of the link fields to have non - zero commutators . it is important to realize that this does not destroy the gauge symmetry of the problem . for each link the above relations can be realized by using the generators of an @xmath4 algebra , with the @xmath36 algebra embedded in it . in the fundamental representation of @xmath4 , for example , we can write @xmath37 the @xmath38 are a set of @xmath39 matrices ( one for each index pair @xmath40 ) of size @xmath20 with @xmath41 . the commutators of real and imaginary parts of the elements of the link matrices take the form @xmath42 = [ \mbox{im } u_{ij},\mbox{im } u_{kl } ] = - i ( \delta_{ik } \ \mbox{im } \vec \lambda_{jl } \cdot \vec r + \delta_{jl } \ \mbox{im } \vec \lambda_{ik } \cdot \vec l ) , \nonumber \\ & & [ \mbox{re } u_{ij},\mbox{im } u_{kl } ] = i ( \delta_{ik } \ \mbox{re } \vec \lambda_{jl } \cdot \vec r - \delta_{jl } \ \mbox{re } \vec \lambda_{ik } \cdot \vec l + \frac{2}{n } \delta_{ik } \delta_{jl } t ) . \end{aligned}\ ] ] in particular , @xmath43 = [ ( u{^\dagger})_{ij},(u{^\dagger})_{kl } ] = 0 $ ] . again , we should emphasize that the commutation relations are local , namely all commutators between operators assigned to different links are zero . the additional generator @xmath44 is given by @xmath45 the real and imaginary parts of the @xmath39 matrix elements @xmath46 are represented by @xmath47 hermitean generators of @xmath4 . together with the @xmath48 generators @xmath32 and @xmath33 of right and left gauge transformations and the generator @xmath44 , these are the @xmath49 generators of @xmath4 . the generator @xmath44 commutes with the generators of @xmath2 gauge transformations , i.e. @xmath50 = [ t,\vec l ] = 0,\ ] ] but not with the elements of the link matrices , because @xmath51 = 2 u , \ [ t , u{^\dagger } ] = - 2 u{^\dagger}.\ ] ] this relation implies that @xmath52 generates an additional @xmath8 gauge transformation , i.e. @xmath53 indeed the hamilton operator of eq.([hamiltonian ] ) is also invariant under the extra @xmath8 gauge transformations and thus describes a @xmath0 lattice gauge theory . the @xmath1 quantum link model of ref.@xcite works with the generators of @xmath54 instead of @xmath55 . that construction does not have the additional @xmath8 gauge symmetry , but it can not be generalized to @xmath2 as will be explained below . the question then arises how quantum link models can be used to represent qcd . the answer is surprisingly simple . all one needs to do is to break the additional @xmath8 gauge symmetry by adding the real part of the determinant of each link matrix to the hamilton operator , such that now @xmath56 + j ' \sum_{x,\mu } \ [ \mbox{det } u_{x,\mu } + \mbox{det } u{^\dagger}_{x,\mu}].\ ] ] since all elements of @xmath21 commute with each other ( although the real and imaginary parts of an individual element do not commute ) the definition of @xmath57 does not suffer from operator ordering ambiguities . by construction the above hamiltonian is invariant under @xmath2 but not under the extra @xmath8 gauge transformations . however , there is a subtlety that needs to be discussed . in the fundamental representation of @xmath4 , that was discussed above , the operator that represents @xmath57 turns out to be zero . hence , in that case even the hamiltonian of eq.([hamdet ] ) has a @xmath0 gauge invariance . on the other hand , since the above commutation relations can be realized with any representation of @xmath4 , we can use higher representations in which the determinant in general does not vanish . in section 4 it will become clear that the @xmath58-dimensional representation of @xmath4 is the smallest with a non - vanishing determinant . for qcd the hilbert space of the corresponding @xmath3 quantum link model then is the direct product of 20-dimensional hilbert spaces associated with each link . specifying the state on a link thus requires just 5 bits ( because @xmath59 ) . in the standard formulation of lattice gauge theory , on the other hand , in the computer an @xmath3 link variable is usually represented by 18 real numbers , which , for example , on a cray corresponds to 912 bits . at this point it is easy to see why there is a special construction for @xmath1 using a 4-dimensional representation @xcite . under gauge transformations the quantum link operator @xmath60 transforms as an @xmath61 tensor . in @xmath1 the fundamental representation @xmath62 and its conjugate @xmath63 are unitarily equivalent . consequently , one can construct another quantum link operator using @xmath64 with the same @xmath1 gauge transformation properties as @xmath60 . a linear combination of the two operators leads to a quantum hamiltonian with @xmath1 but not @xmath65 gauge symmetry . indeed , adding them with equal weight gives the @xmath54 construction of @xcite . it should be clear that the hamiltonians of eqs.([hamiltonian ] ) and ( [ hamdet ] ) represent just the simplest members of a large class of @xmath0 and @xmath2 invariant quantum link models . like in ordinary lattice gauge theories , the action ( in this case the hamiltonian ) can be improved by including more complicated terms , for example , a six - link loop around a double plaquette . further , it is possible to break the extra @xmath8 gauge symmetry by antisymmetrized combinations of various paths connecting two lattice points not just by the determinant operator on a single link . in that way one can construct @xmath2 invariant quantum link hamiltonians even with the fundamental representation of @xmath4 , but with more complicated interactions . in the 15-dimensional representation of @xmath66 one can construct a more complicated plaquette action that leads to an @xmath3 invariant theory , even without the link determinant term . all these options may become important when the question of the most efficient numerical treatment of quantum link models will be addressed . for the moment we limit ourselves to the discussion of one natural extension of the hamilton operators from before . as it was discussed in ref.@xcite , the operators @xmath67 , @xmath29 and @xmath30 resemble abelian and non - abelian 5-d electric fluxes associated with the link @xmath10 . on the other hand , from a 5-d point of view the plaquette term in the above hamilton operators represents the magnetic part of the energy only . hence , it would be natural to add a 5-d electric term as well . this corresponds to @xmath68 + j ' \sum_{x,\mu } \ [ \mbox{det } u_{x,\mu } + \mbox{det } u{^\dagger}_{x,\mu } ] \nonumber \\ & + & j '' \sum_{x,\mu } \ [ \vec r^2_{x,\mu } + \vec l^2_{x,\mu } ] + j ' '' \sum_{x,\mu } \ t^2_{x,\mu}.\end{aligned}\ ] ] this hamiltonian is still gauge invariant , because @xmath69 , @xmath70 and @xmath71 are casimir operators of @xmath72 , which commute with the generators of all gauge transformations . in the fundamental representation of @xmath4 the above combination of casimir operators is proportional to the unit matrix and has no effect on the dynamics . this is not the case for the higher - dimensional representations of @xmath4 . in principle , one can also include the other casimir operators of @xmath72 in the hamiltonian . the above choice of casimir operators is most natural , because it represents abelian and non - abelian 5-d electric field energies . in the following we present universality arguments , which suggest that quantum link models defined with a 4-d hamiltonian ( describing the evolution of the system in a fifth euclidean direction ) get dimensionally reduced to ordinary 4-d yang - mills theories . these arguments are based on symmetry considerations and they apply to all @xmath2 invariant hamiltonians discussed in this section . in the standard formulation of lattice gauge theories , one must include the electric terms in the action in order to get non - trivial dynamics . since the real and imaginary parts of the elements of a quantum link operator do not commute , the dynamics of a quantum link model is already non - trivial without the electric terms . for simplicity , we restrict ourselves to @xmath73 for the rest of this paper . as discussed in detail in ref.@xcite , quantum link models in @xmath5 dimensions are related to ordinary 4-d gauge theories via dimensional reduction . the hamilton operator of the quantum link model , which is defined on a 4-d lattice , describes the evolution of the system in a fifth euclidean direction . the crucial observation is that a genuine 5-d non - abelian gauge theory has deconfined massless gluons and thus an infinite correlation length . when periodic boundary conditions are imposed in the fifth direction and its extent @xmath74 is made finite , the extent of the extra dimension is thus negligible compared to the correlation length . hence , the theory appears to be dimensionally reduced to four dimensions . of course , in four dimensions the confinement hypothesis suggests that gluons are no longer massless . indeed , as it was argued in ref.@xcite , a glueball mass @xmath75 is expected to be generated non - perturbatively . here @xmath76 is the renormalized dimensionful gauge coupling of the 5-d gauge theory . for large @xmath74 the gauge coupling of the dimensionally reduced 4-d theory is given by @xmath77 thus the continuum limit @xmath18 of the 4-d theory is approached when one sends the extent @xmath74 of the fifth direction to infinity . hence , in contrast to naive expectations , dimensional reduction occurs when the extent of the fifth direction becomes large . this is due to asymptotic freedom , which implies that the correlation length @xmath78 grows exponentially with @xmath74 . when a gauge theory is dimensionally reduced , usually the polyakov loop in the extra dimension appears as an adjoint scalar field . here we want to obtain pure 4-d yang - mills theory ( without charged scalars ) after dimensional reduction . this can be achieved if one does not impose gauss law for the states propagating in the fifth dimension , because the polyakov loop is a lagrange multiplier field that enforces the gauss law . formally , this can be realized simply by putting the fifth component of the gauge potential to zero , i.e. @xmath79 in the infinite volume limit of the 5-d theory this restriction has no effect on the dynamics . with finite extent in the fifth direction , however , it deviates from the standard formulation of gauge theories . the leading terms in the low energy effective action of the 5-d gauge theory corresponding to the quantum link model take the form @xmath80 = \int_0^\beta dx_5 \int d^4x \ \frac{1}{2 e^2}[\mbox{tr } \ f_{\mu\nu } f_{\mu\nu } + \frac{1}{c^2 } \mbox{tr } \ \partial_5 a_\mu \partial_5 a_\mu].\ ] ] we expect that the qcd quantum link model leads to a 5-d gauge theory characterized by the `` velocity of light '' @xmath81 . note that here @xmath82 runs over 4-d indices only . at finite @xmath74 the above theory has only a 4-d gauge invariance , because we have fixed @xmath83 , i.e. we have not imposed the gauss law . on the other hand , for @xmath84 a full 5-d gauge symmetry is recovered , although the above action then still is in @xmath83 gauge . since we are interested in dimensional reduction , a 4-d gauge symmetry is sufficient for our purposes . on the level of the quantum link model , not imposing gauss law is achieved by simply writing the quantum statistical partition function as @xmath85 in contrast to the standard formulation of gauge theories we have not included a projection operator on gauge invariant states , i.e. gauge variant states also propagate in the fifth direction . the hamilton operator of the quantum link model is defined on a 4-d space - time lattice and describes the evolution of the system in the fifth unphysical direction . in particular , all the information about the physical spectrum of the 4-d theory is contained in correlation functions in the euclidean time direction , which is part of the 4-d lattice . note that the physical gauss law is properly imposed because the model does contain non - trivial polyakov loops in the euclidean time direction . in the continuum limit @xmath18 , which we approach by increasing the extent @xmath74 of the fifth dimension , we are probing the low lying states in the spectrum of the 4-d hamilton operator of the quantum link model . the space - time correlations in these unphysical states of the 4-d hamiltonian contain the information about the physical spectrum . it is useful to think of the dimensionally reduced 4-d theory as a lattice theory with lattice spacing @xmath86 ( which has nothing to do with the lattice spacing of the quantum link model ) . in fact , one can imagine performing a block renormalization group transformation that averages the 5-d field over cubic blocks of size @xmath74 in the fifth direction and of size @xmath86 in the four physical space - time directions . the block centers then form a 4-d space - time lattice of spacing @xmath86 and the effective theory of the block averaged 5-d field is indeed a 4-d lattice theory . this is illustrated in fig.3 . a similar argument was first used by hasenfratz and niedermayer @xcite in their study of 2-d quantum antiferromagnets . the partition function of eq.([gaugez ] ) can be written as a 5-d path integral of discrete variables in the @xmath3 case the states of the 20-dimensional representation of @xmath66 on each link . in many respects this path integral resembles that of quantum spin systems , which can be simulated by very efficient loop cluster algorithms . due to the discrete nature of the hilbert space , one can even work directly in the continuum for the extra euclidean direction @xcite . indeed for a @xmath8 quantum link model a cluster algorithm operating in the continuum has already been constructed @xcite . it is plausible that cluster algorithms can also be constructed for non - abelian quantum link models , which would allow high - precision simulations of gauge theories . in this section we reformulate quantum link models using anticommuting operators describing fermionic constituents of the gluons . in contrast to composite models @xcite , our rishons are not physical particles , because they are absolutely tightly bound into gluons and there is no kinetic term for the rishons . still , the rishons are convenient mathematical objects that provide theoretical insight into the physics of quantum link models . the rishon representation is interesting both from an analytic and from a computational point of view . first , it may lead to a new way to attack the large @xmath6 limit of qcd and other non - perturbative quantum field theories . in addition , the rishon formulation may also be useful in the construction of cluster algorithms . the algebraic structure of a quantum link model is determined by the commutation relations derived in section 2 . the hilbert space is a direct product of representations of @xmath4 on each link , with the generators on different links commuting with each other . we can thus limit ourselves to a single link , for which the commutation relations are @xmath87 = 2 i f_{abc } r^c , \ [ l^a , l^b ] = 2 i f_{abc } l^c , \ [ r^a , l^b ] = [ t , r^a ] = [ t , l^a ] = 0 , \nonumber \\ & & [ r^a , u_{ij } ] = u_{ik } \lambda^a_{kj } , \ [ l^a , u_{ij } ] = - \lambda^a_{ik } u_{kj } , \ [ t , u_{ij } ] = 2 u_{ij } , \nonumber \\ & & [ \mbox{re } u_{ij},\mbox{re } u_{kl } ] = [ \mbox{im } u_{ij},\mbox{im } u_{kl } ] = - i ( \delta_{ik } \ \mbox{im } \lambda^a_{jl } r^a + \delta_{jl } \ \mbox{im } \lambda^a_{ik } l^a ) , \nonumber \\ & & [ \mbox{re } u_{ij},\mbox{im } u_{kl } ] = i ( \delta_{ik } \ \mbox{re } \lambda^a_{jl } r^a - \delta_{jl } \ \mbox{re } \lambda^a_{ik } l^a + \frac{2}{n } \delta_{ik } \delta_{jl } t ) . \end{aligned}\ ] ] these relations can be realized in a representation using anticommuting rishon operators @xmath88 , @xmath89 with color index @xmath90 . the rishon operators live at the left and right ends of the links and are characterized by a lattice point @xmath23 and a link direction @xmath91 . they obey canonical anticommutation relations @xmath92 under @xmath2 gauge transformations the operators @xmath81 and @xmath93 transform in the fundamental and anti - fundamental representation , respectively . it is straightforward to show that the commutation relations of eq.([commutators ] ) are satisfied when we write @xmath94 actually , the commutation relations are satisfied also when the rishons are quantized as bosons . it is crucial to realize that all operators introduced so far ( including the hamiltonian ) commute with the rishon number operator @xmath95 on each individual link . hence , we can limit ourselves to superselection sectors of fixed rishon number for each link . this is equivalent to working in a given irreducible representation of @xmath4 . let us use the rishon representation to take a closer look at the determinant operator that we used to break the @xmath0 gauge symmetry down to @xmath2 . we have @xmath96 due to the antisymmetry of the @xmath97-tensors this operator would vanish for bosonic rishons . for fermionic rishons , on the other hand , we have @xmath98 only when this operator acts on a state with exactly @xmath99 rishons ( all of a different color ) , it can give a non - zero contribution . in all other cases the determinant vanishes . this means that we can reduce the symmetry from @xmath0 to @xmath2 via the determinant only when we work with exactly @xmath99 fermionic rishons on each link . the number of fermion states per link then is @xmath100 this is the dimension of the @xmath4 representation with a totally antisymmetric young tableau with @xmath6 boxes ( arranged in a single column ) . in the rishon representation the hamilton operator takes the form @xmath101 \nonumber \\ & = & - j \sum_{x,\mu \neq \nu } ( c{^\dagger}_{x+\hat\mu,-\mu } c_{x+\hat\mu,-\nu } ) ( c{^\dagger}_{x+\hat\mu+\hat\nu,-\nu } c_{x+\hat\mu+\hat\nu,-\mu } ) ( c{^\dagger}_{x+\hat\nu,+\mu } c_{x+\hat\nu,-\nu } ) ( c{^\dagger}_{x,+\nu } c_{x,+\mu } ) \nonumber \\ & + & j ' \sum_{x,\mu } \ [ \mbox{det } \ c_{x,+\mu } c{^\dagger}_{x+\hat\mu,-\mu } + \mbox{det } \ c_{x+\hat\mu,-\mu } c{^\dagger}_{x,+\mu}].\end{aligned}\ ] ] the terms in brackets in eq.([hrishon ] ) represent color neutral glueballs formed by two rishons located at the same corner of a plaquette . the plaquette part of the hamilton operator shifts single rishons from one end of a link to the other and simultaneously changes their color . the determinant part , on the other hand , shifts an entire rishon - baryon a color neutral combination of @xmath6 rishons along the link . this is illustrated in fig.2 . the rather simple rishon dynamics may facilitate new analytic approaches to qcd and may also be useful in numerical simulations of quantum link models . it is remarkable that the hamilton operator can be expressed in terms of color neutral glueballs and rishon - baryons . this observation suggests a new approach to the large @xmath6 limit , which is presently under investigation . to represent full qcd , it is essential to formulate quantum link models with quarks . this is more or less straightforward , although some subtleties arise related to the dimensional reduction of fermions . before we discuss the quantum link formulation of full qcd , let us review the standard formulation of lattice gauge theory with fermions . here we concentrate on wilson s method , although it is straightforward to construct quantum link models with staggered fermions along the same lines . the standard wilson action with quarks is given by @xmath102 & = & - \sum_{x,\mu \neq \nu } \mbox{tr } [ u_{x,\mu } u_{x+\hat\mu,\nu } u{^\dagger}_{x+\hat\nu,\mu } u{^\dagger}_{x,\nu } ] \nonumber \\ & + & \frac{1}{2 } \sum_{x,\mu } \ [ \bar\psi_x \gamma_\mu u_{x,\mu } \psi_{x+\hat\mu } - \bar\psi_{x+\hat\mu } \gamma_\mu u{^\dagger}_{x,\mu } \psi_x ] + m \sum_x \bar\psi_x \psi_x \nonumber \\ & + & \frac{r}{2 } \sum_{x,\mu } \ [ 2 \bar\psi_x \psi_x - \bar\psi_x u_{x,\mu } \psi_{x+\hat\mu } - \bar\psi_{x+\hat\mu } u{^\dagger}_{x,\mu } \psi_x].\end{aligned}\ ] ] here @xmath103 and @xmath104 are independent grassmann valued spinors associated with the lattice site @xmath23 , @xmath105 are dirac matrices and @xmath106 is the bare quark mass . the term proportional to @xmath107 is the wilson term that removes unwanted lattice fermion doublers at the expense of explicitly breaking chiral symmetry . in order to reach the continuum limit the bare mass @xmath106 must be tuned appropriately . the guiding principle in the formulation of quantum link models is to replace the classical action of the standard formulation by a hamilton operator that describes the evolution of the system in a fifth euclidean direction . for the quarks we must also replace @xmath103 by @xmath108 . hence , the full qcd quantum link hamiltonian is given by @xmath109 + j ' \sum_{x,\mu } \ [ \mbox{det } u_{x,\mu } + \mbox{det } u{^\dagger}_{x,\mu } ] \nonumber \\ & + & \frac{1}{2 } \sum_{x,\mu } \ [ \psi{^\dagger}_x \gamma_5 \gamma_\mu u_{x,\mu } \psi_{x+\hat\mu } - \psi{^\dagger}_{x+\hat\mu } \gamma_5 \gamma_\mu u{^\dagger}_{x,\mu } \psi_x ] + m \sum_x \psi{^\dagger}_x \gamma_5 \psi_x \nonumber \\ & + & \frac{r}{2 } \sum_{x,\mu } \ [ 2 \psi{^\dagger}_x \gamma_5 \psi_x - \psi{^\dagger}_x \gamma_5 u_{x,\mu } \psi_{x+\hat\mu } - \psi{^\dagger}_{x+\hat\mu } \gamma_5 u{^\dagger}_{x,\mu } \psi_x].\end{aligned}\ ] ] here @xmath110 and @xmath111 are quark creation and annihilation operators with canonical anticommutation relations @xmath112 where @xmath113 , @xmath114 and @xmath115 , which have been suppressed in eq.([qcdaction ] ) , are color , flavor and dirac indices , respectively . of course , we have again replaced the classical link variables @xmath9 by quantum link operators @xmath21 . the generator of an @xmath2 gauge transformation now takes the form @xmath116 and it is again straightforward to show that @xmath117 commutes with @xmath22 for all @xmath23 . the same universality arguments that were used before now suggest that the effective action of the corresponding 5-d gauge theory ( with @xmath83 ) takes the form @xmath118&=&\int_0^\beta dx_5 \int d^4x \ \ { \frac{1}{2 e^2 } [ \mbox{tr } f_{\mu\nu } f_{\mu\nu } + \frac{1}{c^2 } \mbox{tr } \partial_5 a_\mu \partial_5 a_\mu ] \nonumber \\ & + & \bar\psi[\gamma_\mu ( a_\mu + \partial_\mu ) + m_q + \frac{1}{c ' } \gamma_5 \partial_5 ] \psi \}.\end{aligned}\ ] ] the `` velocity of light '' @xmath119 , which characterizes the propagation of a quark in the fifth direction , is in general different from the corresponding quantity @xmath81 for the gluons , because the quantum link formulation has no symmetry between the four physical space - time directions and the extra fifth direction this is no problem , because we are only interested in the 4-d physics after dimensional reduction . to ensure the proper dimensional reduction of the quarks , their boundary conditions in the fifth direction must be chosen appropriately . the standard antiperiodic boundary conditions , which are dictated by thermodynamics in the euclidean time direction , would lead to matsubara modes , @xmath120 , which would limit the physical correlation length of the dimensionally reduced fermion to @xmath121 . the confinement physics of the induced 4-d gluon theory , on the other hand , takes place at a correlation length which is growing exponentially with @xmath74 . in fact , @xmath74 plays the role of the lattice spacing of the dimensionally reduced theory . quarks with antiperiodic boundary conditions in the fifth direction would hence remain at the cut - off and the dimensionally reduced theory would still be a yang - mills theory without quarks . once this problem is understood , one possible solution is obvious . one can simply choose periodic boundary conditions for the quarks in the fifth direction . this gives rise to a matsubara mode , @xmath122 , that survives dimensional reduction . since the extent of the fifth direction has nothing to do with the inverse temperature ( which is the extent of the euclidean time direction ) , one could indeed choose the boundary condition in this way . fortunately , we can do better . the above scenario with periodic boundary conditions for the quarks would suffer from the same fine - tuning problem as the original wilson fermion method . the bare quark mass would have to be adjusted very carefully in order to reach the chiral limit . in practice this is a great problem in numerical simulations . this problem has been solved very elegantly in shamir s variant @xcite of kaplan s fermion proposal @xcite . kaplan studied the physics of a 5-d system of fermions , which is always vector - like , coupled to a 4-d domain wall that manifests itself as a topological defect . the key observation is that under these conditions a zero mode of the 5-d dirac operator appears as a bound state localized on the domain wall . from the point of view of the 4-d domain wall , the zero mode represents a massless chiral fermion . the original idea was to construct lattice chiral gauge theories in this way . shamir has pointed out that the same mechanism can solve the lattice fine - tuning problem of the bare fermion mass in vector - like theories including qcd . he also suggested a variant of kaplan s method that has several technical advantages and that turns out to fit very naturally with the construction of quantum link qcd . in quantum link models we already have a fifth direction for reasons totally unrelated to the chiral symmetry of fermions . we will now follow shamir s proposal and use the fifth direction to solve the fine - tuning problem that we would have with periodic boundary conditions for the quarks . the essential technical simplification compared to kaplan s original proposal is that one now works with a 5-d slab of finite size @xmath74 with open boundary conditions for the fermions at the two sides . this geometry limits one to vector - like theories , because now there are two zero modes one at each boundary which correspond to one left and one right - handed fermion in four dimensions . this set - up fits naturally with our construction of quantum link qcd . in particular , the evolution of the system in the fifth direction is still governed by the hamilton operator of eq.([qcdaction ] ) . the only ( but important ) difference to wilson s fermion method is that now @xmath123 . of course , one could also obtain a left and a right - handed fermion by using a domain wall and an anti - wall with otherwise periodic boundary conditions . in that case the hamiltonian of eq.([qcdaction ] ) would have to be modified in an @xmath124-dependent way . shamir s method is more economical and concentrates on the essential topological aspects , which are encoded in the boundary conditions for the fermions in the fifth direction . it is important that in shamir s construction one also puts @xmath83 . there are some minor differences between the implementations of the method in the standard formulation of lattice gauge theory and in quantum link models . in the standard formulation one works with a 4-d gauge field , which is constant in the fifth direction . in quantum link qcd this is not possible , because the non - trivial dynamics in the fifth direction turns the discrete states of quantum links into the continuous degrees of freedom of physical gluons . however , it is still true that the physical gluon field is essentially constant in the fifth direction , because its correlation length grows exponentially with @xmath74 . this is important for the generation of the fermionic zero modes at the two sides of the 5-d slab . the partition function of the theory with open boundary conditions for the quarks and with periodic boundary conditions for the gluons is simply given by @xmath125 here the trace extends only over the gluonic hilbert space of the quantum link model , thus implementing periodic boundary conditions for the gluons . we decompose the quark spinor into left and right - handed components @xmath126 open boundary conditions for the fermions are realized by taking the expectation value of @xmath127 in the fock state @xmath128 , which is annihilated by all right - handed @xmath129 and by all left - handed @xmath130 @xcite . as a result , there are no left - handed quarks at the boundary at @xmath131 and there are no right - handed quarks at the boundary at @xmath132 . of course , unlike periodic or antiperiodic boundary conditions , open boundary conditions for the fermions break translation invariance in the fifth direction . through the interaction between quarks and gluons , this breaking also affects the gluonic sector . this is no problem , because we are only interested in the 4-d physics after dimensional reduction . for that it is essential that both quarks and gluons have zero modes , which is indeed the case with the boundary conditions from above . of course , as we have argued before , the gluonic correlation length is not truly infinite as long as @xmath74 is finite , but due to confinement it is exponentially large . as we will see now , the same is true for the quarks , but for a totally different reason . in fact , already free quarks pick up an exponentially small mass due to tunneling between the two boundaries , which mixes left and right - handed states , and thus breaks chiral symmetry explicitly . to understand this , let us consider the actual hamiltonian @xmath133 , which describes the evolution of free quarks with spatial lattice momentum @xmath134 in time ( rather than @xmath117 , which describes the evolution of the system in the fifth euclidean direction ) @xmath135.\ ] ] the wave function of a stationary quark state with energy @xmath136 is determined by @xmath137 let us consider states with lattice momenta @xmath138 . the state with @xmath139 describes a physical quark at rest , while the other states correspond to doubler fermions . denoting the number of non - zero momentum components @xmath140 by @xmath141 the above dirac equation ( [ dirac ] ) takes the form @xmath142 due to the boundary conditions introduced before , we must solve these equations with @xmath143 . inserting the ansatz @xmath144 immediately gives @xmath145 this equation has a normalizable solution only if @xmath146 . in that case , for large @xmath74 this implies @xmath147 and hence @xmath148 the important observation is that the bare mass parameter @xmath106 is not the physical quark mass , although we have considered an unrenormalized free theory . let us first consider the doubler fermions , which are characterized by @xmath149 . it is then possible to choose @xmath107 such that @xmath150 , for which no normalizable solution exists . thus , for @xmath151 , i.e. for a sufficiently strong wilson - term with an unconventional sign , the doubler fermions are removed from the physical spectrum . the mass of the physical fermion ( characterized by @xmath152 ) is @xmath153 which is exponentially small in @xmath74 . this situation is illustrated in fig.2 . the above result suggests how the fine - tuning problem of the fermion mass can be avoided . the confinement physics of quantum link qcd in the chiral limit takes place at a length scale @xmath154 which is determined by the 1-loop coefficient of the @xmath74-function of qcd with @xmath155 massless quarks and by the 5-d gauge coupling @xmath76 . as long as one chooses @xmath156 the chiral limit is reached automatically when one approaches the continuum limit by making @xmath74 large . for a given value of @xmath107 one is limited by @xmath157 ( note that @xmath123 ) . on the other hand , one can always choose @xmath158 ( and thus @xmath159 ) such that the above inequality is satisfied . of course , we also want to be able to work at non - zero quark masses . following shamir , we do this by modifying the boundary conditions for the quarks in the fifth direction . instead of using @xmath160 we now demand @xmath161 where @xmath162 is a mass parameter . this reduces to the previous boundary condition for @xmath163 , while it corresponds to antiperiodic boundary conditions for @xmath164 , and to periodic boundary conditions for @xmath165 . solving the above equations with the new boundary condition indeed yields physical quarks of mass @xmath162 in the continuum limit @xmath166 as long as @xmath167 . it has been shown in ref.@xcite that in the interacting theory @xmath162 is only multiplicatively renormalized . on the level of the partition function @xmath168\ ] ] the new boundary condition manifests itself by a mass - dependent operator @xmath169 which was constructed in ref.@xcite . note that in eq.([partitionfunction ] ) the trace is both over the gluonic and over the fermionic hilbert space . in the chiral limit @xmath163 the operator @xmath170 reduces to a projection operator @xmath171 on the fock state introduced before . for @xmath172 , i.e. for antiperiodic boundary conditions , the operator @xmath170 becomes the unit operator and the partition function reduces to the expression well - known from thermodynamics . like any massive fermion action in odd dimensions , the effective action of eq.([effaction ] ) is not parity symmetric . hence , there is the potential danger that the corresponding symmetry breaking terms make their way into the dimensionally reduced 4-d theory . indeed the topological @xmath173-vacuum term violates parity in four dimensions . for the moment we are interested in qcd with @xmath174 and therefore in a parity symmetric theory . later , we will ask how to incorporate @xmath173 . we will now show that the dimensionally reduced theory is indeed parity invariant . for simplicity this discussion will be based on the low energy continuum effective action of eq.([effaction ] ) . it is , however , straightforward to apply the same arguments to the lattice action in the path integral that one obtains from the hamiltonian of eq.([qcdaction ] ) . the 5-d effective action is not invariant under parity as it is usually defined in five dimensions . since we are interested in 4-d physics , this definition of parity is , however , not relevant for us . instead , we now define another transformation that leaves the 5-d action invariant and reduces to ordinary parity in the dimensionally reduced 4-d theory . let us consider @xmath175 it is straightforward to show that this is indeed a symmetry of the 5-d effective action . note that we have changed @xmath124 to @xmath176 . this change is invisible from the point of view of the dimensionally reduced 4-d theory . in fact , in the 4-d world the above transformation reduces to ordinary parity . however , exchanging @xmath124 and @xmath176 is essential as far as the 5-d theory is concerned . this operation exchanges the two domain walls and hence the left and right - handed fermions that are bound to them . this is indeed what parity is supposed to do . a non - perturbative formulation of qcd would be incomplete without a discussion of the vacuum angle @xmath173 . although in the real world this parameter is indistinguishable from zero , it represents a possible parity breaking effect in the qcd lagrangian . in the standard formulation of lattice qcd such a term could be added via the topological charge , for which a lattice regularized expression with adequate topological properties exists @xcite . in quantum link qcd this would not be possible , because the discrete states of the quantum link operators can not directly encode topological properties . here we propose to include @xmath177 as the determinant of the fermion mass matrix and hence via the boundary condition of eq.([boundary ] ) . when one uses a fermion formulation which does not require fine - tuning in the chiral limit , this is the most natural thing to do , even in the standard formulation of lattice qcd . in this section we express the hamiltonian of the @xmath0 quantum link model with quarks in terms of color neutral operators . gauge invariance requires that these operators are local bilinear combinations of rishons and quarks , which we refer to as glueballs , mesons and constituent quarks . all these objects including the constituent quarks are bosons . in fact , the resulting representation of the theory can be viewed as a first step towards a bosonization of @xmath0 qcd . the determinant term in the @xmath2 quantum link hamiltonian would give rise to an additional color - singlet rishon - baryon consisting of @xmath6 rishons . for odd @xmath6 this object would hence be a fermion . to avoid complications related to these objects we limit ourselves to @xmath0 in this section . as in the yang - mills case , we represent the quantum link hamiltonian of @xmath0 qcd in terms of rishons @xmath178 \nonumber \\ & + & m \sum_x \psi{^\dagger}_x \gamma_5 \psi_x \nonumber \\ & + & \frac{r}{2 } \sum_{x,\mu } \ [ 2 \psi{^\dagger}_x \gamma_5 \psi_x - ( \psi{^\dagger}_x c_{x,+\mu } ) \gamma_5 ( c{^\dagger}_{x+\hat\mu,-\mu } \psi_{x+\hat\mu } ) - ( \psi{^\dagger}_{x+\hat\mu } c_{x+\hat\mu,-\mu } ) \gamma_5 ( c{^\dagger}_{x,+\mu } \psi_x ) ] . \nonumber \\ \\end{aligned}\ ] ] note that some of the terms in brackets now represent color neutral bosonic constituent quarks formed by a quark and a rishon located at the same end of a link . in fact , the hamiltonian can be expressed as @xmath179 + m \sum_x \mbox{tr}(m_x \gamma_5 ) \nonumber \\ & + & \frac{r}{2 } \sum_{x,\mu } \ [ 2 \mbox{tr}(m_x \gamma_5 ) - q{^\dagger}_{x,+\mu } \gamma_5 q_{x+\hat\mu,-\mu } - q{^\dagger}_{x+\hat\mu,-\mu } \gamma_5 q_{x,+\mu}].\end{aligned}\ ] ] the traces are over flavor and dirac indices . we have introduced @xmath180 glueball operators @xmath181 which satisfy the local commutation relations of @xmath182 @xmath183 = \delta_{xy } ( \delta_{\pm\nu,\pm\rho } \phi_{x,\pm\mu,\pm\sigma } - \delta_{\pm\mu,\pm\sigma } \phi_{x,\pm\rho,\pm\nu}).\ ] ] we have also defined @xmath184 meson operators @xmath185 which carry flavor and dirac indices and which generate the algebra of @xmath186 @xmath187 = \delta_{xy } ( \delta_{bc } \delta_{\beta\gamma } m^{a \alpha d \delta}_x - \delta_{ad } \delta_{\alpha\delta } m^{b \beta c \gamma}_x).\ ] ] while glueballs and mesons are not related via their commutation relations , i.e. @xmath188 = 0,\ ] ] they are both related with the constituent quark operators @xmath189 via the commutation relations @xmath190 = \delta_{xy } \delta_{\nu\rho } q^{a \alpha}_{x,\pm\mu } , \ [ m^{a \alpha b \beta}_x , q^{c \gamma}_{y,\pm\mu } ] = \delta_{xy } \delta_{ac } \delta_{\alpha\gamma } q^{b \beta}_{x,\pm\mu}.\ ] ] finally , the commutation relations of the constituent quark operators take the form @xmath191 = \delta_{xy}(\delta_{ab } \delta_{\alpha\beta } \phi_{x,\pm\mu,\pm\nu } - \delta_{\pm\mu,\pm\nu } m^{a \alpha b \beta}_x ) , \nonumber \\ & & [ q^{a \alpha}_{x,\pm\mu},q^{b \beta}_{y,\pm\nu } ] = [ q^{a \alpha \dagger}_{x,\pm\mu},q^{b \beta \dagger}_{y,\pm\nu } ] = 0.\end{aligned}\ ] ] thus , the inclusion of the constituent quark operators completes the site - based algebra of @xmath192 to @xmath193 . we expect this representation of the theory to be useful in the investigation of the large @xmath6 limit and in attempts to bosonize quantum link qcd . we have constructed a new lattice formulation of qcd in the framework of quantum link models . in these theories the link - based hilbert space of the gluons is finite . this is possible because the theory is constructed with a fifth euclidean direction that ultimately disappears via dimensional reduction , although in the continuum limit its extent @xmath74 diverges in units of the lattice spacing of the quantum link model . on the other hand , @xmath74 itself plays the role of the lattice spacing of another effective 4-d lattice theory , whose non - perturbatively generated length scales are exponentially large in @xmath74 ( see fig.3 ) . an important question is whether @xmath2 quantum link models on a 4-d space - time lattice possess a continuum limit in the universality class of the coulomb phase of 5-d non - abelian gauge theories . this is essential for their proper dimensional reduction to ordinary 4-d yang - mills theories . one way to investigate this is to study the classical equations of motion of quantum link models . as will be shown in a forthcoming publication @xcite , they indeed resemble those of 5-d yang - mills theories with @xmath83 . this shows that in the classical limit i.e. when one works with a large representation of @xmath4 dimensional reduction does indeed take place . whether dimensional reduction also occurs with the 20-dimensional representation of @xmath66 and hence whether the @xmath3 quantum link model introduced here provides a viable cut - off scheme for the continuum field theory of qcd , is an issue that can only be decided in numerical simulations . this situation is analogous to the one in quantum spin models . based on the classical equations of motion one finds that for large spin the 2-d antiferromagnetic quantum heisenberg model has an ordered ground state and that the corresponding goldstone bosons in that case antiferromagnetic magnons or spin - waves have a relativistic dispersion relation . still , it required high - precision numerical simulations @xcite to be sure that a staggered magnetization is spontaneously generated even in the extreme quantum limit of spin @xmath194 . only this finally justifies the use of a 3-d relativistic low - energy chiral lagrangian with @xmath7 symmetry to describe the dynamics of the spin @xmath194 quantum heisenberg model at low energies @xcite . similarly , at present we can not guarantee that the low - energy excitations of a 4-d @xmath3 invariant quantum link model formulated with the 20-dimensional representation of @xmath66 are 5-d massless gluons . however , this should at least be the case in the classical limit , i.e. when we work with a large representation of @xmath66 . it is remarkable that the introduction of a fifth euclidean direction of finite extent also naturally protects the chiral symmetries of lattice fermions , when open boundary conditions are used for the quarks at the two 4-d boundaries of the 5-d slab ( see fig.2 ) . this set - up first proposed by shamir as a variant of kaplan s fermion method fits very naturally with quantum link models . due to tunneling between the two boundaries , a fermionic correlation length is generated , which is also exponentially large in @xmath74 . quantum link qcd treats quarks and gluons on an equal footing . they both have finite hilbert spaces ( per unit volume ) , both require a fifth euclidean direction that finally disappears via dimensional reduction and they both have correlation lengths that are exponential in @xmath74 without requiring fine - tuning . we believe that these properties point to an interesting connection between fermions and gauge fields in quantum link models . this may help in attempting to formulate chiral or supersymmetric gauge theories on the lattice . due to the fact that their gluonic hilbert space is finite , quantum link models can be expressed in terms of rishons , which are fermionic constituents of the gluons . the rishon dynamics can probably not be formulated as a continuum field theory . therefore we consider the rishons as lattice artifacts in our way to formulate qcd . the gluons in quantum link models are composites of rishons , just like antiferromagnetic magnons can be viewed as composites of electrons that are hopping on a lattice . of course , in that case we know that the electrons are real particles , which are hopping on a real crystal lattice . if our world is a thin five - dimensional slab with a lattice structure at extremely small distances , the rishons in quantum link models could be real particles . we do not want to speculate about that possibility and thus we consider them as mathematical objects only . however , the rishons may turn out to be of great importance for solving qcd in the large @xmath6 limit . we consider this possibility the most promising analytic aspect of the new formulation . at finite @xmath6 we do not see a way to solve quantum link models analytically . however , from a computational point of view quantum link qcd is still very attractive , in particular because the new framework may allow one to construct very efficient cluster algorithms . the five - dimensional set - up is not as inconvenient as it may seem , because due to the discrete nature of the hilbert space the 5-d path integral for quantum link models can be formulated and most important also be simulated directly in the continuum of the fifth euclidean dimension . in fact , even in the standard formulation of lattice qcd it may turn out to be advantageous to work in five dimensions in order to control the chiral properties of the quarks . it remains to be seen if quantum link models provide a more efficient formulation of qcd than standard lattice gauge theory . in any case , quantum link models allow us to attack the long - standing qcd problem from a different perspective . we are indebted to m. basler , b. beard , d. chen , s. levit , p. orland , y. shamir and a. tsapalis for very interesting discussions . one of the authors ( u .- j . w. ) likes to thank the theory group of the weizmann institute in rehovot , where part of this work was done , for its hospitality and the a. p. sloan foundation for its support .

we prove a positive mass theorem for-dimensional asymptotically flat manifolds with a non - compact boundary if either or if and the manifold is spin . this settles , for this class of manifolds , a question posed in a recent paper by the first author in connection with the long - term behavior of a certain yamabe - type flow on scalar - flat compact manifolds with boundary .

1407.0673

let @xmath3 be an oriented riemannian manifold with a non - compact boundary @xmath4 and dimension @xmath2 . we denote by @xmath5 the scalar curvature of @xmath6 . we also assume that @xmath4 is oriented by an outward pointing unit normal vector @xmath7 , so that its mean curvature is @xmath8 . we say that @xmath9 is _ asymptotically flat _ with decay rate @xmath10 if there exists a compact subset @xmath11 and a diffeomorphism @xmath12 such that the following asymptotic expansion holds as @xmath13 : @xmath14 is the coordinate system induced by @xmath15 , @xmath16 , @xmath17 are the coefficients of @xmath18 with respect to @xmath19 , the comma denotes partial differentiation , @xmath20 and @xmath21 . the subset @xmath22 is called the _ end _ of @xmath23 . in the following we use the einstein summation convention with the index ranges @xmath24 and @xmath25 . observe that , along @xmath4 , @xmath26 spans @xmath27 while @xmath28 points inwards . the simplest example , and in fact the model case , of a manifold in this class is the closed half - space @xmath29 endowed with the standard flat metric @xmath30 . this work is devoted to the study of a certain geometric invariant which measures the deviation at infinity of a general asymptotically flat manifold @xmath9 from the model space @xmath31 . [ defasymflat ] suppose that @xmath32 and @xmath5 and @xmath33 are integrable in @xmath23 and @xmath4 , respectively . in terms of asymptotically flat coordinates as above , the _ mass _ of @xmath9 is given by @xmath34 where @xmath35 is a large coordinate hemisphere of radius @xmath36 with outward unit normal @xmath37 , and @xmath38 is the outward pointing unit co - normal to @xmath39 , viewed as the boundary of the bounded region @xmath40 ( see figure [ fig1 ] ) . it is not hard to verify that the limit on the right - hand side of ( [ massadm ] ) exists . moreover , we shall see in section [ massgeo ] that its value does not depend on the particular asymptotically flat coordinates chosen . thus , @xmath41 is an invariant of the asymptotic geometry of @xmath9 . besides having an obvious intrinsic geometric relevance , this invariant appears crucially in @xcite in connection with the global convergence of a certain yamabe - type flow first considered by s. brendle in @xcite , which produces scalar - flat compact domains with constant mean curvature boundary in the long - term limit . as explained in @xcite , the following conjecture is expected to be true . [ conjpmt ] if @xmath9 is asymptotically flat with decay rate @xmath32 as above and satisfies @xmath42 and @xmath43 then @xmath44 , with the equality occurring if and only if @xmath9 is isometric to @xmath45 . this conjecture has been confirmed in some special cases in @xcite . in this work we show more generally that it holds true whenever the standard positive mass conjecture holds ( see @xcite ) . more precisely , the following result holds . [ mainm ] conjecture [ conjpmt ] holds true if either @xmath46 or if @xmath2 and @xmath23 is spin . combined with the main result in @xcite , this guarantees the global convergence of the yamabe - type flow introduced in @xcite for any initial scalar - flat compact manifold with boundary which meets the conditions of the theorem ( i.e. either it is spin or has dimension @xmath47 ) . the following immediate consequence of the rigidity statement in theorem [ mainm ] is also worth noticing . [ mainmcor ] let @xmath9 be as in theorem [ mainm ] and assume further that there exists a compact subset @xmath11 such that @xmath48 is isometric to @xmath49 . then @xmath9 is isometric to @xmath45 . we provide here two proofs of theorem [ mainm ] . in the proof presented in section [ firstproof ] , the first step consists of an improvement of the asymptotics of the given metric in the spirit of the classical proof of the standard positive mass theorem by schoen and yau ( @xcite ) . once this technical step is accomplished in proposition [ conf : flat : assump ] , this proof proceeds by a reduction to the classical cases via a doubling construction . we employ a result by miao ( @xcite ) , which covers the situation in which corners along a compact inner hypersurface appear , to prove that the conjecture above holds true whenever the classical positive mass theorem holds for the doubled manifold . we also use the improvement in the asymptotics to present an alternative proof of theorem [ mainm ] in the case @xmath1 which is more in the spirit of the classical arguments by schoen - yau ( @xcite ) . more precisely , we show that the assumption of negative mass implies the existence of a stable minimal hypersurface without boundary leading to a contradiction as in @xcite . in these proofs , the rigidity statement in the theorem follows by means of the variational characterization of the mass given in proposition [ propvar ] . finally , if @xmath2 and @xmath23 is spin we present in theorem [ wittenform ] the natural extension of witten s celebrated formula for the mass in terms of a suitable harmonic spinor globally defined on @xmath23 . the proof of theorem [ mainm ] in this case is an immediate consequence of this expression . [ gener ] we can conceive a version of theorem [ mainm ] in which the manifold @xmath9 has two collections of finitely many ends , say @xmath50 and @xmath51 , which we assume endowed with diffeomorphisms @xmath52 and @xmath53 such that the expansion ( [ asymflat ] ) holds . to each end @xmath54 we associate the mass given by ( [ massadm ] ) , and to each end @xmath55 we associate its standard adm mass as in @xcite . in this setting , the result says that if @xmath42 and @xmath43 then the mass of each end is non - negative . moreover , if at least one mass vanishes then @xmath9 actually has only one end , being isometric either to @xmath31 or to @xmath56 , according to the type of the end . the proofs of these more general statements follow by straightforward adaptations of the arguments presented here and therefore are omitted . observe that , since we are not assuming that @xmath4 is connected , we allow for the presence of finitely many compact boundary components . if we think of @xmath9 as being the initial data set for a time - symmetric solution of einstein fields equations , then these components may be viewed as ( past or future ) trapped hypersurfaces . in fact , the rigidity statement above actually implies that , in the presence of such compact trapped hypersurfaces , the mass of each end is actually positive . this is of course related to the positive mass theorem for black holes first considered in @xcite ( see also @xcite ) . for more recent results along these lines in the classical spin setting we refer to @xcite and the references therein . this paper is organized as follows . in section [ mass ] we give a motivation for the definition of the mass , by showing that it can be approached from a variational perspective . in section [ massgeo ] we prove that the mass is a geometric invariant in the sense that it does not depend on the asymptotic structure and varies smoothly with the metric . the first proof of theorem [ mainm ] is presented in section [ firstproof ] . as already mentioned , this proof makes use of a result due to miao to reduce our positive mass theorem to the classical version , for manifolds without boundary . in section [ anotherproof ] we provide a second proof of our main theorem by adapting the arguments of schoen and yau , for dimensions up to seven , and the arguments of witten , for spin manifolds . the appendix is devoted to the proof of a technical result concerning some elliptic problems in weighted hlder spaces . * acknowledgements : * we first learned about the notion of mass in ( [ massadm ] ) from professor f. marques . we would like to thank him for suggesting us that theorem [ mainm ] should hold true and for helpful discussions . the first and second authors would like to thank the hospitality of professor a. neves at imperial college london where part of this research was carried out . while at imperial college , the first author was supported by capes / brazil and cnpq / brazil grants and the second one by a cnpq grant . in this section we show how the mass @xmath57 can be approached from a variational perspective . this not only motivates definition [ defasymflat ] but also plays a key role in the proof of the rigidity statement in theorem [ mainm ] . the arguments here are similar to those used in ( * ? ? ? * section 8) for the adm mass and we start by recalling this procedure . we consider a manifold @xmath23 of dimension @xmath2 . recall that @xmath58 , the space of riemannian metrics on @xmath23 , is an open cone in @xmath59 , the space of bilinear symmetric tensors on @xmath23 . thus , if @xmath60 and @xmath61 is small enough then @xmath62 . we recall that the corresponding variation for the scalar curvature @xmath63 is @xmath64 where @xmath65 , @xmath66 is the levi - civita connection of @xmath18 ( extended to act on tensors ) and @xmath67 is the ricci tensor . also , the variation of the volume element is @xmath68 this allows us to compute the variation of the hilbert - einstein action given by @xmath69 we have @xmath70 thus , if no boundary is present the first term in the right - hand side vanishes after integration by parts and we obtain the usual variational formula , namely , @xmath71 in particular , critical metrics for the hilbert - einstein action are precisely ricci - flat metrics . this applies if @xmath23 is closed or , more generally , if the variation @xmath72 is compactly supported . if @xmath23 is asymptotically flat ( with an empty boundary ) then it is natural to consider variations preserving this kind of structure at infinity . this time a boundary contribution appears and , as explained in @xcite , the adm mass is precisely the term that should be subtracted from @xmath73 to restore the expected form of the variational principle . more precisely , if for any such metric @xmath18 on @xmath23 we define the adm mass of @xmath9 as @xmath74 where @xmath37 is the outward unit normal to a large coordinate sphere @xmath75 in the asymptotic region , and set @xmath76 then it follows from ( [ varac ] ) that @xmath77 the obvious analogue of ( [ eulerhe ] ) . let us now assume that @xmath9 is asymptotically flat with a non - compact boundary as in theorem [ mainm ] . we claim that the natural analogue of @xmath73 is the _ gibbons - hawking - york action _ ( @xcite ) given by @xmath78 where @xmath79 and @xmath80 . as before , one must subtract the mass @xmath41 from this in order to restore the expected form of the variational principle . [ propvar ] if @xmath9 is asymptotically flat and @xmath81 then @xmath82 where @xmath83 is the shape operator of @xmath4 . we adapt a classical computation ( @xcite ) to the quantity @xmath84 where @xmath85 is the compact domain whose boundary is @xmath86 ; see figure [ fig1 ] . notice that this is _ not _ the standard ghy action for the compact manifold @xmath85 since the boundary integral over @xmath87 is missing . in order to compute @xmath88 we note that from ( [ varac ] ) the variation of @xmath89 , the hilbert - einstein action evaluated on @xmath85 , is @xmath90 where @xmath91 is the area element of @xmath92 . as usual , we adopt the index ranges @xmath93 and @xmath25 and choose coordinates around @xmath94 so that @xmath26 spans @xmath27 while @xmath28 points inwards . moreover , we assume that @xmath95 at @xmath96 . since @xmath97 the second fundamental form is @xmath98 and the mean curvature is @xmath99 this allows us to compute the variation of @xmath100 . after evaluation at @xmath96 , the final result is @xmath101 where @xmath102 is the induced connection on @xmath4 . from this and ( [ liou ] ) we see that @xmath103 on the other hand , at @xmath96 we have @xmath104 so that @xmath105 thus , if we combine ( [ inter ] ) , ( [ varform ] ) and ( [ eq : deltaa ] ) we get @xmath106 the last integral is clearly a divergence so we can rewrite this as @xmath107 it follows from the results in the next section that the last two integrals converge as @xmath13 to @xmath108 , the variation of the mass . from this , ( [ variform2 ] ) follows easily . we thus see that ricci - flat metrics are again critical for @xmath109 with respect to variations fixing the metric along the boundary ( @xmath110 ) . in this section we give a proof of the geometric invariance of the mass by adapting the standard arguments in @xcite . we also show that this invariant depends smoothly on the asymptotically flat metric , thus justifying the computation leading to ( [ variform2 ] ) . let us define the function @xmath111 as any smooth , positive extension of the asymptotic parameter @xmath112 to @xmath23 . we start by recalling the expansions of the scalar curvature and the mean curvature in the asymptotic region . [ expasymreg ] one has @xmath113 and @xmath114 where @xmath115 is the adm mass density . the expansion ( [ asymexpscal ] ) is well - known ( see @xcite ) . also , ( [ asympexpmean ] ) follows easily from ( [ unitnormalexp ] ) and ( [ meancurvloc ] ) . we now introduce the right functional spaces in order to handle this type of question . given a complete riemannian manifold @xmath23 ( with or without boundary ) , @xmath116 an integer and @xmath117 , we proceed as in @xcite and define the _ weighted _ @xmath118 _ space _ @xmath119 as the set of @xmath118 functions @xmath120 on @xmath23 for which the norm @xmath121 is finite . moreover , if @xmath122 we define the _ weighted hlder space _ @xmath123 as the set of functions @xmath124 such that the norm @xmath125 is finite . here , the supremum is over all @xmath126 such that @xmath127 is contained in a normal coordinate neighborhood of @xmath19 , and @xmath128 is the tensor at @xmath19 obtained by the pararel transport along the radial geodesic from @xmath19 to @xmath127 . we also define the _ weighted lebesgue space _ @xmath129 , @xmath130 , @xmath131 , as the set of locally integrable functions @xmath120 for which the norm @xmath132 is finite . for @xmath116 an integer , @xmath130 and @xmath131 , we define the _ weighted sobolev space _ @xmath133 to be the set of @xmath120 for which @xmath134 for @xmath135 , with the norm @xmath136 notice that for @xmath137 we recover the standard sobolev spaces , denoted simply by @xmath138 . it is easy to check that these are banach spaces whose underlying topologies do not depend on the involved choices . as stated in @xcite for manifolds without boundary , the following weighted sobolev lemma also holds in our context . [ sobolev : lemma ] let @xmath139 , @xmath140 and @xmath141 . then there are continuous embeddings @xmath142 . after fixing asymptotically flat coordinates on the end @xmath143 we consider , for each @xmath10 , the space @xmath144 of all metrics on @xmath23 so that @xmath145 if we fix a background metric @xmath146 and write @xmath147 , it is clear from proposition [ expasymreg ] that we can identify @xmath144 to a subset of the affine space @xmath148 in the topology induced by this identification we have @xmath149 if and only if @xmath150 and @xmath151 the following proposition describes the main technical result on weighted hlder spaces needed in this work . its proof is postponed to appendix [ prooftech1 ] . [ isomorphism ] let @xmath9 be an asymptotically flat manifold @xmath9 with @xmath152 , @xmath10 , and with a nonempty boundary @xmath4 . fix @xmath153 and let @xmath154 be given by @xmath155 where @xmath156 is the laplacian , @xmath7 is the outward unit normal to @xmath4 , @xmath157 and @xmath158 , for some @xmath141 small . if @xmath159 and @xmath160 then @xmath161 is an isomorphism . we can use standard interpolation methods to define @xmath162 for any @xmath163 . in particular , the restriction map @xmath164 extends continuously to the so - called _ trace map _ @xmath165 which is known to be surjective . hence , it makes sense to consider the subspaces of @xmath166 consisting of functions satisfying neumann and dirichlet boundary conditions , namely , @xmath167 and @xmath168 [ sobolev : n ] consider @xmath169 . then + ( a ) @xmath156 is an isomorphism if and only if @xmath170 ; + ( b ) @xmath156 is injective if @xmath171 ; + ( c ) @xmath156 is surjective if @xmath172 . [ sobolev : d ] consider @xmath173 . then + ( a ) @xmath156 is an isomorphism if and only if @xmath170 ; + ( b ) @xmath156 is injective if @xmath171 ; + ( c ) @xmath156 is surjective if @xmath172 . we consider the the double @xmath174 of @xmath9 along @xmath4 defined by @xmath175 , where @xmath176 for all @xmath177 , and @xmath178 for all @xmath179 and @xmath180 . although @xmath181 is not smooth on @xmath182 , it satisfies the hypotheses in ( * ? ? ? * definition 2.1 ) . then both proofs follow from ( * ? ? ? * proposition 2.2 ) by means of reflection arguments of functions on @xmath23 . the details are left to the reader . [ remsurj ] suppose @xmath183 , @xmath184 . as a consequence of proposition [ sobolev : n ] , standard arguments show that the neumann problem @xmath185 has a solution @xmath186 for any @xmath187 and @xmath188 . in fact , we can solve the cases @xmath189 and @xmath190 separately . the latter case follows directly from proposition [ sobolev : n ] . in order to solve the case @xmath189 , we choose @xmath191 such that @xmath192 . then we use proposition [ sobolev : n ] to find @xmath193 satisfying @xmath194 . thus , @xmath195 is a solution to ( [ sistsobolev ] ) when @xmath189 . a similar result holds for the dirichlet problem in ( [ sistsobolev ] ) as a consequence of proposition [ sobolev : d ] . the geometric invariance of the mass is described in the next proposition . [ massinv ] if @xmath9 is asymptotically flat with @xmath196 , @xmath32 , then the mass @xmath41 only depends on the metric @xmath18 . moreover , this dependence is smooth with respect to the topology on @xmath197 described above . the rest of this section is devoted to the proof of proposition [ massinv ] . the first step is to integrate ( [ asymexpscal ] ) over the region @xmath198 determined by two coordinates hemispheres , say @xmath87 and @xmath199 , with @xmath200 . we find that @xmath201 where @xmath202 is the portion of the boundary of @xmath198 lying on @xmath4 . on the other hand , from ( [ asympexpmean ] ) we get @xmath203 hence , @xmath204 where @xmath205 . taking into account that @xmath206 and @xmath207 , this clearly shows that the limit in the right - hand side of ( [ massadm ] ) exists and is finite for any given asymptotically flat coordinate system . if we repeat the above computation using @xmath208 instead of @xmath209 , where @xmath210 is a cutoff function which equals @xmath211 in a neighborhood of infinity , then we easily see that the mass is continuous as a function on @xmath144 , for the fixed asymptotically flat chart . since it is obviously affine , its smoothness follows at once . thus , it remains to check that the mass does not depend on the asymptotically flat chart used to compute it . for this we need to show that @xmath29 is rigid at infinity in a suitable sense . this uses harmonic coordinates as in @xcite . [ exisharmcoord ] suppose @xmath9 is asymptotically flat with @xmath196 , @xmath32 , and @xmath212 are asymptotically flat coordinates defined on @xmath143 . then there exist smooth functions @xmath213 on @xmath23 satisfying @xmath214 for @xmath215 , @xmath216 and @xmath217 moreover , the functions @xmath213 form an asymptotic flat coordinate system in a neighborhood of infinity . let us first extend @xmath218 arbitrarily to smooth functions on @xmath23 satisfying @xmath219 on @xmath220 . if @xmath221 , then @xmath222 is negative and the result follows from proposition [ isomorphism ] . indeed , we use the fact that @xmath223 and @xmath224 , for @xmath225 , to solve for @xmath226 the equations @xmath227 and @xmath228 with boundary conditions @xmath229 and @xmath230 , respectively . it is clear that @xmath231 meets the conditions of the proposition . if @xmath232 , @xmath222 may be positive , so we will need to make use of propositions [ sobolev : n ] and [ sobolev : d ] to find @xmath233 as above . by proposition [ sobolev : lemma ] , @xmath234 and @xmath235 , for any @xmath139 and @xmath141 . assuming that @xmath236 is chosen such that @xmath237 and @xmath184 , it follows from remark [ remsurj ] that there exist @xmath238 satisfying @xmath227 and @xmath239 with boundary conditions @xmath240 and @xmath230 , respectively . for @xmath241 , proposition [ sobolev : lemma ] implies that @xmath242 , and lemma [ lemma2](a ) ensures that @xmath243 . finally , we set again @xmath231 . that @xmath213 form a coordinate system in some neighborhood of infinity follows from the fact that @xmath244 . we now consider two asymptotically flat coordinate systems @xmath212 and @xmath245 on the same manifold and let @xmath213 and @xmath246 be the corresponding harmonic coordinate systems as in proposition [ exisharmcoord ] . the following result describes the relationship between these coordinate systems . [ relatcoordharm ] if @xmath9 is as in proposition [ exisharmcoord ] then there exists an orthogonal matrix @xmath247 and constants @xmath248 , @xmath249 , so that @xmath250 with @xmath251 , for @xmath215 . we consider the double @xmath174 as in the proofs of propositions [ sobolev : n ] and [ sobolev : d ] . since @xmath252 for all @xmath253 and @xmath130 , we can define functions @xmath254 by @xmath255 and @xmath256 for @xmath180 . we define @xmath257 in a similar way . although the coordinates @xmath218 and @xmath258 define different spaces @xmath259 for @xmath260 , @xmath261 is independent of the chosen coordinates as observed in @xcite . since @xmath262 and the set @xmath263 is linearly independent , we can write @xmath264 and then the result follows by using the boundary conditions on @xmath233 and @xmath265 . we observe that @xmath266 is orthogonal because the metric @xmath18 is asymptotially flat with respect to both @xmath233 and @xmath265 . by eventually composing the coordinates with rigid motions of @xmath267 we may assume that @xmath268 with @xmath269 this is the promised rigidity at infinity of @xmath29 , which we now explore to complete the proof of proposition [ massinv ] . as in @xcite we write @xmath270 in a given coordinate system @xmath212 , where @xmath271 , with @xmath272 and @xmath273 , and @xmath274 . observe that the mass density @xmath275 ( defined in proposition [ expasymreg ] ) in this same coordinate system satisties @xmath276 proceeding as in @xcite it follows from ( [ trans1])-([trans2 ] ) that @xmath277 that is , the adm densities differ by the sum of a total differential and a term that integrates to zero as @xmath13 . this is the simple but curious cancellation mentioned in @xcite . observing the chosen orientation for @xmath278 in definition [ defasymflat ] , we find that @xmath279 on the other hand , by ( [ trans1])-([trans2 ] ) we compute that @xmath280 this completes the proof of proposition [ massinv ] . [ keyrem ] in a coordinate system where @xmath281 along @xmath4 in the asymptotic region , the expression ( [ massadm ] ) simplifies to @xmath282 in particular , this takes place if the metric is conformally flat near infinity , which is the case of the metric @xmath283 constructed in proposition [ conf : flat : assump ] below . as a concrete example , consider the _ half schwarzschild space _ , which is @xmath284 endowed with the conformal metric @xmath285 thus , @xmath286 is scalar - flat with a non - compact totally geodesic boundary given by @xmath219 and a straightforward computation using ( [ massadm ] ) shows that @xmath287 this means that @xmath288 is half the adm mass of the standard schwarzschild space , which is the double of @xmath289 along its totally geodesic boundary . a similar remark applies for the mass invariants of the manifolds appearing in the doubling construction used in the proof of theorem [ mainm ] presented in the next section . in the following we denote the dependence of geometric invariants on the underlying metric by a subscript . in particular , we consider the conformal operators @xmath290 and @xmath291 acting on function on @xmath23 and @xmath4 , respectively . we also recall the function @xmath111 defined in section [ massgeo ] as any smooth , positive extension of the asymptotic parameter @xmath112 to @xmath23 . the following result shows that , under the conditions of theorem [ mainm ] , the asymptotics of the metric @xmath18 can be substantially improved . this follows an idea first put forward by schoen and yau in their celebrated proof of the classical positive mass theorem ( @xcite ) . [ conf : flat : assump ] let @xmath9 be an asymptotically flat manifold with @xmath292 , where @xmath32 , and assume that @xmath42 and @xmath43 . then for any @xmath293 small enough there exists an asymptotically flat metric @xmath294 satisfying : + i ) @xmath295 and @xmath296 , with @xmath297 and @xmath298 near infinity ; + ii ) @xmath283 is conformally flat near infinity ; + iii ) @xmath299 . our argument uses the conformal method and proceeds similarly to the proof of ( * ? ? ? * lemma 10.6 ) . let @xmath300 $ ] be a smooth cutoff function such that @xmath301 for @xmath302 and @xmath303 for @xmath304 . for @xmath305 large define @xmath306 and set @xmath307 . we are going to solve @xmath308 for @xmath309 and @xmath310 large enough , and check that the conformal metric @xmath311 has all the desired properties . we write @xmath312 and set @xmath313 where @xmath314 and @xmath315 . thus , ( [ sistema : ur ] ) is equivalent to @xmath316 for any @xmath141 we have @xmath317 and @xmath318 as @xmath319 . in what follows we are going to solve ( [ sistema : psir ] ) uniquely for @xmath320 , with @xmath321 as @xmath319 . let us fix @xmath322 . according to proposition [ isomorphism ] , the operator @xmath323 given by @xmath324 is an isomorphism . let us consider @xmath325 . it follows from the easily established estimates @xmath326 @xmath327 @xmath328 and @xmath329 that @xmath330 is arbitrarily small in the operator norm as @xmath331 . from this we conclude that @xmath332 is also an isomorphism for large @xmath310 , which provides a unique solution @xmath333 to ( [ sistema : psir ] ) . now we can choose @xmath334 for @xmath310 large , proving ( i ) and ( ii ) . it is easy to prove that @xmath335 in @xmath336 , as @xmath331 , so that the property ( iii ) also holds . we will now present our first proof of theorem [ mainm ] . we first observe that in order to prove the inequality @xmath44 , it suffices to assume that @xmath18 satisfies the conclusion of proposition [ conf : flat : assump ] . thus , we may assume that @xmath42 and @xmath43 everywhere with @xmath337 , @xmath338 and @xmath18 conformally flat outside a compact set @xmath339 . since @xmath4 is umbilic and @xmath340 outside @xmath341 it follows that @xmath342 is totally geodesic . this suggests to consider the double @xmath174 of @xmath9 along @xmath4 . more precisely , @xmath175 , where @xmath176 for all @xmath177 , and @xmath178 for all @xmath179 and @xmath180 . it is easy to check that @xmath181 is @xmath343 on @xmath344 , where @xmath345 is the double of @xmath341 . observe that , if we consider the compact hypersurface ( with boundary ) @xmath346 , we have that both @xmath347 and on @xmath348 induce the same metric on @xmath4 and hence on @xmath349 . also , since @xmath349 has nonnegative mean curvature @xmath33 with respect to @xmath350 and the unit normal @xmath7 , it has nonpositive mean curvature @xmath351 with respect to @xmath352 and the same unit normal vector @xmath7 . we can extend @xmath349 to a closed hypersurface @xmath353 in such a way that @xmath354 and @xmath7 points to the unbounded connected component of @xmath355 ; see figure [ fig2 ] . let @xmath356 be the mean curvature of @xmath353 with respect to this unbounded component and @xmath357 be the one with respect to the bounded component , both calculated using a smooth extension of @xmath7 normal to @xmath353 . observe that @xmath358 on @xmath349 , the region where @xmath359 is possibly nonsmooth . on the other hand , since @xmath181 is @xmath343 in @xmath360 , we see that @xmath361 on @xmath362 . thus , @xmath182 together with @xmath353 satisfy the assumptions of ( * ? ? ? * theorem 1 ) , which allows us to infer that the adm mass @xmath363 of the doubled manifold is nonnegative . as in remark [ keyrem ] , this mass is precisely @xmath364 , so we conclude that @xmath44 , as desired . although originally stated in a more restrictive setting , the result in @xcite also holds in our context where @xmath353 may have finitely many connnected components and the unbounded region @xmath365 determined by this hypersurface does not need to be diffeomorphic to @xmath366 minus a ball . in fact , it is enough that @xmath365 is an asymptotically flat manifold in the classical sense . we can extend this result even further to the case where @xmath367 has a finite number of ends , allowing the generalization stated in remark [ gener ] . we will now prove the rigidity statement in theorem [ mainm ] . the key result is the lemma below , which says that a manifold @xmath9 as in theorem [ mainm ] which has minimal mass necessarily is ricci - flat and has totally geodesic boundary . [ scalar : rigidity ] under the conditions of theorem [ mainm ] , if @xmath368 then @xmath369 and @xmath370 . as in the proof of ( * lemma 10.7 ) , we will use the variational characterization of the mass ; see proposition [ propvar ] . we set @xmath371 , where @xmath372 is a compactly supported symmetric @xmath373-tensor on @xmath23 . these metrics do not necessarily satisfy the positivity conditions on the scalar and mean curvatures but we remedy this as follows . we consider the linear boundary value problem @xmath374 for @xmath375 . if we write @xmath376 this is equivalent to @xmath377 where @xmath378 and @xmath379 . since @xmath380 and @xmath381 are both compactly supported and converge to zero in @xmath118 as @xmath382 , we see that @xmath383 and @xmath384 as @xmath382 . arguing as in the proof of proposition [ conf : flat : assump ] , for small @xmath385 we can solve ( [ sistema : psi ] ) uniquely for @xmath386 , with @xmath387 as @xmath382 . thus , if we set @xmath388 , it follows from ( [ sistema : phi ] ) that @xmath389 notice that we have already proved that @xmath44 for any @xmath9 as in theorem [ mainm ] . hence , the assumption @xmath390 means that @xmath18 is a minimum for the mass among the metrics @xmath391 , @xmath392 small . on the other hand , if @xmath393 we easily see that @xmath394 and @xmath395 using propositions [ propvar ] and [ massinv ] , and putting all these facts together , a straightforward computation gives @xmath396 since this holds for any compactly supported @xmath372 , the conclusion follows . the proof of theorem [ mainm ] is completed using the following lemma . [ ricci : rigidity ] if @xmath9 as in theorem [ mainm ] satisfies @xmath397 and @xmath370 then it is isometric to @xmath267 with the standard flat metric . the double @xmath398 of @xmath9 along its totally geodesic boundary @xmath4 is a complete @xmath399 asymptotically flat manifold ( with an empty boundary ) satisfying @xmath400 , which is well - known to be isometric to @xmath401 with the standard flat metric . the result follows . in this section we provide another proof of theorem [ mainm ] which is more in line with the classical arguments . thus , in subsection [ nleq7 ] we treat the case @xmath47 and in subsection [ setup ] we treat the case when @xmath23 is spin . we present below another proof of theorem [ mainm ] for this case , following the ideas in @xcite . let us assume by contradiction that the mass @xmath57 is negative . we will construct an embedded asymptotically flat minimal hypersurface @xmath402 and obtain a contradiction . the boundary @xmath4 will work as a barrier when constructing @xmath403 . by proposition [ conf : flat : assump ] we can assume that @xmath9 is asymptotically flat satisfying @xmath404 and @xmath405 near infinity . hence , @xmath406 for @xmath112 large . thus we can write @xmath407 where @xmath408 and @xmath409 is a positive constant so that @xmath410 by our assumptions . ( the proof of this asymptotic expansion is a simplified version of the arguments used for the function @xmath411 below . an alternative argument is a modification of the one in @xcite , p.83 . ) if @xmath412 and @xmath413 are positive functions decaying rapidly on @xmath23 and @xmath4 respectively , then we can find a solution to @xmath414 in the form @xmath415 , where we can make @xmath141 arbitrarily small . ( see section [ firstproof ] for the definition of @xmath416 and @xmath417 . ) in fact , as in the proof of proposition [ conf : flat : assump ] we rewrite the system ( [ eq : f1:f2 ] ) by making @xmath418 , @xmath419 , and obtain a new system in terms of @xmath420 , which is solvable if @xmath421 and @xmath422 . the solvability of this problem relies on the fact that @xmath423 , which ensures its uniqueness and allows us to use proposition [ isomorphism ] . observe also that @xmath424 has the same decay rate of @xmath425 , @xmath426 , because both @xmath5 and @xmath33 have compact support . in order to obtain the expansion for @xmath411 , we rewrite the system for @xmath420 in terms of the background metric @xmath427 , which is euclidean outside a compact set . then we proceed as in the proof of lemma [ lemma4 ] to estimate @xmath428 by a constant which can be assumed arbitrarily small by choosing @xmath429 and @xmath430 small with appropriate decay at infinity . the details are left to the reader . in particular , in what follows we can assume that @xmath431 in @xmath23 , @xmath432 in @xmath4 , @xmath404 in the end @xmath143 , and @xmath120 has the asymptotic expansion ( [ exp : u ] ) with @xmath410 . let us consider the unit vector field @xmath433 defined on @xmath143 and inward normal to @xmath434 . we can extend @xmath435 to @xmath23 in such a way that it is still a unit vector on @xmath23 and still normal to @xmath4 outside @xmath143 . as in @xcite we compute the divergence of @xmath435 to find that @xmath436 near infinity . so we can choose @xmath437 sufficiently large such that @xmath438 whenever @xmath439 . for @xmath440 large and @xmath441 , we set @xmath442 let @xmath443 be the @xmath444-hypersurface of least area having @xmath445 as its boundary . this is possible because @xmath446 along @xmath4 . we define @xmath447\}.\ ] ] we set @xmath448 for @xmath449 $ ] such that @xmath450 . for @xmath305 large , let us set @xmath451 . as in @xcite , we can choose @xmath452 large such that @xmath453 is a convex function for @xmath454 . for each @xmath455 large we choose @xmath456 such that @xmath457 by the maximum principle we see that @xmath458 . since we have changed the metric @xmath18 in such a way that @xmath4 has positive mean curvature , we have @xmath459 along @xmath4 . by continuity , one can choose @xmath460 small such that @xmath461 for any @xmath462 such that @xmath463 . in particular we can conclude that @xmath464 for some @xmath465 . this is done by means of the divergence theorem and the minimizing properties of @xmath466 . if @xmath23 has more than one end , the same arguments apply observing that @xmath467 , @xmath310 large , works as a barrier in the other ends as well , preventing @xmath468 from escaping to infinity . now the proof follows as in @xcite , observing that ( [ epsilon ] ) ensures that the boundary does not interfere when calculating the area first and second variations formula for each @xmath468 . the limiting area minimizing hypersurface @xmath403 , obtained as @xmath469 , is asymptotically flat without boundary , and we will be able to change conformally the induced metric on @xmath403 to a scalar - flat metric . in dimensions @xmath470 , by analyzing the second variation formula of area on @xmath403 we conclude that its mass is negative , contradicting the classical version of the positive mass theorem . in dimension @xmath232 , the contradiction is given by analyzing the integral of the gaussian curvature of the surface @xmath403 as well as its euler number . the rigidity statement follows as in the previous section . in this subsection we establish a witten - type formula for the mass of asymptotically flat spin manifolds in any dimension @xmath2 . as a consequence , we obtain another proof of theorem [ mainm ] in the spin setting . [ wittenform ] if @xmath9 is an asymptotically flat spin manifold of dimension @xmath2 as in theorem [ mainm ] then @xmath471 where @xmath420 is a suitable nontrivial harmonic spinor globally defined on @xmath23 . we start by recalling some basic facts on spinors in manifolds with boundary . the reader will find a more detailed account of this preparatory material in @xcite . we consider a spin manifold @xmath472 of dimension @xmath2 endowed with a riemannian metric @xmath18 . we denote by @xmath473 the spin bundle of @xmath472 and by @xmath66 both the levi - civita connection of @xmath474 and its lift to @xmath473 . the corresponding dirac operator @xmath475 is locally given by @xmath476 where @xmath477 is a local orthonormal frame and @xmath478 is the clifford product . if @xmath479 is the boundary of @xmath472 , which we assume oriented by its inner unit normal @xmath480 , then , given a spinor @xmath481 , a well - known computation gives the integral lichnerowicz formula : @xmath482 where @xmath483 for our purposes it will be convenient to rewrite the left - hand side above in terms of the mean curvature @xmath100 of @xmath479 . we first note that @xmath479 carries the spin bundle @xmath484 , obtained by restricting @xmath473 to @xmath479 . this becomes a dirac bundle if its clifford product is @xmath485 and its connection is @xmath486 where @xmath83 is the shape operator of @xmath479 . the corresponding dirac operator @xmath487 is @xmath488 a well - known computation shows that @xmath489 so that ( [ parts ] ) is equivalent to @xmath490 here we follow @xcite and discuss a certain boundary value problem for spinors on an asymptotically flat manifold with a non - compact boundary @xmath4 . we start by observing that if @xmath435 is the inward unit normal to @xmath4 then the linear map @xmath491 is a self - adjoint involution . thus , we have a ( pointwise ) decomposition , @xmath492 corresponding to the eigenbundles of @xmath493 , that is , @xmath494 . we denote the corresponding projections by @xmath495 , @xmath496 and we set @xmath497 , @xmath498 . it is easy to check that @xmath499 . [ nice ] if @xmath500 then @xmath501 . we compute @xmath502 where in the last step we used that the decomposition ( [ ortho ] ) is orthogonal . it turns out that the projections @xmath503 define nice boundary conditions for the dirac operator @xmath504 of @xmath23 . more precisely , the following result holds . [ bdp ] if @xmath9 is as in theorem [ mainm ] and @xmath505 satisfies @xmath506 , then there exists a unique @xmath507 solving the boundary value problem @xmath508 the assumption @xmath509 implies , via ( [ defdirac ] ) and cauchy - schwarz , that @xmath510 . the result is then an immediate consequence of ( * ? ? ? * corollary 3.16 ) . in this subsection we prove theorem [ wittenform ] by showing that the mass formula ( [ wittenform2 ] ) holds true . as already mentioned above , our proof adapts witten s well - known argument as reported in @xcite to the class of asymptotically flat manifolds we consider here . we first claim that , starting from an arbitrary asymptotically flat coordinate system @xmath212 , we can always produce another such coordinate system @xmath511 such that @xmath281 along @xmath4 near infinity ; see remark [ keyrem ] . in effect , it follows from ( [ unitnormalexp ] ) that @xmath512 for @xmath513 . also , from ( [ qual ] ) we see that @xmath514 and @xmath515 . thus , the claim is verified if we choose @xmath511 so that @xmath516 and @xmath517 along @xmath4 and extend this to the whole asymptotic region in the obvious manner . in such a coordinate system , we can use the simplified expression ( [ keyremeq ] ) to compute @xmath41 . with this preliminary remark at hand , we start the proof by fixing a _ constant _ spinor @xmath210 with respect to the given asymptotically flat chart , which means that @xmath518 in the asymptotic region . moreover , we may assume that @xmath519 and @xmath520 along the boundary of this region . we extend @xmath210 as zero to the rest of @xmath4 , so that @xmath520 everywhere , and finally we extend @xmath210 to the rest of @xmath23 in an arbitrary manner . the well - known formula for the spin connection shows that @xmath509 , so we may find @xmath507 as in proposition [ bdp ] . it is immediate that @xmath521 satisfies @xmath522 we now apply ( [ parts])-([parts2 ] ) in the usual way to the region @xmath85 with boundary @xmath86 to obtain @xmath523 where @xmath524 is the outward unit normal and @xmath525 denotes real part . the boundary condition @xmath526 implies , via proposition [ nice ] , that the first integral in the right - hand side vanishes . by sending @xmath13 we get @xmath527 so we must check that @xmath528 we use ( [ decompo ] ) to split the integral as @xmath529 as explained in @xcite , algebraic cancellations and the decay properties of @xmath530 and @xmath531 imply that the first three terms eventually vanish at infinity , so we must evaluate the fourth one as @xmath13 . in order to handle this limit we note that asymptotic flatness means that @xmath532 hence , the coordinate frame @xmath533 can be orthonormalized to yield @xmath534 which gives @xmath535 where from now on we represent clifford product by a dot . following @xcite we introduce the @xmath536-form @xmath537\phi,\xi\right\rangle \eh_l\lrcorner\eh_m\lrcorner dm.\ ] ] a straightforward computation gives @xmath538 where @xmath539\nabla_{\eh_m}.\ ] ] in particular , @xmath540 in witten s original argument , the boundary term in the right - hand side of ( [ lem4 ] ) vanishes because the integration in the left - hand side is performed over a closed sphere . in our case , @xmath87 is a hemisphere and this terms contributes with an integral over @xmath39 which , as we shall see , vanishes at infinity . to see this we integrate by parts to get @xmath541 where in the last step we used ( [ decompo ] ) and the fact that @xmath420 is harmonic . again due to the decay properties , the first integral in the right - hand side above vanishes at infinity . also , the standard computation as in @xcite shows that @xmath542 thus , it remains to check that @xmath543 using ( [ commut ] ) and restricting to @xmath4 in the asymptotic region we have @xmath544 since @xmath545 we see that the last term in the right - hand side integrates to zero at infinity . on the other hand , if @xmath546 means transpose conjugation then @xmath547 and hence @xmath548 by ( [ commut ] ) . by using clifford relations in the asymptotic region we get @xmath549 which gives @xmath550 . thus , the first term in the right - hand side of ( [ intzero ] ) also integrates to zero as @xmath551 . this completes the proof of theorem [ wittenform ] . with theorem [ wittenform ] at hand , we can easily produce a proof of theorem [ mainm ] in the spin case . first , it is immediate from ( [ wittenform2 ] ) that a spin manifold as in theorem [ mainm ] satisfies the mass inequality @xmath44 . moreover , if @xmath552 then @xmath9 carries a non - trivial parallel spinor , say @xmath420 . in particular , @xmath18 is ricci flat . also , since @xmath553 along @xmath4 , we see after differentiation that @xmath554 for any @xmath555 tangent to @xmath4 . since @xmath9 actually carries as many parallel spinors as the model space @xmath31 , we conclude that @xmath4 is totally geodesic and the rigidity statement follows from lemma [ ricci : rigidity ] . in this technical appendix we present a proof of proposition [ isomorphism ] . the argument follows from a series of lemmas which taken together establish the mapping properties of the operator @xmath161 appearing in the proposition . our proof is inspired by the ideas in @xcite , where the case of manifolds without boundary is treated . we set @xmath560 , where @xmath561 if @xmath562 . observe that @xmath563 for any @xmath564 , and @xmath565 for any @xmath566 and @xmath567 . then , for any @xmath567 with @xmath568 , green s formula yields @xmath569 choosing @xmath570 and using the fact that @xmath558 , one can check that @xmath571 also , since @xmath572 for any @xmath573 with @xmath570 , we get @xmath574 hence , taking the limit as @xmath331 in green s formula above and using the hypothesis @xmath575 we obtain @xmath576 since @xmath577 , we can use the fact that @xmath578 for any @xmath579 , so that it follows from ( [ lemma1:2 ] ) that @xmath580 which proves the lemma . [ lemma2 ] let @xmath9 be an asymptotically flat manifold with @xmath196 , @xmath10 , and boundary @xmath4 , and let @xmath581 . then the following assertions hold : + ( a ) there exists @xmath582 such that , if @xmath583 , @xmath584 and @xmath585 , then @xmath586 and we have @xmath587 + ( b ) assume that @xmath588 outside a compact set and @xmath153 . then there exists @xmath582 and a compact set @xmath11 such that , for all @xmath589 , @xmath590 in particular , if @xmath588 outside a compact set and @xmath153 , then there exists @xmath582 and a compact set @xmath11 such that , if @xmath591 , then @xmath586 and we have @xmath592 we can identify the end @xmath143 with @xmath593 under the given asymptotically flat chart . for @xmath594 we will denote by @xmath595 the compact set @xmath596 . finally , for any subset @xmath597 , we define @xmath598 . for the proof of item ( a ) , let us set @xmath599 and @xmath600 . for @xmath594 we also set @xmath601 and @xmath602 so that @xmath603 and @xmath604 . let @xmath605 be a smooth cutoff function satisfying @xmath606 in @xmath607 and @xmath608 in @xmath609 , and let @xmath583 . we set @xmath610 for @xmath611 and define a metric @xmath612 on @xmath83 by @xmath613 . it follows from elliptic regularity that @xmath614 and @xmath615 for some @xmath616 . observe that @xmath617 @xmath618 and @xmath619 so that @xmath620 expanding this in terms of @xmath621 norms , multiplying by @xmath622 and rewriting the result in terms of @xmath120 and @xmath18 , we get @xmath623 since this holds for r arbitrarily large , we conclude that @xmath586 . let us now consider a smooth cutoff function @xmath628 satisfying @xmath629 in @xmath630 and @xmath631 in @xmath632 . by elliptic regularity , @xmath633 so that @xmath634 the estimate in item ( a ) follows immediately from ( [ lemma2:5 ] ) and ( [ lemma2:6 ] ) . in order to prove item ( b ) , assume that @xmath588 in @xmath635 , @xmath594 , and consider a smooth cutoff function @xmath636 satisfying @xmath629 in @xmath595 and @xmath631 in @xmath637 . since @xmath638 has support in @xmath143 , the restriction of @xmath638 to @xmath143 can be seen as a function @xmath639 . hence , according to lemma [ lemma1 ] , there exists @xmath557 such that @xmath640 since @xmath641 we obtain @xmath642 which clearly implies the estimate in item ( b ) . [ lemma3 ] let @xmath9 be as in lemma [ lemma2 ] and consider the operators @xmath643 and @xmath644 where @xmath157 and @xmath158 . if @xmath153 , we define by @xmath645 the operator @xmath646 if @xmath161 is injective then there holds @xmath647 for all @xmath586 and some @xmath648 . we retain the notation in the proof of lemma [ lemma2 ] . we consider a smooth cutoff function @xmath636 satisfying @xmath629 in @xmath649 and @xmath631 in @xmath650 . since the support of @xmath651 is contained in @xmath652 , it makes sense to define @xmath653 , @xmath594 , by @xmath654 , which is supported in @xmath655 . also , let us define the metric @xmath612 on @xmath23 by @xmath656 , so that @xmath657 in @xmath637 . for later use we observe that @xmath658 by the last assertion in lemma [ lemma2 ] there exists @xmath659 such that @xmath660 for some large @xmath661 . let us first estimate @xmath662 . using the standard coordinate expression for the laplacian we can verify that @xmath663 also , using ( [ lemma3:1 ] ) we have @xmath664 where the constant @xmath665 is independent of @xmath310 , which implies @xmath666 on the other hand , writing @xmath667 we have @xmath668 where the last constant @xmath665 depends on @xmath669 . thus , @xmath670 similarly , if we make use of ( [ unitnormalexp ] ) we obtain @xmath671 where @xmath665 depends on @xmath672 , so that ( [ lemma3:2 ] ) , ( [ lemma3:5 ] ) and ( [ lemma3:8 ] ) lead to @xmath673 hence , if we choose @xmath310 large we finally obtain the key estimate @xmath674 the rest of the proof of lemma [ lemma3 ] will follow by a contradiction argument using the injectivity assumption and ( [ lemma3:9 ] ) . indeed , assuming that ( [ lemma3:0 ] ) does not hold we can choose @xmath675 satisfying @xmath676 in particular , as @xmath677 , @xmath678 since @xmath679 we can assume that @xmath680 converges in @xmath681 . then , using ( [ lemma3:9 ] ) with @xmath682 , we see that @xmath680 is a cauchy sequence in @xmath683 . hence , this sequence converges in @xmath683 to some @xmath684 with @xmath685 . the fact that @xmath686 is a continuous operator together with ( [ lemma3:10 ] ) implies that @xmath687 and @xmath688 . thus , @xmath689 by the injectivity hypothesis . this contradicts the fact that @xmath685 and concludes the proof of lemma [ lemma3 ] . [ lemma4 ] let @xmath9 be an asymptotically flat manifold as in lemma [ lemma3 ] . if @xmath153 consider the operator @xmath690 given by @xmath691 . if @xmath588 outside a compact subset of @xmath23 then @xmath161 is an isomorphism . we use the notation in the proof of lemma [ lemma3 ] . we choose @xmath310 large so that @xmath588 in @xmath692 and consider the set @xmath693 , which we still denote by @xmath143 . then the diffeomorphism @xmath694 extends to a coordinate system @xmath695 . moreover , if @xmath696 is the canonical frame on @xmath697 , then @xmath210 induces the coordinate frame @xmath698 on @xmath699 . let us define a metric @xmath700 on @xmath23 , where @xmath701 is a positive smooth function on @xmath23 satisfying @xmath702 for all @xmath703 . observe that @xmath181 can be extended to a smooth metric on @xmath704 , still denoted by @xmath181 . in fact , for any @xmath703 , we have @xmath705 . by eventually multiplying @xmath709 and @xmath710 by real constants and using the notation of section [ firstproof ] , this is equivalent to @xmath711 and using that @xmath712 and @xmath713 , this becomes @xmath714 where @xmath715 . now we shall find @xmath716 solving @xmath717 so that it solves ( [ lemma4:3 ] ) in particular . to that end it suffices to prove uniqueness for this last problem . thus , suppose that @xmath411 satisfies @xmath718 then elliptic regularity implies that @xmath716 and we shall see that actually @xmath719 . since multiplication is continuous in weighted hlder spaces ( see ( * ? ? ? * lemma 1 ) ) , it follows that @xmath720 for any @xmath116 . but @xmath721 satisfies @xmath722 which implies @xmath689 , as we can check by a simple integration by parts . hence , @xmath719 and we have uniqueness for the problem ( [ lemma4:3 ] ) . ( in particular , @xmath161 is injective . ) thus , we can always find a solution @xmath716 to ( [ lemma4:3 ] ) and hence a solution @xmath586 to ( [ lemma4:1 ] ) in case both @xmath709 and @xmath723 are compactly supported . let us now consider the general case where @xmath707 and @xmath708 . we want to find @xmath586 such that @xmath724 . if @xmath725 and @xmath726 we can find sequences @xmath727 and @xmath728 such that , as @xmath729 , @xmath730 and @xmath731 by the special case already proved , we can find @xmath732 so that @xmath733 . it follows from lemma [ lemma3 ] that , as @xmath734 , @xmath735 and @xmath736 hence , we can assume that @xmath737 in @xmath738 for some @xmath586 . as a consequence , @xmath739 in @xmath740 and @xmath741 in @xmath742 , as @xmath677 , and the result follows from the fact that @xmath743 in @xmath740 and @xmath744 in @xmath742 . first observe that all the operators @xmath161 as in the proposition are injective , as we can see by applying the maximum principle . let @xmath745 be the set of all these operators and let @xmath746 be the subset of isomorphisms . we consider @xmath745 with the operator norm topology . it follows from the implicit function theorem that @xmath209 is open in @xmath745 . we will prove that it is also closed . we set @xmath747 , @xmath748 and consider @xmath749 with the norm @xmath750 let @xmath751 be a sequence converging to some @xmath752 under the operator norm @xmath753 . we shall prove that @xmath161 is surjective . given @xmath754 we must find @xmath586 such that @xmath755 . let us write @xmath756 . by hypothesis , there exists @xmath757 satisfying @xmath758 , so that , by lemma [ lemma3 ] , there exists @xmath659 such that @xmath759 for all @xmath760 . in particular , @xmath761 is uniformly bounded in @xmath762 . we just need to prove that @xmath765 in @xmath766 and @xmath767 in @xmath768 to conclude that @xmath769 . observe that @xmath770 implies that @xmath771 and @xmath772 . we also have @xmath773 the first term on the right - hand side of ( [ propo1:1 ] ) converges to zero because @xmath737 in @xmath738 . as for the second one , @xmath774 since @xmath775 is uniformly bounded . this proves that @xmath776 . the proof that @xmath777 is similar , which proves that @xmath209 is closed in @xmath745 . using the notation in the proof of lemma [ lemma3 ] , we consider the family of metrics @xmath612 for @xmath594 , and observe that the operators of the form @xmath778 are isomorphisms , according to lemma [ lemma4 ] . thus , we just need to find a continuous family of injective operators connecting one of those operators to @xmath779 . this is easily accomplished if we set @xmath780 , @xmath781 and define @xmath782 for @xmath783 and @xmath784 . schoen , r. : variational theory for the total scalar curvature functional for riemannian metrics and related topics . `` topics in calculus of variations '' , lecture notes in mathematics , springer - verlag , new york , v. 1365 ( 1989 )

given an-tuple of self - adjoint operators in a finite von neumann algebra and a faithful , normal tracial state on , we define a map from to by . the image of the positive part of the unit ball under is called the * spectral scale * of relative to and is denoted by . in a previous paper with nik weaver we showed that the geometry of reflects spectral data for real linear combinations of the operators \{}. for example , we showed that an exposed face in is determined by a certain pair of spectral projections of a real linear combination of the s . in the present paper we extend this study to faces that are not exposed . in order to do this we need to introduce a recursive method for describing faces of compact convex sets in . using this new method , we completely describe the structure of arbitrary faces of in terms of and . we also study faces of convex , compact sets that are exposed by more than one hyperplane of support ( we call these * sharp faces * ) . when such faces appear on , they signal the existence of commutativity among linear combinations of the operators \{}. although many of the conclusions of this study involve too much notation to fit nicely in an abstract , there are two results that give their flavor very well . theorem 6.1 : if the set of extreme points of is countable , then is abelian . corollary 5.6 : has a finite number of extreme points if and only if is abelian and has finite dimension .

math0010271

in this work we continue the investigation begun in @xcite of the spectral relationships within an @xmath0-tuple of self adjoint operators in a finite von neumann algebra as reflected in the geometry of the corresponding spectral scale . let us begin by introducing some notation and reviewing some of the results in @xcite . * the following notation will apply throughout the rest of the paper without explicit reference . * let @xmath2 denote a finite von neumann algebra equipped with a faithful normal tracial state @xmath3 , let @xmath0 denote a positive integer , and let @xmath13 denote an @xmath0tuple of self adjoint operators in @xmath2 . we define a map @xmath4 from @xmath2 to @xmath14 by the formula @xmath15 and write @xmath16 , where @xmath17 . since @xmath3 is normal and @xmath18 is weak * compact and convex , it follows that @xmath8 is a compact , convex subset of @xmath14 . we call @xmath8 the * spectral scale * of the @xmath10 s * relative to @xmath3*. we use parentheses [ @xmath19 to denote vectors in @xmath5 and angle brackets [ @xmath20 to denote inner products . 1 . we let @xmath12 denote the von neumann subalgebra of @xmath2 generated by @xmath13 and the identity . if @xmath21 and @xmath22 are projections in @xmath2 and @xmath23 , then the * order interval * that they determine is @xmath24 = \{a\in m^+_1 : p \le a \le q\}.\ ] ] 3 . for each nonzero vector @xmath25 we write @xmath26 4 . by a * spectral pair * we mean a pair of the form @xmath27 , where @xmath28 is a real number and @xmath29 is a nonzero vector in @xmath5 . our standard way of embedding @xmath5 into @xmath14 is @xmath30 . if @xmath27 is a spectral pair , then @xmath31 and @xmath32 denote the spectral projections of @xmath33 determined by the intervals @xmath34 $ ] and @xmath35 . we call @xmath36 the * spectral interval projections * determined by @xmath28 and @xmath37 . 6 . if @xmath27 is a spectral pair and @xmath38 is a real number , then @xmath39 denotes the hyperplane in @xmath14 defined by the formula @xmath40 and we write @xmath41for the half space defined by the inequality @xmath42 7 . if @xmath43 , then the * isotrace slice * of @xmath8 at @xmath28 is by definition @xmath44 the notion of an isotrace slice may seem artificial , but , as we shall show in a forthcoming paper , for @xmath45 it essentially contains the more familiar concept of the numerical range of the operator @xmath46 . isotrace slices are particularly useful when there are two operators so that @xmath47 as will be seen in 6 of this paper . as shown in @xcite , the geometry ( i.e. the facial structure ) of the spectral scale @xmath8 contains information about the spectrum and spectral projections of real linear combinations of the @xmath13 . in this paper we complete this part of the theory by giving an exact description of an arbitrary face of @xmath8 in terms of the @xmath10 s . in @xcite this was done only for the case @xmath48 , and it is useful to begin by presenting a description of these results . assume @xmath49 and @xmath50 . thus , @xmath8 is a convex , compact subset of the plane so that a face in @xmath8 is either an extreme point or a line segment . also , for each real @xmath28 write @xmath51 , ( resp . , @xmath52 ) for the spectral projection of @xmath53 corresponding to the interval @xmath54 $ ] ( resp . , @xmath55 ) . finally write @xmath56 for the spectrum of @xmath53 . the results obtained in ( * ? ? ? * theorems 1.5 and 1.6 ) may be summarized as follows . 1 . @xmath8 lies between the lines @xmath57 and @xmath58 and has `` sharp points '' at @xmath59 and @xmath60 . thus , the boundary of @xmath8 is divided into an upper and lower boundary . the spectrum of @xmath53 is exactly the set of slopes of the tangent lines to the lower ( or upper ) boundary of @xmath8 . a point @xmath61 in @xmath8 is an extreme point on the lower boundary if and only if @xmath61 has the form @xmath62 for some @xmath63 . extreme points on the upper boundary have the form @xmath64 . 4 . the line segments in the lower ( or upper ) boundary of @xmath8 are in one - to - one correspondence with the eigenvalues of @xmath53 . if @xmath28 is an eigenvalue for @xmath53 so that @xmath65 and @xmath66 is the line segment on the lower boundary of @xmath8 that it determines , then @xmath67)$ ] . the corners in the lower ( or upper ) boundary of the spectral scale are in one - to - one correspondence with gaps in the spectrum of the operator @xmath53 . let us now review the basic facts obtained in @xcite for the higher dimensional case . since we shall refer to these facts repeatedly , it is convenient to give them local numbers . the following is a restatement of the results in theorems 2.3 and 2.4 in @xcite . the following statements hold . 1 . if @xmath61 is an extreme point of @xmath8 , then there is a projection @xmath21 in @xmath12 such that @xmath68 further , @xmath69 and @xmath70 are extreme points of @xmath8 for every spectral pair @xmath27 . we have @xmath71 . the hyperplane @xmath72 is a hyperplane of support for @xmath8 with @xmath73 if and only if @xmath74 in this case we have @xmath75 . 3 . if @xmath76 , then @xmath77 is a face in @xmath8 . moreover , @xmath78\ ] ] and @xmath79).\ ] ] recall that a face @xmath66 of a convex set @xmath80 is said to be an * exposed face * if there is a hyperplane of support @xmath81 for @xmath80 such that @xmath82 . if the exposed face @xmath66 is a single point @xmath61 , and so necessarily an extreme point of @xmath8 , we call @xmath61 an * exposed point*. ( see 2 for a more detailed review of these notions ) . although it was not specifically noted in @xcite , theorem 0.2 allows the complete classification of the exposed faces in the spectral scale which we record below . the following statements hold . 1 . @xmath66 is an exposed face in @xmath8 if and only if there is a spectral pair @xmath27 such that @xmath83).\ ] ] 2 . if @xmath61 is an extreme point in the spectral scale , then @xmath61 is an exposed point if and only if there is a spectral pair @xmath27 such that @xmath84 and @xmath85 . if @xmath66 is a face that is not an extreme point , then @xmath66 is an exposed face if and only if there is a spectral pair @xmath27 such that @xmath28 is an eigenvalue for @xmath37 . thus , in this case @xmath86 and @xmath87 . the first conclusion follows immediately from parts @xmath88 and @xmath89 of theorem 0.2 . suppose @xmath61 is an exposed point so that by theorem 0.2 and part 1 of this corollary we have @xmath90 ) = \psi({p_{s,{\mathbf t}}^{\pm}}),\ ] ] where @xmath27 is a spectral pair . write @xmath91 and @xmath92 if @xmath93 , then there is a nontrivial projection @xmath94 in @xmath95 such that @xmath96 . but in this case , we have @xmath97 so that @xmath98 and @xmath99 ) \ne \{{\mathbf x}\}$ ] . as this is a contradiction , we get @xmath84 , as desired . conversely , if @xmath27 is a spectral pair such that @xmath84 and @xmath85 , then @xmath100)\ ] ] is a singleton and so this face is an exposed point . finally , by part 1 and the fact that @xmath4 is faithful , we have that @xmath101 exactly when the face @xmath102)$ ] has positive dimension . the results to be presented below concern the following topics . @xmath103 a general analysis of a face of a compact , convex set in @xmath5 . ( 2 and 5 ) . @xmath103 a complete description of the facial structure of the spectral scale in the general case . ( 3 ) . @xmath103 an analysis of the `` corners '' of the spectral scale in higher dimensions . ( 4 ) . @xmath103 applications of the results in 4 to show how geometric properties of @xmath8 imply the existence of central projections in @xmath12 . ( 5 and 6 ) . let us now describe our results in more detail . although we defined exposed faces of convex compact subsets in @xmath5 using hyperplanes of support , there is also an equivalent formulation in terms of linear functionals . from this point of view , a face @xmath66 in the ( convex , compact ) set @xmath80 is said to be * exposed * if there is linear functional @xmath104 and a scalar @xmath105 such that @xmath106 for each @xmath61 in @xmath66 and @xmath107 for each @xmath61 in @xmath108 . in this case , we say that @xmath104 * exposes * @xmath66 . since the former definition is more geometric it fits better with the emphasis of this paper , but it will sometimes be convenient to use the latter , more algebraic , definition . observe that not every face need be exposed . for example , it is easy to construct a convex set in two dimensions that has a face in its boundary of dimension one such that the end points of this line segment ( which are extreme points ) are not exposed . we wish to make distinctions among exposed faces as follows . the exposed face @xmath66 is said to be have * degree * @xmath109 if @xmath110 is the cardinality of the largest linearly independent set of linear functionals that expose @xmath66 . in terms of hyperplanes , this means that there are @xmath110 hyperplanes of support for @xmath66 such that their normal vectors are linearly independent . we study such faces in 4 . in 3 we show that a general face of the spectral scale provides spectral information for linear combinations of the defining @xmath10 s and certain cut downs of these operators . if @xmath66 is not exposed , then a condition similar to that of corollary 0.1 ( 1 ) holds , except that it is first necessary to cut down by a certain spectral projection . this analysis allows us to complete the characterization of the extreme points of the spectral scale which was begun in @xcite . generalized `` corners '' are studied in 4 . a corner of a convex planar set is a point on the boundary that admits more than one tangent line of support . the natural generalization of this notion in higher dimensions is a face that admits more than one hyperplane of support , i.e. , a face of degree @xmath111 for some @xmath112 . we call such faces * sharp faces * since they generalize sharp ( i.e. , nondifferentiable ) corners on the boundary of a 2 dimensional convex set . our main result on this topic can be described as follows . in two dimensions , a corner of a convex set admits precisely two tangent lines of support with linearly independent normal vectors . in higher dimensions , there can be much wider variation and a new phenomenon occurs . specifically , in the case of a spectral scale formed by a single operator , a tangent line of support for a face of the spectral scale is determined by real numbers @xmath28 and @xmath38 , where @xmath113 . in the general case where there are @xmath0 @xmath10 s , the hyperplanes of the support for a sharp face are determined by a sequence @xmath114 of spectral pairs and real numbers @xmath115 . for example , suppose @xmath66 is a sharp face which is contained in hyperplanes of the form @xmath116 and @xmath117 , where the spectral pairs @xmath118 and @xmath119 are linearly independent . if @xmath120 and @xmath121 , then , just as in two dimensions , the interval @xmath122 lies in a gap in the spectrum of @xmath37 . on the other hand if @xmath123 is a linearly independent set , then something new occurs . in this case , there is a projection @xmath124 that commutes with @xmath125 and @xmath126 and such that @xmath127 if there are @xmath111 hyperplanes containing @xmath66 , with linearly independent vectors @xmath128 , then the projection @xmath124 commutes with each @xmath129 and the compression of each @xmath129 to @xmath124 is a scalar . next , we observe in 5 that if @xmath130 for the sharp face @xmath66 of the previous paragraph , then the projection @xmath124 commutes with each @xmath10 and so @xmath124 is a nontrivial central projection in @xmath12 . put colloquially , if we can `` wobble '' a sharp face in all @xmath0 `` @xmath29directions '' , then the center of @xmath12 is not trivial . 6 is devoted to showing that if @xmath8 has a countable number of extreme points , then @xmath12 is abelian . as noted in the introduction , one can `` see '' if @xmath12 is abelian from the shape of the spectral scale in the finite dimensional case , but the situation in infinite dimensions is much more complicated . in this section we present several examples that illustrate this point . other examples of the spectral scale in @xmath131 were presented in @xcite at the end of 1 . examples for which @xmath12 is finite dimensional and noncommutative appear in a forthcoming paper . we begin with an example where @xmath12 is abelian and @xmath8 has a countable number of extreme points . let @xmath53 denote the diagonal operator on @xmath132 defined by @xmath133 , and let @xmath2 be the von neumann algebra generated by @xmath53 and the identity ( so that @xmath134 ) . also , define a trace @xmath3 on @xmath2 by @xmath135 . since the spectrum of @xmath53 is exactly @xmath136 together with 0 , it is countable . by theorem 2.2 the spectral scale of @xmath137 is a 2 dimensional , convex , compact set with upper boundary curve consisting of line segments with decreasing slopes specifically the slopes @xmath136 . thus the algebra @xmath12 is abelian ( it equals @xmath2 ) and infinite dimensional , and the set of extreme points of @xmath8 is countable . as the next example shows , even if @xmath12 is abelian and all linear combinations of @xmath138 have countable spectrum , @xmath8 may still have uncountably many extreme points . let @xmath2 and @xmath3 be as in example 1.1 and select positive real sequences @xmath139 and @xmath140 , each of which converges to 0 , and such that the sequence @xmath141 takes on every rational value between 0 and infinity . also , let @xmath142 and @xmath143 denote the corresponding diagonal operators in @xmath144 . since the two sequences converge to 0 , every linear combination of @xmath142 and @xmath143 is compact . in particular , each linear combination of @xmath142 and @xmath143 has countable spectrum . to see that @xmath8 has an uncountable number of extreme points , fix an irrational number @xmath145 , write @xmath146 and consider the operator @xmath147 . observe that @xmath148 thus , if we write @xmath149 then @xmath150 is the projection onto the subspace of @xmath151 spanned by the corresponding standard basis vectors and we have @xmath152 hence , if @xmath153 and we write @xmath154 , then @xmath155 , @xmath156 and so @xmath157 hence there are an uncountable number of spectral projections of this form and therefore by part @xmath158 of theorem 0.2 , the set of extreme points of @xmath8 is uncountable . as the next two examples show , it is not always obvious " from the geometry of b whether or not @xmath12 is abelian . finding a condition on @xmath8 that is necessary and sufficient for commutativity of @xmath12 is one of the major unsolved problems in the theory . the pictures of the spectral scales in these examples are rather complicated . the details are much clearer in color . to view the pictures , go to http://www.math.psu.edu/anderson/pictures . suppose that @xmath159 , @xmath142 is multiplication by @xmath160 , @xmath143 is the projection determined by the characteristic function of the interval @xmath161 $ ] and the trace @xmath3 is integration . thus , in this case @xmath12 is abelian . the spectral scale for these operators resembles two bed sheets , tied at their corners and billowing outwards from one another . the coinciding edges of the bed sheets form the borders of 4 two dimensional faces with parabolic boundaries that meet at the corners of the bed sheets . these faces have the same shape as the spectral scale for multiplication by @xmath160 . ( see ( * ? ? ? * end of 1 ) ) . thus , the spectral scale of @xmath142 is visible in @xmath8 . the spectral scale of @xmath143 ( which is just a parallelogram ) is also visible as the convex hull of the corners of the bed sheets . now let us consider a non commutative example . put @xmath162 and write @xmath163 where @xmath164 is multiplication by @xmath160 on @xmath161 $ ] . thus , @xmath142 is still multiplication by @xmath160 on @xmath165 $ ] , but the projection @xmath143 now does not commute with @xmath142 . in this case the spectral scale is quite similar to the scale in the previous example , although the `` billowing '' of the bed sheets is more pronounced . so , based on this evidence , there does not seem to be an obvious visual method for determining when @xmath12 is abelian in the infinite dimensional case . how should one describe a face @xmath66 of a convex set @xmath80 in @xmath5 ? assuming that we have already some description of @xmath80 , it is natural to add some conditions to that description in order to distinguish the points of @xmath66 from the other points of @xmath80 . the most natural conditions , from the convexity viewpoint , should be expressed in terms of linear functionals . e.g. @xmath166 . this only works if the linear functional @xmath104 exposes the face @xmath66 . many faces are not exposed , so how can we describe them in some systematic way using linear functionals ? the answer to that question is to use more than one linear functional via the recursive process described in this section . we begin by reviewing some of the standard notions in the theory of convexity and then present some general results that will be used in the sequel . although some of the material that appears here does not seem to be in any of the standard texts on convexity , it is quite possible that some of the unattributed results are known . if @xmath167 is a nonzero vector in @xmath5 and @xmath38 is a real number then we write @xmath168 for the * hyperplane * in @xmath5 determined by @xmath29 and @xmath38 . the half spaces of @xmath5 determined by the hyperplane @xmath169 are given as follows : @xmath170 thus , @xmath169 is the unique hyperplane in @xmath5 passing through @xmath61 and orthogonal to @xmath29 and so it is natural to call the vector @xmath29 a * normal vector * for this hyperplane . now fix a convex subset @xmath80 of @xmath5 . the hyperplane @xmath169 is a * hyperplane of support * for @xmath80 if there is a vector @xmath61 in the boundary of @xmath80 with @xmath171 and we have @xmath172 note that in order for @xmath169 to be a hyperplane of support we just as easily could have required @xmath173 . in fact this is often the definition used in many texts on convexity . we prefer to use the positive half space since it is more natural in our applications . note that @xmath169 is a hyperplane of support for @xmath80 containing the vector @xmath174 if and only if @xmath175 if @xmath80 is compact and @xmath29 is any nonzero vector in @xmath5 , then it is straightforward to show that there is a unique real number @xmath38 such that @xmath169 is a hyperplane of support for @xmath80 . further , if @xmath169 is a hyperplane of support for @xmath80 , then @xmath176 is a face in the boundary of @xmath80 . let us now record a few general results on faces in convex sets . as above , the symbol @xmath80 stands for a convex set in @xmath5 . the proof of the following theorem is essentially the same as the proof of theorem 2.6.17 in @xcite , although its statement is a bit more detailed than the one given there . if @xmath177 are hyperplanes of support for @xmath80 such that @xmath178 @xmath179 are positive real numbers and we write @xmath180 then @xmath181 is a hyperplane of support for @xmath80 and @xmath182 in particular , @xmath183 is an exposed face of @xmath80 . fix @xmath184 . we have then that @xmath185 for each @xmath186 and so @xmath187 hence @xmath188 and therefore , @xmath189 now fix @xmath188 so that @xmath190 if we had @xmath191 for some @xmath186 , then since @xmath192 we would have @xmath193 and therefore @xmath194 contradicting @xmath195 . hence , @xmath196 and so @xmath197 finally , since @xmath198 by hypothesis , we get that @xmath183 contains a point on the boundary of @xmath80 . further , if @xmath192 , then we have @xmath199 and so @xmath200 . hence , @xmath201 @xmath181 is a hyperplane of support for @xmath80 , and @xmath183 is an exposed face of @xmath80 . @xmath158 observe that if @xmath202 , then @xmath203 . thus , if we let @xmath204 and replace @xmath29 and @xmath38 with @xmath205 and @xmath206 , then we have that @xmath207 is a convex combination of the @xmath208 s . hence , the proposition is equivalent to one that only considers convex combinations of vector in @xmath5 . @xmath88 if @xmath66 is a face in the boundary of @xmath80 , then there is a hyperplane of support @xmath81 for @xmath80 that contains @xmath66 and such that @xmath81 is disjoint from the relative interior of @xmath80 . ( see ( * ? ? ? * theorem 11.3 ) , for example ) . the following result is merely a restatement of corollary 2.6.9 in @xcite the intersection of any family of faces of @xmath80 can be expressed as an intersection of @xmath209 or fewer members of the family . if @xmath66 is any nonempty face of @xmath80 , then there is a ( unique ) minimal exposed face @xmath210 such that @xmath211 let @xmath212 denote the family of all exposed faces that contain @xmath66 . we have that @xmath213 by remark @xmath88 above . write @xmath214 we have that @xmath210 is the intersection of a finite number of faces in @xmath212 by theorem 2.3 . since @xmath215 this intersection is not empty and so @xmath210 is an exposed face by proposition 2.1 . it is clear from its definition that @xmath210 is the unique minimal exposed face containing @xmath66 . if @xmath66 is a face in a convex subset @xmath80 of @xmath5 , then there is a unique sequence @xmath216 such that 1 . @xmath217 is the unique minimal exposed face in @xmath80 containing @xmath66 . if @xmath218 , then @xmath219 is the unique minimal exposed face of @xmath220 that contains @xmath66 . 3 . @xmath221 . the assertions are easily established using corollary 2.3 and a standard induction argument . let us call the chain @xmath222 described above the * minimal exposed facial chain * determined by @xmath66 . with these preparations , we may now describe how general ( i.e. , non exposed ) faces of @xmath8 are determined by spectral properties of certain operators in @xmath12 . it turns out that these faces are described exactly as exposed faces are except that it is necessary to work in a cut down " von neumann algebra . the key result that makes this analysis possible is the fact that a face ( of positive dimension ) in @xmath223 may be identified as a spectral scale in its own right . this scale is determined by cut down operators of the form @xmath224 in the cut down von neumann algebra @xmath225 . in order to describe the situation more precisely , some new notation is required . suppose that @xmath66 is a proper face of @xmath8 with dimension greater than zero ; i.e. @xmath66 is not an extreme point . in the next lemma , it will be shown that @xmath66 is an affine translate of a certain related spectral scale . the new spectral scale is determined as follows . by ? * theorems 1.1 and 2.4 ) , we have @xmath226)$ ] where @xmath227 are projections in @xmath12 . write @xmath228 and note that since @xmath66 is a proper face that is not an extreme point , @xmath229 and therefore @xmath230 . next put @xmath231 , write @xmath232 and define @xmath233 on @xmath95 by the formula @xmath234 finally , let @xmath235 denote the spectral scale determined by @xmath233 relative to @xmath236 . suppose @xmath237)$ ] is a proper face in @xmath8 that is not an extreme point . using the notation developed in the preceding paragraph , the map @xmath238 defined by the formula @xmath239 implements an affine isomorphism of @xmath66 onto @xmath235 . hence we have @xmath240 observe that we have the relation @xmath241 now fix @xmath242 so that we have @xmath243 for some @xmath244 $ ] . if we write @xmath245 then we have @xmath246 and therefore @xmath247 . further , since every vector in @xmath235 has the form @xmath248 for some @xmath249 , this same calculation shows that @xmath238 is surjective . finally , suppose @xmath250 and @xmath251 are vectors in @xmath66 such that @xmath252 . as @xmath250 and @xmath251 are in @xmath66 , we may find operators @xmath253 and @xmath254 in @xmath255 $ ] such that @xmath256 and we may write @xmath257 where @xmath258 . with this we have @xmath259 by @xmath195 . as @xmath260 we get @xmath261 and so @xmath238 is injective . hence , @xmath238 implements an affine isomorphism of @xmath66 onto @xmath235 and therefore @xmath262 the next lemma shows that if b is contained in one or more hyperplanes , then this geometric condition is equivalent to the existence of linear dependence relationships ( corresponding to the hyperplanes that contain @xmath8 ) among @xmath263 . if @xmath264 , and the span of @xmath265 has dimension @xmath110 , then there are real numbers @xmath266 and linearly independent vectors @xmath267 such that @xmath268 for @xmath269 . conversely , if there are real numbers @xmath266 and linearly independent vectors @xmath267 such that @xmath268 for @xmath269 , then the subspace spanned by @xmath8 has dimension at most @xmath110 . suppose @xmath270 is a hyperplane in @xmath14 that contains @xmath8 and observe that since @xmath271 , we may take @xmath272 . in this case , we have @xmath273 for each @xmath274 . now select @xmath145 such that @xmath275 . setting @xmath276 we get @xmath277 and since @xmath3 is faithful @xmath278 . similarly , we get @xmath279 and so @xmath280 . if @xmath8 has dimension @xmath110 , then there are @xmath281 hyperplanes of the form @xmath282 that contain @xmath8 and such that their normal vectors @xmath283 are linearly independent . in fact we must have that the @xmath208 s are linearly independent . to see this , suppose that there are real numbers @xmath284 such that @xmath285 in this case we get @xmath286 but , then @xmath287 would be a linear combination of the remaining normal vectors contradicting their linear independence . for the converse , if real scalars @xmath288 and linearly independent vectors @xmath289 exist such that @xmath268 for all @xmath290 , then it immediately follows that every @xmath61 in @xmath8 is orthogonal to each vector @xmath291 and so the subspace spanned by @xmath8 has dimension at most @xmath110 . using the notation developed for lemma 3.1 and letting @xmath292 be the von neumann algebra generated by @xmath293 , the following statements hold . 1 . @xmath66 has dimension one if and only if @xmath292 has dimension one . 2 . if @xmath66 has dimension two , then @xmath292 is abelian . we have @xmath294 . suppose that @xmath292 has dimension one , i.e. that there are scalars @xmath295 such that @xmath296 for each @xmath186 . in this case , if @xmath297 so that @xmath298 , where @xmath299 , then we have @xmath300 since @xmath301 , we get that @xmath66 is the line segment with end points @xmath302 and @xmath303 and so @xmath66 has dimension one . if @xmath66 has dimension one , then by lemma 3.1 @xmath235 has dimension one . thus by lemma 3.2 there are @xmath0 linearly independent vectors @xmath304 and real numbers @xmath305 such that @xmath306 . since the @xmath208 s are linearly independent , we get that each @xmath307 is a multiple of @xmath124 . hence assertion @xmath158 holds . now suppose @xmath66 has dimension two so that @xmath235 has dimension two by lemma 3.1 . in this case , by lemma 3.2 , there are @xmath308 linearly independent vectors @xmath208 and real numbers @xmath309 such that @xmath306 for @xmath310 . this means that there is a vector @xmath29 such that @xmath311 and so @xmath292 is abelian . observe that the spectral scale described in example 1.3 has two dimensional faces such that @xmath95 is abelian ( in fact @xmath2 is abelian in this case ) , but @xmath95 is infinite dimensional since it is isomorphic to the von neumann algebra generated by multiplication by @xmath160 on a suitable interval . let us return now to the description of all of the faces of @xmath8 . by identifying a face @xmath66 with @xmath235 , we may view all subfaces of @xmath66 as faces in @xmath235 . in particular , we may apply the results in theorem 0.2 to the exposed subfaces of @xmath235 to obtain data about the corresponding subfaces of @xmath66 . this process involves a good deal of new notation and is a bit complex . we now gather this information in a definition . in what follows , we shall consider cut downs of operators in @xmath2 of the form @xmath312 where @xmath124 is a projection and @xmath53 is self - adjoint . note that we may view @xmath313 as an element of @xmath2 or the von neumann algebra @xmath225 . if @xmath28 is a real number then the * cut down spectral interval projections * determined by @xmath314 and @xmath28 are the spectral projections of @xmath313 determined by the intervals @xmath55 and @xmath54 $ ] . we compute the spectrum and the corresponding spectral projections in the algebra @xmath95 where the projection @xmath124 is the identity . if @xmath315 is a sequence of spectral pairs , then its associated * facial complex * is defined inductively as follows . write @xmath316 let @xmath317 denote the associated cut down spectral interval projections determined by @xmath318 and @xmath319 and write @xmath320 now suppose that for some @xmath321 we have defined sequences @xmath322 such that if @xmath323 , then @xmath324 where @xmath325 are the cut down spectral interval projections determined by @xmath326 and @xmath327 . we may then set @xmath328 let @xmath329 denote the spectral interval projections determined by @xmath330 and @xmath331 and write @xmath332 this completes the induction . thus the facial complex determined by the spectral pairs @xmath114 consists of the projections @xmath333 together with the sequences @xmath334 as defined above . if @xmath66 is a face in the spectral scale and @xmath335 is its associated minimal exposed facial chain , then there is a sequence @xmath315 of spectral pairs such that if @xmath336 denote the spectral interval projections in its associated facial complex , then @xmath337)$ ] and for @xmath338 we have @xmath339 ) = \psi({q_{s_{i-1},{\mathbf t}_{i-1}}^- } ) + \psi([{q_{s_{i},{\mathbf t}_{i}}^- } , { q_{s_{i},{\mathbf t}_{i}}^+}])\ ] ] where @xmath340 let us proceed by induction . as @xmath217 is an exposed face in @xmath8 , there is a spectral pair @xmath341 such that @xmath342).\ ] ] if @xmath343 , we are done . otherwise , write @xmath344 and let @xmath345 denote the affine map defined in lemma 3.1 for the face @xmath217 of @xmath8 . we have that @xmath346 is an exposed face in @xmath347 and so by theorem 0.2 , there is a spectral pair@xmath348 such that if @xmath349 and @xmath317 denote the associated cut down spectral interval projections , then we have @xmath350).\ ] ] hence , by lemma 3.1 we have @xmath351 ) ) = \psi({p_{s_{1},{\mathbf t}_{1}}^- } ) + \psi((m_{r_1})_1^+)\ ] ] if @xmath352 , then we are done . otherwise , suppose that spectral pairs @xmath353 and sequences @xmath354 have been determined as above such that if @xmath355 and @xmath356 then we have @xmath357 ) = \psi({q_{s_{j-1},{\mathbf t}_{j-1}}^- } ) + \psi([{q_{s_{j},{\mathbf t}_{j}}^-},{q_{s_{j},{\mathbf t}_{j}}^+}]).\ ] ] with this , we may write @xmath358 , @xmath359 and let @xmath329 denote the associated spectral interval projections . as @xmath360 is exposed in @xmath220 , we may apply lemma 3.1 to get the affine map @xmath361 such that @xmath362 and @xmath363 is an exposed face in @xmath364 . hence , applying theorem 0.2 we get a spectral pair @xmath365 such that if we write @xmath366 and let @xmath367 denote the associated spectral interval projections , then we have @xmath368).\ ] ] applying @xmath369 to @xmath370 , yields the desired result . in the case of a single operator the corners on the boundary of the spectral scale are in one - to - one correspondence with the gaps in the spectrum of @xmath53 . in this section we study analogous faces in higher dimensions . if @xmath80 is a convex set in a vector space and @xmath66 is a face in @xmath80 , then we say @xmath66 is a * sharp * face if it is contained in two or more hyperplanes of support . recall that if @xmath238 is an affine space in @xmath5 , then the dimension of @xmath238 is the dimension of @xmath371 , where @xmath61 is any vector in @xmath238 . the * dimension * of the convex set @xmath80 is by definition the dimension of the affine space that @xmath80 generates . the dimension is denoted by @xmath372 . now fix a nonempty face @xmath66 of @xmath80 . recall that a vector @xmath373 is a * normal vector * for @xmath66 if @xmath374 is a hyperplane of support for @xmath80 and @xmath375 . the * normal cone * of @xmath66 is by definition @xmath376 the following result is basic to our analysis of sharp faces in the spectral scale . if @xmath237)$ ] is a face in the spectral scale @xmath8 and @xmath27 is a spectral pair , then the following statements are equivalent . 1 . the vector @xmath377 is in @xmath378 . we have @xmath379 . 3 . if we write @xmath380 set @xmath381 and @xmath382 , then + @xmath383 + and @xmath384 . + in other words , the matrix for @xmath37 determined by the projections @xmath385 and @xmath386 has the form @xmath387 in particular , the @xmath388 s commute with @xmath37 . suppose @xmath158 holds and @xmath72 is the hyperplane of support that contains @xmath66 . since @xmath237 ) \subset \psi([{p_{s,{\mathbf t}}^-},{p_{s,{\mathbf t}}^+}])$ ] we get @xmath389 by parts @xmath88 and @xmath89 of theorem 0.2 . hence @xmath390 . it is also clear from the above that @xmath391 . now suppose @xmath88 holds . in this case since , @xmath36 are the spectral interval projections for @xmath37 and @xmath379 we get @xmath392 further we have @xmath393 and @xmath394 and so @xmath37 has the desired form . thus , @xmath395 . now suppose @xmath89 holds . as @xmath396 , we get @xmath397 and so @xmath398 . if @xmath399)$ ] is a face in the spectral scale , then @xmath66 is a sharp face if and only if there are linearly independent spectral pairs @xmath118 and @xmath119 such that @xmath400 in this case if we write @xmath401 then we have 1 . @xmath402 , 2 . @xmath403 , 3 . @xmath404 , and 4 . @xmath405 for @xmath406 and the @xmath388 s commute with @xmath37 . the assertions are easily established using the definition of a sharp face and theorem 4.2 . if @xmath237)$ ] is a proper face of @xmath8 and @xmath210 is the minimal exposed face containing @xmath66 , so that @xmath407)$ ] , where @xmath408 are spectral interval projections of a linear combination of the @xmath10 s , then we have @xmath409 further , if @xmath27 is a spectral pair such that @xmath410)$ ] is an exposed face of @xmath8 that contains @xmath210 , then @xmath411 all of the assertions are simple consequences of part @xmath158 and @xmath88 of theorem 4.2 . using the notation of theorem 4.4 , a face @xmath412)$ ] of @xmath8 is an extreme point of @xmath8 if and only if @xmath413 . in this case we have @xmath414 where @xmath415 and @xmath416 . the face @xmath417 is an exposed point if and only if @xmath418 . we have that @xmath66 is a singleton if and only if @xmath413 . in this case , in the notation of theorem 4.2 , we get @xmath419 and so @xmath37 has the desired matrix decomposition by part @xmath89 of theorem 4.2 . we have that @xmath61 is an exposed point if and only if @xmath420 , where @xmath421 . since @xmath407)$ ] by corollary 4.4 , the final assertion follows . before presenting the next result , it is convenient to introduce some new notation . suppose @xmath66 is a face in a convex subset of @xmath5 . the * degree * of @xmath66 is denoted by @xmath422 and is by definition the dimension of the normal cone @xmath378 . recall that , the subspace spanned by @xmath378 is @xmath423 and so the degree of @xmath66 is just the dimension of @xmath424 . the following assertions hold . 1 . the degree of @xmath66 is equal to the maximal number of linearly independent vectors in @xmath378 . we have @xmath425 . 3 . @xmath66 is a sharp face if and only if @xmath426 . for @xmath158 , it is clear that there can be no more than @xmath422 such vectors . now suppose @xmath427 is a maximal set of linearly independent vectors in @xmath378 and fix any vector @xmath29 in @xmath424 we must have @xmath428 , where @xmath429 . this means that both @xmath430 and @xmath431 are linear combinations of @xmath427 and so @xmath29 is a linear combination of @xmath432 . thus , @xmath433 is a basis for @xmath424 . next , since each vector in @xmath378 is orthogonal to the difference of any two vectors in @xmath66 , assertion @xmath88 is clear . finally , assertion @xmath89 follows immediately from the definitions of a sharp face and the degree of @xmath66 . if @xmath434 , @xmath435)$ ] is a face in @xmath8 and there are spectral pairs @xmath436 such that 1 . @xmath437)$ ] for @xmath438 and 2 . @xmath304 are linearly independent , then @xmath439 and @xmath440 are central projections in the von neumann algebra generated by @xmath441 . applying theorem 4.3 , we get that @xmath439 and @xmath440 commute with each @xmath442 . since @xmath304 are linearly independent , each @xmath10 is a linear combination of the @xmath442 s and so @xmath443 and @xmath440 commute with each @xmath10 . in the notation of the last corollary , @xmath444 , so that by part @xmath88 of theorem 4.6 , we get that @xmath445 . thus , in this case , @xmath66 is either a sharp extreme point or a sharp line segment in @xmath8 . if @xmath237)$ ] is a face in the spectral scale and @xmath446 and @xmath447 are distinct vectors in @xmath378 with @xmath448 , then 1 . the interval @xmath122 lies in a gap in the spectrum of @xmath37 and 2 . @xmath66 is an exposed point . applying theorem 4.3 ( and using its notation ) we get that if @xmath449 , then @xmath450 and so @xmath451 and @xmath452 . next as @xmath453 for @xmath406 , we get @xmath454 for @xmath406 . on the other hand , since @xmath121 we have @xmath455 so that @xmath456 and therefore @xmath457 hence there must be a gap in the spectrum of @xmath37 that contains the interval @xmath122 . next , if we select any @xmath28 such that @xmath458 , then we have @xmath459 and so @xmath460 . thus , @xmath66 is an exposed point by corollary 0.3 . we say that an element @xmath61 in a convex set @xmath80 is an * isolated extreme point * of @xmath80 if it is an extreme point and it is isolated in the set @xmath461 of all extreme points of @xmath80 . our next goal is to show that isolated extreme points in @xmath8 are images of central projections in @xmath12 and further that this condition characterizes isolated extreme points in the finite dimensional case . it is convenient to begin by presenting three lemmas . the first lemma simply records an easy fact about convex subsets in @xmath5 . the ( slick ) proof of the second is due to robert phelps ( private communication ) . recall that if @xmath345 and @xmath462 are disjoint , compact , convex subsets in a real topological vector space , then there is a continuous linear functional @xmath104 that * separates * @xmath345 and @xmath462 in the sense that there are real real numbers @xmath463 such that @xmath464 for all @xmath465 and @xmath466 ( * ? ? ? * theorem 3.4 ( b ) ) . in this case we say that @xmath28 and @xmath105 are separation constants for @xmath104 . if @xmath467 is a compact , convex subset of @xmath5 , @xmath61 is a point in @xmath5 disjoint from @xmath467 and @xmath468 , then a linear functional @xmath104 exposes @xmath61 in @xmath80 if and only if @xmath104 separates @xmath467 and @xmath469 . suppose that @xmath104 exposes @xmath61 in @xmath80 so that there is a number @xmath105 such that @xmath106 and @xmath470 if @xmath471 and @xmath472 . write @xmath473 and suppose @xmath474 is a sequence in @xmath467 such that @xmath475 . as @xmath467 is compact , we may assume that @xmath474 converges to a vector @xmath476 . since @xmath477 , @xmath104 exposes @xmath61 and @xmath478 , we must have that @xmath479 . hence , we get that @xmath104 separates @xmath467 and @xmath61 . now suppose that @xmath104 separates @xmath467 and @xmath61 with separation constants @xmath463 and fix @xmath480 . since @xmath468 , there is a number @xmath124 with @xmath481 and a vector @xmath476 , such that @xmath482 . hence , @xmath483 and so @xmath104 exposes @xmath61 . if @xmath80 is a convex , compact subset in @xmath5 and @xmath61 is an isolated extreme point in @xmath80 , then the set of linear functionals that expose @xmath61 is a nonempty open set . further , there are linearly independent vectors @xmath304 and real numbers @xmath484 such that the hyperplanes @xmath485 are hyperplanes of support for @xmath80 and @xmath486 write @xmath238 for the closure of @xmath487 and observe that since @xmath61 is an isolated point in @xmath461 , @xmath488 . further , if @xmath467 denotes the closed convex hull of @xmath238 , then @xmath489 . indeed , if we had @xmath490 , then @xmath61 would be an extreme point of @xmath467 . but in this case we would get that @xmath491 by choquet s converse " to the krein milman theorem ( * ? ? ? * appendix b14 ) . as noted in the paragraph before lemma 5.1 , there is a continuous linear functional @xmath104 that separates @xmath467 and @xmath61 and so , by lemma 5.1 @xmath104 exposes @xmath61 in @xmath80 . hence the set of linear functionals that expose @xmath61 is not empty . now suppose that @xmath104 is a linear functional that exposes @xmath61 . applying lemma 5.1 , we get that there are real numbers @xmath492 such that @xmath104 separates @xmath467 and @xmath61 with separation constants @xmath28 and @xmath105 . now select @xmath493 so that if @xmath494 and let @xmath495 denote a linear functional with @xmath496 . with this we get @xmath497 if @xmath476 , then we have @xmath498 and so @xmath104 separates @xmath467 and @xmath61 with separation constants @xmath499 and @xmath500 . applying lemma 5.1 we get that @xmath495 also exposes @xmath61 and so the set of all such linear functionals is open . to get the final conclusion , observe that since the set of linear functionals that expose @xmath61 is a nonempty open set , it contains a linearly independent set of @xmath0 linear functionals , @xmath501 . thus there are linearly independent vectors @xmath304 such that @xmath502 is given by the inner product with @xmath208 , and the conclusion follows . observe that the proofs of the last two lemmas go through in any normed linear space . the following statements hold . 1 . if @xmath61 is an isolated extreme point of @xmath8 , then there is a central projection @xmath503 such that @xmath504 . if @xmath12 is finite - dimensional and @xmath61 is an extreme point of @xmath8 , then @xmath505 for some central projection @xmath506 , if and only if @xmath61 is an isolated extreme point of @xmath8 . \(1 ) suppose @xmath61 is an isolated extreme point in @xmath8 . by part @xmath158 of theorem 0.2 , there is a unique projection @xmath503 such that @xmath505 . next , applying lemma 5.2 and parts @xmath88 and @xmath89 of theorem 0.2 , we get linearly independent spectral pairs@xmath507 such that @xmath508),\quad k = 1,\dots , n+1.\ ] ] since @xmath509 , we have @xmath510 since these spectral pairs are linearly independent , a straightforward argument shows that the subspace of @xmath5 spanned by @xmath511 has dimension @xmath0 . hence , relabeling if necessary , we may assume that @xmath304 are linearly independent . therefore , by corollary 4.7 @xmath512 is a central projection in @xmath12 . \(2 ) suppose that @xmath504 , where @xmath21 is a central projection in @xmath12 and let @xmath513 denote a sequence of extreme points of @xmath8 that converges to @xmath61 . applying part @xmath158 of theorem 0.2 we get a sequence @xmath514 of unique projections in @xmath12 such that @xmath515 for each @xmath516 . since @xmath2 is finite dimensional , we may pass to a subsequence and assume that the sequence @xmath517 is norm convergent with limit @xmath22 . since @xmath4 is continuous , we have @xmath518 and since @xmath519 , we get @xmath520 . now , since @xmath21 is central and the central projections are isolated in the set of all projections in @xmath12 , we must have that @xmath521 ( and hence @xmath522 ) for all sufficiently large @xmath516 . hence @xmath61 is an isolated extreme point . if @xmath61 is an isolated extreme point in @xmath8 , then by part @xmath158 of this theorem , there is a central projection @xmath21 in @xmath12 such that @xmath504 @xmath158 let @xmath53 denote multiplication by @xmath160 on the interval @xmath165 $ ] and write @xmath523 . in this case every projection in @xmath12 is central but @xmath8 has no isolated extreme points . ( see the example at the end of section 1 in @xcite ) . hence , the converse to part @xmath158 of theorem 5.3 is false . @xmath88 if @xmath524 and @xmath12 are as in remark @xmath158 above , then there are linearly independent spectral pairs @xmath341 and @xmath348 such that @xmath525),\quad i = 1,2,\ ] ] but @xmath526 is not an isolated extreme point in @xmath8 . ( see @xcite ) . thus , the converse to lemma 5.2 is also false . @xmath12 is abelian and has finite dimension if and only if @xmath8 has a finite number of extreme points . if @xmath12 is abelian and has finite dimension , then each projection in @xmath12 is diagonal . hence @xmath12 contains a finite number of projections and since every extreme point in @xmath8 is the image under @xmath4 of a projection @xmath12 , @xmath527 is finite . for the converse , if @xmath8 has a finite number of extreme points , then each extreme point in @xmath8 is an isolated extreme point and so by part @xmath158 of theorem 5.3 , each extreme point of @xmath8 has the form @xmath528 , where @xmath21 is a central projection . on the other hand , if @xmath529 is any spectral pair , then @xmath530 is an extreme point by part @xmath158 of theorem 0.2 and so @xmath36 is a central projection . thus , for each fixed @xmath29 the von neumann algebra @xmath531 is central and therefore each @xmath37 is central . hence @xmath12 is abelian . since the projections of the form @xmath36 generate @xmath12 , this algebra has finite dimension . our goal in this section is to establish the following theorem . suppose that the closure of the isolated points of @xmath532 is not all of @xmath532 . by ( * ? ? ? * theorem 2.7 ) each open subset of @xmath532 is a locally compact hausdorff space . hence , there is a nonempty open set @xmath533 in @xmath532 with no isolated points . since @xmath532 is countable , @xmath533 is countable and we may write @xmath534 . set @xmath535 for each @xmath0 . as @xmath532 is hausdorff , each @xmath536 is open and since @xmath537 is not isolated , each @xmath536 is dense in @xmath533 . hence by the baire category theorem ( * ? ? ? * theorem 2.2 ( b ) ) , @xmath538 is dense in @xmath533 . since , @xmath539 , this is impossible . hence , no such @xmath533 exists and so the isolated extreme points in @xmath532 are dense . next suppose @xmath527 is countable and fix @xmath186 and @xmath547 . for each extreme point @xmath61 in @xmath548 write @xmath549 so that @xmath550 is a closed face of @xmath8 . now each @xmath550 is nonempty and so contains extreme points . since @xmath551 and @xmath552 are disjoint if @xmath250 and @xmath251 are distinct extreme points , the extreme points of @xmath553 must be countable . thus , if @xmath527 is countable , then @xmath554 is countable for each pair @xmath555 . hence , if the theorem is true when @xmath540 , then each @xmath556 is abelian and therefore @xmath12 is abelian . for the remainder of this section , we shall assume that @xmath12 is the von neumann algebra generated by @xmath557 and the identity and that @xmath558 denotes the associated spectral scale . as @xmath47 each face in @xmath8 must have dimension zero , one or two . recall that for each real @xmath28 , with @xmath559 @xmath560 denotes the * isotrace slice * of @xmath8 at @xmath28 . 1 . if @xmath66 has dimension greater than zero , then @xmath66 is transverse to the isotrace slices of @xmath8 . if @xmath66 has dimension one , then @xmath66 meets each isotrace slice in at most one point . 3 . if @xmath66 has dimension two , and @xmath561 contains a point in the relative interior of @xmath66 , then @xmath562 is a proper line segment . we have that @xmath102)$ ] for some spectral pair @xmath27 by part @xmath89 of theorem 0.2 . now suppose that @xmath66 has dimension greater then zero . in this case we must have @xmath563 and so @xmath564 . hence , @xmath66 is transverse to the isotrace slices of @xmath8 and so @xmath158 is true . 1 . if @xmath565 is a proper closed line segment in the boundary of @xmath561 , then there is a two dimensional face @xmath66 of @xmath8 such that @xmath566 . if @xmath61 is an extreme point of @xmath561 , then either @xmath61 is an extreme point of @xmath8 or else it lies in the relative interior of a one dimensional face of @xmath8 . if @xmath61 is an extreme point in @xmath561 that is an isolated point in @xmath567 , then there are two distinct two dimensional faces of @xmath8 that contain @xmath61 . 4 . if the set of extreme points @xmath8 is countable then the set of extreme points of @xmath561 is countable . fix a closed proper line segment @xmath565 in the boundary of @xmath561 . as @xmath565 is in the boundary of @xmath561 , it is in the boundary of @xmath8 and so is contained in a minimal face @xmath66 in @xmath8 . since faces in @xmath8 of dimension greater than zero are transverse to the isotrace slices by lemma 6.4(1 ) , @xmath66 must be two dimensional . hence @xmath158 is true . next , fix an extreme point @xmath61 of @xmath561 . if @xmath61 is not an extreme point of @xmath8 , then it is properly contained in a minimal face @xmath66 of @xmath8 . if @xmath66 were two dimensional , then by its minimality , we would get that @xmath61 lies in the relative interior of @xmath66 . but in this case @xmath562 would be a line segment by lemma 6.4(2 ) and @xmath61 would be in its interior , which is impossible since @xmath61 is extreme in @xmath561 . hence , @xmath66 has dimension less than two . since @xmath61 is properly contained in @xmath66 , this face must have dimension one . finally , since @xmath61 is not extreme in @xmath8 , it must be contained in the relative interior of @xmath66 and so @xmath88 is true . now suppose @xmath61 is an isolated extreme point in @xmath561 . since @xmath561 has dimension 2 , its faces consist of extreme points and line segments . since @xmath61 is an isolated extreme point , it must be the endpoint of adjacent line segments in the boundary of @xmath561 . hence @xmath61 lies in two distinct two dimensional faces of @xmath8 by part @xmath158 of this lemma . thus part @xmath89 is true . finally , assume that the set @xmath527 of extreme points of @xmath8 is countable . observe that each face in @xmath8 of dimension one is uniquely determined by its end points , which are extreme points in @xmath8 by theorem 0.2 . hence @xmath8 has a countable number of faces of dimension one . now fix @xmath28 with @xmath568 . we have that @xmath569 is countable . each remaining extreme point of @xmath561 is contained in a unique face of @xmath8 with dimension one by part @xmath88 of this lemma . since there are a countable number of one dimensional faces in @xmath8 , the set of extreme points of @xmath561 is countable . as @xmath61 is isolated among the extreme points of @xmath561 , there are distinct two dimensional faces @xmath217 and @xmath575 of @xmath8 such that @xmath576 by part @xmath89 of lemma 6.5 . next , since @xmath217 and @xmath575 have maximal dimension in @xmath8 , they must be exposed and so we may write @xmath577 ) , \quad i = 1,2.\ ] ] further , since the @xmath219 s intersect @xmath561 in adjoining line segments , we must have that @xmath578 . since neither of the @xmath208 s is zero , these vectors are linearly independent . hence , by corollary 4.7 , we must have @xmath579),\ ] ] where @xmath570 and @xmath571 are central projections in @xmath12 . finally , since the @xmath219 s have dimension two , @xmath580 must be a ( possibly degenerate ) line segment . hence elements of @xmath581 are the images under @xmath4 of convex combinations of @xmath570 and @xmath571 , as desired . write @xmath582 and @xmath583 . to show that @xmath12 is abelian , it suffices to show that @xmath584 is central for each spectral pair of the form @xmath585 . indeed in this case , by a standard argument in measure theory , we get that the spectral projections for @xmath142 and @xmath143 are central and so @xmath142 and @xmath143 commute . so now let @xmath21 denote such a projection , write @xmath504 and let @xmath561 denote the isotrace slice containing @xmath61 . by part @xmath158 of theorem 0.2 , @xmath61 is an extreme point of @xmath8 and so it is also an extreme point of @xmath561 . since @xmath527 is countable , the extreme points of @xmath561 are countable by part @xmath586 of lemma 6.5 and since @xmath561 has dimension 2 , these extreme points form a closed set . hence , the isolated extreme points of @xmath561 are dense in @xmath561 by lemma 6.2 . now let @xmath587 denote a sequence of isolated extreme point in @xmath561 that converges to @xmath61 . applying lemma 6.6 , we get sequences @xmath588 of central projections in @xmath12 and a sequence @xmath589 of real numbers in @xmath165 $ ] such that @xmath590 write @xmath591 for each @xmath0 . as the sequence @xmath592 lies in @xmath593 , it is bounded and so it admits a subnet @xmath594 that converges to an element @xmath164 in @xmath593 in the weak@xmath595-topology . since @xmath596 is normal , we have since @xmath519 by part @xmath158 of theorem 0.2 , we get @xmath598 . since each @xmath599 is central and the center of @xmath12 is closed in the weak@xmath595topology , we get that @xmath21 is central , as desired .

the study and analysis of social graphs impacts on a wide range of applications , such as community decision making support and recommender systems . with the boom of _ online _ social networks , such analyses are benefiting from a massive collection and publication of social graphs at large scale . unfortunately , individuals privacy right might be inadvertently violated when publishing this type of data . in this article , we introduce-anonymity ; a novel privacy measure aimed at evaluating the resistance of social graphs to active attacks . -anonymity is based on a new problem in graph theory , the _ -metric antidimension _ defined as follows . let be a simple connected graph and an ordered subset of vertices . the metric representation of a vertex with respect to is the-vector , where represents the length of a shortest path in . we call a-antiresolving set if is the largest positive integer such that for every vertex there exist other different vertices with . the-metric antidimension of is the minimum cardinality among all the-antiresolving sets for . we address the-metric antidimension problem by proposing a true - biased algorithm with success rate above when considering random graphs of size at most . the proposed algorithm is used to determine the privacy guarantees offered by two real - life social graphs with respect to-anonymity . we also investigate theoretical properties of the-metric antidimension of graphs . in particular , we focus on paths , cycles , complete bipartite graphs and trees . anonymity , active attack , social network , graph , resolving set ,-metric antidimension

1408.2154

social networking services are widely used in modern society as illustrated by the alexa s top 500 global sites statistics where _ facebook _ and _ linkedin _ rank @xmath18nd and @xmath19th respectively in 2014 . such popularity has enabled governments and third - party enterprises to massively collect social network data , which eventually can be released for mining and analysis purposes . the power of social network analysis is questionless . it might uncover previously unknown knowledge such as community - based problem , media use , individual engagement , amongst others . sociology is a trivial example of a field that certainly benefits from social graphs publication . many other fields ( _ e.g. , _ economics , geography , or political science ) and systems ( _ e.g. , _ service - oriented systems , advertisers , or recommended systems ) improve their decisions , processes , and services , based on users interaction . however , all these benefits are not cost - free . an adversary can compromise users privacy using the published social network , which results in the disclosure of sensitive data such as e - mails , instant messages , or relationships . a simple and popular approach to prevent this privacy problem is _ anonymization _ by means of removing potential identifying attributes . doing so , aggregate knowledge still can be inferred ( _ e.g. , _ connectivity , distance , or node degrees ) while the `` who '' information has been removed . in practice , however , this naive approach is not enough for protecting users privacy . what makes social network anonymization a challenging problem is the combination of the adversary s background knowledge with the released structure of the network . considering a social network as a simple graph , in which individuals are represented by vertices and their bidirectional relationships by edges , the adversary s background knowledge about a victim may take many forms , _ e.g. , _ vertex degrees , connectivity , or local neighborhood . this structural knowledge , together with the released graph , is often enough to perform _ passive attacks _ where the users and their relationships are re - identified @xcite . other privacy attacks exist . in 2007 , backstrom _ et al . _ @xcite introduced _ active attacks _ based on the creation and insertion in the network of _ attacker nodes _ controlled by the adversary . the attacker nodes could be either new accounts with pseudonymous or spoofed identities ( sybil nodes ) , or legitimate users in the network who collude with the adversary . attacker nodes establish links with other nodes in the network ( also between themselves ) aiming at creating a sort of fingerprint in the network . once the social graph is released , the adversary just need to retrieve such a fingerprint ( the attacker nodes ) and use it as a hub to re - identify other nodes in the network . backstrom et . al . proved that @xmath20 attacker nodes in the network can compromise the privacy of arbitrary targeted nodes with high probability , which makes active attack particularly dangerous . several active attacks to social graphs have been proposed . they could even target random nodes in the network as recently shown in @xcite . however , to the best of our knowledge , no privacy measure aimed at evaluating the resistance of a social graph to this kind of attack exists . the lack of such a measure prevents the development of privacy - preserving methods with theoretically proven privacy guarantees . in this article we define _ @xmath0-anonymity _ ; a privacy measure that can be applied to real - life social graphs in order to measure their resistance to active attacks . the proposed privacy measure copes with adversaries whose background knowledge concerning a node @xmath21 and a subset @xmath5 of attacker nodes is the metric representation of @xmath21 with respect to @xmath5 . @xmath0-anonymity turns out to be based on a new problem in graph theory : the _ @xmath1-metric antidimension_. we propose a true - biased algorithm whose computational complexity and success rate can be adjusted . empirical results show that our algorithm finds @xmath1-antiresolving basis in random graphs of order at most @xmath16 with a success rate above @xmath15 . our algorithm has been also used to determine the privacy offered by two real - life social graphs against active attacks . finally , we provide theoretical results on the @xmath1-metric antidimension of graphs , such as paths , cycles , complete bipartite graphs and trees . the rest of this article is structured as follows . section [ sec_related_work ] briefly reviews the literature on privacy - preserving publication of social network data . section [ sec_antiresolving_sets ] presents the metric representation as a reasonable definition of the adversary s background knowledge . it also introduces the @xmath1-metric antidimension as the basis for the privacy measure @xmath0-anonymity . in section [ sect - algo ] we present a true - biased algorithm for computing the @xmath1-metric antidimension of a graph , and evaluate the proposed algorithm through experiments . preliminary results ( mathematical properties ) on the new problem ( the @xmath1-metric antidimension ) are provided in sections [ sec_metric_antidimension ] and [ section - trees ] ( the later specifically addresses the case of tree graphs ) . section [ sec_conclusions ] draws conclusions and future work . a social graph @xmath2 is a simple graph where @xmath22 represents the set of social actors and @xmath23 their relationships . both vertices and edges could be enriched with attribute values such as weights representing trustworthiness or labels providing meaning . we consider , however , social network data in its most `` simplest '' form , _ i.e. _ , a simple graph without further annotation . privacy breaches in social networks are mainly categorized in identity disclosure or link disclosure @xcite . to perform such attacks adversaries rely on background knowledge , which is usually defined as structural knowledge such as vertex degrees @xcite or neighborhoods @xcite . the assumptions on the adversary s background knowledge determine the type of privacy attacks and the corresponding countermeasures . privacy - preserving methods for the publication of social graphs are normally based on the well - known concept @xmath1-anonymity @xcite adapted to graphs . @xmath1-anonymity , initially proposed for microdata , aims at ensuring that no record in a database can be re - identified with probability higher than @xmath24 . to do so , identifying attributes should be obviously removed , and any combination of non - identifying attribute values should not be unique in the database . in practice , not all the attributes need to be combined , because they do not belong to the adversary s knowledge . this leads to the concept of _ quasi - identifier _ , that is , an attribute that can be found in external source of information and , combined with other quasi - identifiers , can uniquely identify a record in the database . even though graphs can be represented in tabular form and , thus , graph @xmath1-anonymity can be defined in terms of quasi - identifying attributes @xcite , graph @xmath1-anonymity is typically defined in terms of structural properties of the graph rather than on attributes . for instance in @xcite , the adversary s background knowledge is defined as a _ knowledge query _ @xmath25 evaluated for a given target node of the original graph @xmath10 . the knowledge query @xmath25 allows the creation of a candidate set consisting of @xmath26 . in other words , all the nodes in the network matching the query @xmath27 are equally likely to be the target node @xmath28 . this simple concept is the basis of several passive attacks and privacy - preserving methods in the publication of social graphs @xcite . other privacy notions based on entropy rather than on @xmath1-anonymity have been proposed @xcite . this type of privacy measure is better suited for methods based on random addition , deletion , or switching of edges . the perturbation could be made in such a way that the number of edges or the degree of the vertices are preserved @xcite . however , empirical results obtained in @xcite suggest that random obfuscation poorly preserves the topological features of the network . passive attacks to social networks can be combined with active attacks . in addition to structural knowledge , in an active attack the adversary manages to control a subset of nodes ( attacker nodes ) of the original graph @xmath10 @xcite . the attacker nodes aim at creating links with their victims by either identity theft or cloning of existing users profiles @xcite . they also establish links between themselves so as to build a subgraph @xmath29 of attacker nodes with the following properties : i ) @xmath29 can be efficiently identified in @xmath10 and ii ) @xmath29 does not have a non - trivial automorphisms . once @xmath29 has been identified , the adversary is able to re - identify neighbor nodes of @xmath29 @xcite or even arbitrary nodes in the network @xcite . performing active attacks is not easy , given that there exist several detection mechanisms of attacker or sybil nodes in a network @xcite . however , such defenses strongly depend on assumptions on the topological structure of the social network , which does not hold in many real - world scenarios @xcite . actually , recent works aim at mitigating , instead of preventing , the impact of sybil attacks @xcite . furthermore , a group of users who collude in order to breach the privacy of other users in the network can be also regarded as attacker nodes . other types of active attacks exist . for instance , the maximal vertex coverage ( mvc ) attack consists in attacking a few nodes so as to delete as many edges of the network as possible . in this attack , the attacker tries to convince some users to leave the social network in order to reduce the number of residual social ties . metrics to quantify the impact of mvc attacks have been studied in @xcite . mvc is not a privacy attack , though . while there exist several published active attacks to social graphs , there does not exist yet a rational privacy metric for evaluating the resistance of social graphs to this type of privacy attack . to overcome this problem , in this article we introduce @xmath0-anonymity ; a privacy notion based on @xmath1-anonymity and the metric representation of nodes in a graph . note that , privacy notions with the same name has been already proposed . for instance , feder and nabar proposed @xmath0-anonymity where @xmath30 represents the number of common neighbors of two nodes @xcite . this notion was later generalized by stokes and torra in @xcite . in our privacy notion , however , @xmath30 represents an upper bound on the @xmath1-metric antidimension of the graph . interested readers could refer to @xcite for further reading on privacy - preserving publication of social graphs . in this section we define the metric representation of nodes with respect to a set of attacker nodes @xmath5 as the adversary s background knowledge . we also introduce the resulting privacy measure , named @xmath0-anonymity , and its related mathematical problem : the @xmath1-metric antidimension . vulnerabilities in an anonymized social graphs are better understood once the adversary s knowledge has been properly modeled . this knowledge can be acquired from public information sources and through malicious actions . in practice , the adversary could even be a close friend , which makes the publication of social network where users can not re - identify themselves a reasonable privacy goal . adversary s background information in passive attacks is typically modeled as structural knowledge on the network . this is a sort of _ global _ view that provides adversaries with the ability to partition the set of nodes into equivalence classes of structurally equivalent nodes . the strongest of those structural relations is automorphism @xcite . two vertices @xmath21 and @xmath31 are automorphically equivalent if there exists an isomorphism from the graph to itself such that @xmath21 maps to @xmath31 . other types of structural relations are based on vertex degrees , connectivity , or local neighborhood . intuitively , structurally equivalent vertices are indistinguishable with respect to the considered structural property . however , adversaries controlling attacker nodes in a network are undoubtedly more powerful . in addition to the global view , they have a _ local view _ determined by the relationship of the attacker nodes with the network . to illustrate this let us consider the graph shown in figure [ fig : graph : path ] . with respect to the vertex degree property , @xmath32 and @xmath33 are indistinguishable . they are easily re - identifiable by either an adversary or a legitimate user owning the vertex @xmath34 and knowing its distance to @xmath32 and @xmath33 , though . a first step towards modeling such local view was given by hay et al . @xcite , who defined the concept of _ hub fingerprint queries_. a hub is a relevant node in the network with high degree and high centrality , and a hub fingerprint for a target node @xmath28 is a vector of distances from @xmath28 to hub vertices . although not explicitly mentioned in @xcite , the largest hub fingerprint for a target node @xmath28 is indeed the _ metric representation _ of @xmath28 with respect to the hub vertices . we formally define this concept as follows . [ def_vertex_representation ] let @xmath2 be a simple connected graph and @xmath8 be the length of the shortest path between the vertices @xmath21 and @xmath31 in @xmath10 . for an ordered set @xmath35 of vertices in @xmath22 and a vertex @xmath31 , we call @xmath36 the _ metric representation _ of @xmath31 with respect to @xmath5 . similarly to hay et al . work @xcite , we define the adversary s background knowledge about a target node @xmath21 as the metric representation of @xmath21 with respect to @xmath5 . in this article , however , we assume @xmath5 to be any subset of attacker nodes rather than hub vertices only . it is worth mentioning that the concept of metric representation is also the basis of two weel - known concepts : resolving sets and metric dimension ( cf . definition [ def : resolving ] ) . both have been already motivated by problems related to unique recognition of an intruder position in a network @xcite , where resolving sets were called _ locating sets_. the name `` resolving set '' is due to harary and melter @xcite , who introduced the concept in 1976 . [ def : resolving ] let @xmath2 be a simple connected graph . a set @xmath37 is said to be a _ resolving set _ for @xmath10 if any pair of vertices of @xmath10 have different metric representations with respect to @xmath5 . a resolving set of the smallest possible cardinality is called a _ metric basis _ , and its cardinality the _ metric dimension _ of @xmath10 . @xmath0-anonymity is a privacy measure that evolves from the adversary s background knowledge defined previously . it is based on the concept of _ @xmath1-antiresolving set _ defined as follows . [ def_antiresolving_set ] let @xmath2 be a simple connected graph and let @xmath38 be a subset of vertices of @xmath10 . the set @xmath5 is called a @xmath1-_antiresolving set _ if @xmath1 is the greatest positive integer such that for every vertex @xmath11 there exist at least @xmath12 different vertices @xmath39 with @xmath14 , _ i.e. _ , @xmath31 and @xmath40 have the same metric representation with respect to @xmath5 . the following concepts derive from definition [ def_antiresolving_set ] , whose study is one of the goals of this article . [ def_metric_antidimension ] the @xmath1-_metric antidimension _ of a simple connected graph @xmath2 is the minimum cardinality amongst the @xmath1-antiresolving sets in @xmath10 and is denoted by @xmath41 . a @xmath1-antiresolving set of cardinality @xmath41 is called a @xmath1-_antiresolving basis _ for @xmath10 . it is easy to prove that if the set of attacker nodes @xmath5 is a @xmath1-antiresolving set , the adversary can not uniquely re - identify other nodes in the network with probability higher than @xmath24 . however , given that @xmath5 is unknown , the privacy measure should quantify over all possible subsets @xmath5 as follows . [ def_graph_anonymity ] a graph @xmath10 meets @xmath0-_anonymity _ with respect to active attacks if @xmath1 is the smallest positive integer such that the @xmath1-metric antidimension of @xmath10 is lower than or equal to @xmath30 . in definition [ def_graph_anonymity ] the parameter @xmath1 is used as a privacy threshold , whilst @xmath30 is an upper bound on the expected number of attacker nodes in the network . because attacker nodes are difficult to enrol in a network without been detected @xcite , @xmath30 can be estimated through statistical analysis . a fair assumption , for example , is that the number of attacker nodes is significantly lower than the total number of nodes in the network . to further explain the role of @xmath1 and @xmath30 in definition [ def_graph_anonymity ] we provide the following example result . [ def_simple_theorem ] for every @xmath42 and @xmath43 , the graph @xmath44 meets @xmath45 . since all the vertices in a complete graph @xmath44 are connected , every subset @xmath5 of vertices of @xmath44 is an @xmath46-antiresolving set . therefore , the @xmath1-metric antidimension of @xmath44 is @xmath47 . according to definition [ def_graph_anonymity ] , the @xmath1-metric antidimension should be lower than or equal to @xmath30 , which implies that @xmath48 . moreover , @xmath1 should be the smallest positive integer satisfying the previous condition . therefore , @xmath44 holds @xmath49-_anonymity_. [ def_simple_corollary ] a social graph @xmath44 guarantees that a user can not be re - identified with probability higher than @xmath50 by an adversary controlling @xmath30 attacker nodes . these simple and intuitive result obtained in theorem [ def_simple_theorem ] and corollary [ def_simple_corollary ] shows the role of the privacy measure @xmath17-anonymity in privacy - preserving publication of social graphs . before releasing a social graph @xmath10 , the goal is to find @xmath1 such that @xmath10 satisfies @xmath51-anonymity . to do so , theoretical results and efficient algorithms on the @xmath1-metric antidimension of a graph need to be investigated . computing the @xmath1-metric antidimension of a graph seems to be a challenging problem , whose hardness ought to be investigated . it should be remarked that its counterpart : the @xmath1-metric dimension is an np - complete problem @xcite ( and @xcite for @xmath52 ) . particularly , we observe that any resolving set is also a @xmath53-antiresolving set , which gives some intuitive idea on the hardness of computing the @xmath1-metric antidimension of graphs . thus , we address the @xmath1-metric antidimension problem by proposing a true - biased algorithm whose success rate and computational cost can be balanced . a true - biased algorithm is always correct when it returns ` true ` , it might fail with some small probability when its output is ` false ` . true - biased algorithms normally are monte carlo algorithms with deterministic running time and randomized behavior . the algorithm we introduce in this section resembles to a monte carlo algorithm in the sense that it is deterministic and has the true - biased property . the proposed algorithm is not randomized , though . the mathematical foundation of our algorithm requires the introduction of notation as follows . for a given subset of vertices @xmath54 , we denote @xmath55 to the symmetric , reflexive and transitive relation satisfying that @xmath56 . the set of equivalence classes created by @xmath57 over the subset of vertices @xmath58 is denoted as @xmath59 . we deliberately abuse notation and use @xmath60 and @xmath61 instead of @xmath62 and @xmath63 for every vertex @xmath64 . [ prop_1 ] let @xmath65 and @xmath66 : * @xmath67 * @xmath68 there exists @xmath69 such that @xmath70 * @xmath68 and @xmath71 , @xmath72 [ theo_1 ] let @xmath5 be a @xmath1-antiresolving set and let @xmath66 . let @xmath73 , then @xmath74 . by proposition [ prop_1 ] , for every @xmath75 there exists @xmath76 such that @xmath77 , which implies that @xmath78 due to the definition of @xmath1-antiresolving set . consequently , @xmath79 implies that there does not exist @xmath80 such that @xmath77 , meaning that there does not exist @xmath80 such that @xmath81 according to proposition [ prop_1 ] . therefore , @xmath82 and thus @xmath83 . in the spirit of lemma [ theo_1 ] , let @xmath84 be the function defined recursively as follows : @xmath85 according to lemma [ theo_1 ] , if @xmath5 is a subset of a @xmath1-antiresolving set , so is @xmath86 . we therefore give some useful properties of the function @xmath87 in theorem [ theo_3 ] below . [ theo_3 ] the function defined in equation [ eq_1 ] satisfies the following properties . 1 . @xmath88 2 . @xmath89 3 . @xmath90 4 . @xmath91 the first property comes straightforwardly from equation [ eq_1 ] . in order to prove the second property , let @xmath66 and @xmath92 . if @xmath93 , then @xmath94 by definition . let us thus assume that @xmath95 . given that @xmath92 , there exist @xmath69 such that @xmath96 and @xmath97 . let @xmath98 such that @xmath99 . note that , such an @xmath100 exists because @xmath101 . according to proposition [ prop_1 ] , since @xmath81 and @xmath102 , then @xmath77 , which means that @xmath103 and that @xmath104 , which proves the second property . the third property can be proven by using the first property . given that @xmath105 and @xmath106 , then @xmath107 and @xmath108 , hence , @xmath109 . similarly , @xmath110 and @xmath111 by definition , which implies that @xmath112 . again , applying the first property we obtain that @xmath113 . the two results lead to @xmath114 . finally , the last property is proven as follows . if @xmath115 , then @xmath116 by applying the second property . the proof is concluded by simply considering the first property . the function @xmath117 is the basis of algorithm [ alg_antiresolving ] , which aims to find a @xmath1-antiresolving set in a graph . algorithm [ alg_antiresolving ] is an optimized version supported by theorem [ theo_3 ] of the following algorithm . let us consider all subsets @xmath5 of @xmath118 with cardinality lower than or equal to @xmath119 . if @xmath86 is a @xmath1-antiresolving set , then a positive output is provided . if not , a proof that a @xmath1-antiresolving set does not exist is found when @xmath120 for every @xmath65 such that @xmath121 . note that , this impossibility result comes from the monotonicity of the function @xmath87 , _ any other case leads to the @xmath122 state where neither a proof nor a disproof of the existence of a @xmath1-antiresolving set can be found . a graph @xmath10 , an integer value @xmath119 to control the exponential explosion , and the integer value @xmath1 . let @xmath123 let @xmath124 [ step_1 ] true [ alg : line : return : true:1 ] [ step_2 ] let @xmath125 be an empty set let @xmath126 be the @xmath127th element of @xmath128 let @xmath129 be the @xmath130th element of @xmath128 [ step_3 ] @xmath131 ` true`[alg : line : return : true:2 ] add @xmath5 to @xmath125 ` false ` ` unknown ` algorithm [ alg_antiresolving ] can be considered a true - biased algorithm if the unknown state is regarded as a negative result . its computational complexity is clearly exponential in terms of @xmath119 . more precisely , for every @xmath132 we obtain that @xmath133 , because @xmath134 is formed by joining every pair of elements of @xmath135 . this means that the computational complexity of algorithm [ alg_antiresolving ] is determined by the size of @xmath136 . given that , in the worst case , the cardinality of @xmath136 quadratically increases with respect to @xmath137 , we obtain that the worst - case computational complexity of this algorithm is @xmath138 . although @xmath138 is double exponential in terms of @xmath119 , when @xmath139 it becomes significantly lower than the computational complexity of a brute force algorithm that considers the @xmath140 subsets of @xmath118 . for example , for @xmath141 , @xmath142 , and @xmath143 , the computational complexity becomes @xmath144 , @xmath145 , and @xmath146 , respectively . moreover , given that the search space monotonically increases with @xmath119 , the accuracy of the algorithm also increases with @xmath119 . in this sense , @xmath119 provides a trade - off between false negatives and computational cost . it is worth remarking that a theoretical lower bound , although not considered in the analysis , of the computational complexity of algorithm [ alg_antiresolving ] is @xmath147 , which is the computational complexity of the _ classic _ floyd - warshall algorithm required to compute the metric representation of all vertices . this prevents our method to be used on large graphs even when @xmath141 . in this case , more efficient implementations of both algorithm [ alg_antiresolving ] and the floyd - warshall algorithm ought to be considered , _ e.g. , _ @xcite . algorithm [ alg_antiresolving ] can be adapted to find a @xmath1-antiresolving basis rather than a @xmath1-antiresolving set . to that aim , we rely on proposition [ prop_2 ] below . proposition [ prop_2 ] gives a sufficient condition for the presence of a @xmath1-antiresolving basis . this implies just a small modification to algorithm [ alg_antiresolving ] . in particular , the conditional statements in lines [ alg : line : return : true:1 ] and [ alg : line : return : true:2 ] should take into account that such sufficient condition is satisfied . the full pseudo - code considering this modification is presented in algorithm [ alg_antiresolving_basis ] . [ prop_2 ] let @xmath5 be the subset of smaller cardinality in @xmath118 such that @xmath86 is a @xmath1-antiresolving set . then , @xmath86 is a @xmath1-antiresolving basis if @xmath148 such that @xmath149 it follows that @xmath150 . a graph @xmath10 , an integer value @xmath119 to control the exponential explosion , and the integer value @xmath1 . let @xmath123 let @xmath124 let @xmath151 ` true ` let @xmath125 be an empty set let @xmath126 be the @xmath127th element of @xmath128 let @xmath129 be the @xmath130th element of @xmath128 add @xmath152 to @xmath125 let @xmath153 be the minimum cardinality of a set in @xmath125 ` true ` ` false ` ` unknown ` in order to show the feasibility of both algorithm [ alg_antiresolving ] and algorithm [ alg_antiresolving_basis ] , we ran experiments considering @xmath154 and random graphs as input . the aim of the experiments is to provide statistically sound data on the ratio between positive , negative , and unknown results of the proposed algorithms . further below in this section we also show results on real - life social graphs . a random graph is created by choosing integer values uniformly distributed in the interval @xmath155 $ ] as the number of vertices @xmath156 ; where @xmath1 is a privacy threshold . the number of edges also distributes uniformly in the interval @xmath157 $ ] , and the edges are added randomly to the graph . for each pair @xmath158 , we created @xmath159 random graphs and executed algorithm [ alg_antiresolving ] and algorithm [ alg_antiresolving_basis ] in order to look for a @xmath1-antiresolving set and a @xmath1-antiresolving basis , respectively . the success rate of both algorithms considering @xmath160 is shown in figure [ fig : rates ] . we define a success as either a positive or a negative result , _ i.e. , _ whenever a @xmath1-antiresolving set ( basis ) or a proof that it does not exist is found . according to figure [ fig : rates ] , both algorithms perform poorly for @xmath161 . however , when @xmath142 they already achieve a success rate above @xmath15 , which is further improved by the more computationally demanding versions of algorithm [ alg_antiresolving ] and algorithm [ alg_antiresolving_basis ] that consider @xmath143 . the difference between figure [ figure : set ] and figure [ figure : basis ] suggests , as expected , that finding a @xmath1-antiresolving basis is harder than finding a @xmath1-antiresolving set . notwithstanding , algorithm [ alg_antiresolving_basis ] performs above @xmath15 when @xmath142 and @xmath143 . it is also worth remarking that , even though figure [ fig : rates ] hints that the success rate of both algorithms monotonically decreases with @xmath1 , our algorithms have @xmath162 of success rate if @xmath1 is equal to the order of the graph . this is because @xmath163 and , thus , all the nodes in the graph should be contained in a @xmath1-antiresolving set according to lemma [ theo_1 ] . in the previous section we provided a theoretical impossibility result whereby a graph can be proven to not contain a @xmath1-antiresolving set . in figure [ fig : ratio ] , we show that such impossibility result can be achieved by random graphs ; with small probability though . it seems also that the percentage of negative results monotonically increases with @xmath1 . indeed , it is easy to prove that this percentage reaches its minimum ( @xmath164 ) and maximum ( @xmath162 ) when @xmath1 takes its minimum ( @xmath165 ) and maximum ( @xmath166 ) value , respectively . however , proving the monotonicity of the percentage of negative results with respect to @xmath1 looks challenging and cumbersome . figure [ fig : ratio ] also shows that the increase of the success rate of both algorithms when @xmath119 grows is due to an increase on both the number of positive and negative results . this section ends with the evaluation of two real - life social graphs with respect to the proposed privacy measure . the first graph , named _ facebook graph _ in what follows , consists of @xmath167 ego - networks from facebook @xcite . it contains @xmath168 users , @xmath169 edges , and @xmath170 circles . the second graph describes an online community of students at the university of california @xcite . in total , @xmath171 students were registered in the network and @xmath172 links were created . we refer to this graph as _ panzarasa graph_. both graphs have been analyzed in order to determine the values of @xmath1 and @xmath30 such that they @xmath0-anonymity . taking into account the previously presented empirical results on synthetic data , we used for the analyses algorithm [ alg_antiresolving_basis ] with @xmath142 as a good trade - off between performance and success rate . the results are as follows . the panzarasa graph contains a @xmath53-antiresolving basis of size @xmath53 . this means that this graph does not satisfy @xmath51-anonymity for @xmath173 unless @xmath174 , which is a meaningless scenario . similarly , the @xmath1-metric antidimension of the facebook graph is @xmath53 for @xmath175 . hence , it satisfies @xmath176-anonymity only ; the lowest privacy guarantee with respect to our privacy measure . our results show that neither the panzarasa nor the facebook graph provide privacy guarantees against active attack . this is not surprising since these graphs have not been anonymized to prevent any type of structural attack . future work thus should be oriented to anonymization methods that consider @xmath51-anonymity as a privacy goal . in the next two sections we provide some primary theoretical results on the @xmath1-metric antidimension problem . we focus on giving mathematical properties that , supported by the results in section [ sect - algo ] , can determine or bound the @xmath1-metric antidimension for some families of graphs . in particular in this section , we study the @xmath1-metric antidimension of cycles , paths , complete bipartite graphs , and other graph families satisfying some specific conditions . to do so , we first observe some basic properties of @xmath1-antiresolving sets , which will be further used . [ basic - prop]@xmath177 1 . any resolving set is also a @xmath53-antiresolving set . 2 . there does not exist @xmath173 such that all the vertices of a graph form a @xmath1-antiresolving set . there does not exist any @xmath178-antiresolving set in a graph of order @xmath178 . 4 . not for every graph @xmath10 of order @xmath178 and every @xmath179 , there exist a @xmath1-antiresolving set in @xmath10 . for instance , if @xmath10 is a path graph , for @xmath180 there does not exist a @xmath1-antiresolving set in @xmath10 . in order to continue with our study we need to introduce some terminology and notation . for a graph @xmath10 and a vertex @xmath181 , the set @xmath182 is the _ open neighborhood _ of @xmath31 and the set @xmath183 = n_g(v)\cup \{v\}$ ] is the _ closed neighborhood _ of @xmath31 . two vertices @xmath28 , @xmath184 are called ( _ false _ ) _ true twins _ if ( @xmath185 ) @xmath186 = n_g[y]$ ] . in this sense , a vertex @xmath28 is a _ twin _ if there exists @xmath187 such that they are either true or false twins . the diameter of @xmath10 is defined as @xmath188 . as mentioned before , not for every integer @xmath1 it is possible to find a @xmath1-antiresolving set in a graph @xmath10 . thus , it is desirable to analyze first the interval of suitable values for @xmath1 satisfying that @xmath10 contains at least one @xmath1-antiresolving set . according to definition [ def_antiresolving_set ] we present the following concept , which is relevant in the study of the @xmath1-metric antidimension of graphs . a simple connected graph @xmath2 is @xmath1-_metric antidimensional _ , if @xmath1 is the largest integer such that @xmath10 contains a @xmath1-antiresolving set . in order to study the @xmath1-metric antidimension of graphs , we first focus into obtaining the values of @xmath1 for which a given graph is @xmath1-metric antidimensional . first notice that any graph @xmath10 is always @xmath1-metric antidimensional for some @xmath189 , and a natural upper bound for @xmath1 which makes that @xmath10 would be @xmath1-metric antidimensional is clearly the maximum degree of the graph , since the number of vertices at distance one from any vertex is at most the maximum degree of the graph . [ remark - delta ] if @xmath10 is a connected @xmath1-metric antidimensional graph of maximum degree @xmath190 , then @xmath191 . since the maximum degree of a graph is at most the order of the graph minus one , a particular case of the above result is the next one . if @xmath10 is any connected @xmath1-metric antidimensional graph of order @xmath178 , then @xmath192 . moreover , @xmath10 is @xmath193-metric antidimensional if and only if @xmath10 has maximum degree @xmath194 . the upper bound is a particular case of remark [ remark - delta ] . now , it is straightforward to observe that if @xmath31 is a vertex of @xmath10 of degree @xmath194 , then for every vertex @xmath195 it follows that @xmath196 , _ i.e. _ , every vertex different from @xmath31 has the same metric representation with respect to @xmath197 . thus , @xmath197 is a @xmath193-antiresolving set , since there exists no @xmath178-antiresolving sets in @xmath10 . thus , @xmath10 is @xmath193-metric antidimensional . on the contrary , we assume that @xmath10 is @xmath193-metric antidimensional . hence , if @xmath5 is a @xmath193-antiresolving set , then @xmath198 . thus , the only possibility is that @xmath5 is formed by a single vertex and that every vertex is adjacent to it . to continue with our study we need some extra notation . the _ eccentricity _ @xmath199 of a vertex @xmath31 in a connected graph @xmath10 is the maximum length of a shortest path between @xmath31 and any other vertex @xmath21 of @xmath10 . notice that the maximum of the eccentricities of any vertex of @xmath10 is the diameter of @xmath10 and the minimum of the eccentricities is the radius of @xmath10 . for more information on vertex eccentricity in graphs , see for instance @xcite . figure [ example - ecc ] shows a graph @xmath10 and a table with the eccentricities of all its vertices . given a vertex @xmath21 of a graph @xmath10 , we consider the following local parameter . for every @xmath200 , let @xmath201 . now , for every @xmath202 , let @xmath203 [ phi - page ] and for any graph @xmath10 , let @xmath204 . the table and the graph of figure [ example - ecc ] clarify the notation above . and a table which shows that @xmath205.,title="fig:",scaledwidth=14.0% ] and a table which shows that @xmath205.,title="fig:",scaledwidth=70.0% ] [ dimensional - phi ] any connected graph @xmath10 is @xmath1-metric antidimensional for some @xmath206 . assume @xmath28 is a vertex of degree at least two in @xmath10 such that @xmath207 . thus , for any vertex @xmath187 , there exist at least @xmath208 vertices @xmath209 in @xmath210 such that @xmath211 . moreover , since @xmath207 , there exists at least one vertex @xmath212 such that there are exactly @xmath213 different vertices satisfying the above mentioned . so , @xmath214 is a @xmath215-antiresolving set and @xmath10 is @xmath1-metric antidimensional for some @xmath206 . now , notice that if a graph @xmath10 is a @xmath53-metric antidimensional , then every subset of vertices @xmath5 ought to be a @xmath53-antiresolving set , implying that @xmath216 . according to that fact , in this work we are mainly interested in those graphs being @xmath1-metric antidimensional for some @xmath217 . an example of a graph being @xmath53-metric antidimensional is for instance the path graph of even order . to begin with the description of some families of graphs being @xmath1-metric antidimensional for some @xmath217 we define the _ radius _ and the _ center _ of a graph as follows . the _ radius _ @xmath218 of @xmath10 is the minimum eccentricity of any vertex in @xmath10 . the _ center _ of @xmath10 is the set @xmath5 of vertices of @xmath10 having eccentricity equal to the radius of @xmath10 . if the center of a graph @xmath10 is only one vertex , then @xmath10 is @xmath1-metric antidimensional for some @xmath217 . let @xmath31 be the center of @xmath10 . hence , there exist two diametral vertices @xmath219 such that @xmath220 . since @xmath31 has eccentricity @xmath218 , there is no vertex @xmath221 in @xmath10 such that @xmath222 . thus , it follows that for any vertex @xmath223 there exists at least a vertex @xmath184 belonging to the @xmath224 path such that @xmath225 . therefore , @xmath197 is a @xmath1-antiresolving set in @xmath10 for some @xmath217 . if a path graph has odd order , then its center is formed by only one vertex . also , for every vertex of any path , there exists at most other different vertex having equal distance to a third vertex of the path . thus , it is clear the following consequence of the remark above . [ paths ] if a path @xmath226 has odd order , then it is @xmath18-metric antidimensional . another example of @xmath18-metric antidimensional are the cycle graphs as we next see . [ cycles ] any cycle graph @xmath227 is @xmath18-metric antidimensional . we assume first that @xmath178 is odd . let @xmath31 be any vertex of @xmath227 . hence , for any vertex @xmath223 of @xmath227 , there exists only one @xmath228 such that @xmath229 . thus , @xmath197 is a @xmath18-antiresolving set in @xmath227 . assume now that @xmath178 is even and let @xmath230 be any two diametral vertices of @xmath227 . we observe that for any vertex @xmath231 of @xmath227 , there exists only one @xmath232 such that @xmath233 and @xmath234 . thus , @xmath230 is a @xmath18-antiresolving set in @xmath227 . on the other hand , there does not exists @xmath235 such that @xmath227 contains a @xmath1-metric antiresolving set , since for any vertex of @xmath227 , there exists at most another different vertex having equal distance to a third vertex of the path . therefore , @xmath227 is @xmath18-metric antidimensional . if the vertices of a set @xmath5 are pairwise twins in a graph @xmath10 , then it is clear that they have the same distance to every other vertex @xmath236 . so , @xmath237 is a @xmath238-antiresolving set for @xmath10 . hence , the following result . [ obs : twin - vertices ] if the vertices of a set @xmath5 are pairwise twins in a graph @xmath10 , then @xmath10 is @xmath1-metric antidimensional for some @xmath239 . complete bipartite graph is complete bipartite if its vertex set can be divided into two disjoint sets @xmath240 and @xmath22 such that every vertex in @xmath240 is adjacent to every vertex in @xmath22 and no more . ] are special kind of graphs , since they have a bipartition of the vertex set in which all the vertices belonging to one of the sets of the bipartition are pairwise twin vertices . let @xmath241 be a complete bipartite graph . next we analyze the suitable values @xmath1 making a complete bipartite graph @xmath1-metric antidimensional . [ bipartite ] any complete bipartite graph @xmath241 with @xmath242 is @xmath243-metric antidimensional . let @xmath240 and @xmath22 be the two disjoint sets of @xmath241 with @xmath244 and @xmath245 . notice that @xmath240 ( respectively @xmath22 ) is a set of pairwise twin vertices . thus , by observation [ obs : twin - vertices ] we have that @xmath241 is @xmath1-metric antidimensional for some @xmath246 . suppose that @xmath247 and let @xmath5 be a @xmath1-antiresolving set for @xmath241 . since every vertex of @xmath241 has distance either one or two to any other vertex of @xmath241 it is not possible to find @xmath12 vertices having the same distance to every vertex of @xmath5 , a contradiction . so , @xmath248 and the proof is complete . in this subsection we compute the @xmath249-metric antidimension of some graphs which were already described to be @xmath1-metric antidimensional for some value @xmath250 . it is clear that the first natural bound which follows for the @xmath1-metric antidimension of a graph of order @xmath178 is @xmath251 . such a bound is tight . it is achieved , for instance , for the complete bipartite graphs @xmath241 as we can see at next by taking the case @xmath252 . let @xmath253 be two positive integers with @xmath242 . 1 . if @xmath252 , then @xmath254 . 2 . if @xmath255 , then @xmath256 . from proposition [ bipartite ] we know that @xmath241 is a @xmath243-metric antidimensional graph . let @xmath240 and @xmath22 be the two partite sets of @xmath241 with @xmath244 and @xmath245 . we assume first that @xmath252 . let @xmath257 with @xmath258 and let be @xmath259 . notice that if @xmath248 , then @xmath260 and so , @xmath261 . since any vertex @xmath262 ( or equivalently @xmath263 ) is adjacent to every vertex of @xmath22 and it has distance two to every vertex in @xmath264 , we have that all the vertices of @xmath265 have the same metric representation with respect to @xmath5 . as @xmath258 , it follows that @xmath5 is a @xmath1-antiresolving set and @xmath266 . now , suppose @xmath267 and let @xmath268 be a @xmath1-antiresolving set for @xmath241 . so , we have either one of the following situations . * there exist more than @xmath1 vertices of @xmath240 not in @xmath268 . hence , for any vertex @xmath269 there exist at least @xmath1 vertices not in @xmath268 which , together with @xmath21 , have the same metric representation with respect to @xmath268 . so , @xmath268 is not a @xmath1-antiresolving set , but a @xmath249-antiresolving set for some @xmath270 , a contradiction . * there exists at least one vertex of @xmath22 not in @xmath268 . it is a direct contradiction , since @xmath271 . therefore , we obtain that @xmath254 . on the other hand , we assume that @xmath272 . let @xmath273 with @xmath274 , let @xmath275 with @xmath276 and let @xmath277 . hence , for any vertex @xmath278 ( or equivalently @xmath279 ) , there exist exactly @xmath12 vertices , such that all of them , together with @xmath31 , have the same metric representation with respect to @xmath280 . thus , @xmath280 is a @xmath1-antiresolving set and @xmath281 . now , suppose that @xmath282 and let @xmath283 be a @xmath1-antiresolving set in @xmath241 . hence , either there exist more than @xmath1 vertices of @xmath240 not in @xmath283 or there exist more than @xmath1 vertices of @xmath22 not in @xmath283 . as above , in any of both possibilities we obtain that @xmath283 is not @xmath1-antiresolving set , but a @xmath249-antiresolving set for some @xmath270 , a contradiction . as a consequence , we obtain that @xmath284 . next we study the @xmath1-metric antidimension of some other families of basic graphs . according to remark [ cycles ] we know that the cycles @xmath227 are @xmath18-metric antidimensional and , by corollary [ paths ] , that the paths @xmath226 are @xmath18-metric antidimensional only in the case @xmath178 is odd . next we compute its @xmath18-metric antidimension . let @xmath285 be an integer . then @xmath286 if @xmath31 is the center of a path @xmath287 , then for any other vertex @xmath288 there exists exactly one vertex @xmath289 such that @xmath290 have the same metric representation with respect to @xmath197 . thus @xmath291 . suppose @xmath178 is even and let @xmath219 be two diametral vertices in @xmath227 . we observe that for any vertex @xmath231 there exists exactly one vertex @xmath292 in @xmath227 , such that @xmath293 have the same metric representation with respect to @xmath230 . thus , @xmath294 . to see that @xmath295 we can observe that any set with only one vertex @xmath296 is not a @xmath53-antiresolving set , since for the vertex @xmath87 being diametral with @xmath296 there does not exist any other vertex @xmath297 having the same metric representation with respect to @xmath298 . on the other hand , if @xmath178 is odd and @xmath299 is any vertex of @xmath227 , then we can check that for any vertex @xmath300 , there exists exactly one vertex @xmath301 , such that @xmath302 have the same metric representation with respect to @xmath303 . thus , @xmath304 . let @xmath305 be a tree and let @xmath21 be a vertex of @xmath305 of degree at least two . let @xmath31 be a neighbor of @xmath21 . a @xmath31-branch of @xmath305 at @xmath21 is the subtree @xmath306 obtained from the union of all length maximal paths beginning in @xmath21 , passing throughout @xmath31 and finishing at a vertex of degree one in @xmath305 . given a @xmath184-branch @xmath307 at @xmath28 , we say that @xmath308 is the eccentricity of the vertex @xmath28 in the @xmath184-branch @xmath307 . two branches @xmath309 and @xmath310 at @xmath28 are @xmath311-_equivalent _ if @xmath312 . for every vertex @xmath28 of @xmath305 , let @xmath313 represents the maximum number of pairwise @xmath311-equivalent branches at @xmath28 and let @xmath314 equals the length of any @xmath311-equivalent branch . now , for any tree @xmath305 , we define the following parameter : @xmath315 an example which helps to clarify the above definitions is given in figure [ example ] . there we have a tree @xmath305 satisfying the following . the vertex @xmath316 has 4 branches : @xmath317 , @xmath318 , @xmath319 and @xmath320 . for instance @xmath321 . we observe that @xmath322 , @xmath323 , @xmath324 and @xmath325 . so , @xmath316 has no @xmath326-equivalent branches and @xmath327 . similarly , it can be noticed that @xmath33 and @xmath328 are the only vertices of @xmath305 which have equivalent branches . that is , @xmath329 and @xmath330 are @xmath331-equivalent , since @xmath332 . thus @xmath333 and @xmath334 . analogously , @xmath335 and @xmath336 , @xmath337 . therefore @xmath338 . -metric antidimensional tree @xmath305.,scaledwidth=95.0% ] now , for the particular case of trees , we next use the definition of @xmath339 already presented in section [ sec_metric_antidimension ] . as an example , for the tree of figure [ example ] we have that , for instance , @xmath340 , @xmath341 and @xmath342 . also , some calculations give that @xmath343 . now , with the definitions above we present the following result . [ dimensional - trees ] any tree @xmath305 is @xmath1-metric antidimensional for some @xmath344 . from theorem [ dimensional - phi ] it follows that @xmath345 . now , let @xmath28 be a vertex of degree at least two in @xmath305 such that @xmath346 . hence , there exist @xmath347 disjoint paths beginning in @xmath28 , passing throughout a vertex @xmath348 ( neighbor of @xmath28 ) , and ending in a vertex @xmath349 of degree one in @xmath305 with @xmath350 . moreover , every @xmath348-branch @xmath351 does not contain any other vertex further away from @xmath28 than @xmath349 . we consider now the set @xmath352 notice that for any vertex @xmath353 , there exist at least @xmath354 different vertices @xmath355 in @xmath356 such that @xmath357 for every @xmath358 . moreover , since @xmath346 , it follows that there exists a vertex @xmath359 such that there are exactly @xmath354 different vertices satisfying the above mentioned . thus , @xmath265 is a @xmath347-antiresolving set and , as a consequence , @xmath305 is @xmath1-metric antidimensional for some @xmath360 . therefore we obtain that @xmath305 is @xmath1-metric antidimensional for some @xmath361 and the proof is complete . according to theorem [ dimensional - trees ] , we conclude that the tree @xmath305 in figure [ example ] is @xmath1-metric antidimensional for some @xmath180 , since @xmath343 and @xmath338 ; that tree is indeed @xmath362-antidimensional . if we add some extra vertices to this mentioned tree , like in figure [ example2 ] , we obtain that @xmath363 ( since @xmath364 ) , and it remains @xmath338 . thus , this new tree is @xmath1-metric antidimensional for some @xmath365 , and by remark [ remark - delta ] we have that @xmath366 . -metric antidimensional tree @xmath305.,scaledwidth=95.0% ] notice that @xmath367 holds for some trees . for instance , if we take a star graph @xmath368 , @xmath369 , and we add an extra vertex @xmath28 connected by an edge with one leaf @xmath184 of @xmath368 , then we have a tree @xmath305 such that @xmath370 ( for the vertex @xmath184 , @xmath371 ) and @xmath372 . moreover , there are graphs in which the bound of theorem [ dimensional - trees ] is not achieved . an example of this appears in figure [ example3 ] . there we have a tree @xmath305 such that @xmath338 ( @xmath373 ) and @xmath343 ( @xmath374 ) . nevertheless the set @xmath375 is a @xmath376-antiresolving set . is a @xmath376-metric antiresolving set.,scaledwidth=95.0% ] once we have a lower bound for the integer @xmath249 for which a given tree @xmath305 is @xmath249-metric antidimensional , we are able to compute its @xmath1-metric antidimension for a suitable value @xmath377 . we first notice that if @xmath305 is @xmath53-metric antidimensional , then any @xmath53-metric antiresolving set is an standard resolving set , as defined in @xcite and , in such a case , @xmath378 . since it is not our goal to study such a case , from now on we consider only those trees being @xmath1-metric antidimensional for some @xmath217 . [ remark - phi - equal-1 ] let @xmath305 be a @xmath1-metric antidimensional tree and let @xmath379 such that @xmath380 . if @xmath381 , then @xmath382 . since @xmath381 , then for every vertex @xmath187 , there exist at least @xmath383 vertices @xmath384 such that @xmath211 . moreover , there exists a vertex @xmath212 such that there are exactly @xmath383 different vertices satisfying the above mentioned . thus , @xmath214 is a @xmath6-metric antiresolving set and , as a consequence , @xmath382 , since on the other hand , @xmath385 for any graph @xmath10 . for any tree @xmath305 such that @xmath386 , @xmath387 . according to the results above , it remains to study the @xmath1-metric dimension of trees for the case in which every vertex @xmath31 of @xmath305 satisfies that @xmath388 . to do so , we need to introduce some notations . we denote by @xmath389 , for some @xmath390 , the set of vertices @xmath391 such that @xmath392 . now , for every @xmath393 , let @xmath394 and for the set of vertices @xmath395 such that @xmath396 , let @xmath397 be the maximum cardinality among all possible sets obtained as the union of @xmath1 vertex sets of the branches @xmath398 where @xmath399 minus the vertex @xmath31 itself . as an example , we consider the tree of figure [ example ] . for @xmath400 , there we have that @xmath401 , @xmath402 , @xmath403 , @xmath404 and @xmath405 ( notice that @xmath406 can be different from this set , but it always has five vertices ) . with this definition we are able to present the following result , where we analyze only those graphs being @xmath249-metric antidimensional for @xmath407 . [ dimension - trees - bound ] let @xmath305 be a @xmath249-metric antidimensional of order @xmath178 with @xmath407 . then for any @xmath377 , @xmath408 we consider a set @xmath409 given by @xmath410 in this sense , for any vertex @xmath236 , there exists at least @xmath12 vertices @xmath411 not in @xmath5 such that @xmath412 for every @xmath413 . moreover , if there exists at least one vertex @xmath414 for which there are exactly @xmath12 vertices not in @xmath5 satisfying the above mentioned , then @xmath5 is a @xmath1-metric antiresolving set and the result follows since the cardinality of @xmath5 is given by the formula of the theorem . on the contrary , if such a vertex does not exist , then @xmath5 is a @xmath415-metric antiresolving set for @xmath10 for some @xmath416 . since in this case , @xmath417 we obtain the result . consider now the example of figure [ example ] . according to the result above , we have that the set @xmath418 is a @xmath18-metric antiresolving set for such a tree @xmath305 and @xmath419 . nevertheless , since @xmath342 , from remark [ remark - phi - equal-1 ] we have that @xmath420 . next we present a family of trees , where the bound of theorem [ dimension - trees - bound ] is achieved . we consider the family @xmath421 of trees @xmath422 satisfying the following conditions . * the center of @xmath422 is formed by two adjacent vertices , say @xmath293 . * @xmath422 is `` rooted '' in @xmath293 . * @xmath422 is a complete @xmath243-ary tree ( each vertex of degree greater than one has @xmath243 children ) * any two leaves being descendants of the same root ( @xmath28 or @xmath184 ) , have the same distance to this root . an example of a tree @xmath423 of the family @xmath421 is given in figure [ example4 ] . of the family @xmath421 . the set @xmath424 is a @xmath362-metric antiresolving set of minimum cardinality.,scaledwidth=95.0% ] it is straightforward to observe that @xmath425 and @xmath426 . thus , @xmath422 is @xmath249-metric antidimensional for some @xmath247 , and by remark [ remark - delta ] we have that @xmath427 . now on , we compute the @xmath243-metric antidimension of @xmath422 . according to the construction of the family @xmath421 , we see that the root vertices @xmath293 of a tree @xmath428 satisfy that @xmath429 . also , @xmath430 and the sets @xmath431 , @xmath432 are formed by the set of all their corresponding descendants ( this fact makes unnecessary to consider other vertices of @xmath422 ) . as a consequence of this , by theorem [ dimension - trees - bound ] we have that @xmath424 is a @xmath243-metric antiresolving set and @xmath433 . since , for any non - leaf vertex @xmath21 of @xmath422 satisfies that @xmath434 , we have that any singleton vertex ( being not a leaf ) is a @xmath435-metric antiresolving set . thus , @xmath436 and we have that @xmath437 , which makes that the bound of theorem [ dimension - trees - bound ] is tight . in this article we have introduced a new problem in graph theory ( the @xmath1-metric antidimension problem ) that resembles to the well - known metric dimension problem . the @xmath1-metric antidimension is the basis of our novel privacy measure @xmath17-anonymity . this measures quantifies the level of privacy offered by an outsourced social graph against active attacks . consequently , privacy - preserving methods for the publication of social networks ought to consider @xmath0-anonymity as one of their privacy goal . we have proposed a true - biased algorithm aimed at finding both a @xmath1-antiresolving set and a @xmath1-antiresolving basis in a graph . the algorithm , although computationally demanding , reached a success rate above @xmath15 during the executed experiments when looking for a @xmath1-antiresolving basis . we expect future experiments to be conducted over real - life social graphs so that privacy - preserving methods satisfying @xmath0-anonymity can be empirically evaluated in terms of utility and resistance to active attacks . we have also began the study of mathematical properties of the @xmath1-antiresolving sets and the @xmath1-metric antidimension of graphs . we have studied some particular graph families like cycles , paths , complete bipartite graphs and trees . for instance , we have obtained that for any path @xmath226 of odd order , @xmath438 and for any cycle @xmath227 it follows that @xmath439 if @xmath178 is odd , and @xmath440 if @xmath178 is even . also , for every complete bipartite graph @xmath241 , @xmath254 if @xmath441 , and @xmath256 if @xmath255 . for the case of trees we have presented a tight lower bound for its @xmath1-metric antidimension in terms of the order of the tree and the order of some subtrees satisfying some specific conditions . we have also described an infinite family of @xmath1-ary trees which achieve this bound . finally , this article opens new and challenging open problems related to the @xmath1-metric antidimension of graphs and the privacy concept @xmath17-anonymity . for instance , it would be interesting to characterize the family of graphs such that they are @xmath53-metric antidimensional , as well as looking for a close relationship between the @xmath1-metric antidimension and the @xmath1-metric dimension of a graph . in particular , those families of graphs that resemble to social graphs must be considered . the computational complexity of computing the @xmath1-metric antidimension should also be addressed . in case the problem is np - complete , efficient heuristics and privacy - preserving methods need to be developed so as to compute the @xmath1-metric antidimension and , ultimately , transform a social graph into a @xmath0-anonymous graph for given values of @xmath1 and @xmath30 . 10 abedelaziz mohaisen , tamer abuhmed , hun jeong kang , yongdae kim , and daehun nyang . mistaking friends for foes : an analysis of a social network - based sybil defense in mobile networks . , pages 28:128:8 , 2011 .

we analyse some properties of circumplanetary discs . flow through such discs may provide most of the mass to gas giant planets , and such discs are likely sites for the formation of regular satellites . we model these discs as accretion discs subject to the tidal forces of the central star . the tidal torques from the star remove the disc angular momentum near the disc outer edge and permit the accreting disc gas to lose angular momentum at the rate appropriate for steady accretion . circumplanetary discs are truncated near the radius where periodic ballistic orbits cross , where tidal forces on the disc are strong . this radius occurs at approximately for the planet hill radius . during the t tauri stage of disc accretion , the disc is fairly thick with aspect ratio and the disc edge tapering occurs over a radial scale . the disc fluid equations can be rescaled in the hill approximation to a form similar to the flow equations for a disc in a binary star system with a mass ratio of unity . for a circular or slightly eccentric orbit planet , no significant resonances lie within the main body of the disc . tidally driven waves involving resonances nonetheless play an important role in truncating the disc , especially when it is fairly thick . we model the disc structure using one dimensional time - dependent and steady - state models and also two dimensional sph simulations . the circumplanetary disc structure depends on the variation of the disc turbulent viscosity with radius and is insensitive to the angular distribution of the accreting gas . dead zones may occur within the circumplanetary disc and result in density structures . if the disc is turbulent throughout , the predicted disc structure near the location of the regular jovian and saturnian satellites is smooth with no obvious feature that would favor formation at their current locations . it may be possible that substructure , such as due to variations in the disc turbulence , could lead to the trapping of migrating satellites . [ firstpage ] accretion , accretion discs planets and satellites : formation planetary systems planet - disk interactions

1012.4102

in the core accretion model of planet formation , at the earliest stages of giant planet formation , the planet is fully embedded in the gaseous disc that orbits the central star @xcite . at later stages , during run - away - gas accretion , tidal forces due to the planet open a gap in the disc , typically when the planet reaches a mass of order neptune s mass @xcite . the gap opening does not necessarily imply that the gas flow on to the planet has ceased @xcite . the gap structure is determined by the properties of dynamical flow within it . the relatively low density of gas in the gap results from the relatively high flow radial velocities towards the star as a consequence of the disturbance caused by the planet . outside the gap region , the gas radial drift occurs on the much slower viscous timescale , resulting in much higher densities than occur in the gap . some studies suggest that most of the inflowing gas just outside the gap is accreted by a planet whose mass is comparable to or less than jupiter s for typical disc parameters @xcite . during this gap phase , the planetary radius is typically much smaller than its hill ( tidal ) radius . circumplanetary discs may form as gas flowing through the gap on to a planet carries some angular momentum about the planet @xcite . as a result , the inflowing gas has too much angular momentum to directly strike the planet and instead forms a disc about it . most of the mass of jupiter may have been acquired in the gap stage through gas flow that involves a circumplanetary disc . there are at least two motivations for studying circumplanetary discs . the observational detection of a disc orbiting an extra - solar gas giant planet would provide important evidence about the process of planet formation . the determination of their expected properties provides constraints for their detection . a second motivation is to better understand satellite formation , since circumplanetary discs are likely sites of satellite formation . satellites that orbit jupiter and saturn are classified into two groups . the regular satellites have low eccentricity and low orbital inclination to the equatorial plane of their planet . they are thought to have formed in the circumplanetary disc @xcite . the second type , the irregular satellites , can have a high eccentricity and high orbital inclinations . they can orbit either progradely or retrogradely with respect to the planet spin @xcite . the circumplanetary disc may capture these satellites though gas drag @xcite , although the disc does not play such a role in the nice model ( e.g. * ? ? ? the circumplanetary accretion disc plays a vital role in satellite formation . the regular satellite systems of jupiter and saturn extend over a small fraction of their respective hill radii , to less than @xmath4 . this fact has motivated a dynamical explanation in terms of circumplanetary disc sizes ( e.g. * ? ? ? * ; * ? ? ? we discuss this in more detail in section 8 . previous studies have considered the role of the angular momentum ( about the planet ) of the inflowing gas in determining the disc structure , in particular the disc outer radius . this picture has some intuitive appeal , since no disc would form , i.e. , the disc radius would be zero , if the inflowing gas has zero angular momentum . @xcite suggested that after the planet has opened a gap , the circumplanetary disc extends to @xmath5 . this result is based on the initial angular momentum of gas whose flow relative to the planet is slow near the planet s hill sphere and subsequently accelerates inward towards the planet while conserving its angular momentum about the planet . @xcite and @xcite considered cases where the angular momentum of the inflowing gas is considerably smaller . @xcite ( see also @xcite ) suggested that the circumplanetary disc structure prior to or during gap opening could account for the location of regular satellites . this model was also based on the concept that the disc structure is determined by the angular momentum of infalling gas about the planet , with an implied smaller value of disc radius of @xmath6 . this paper concentrates on the former case , the case of a circumplanetary disc in the presence of gap opening . the results of this paper may also apply to this case . we briefly discuss this point again below equation ( [ hr ] ) and in section [ concs ] . this radius characteristic of the angular momentum of the accreting gas is useful for providing an estimate of the conditions required for circumplanetary disc formation . for a disc to form about a planet with a gap , the planetary radius must be smaller than this value . violating this constraint for disc formation would require very short orbital periods , of order @xmath7 , for a planet of jupiter s radius . however , it is not clear that this radius characteristic of the angular momentum of the accreting gas is important for determining the disc radius or even has much influence of the disc structure . the reason has to do with the requirements of angular momentum loss from a steady - state disc . the process of angular momentum transport in circumplanetary discs has up to now been largely ignored . the work by @xcite and @xcite did consider the angular momentum issue , as will be discussed further in section 6.5 . @xcite briefly discuss the possible role of resonances in truncating the disc . for the disc to accrete in a steady - state manner , gas within it must continuously lose angular momentum . these studies recognized that the disc would extend beyond the radius in which mass is injected in order to account for the requirements of angular momentum conservation . the angular momentum issue has been previously encountered in the case of discs in mass exchange binary star systems ( e.g. * ? ? ? the inflowing gas from the mass losing star is captured within the roche lobe of the companion star . this gas has some angular momentum that often results in the formation of disc about the companion . the picture that has resulted is that the disc spreads outward by viscous torques until it reaches a radius at which tidal torques from the companion are sufficient to remove the angular momentum at the rate required for steady state accretion . the disc structure is insensitive to the angular momentum of the accreting gas . the case of a planet accreting gas can be considered to be a binary of extreme mass ratio . the accretion disc model in the binary star case suggests a somewhat different picture of how the accretion process operates than has been previously considered for circumplanetary discs . we consider the accretion disc model of circumplanetary discs in this paper . some recent three - dimensional simulations have analysed the flow in circumplanetary discs . for example , @xcite simulated gas accretion by protoplanets in three - dimensions with gas self - gravity and radiation transfer to investigate the properties of circumplanetary discs . they determined the disc scale height to be @xmath8 and found the disc size to be about @xmath9 . they attributed this value of the disc radius as following from the estimate of @xcite , for which the radius is determined by the angular momentum of the accreting gas . @xcite also performed three - dimensional calculations of circumplanetary discs . for a jupiter mass planet , he found a peak in the surface density at about 21 protoplanetary radii ( or @xmath10 ) from the protoplanet centre . according to these results , the peak is due to the balance between the centrifugal force and the gravity of the protoplanet . such small scale features might then correspond to the formation sites of the regular satellites around jupiter and saturn . however , this peak in the surface density was not reproduced by @xcite who found that the surface density decreases monotonically with radius from the protoplanet . we analyse the general properties of circumplanetary discs by the use of some simplified models , in order to understand the physical processes that affect disc structure . in section [ properties ] , we estimate the scale height and temperature of the disc . in section [ test ] we consider a ballistic disc model consisting of periodic particle orbits around a planet . in section [ eqs ] we examine the circumplanetary fluid disc evolution equations in the hill approximation and discuss their relationship to the binary star case . in section [ res ] we consider whether resonances could lie within a circumplanetary disc . in section [ tidal ] , we analyse the effects of a tidal torque on a purely viscous disc . in section [ sph ] we describe results of a model that includes the effects of gas pressure by means of some sph simulations . section [ concs ] contains a discussion and the conclusions . we consider a system with a planet of mass @xmath11 in a circular orbit around a star of mass @xmath12 at a separation @xmath13 . the hill sphere is the approximate region where the planet s gravity dominates that of the star and its radius is given by @xmath14 where @xmath15 . the sound speed in an ideal gas is @xmath16 where @xmath17 is the boltzmann constant , @xmath18 is the temperature , @xmath19 is the mean molecular weight and @xmath20 is the mass of a hydrogen atom . the surface temperature of a steady state accretion disc around an object of mass @xmath21 at a radius @xmath22 is given by @xmath23 @xcite where @xmath24 is the stefan - boltzmann constant . we estimate the temperature of the accreting gas with the disc . similar approaches were taken by @xcite . consider a disc orbiting a solar mass star that is accreting gas at a rate of @xmath25 , a typical value inferred from observations of t tauri stars . equation [ temp ] implies that at a distance from the central star of @xmath26 the disc has a temperature of about @xmath27 , ignoring effects of stellar heating . this temperature corresponds to a disc aspect ratio @xmath28 . the aspect ratio of the circumplanetary disc taken with respect to the planet is @xmath29 where @xmath30 is the mass of jupiter and @xmath22 is the distance from the planet . ( there is a small correction due to the vertical gravity of the star that we ignore . ) this aspect ratio is much larger than the value for the circumstellar disc , @xmath31 . we have taken the disc sound speed @xmath32 to be given by its value at the disc surface and have ignored any increase at the midplane with optical depth . the relatively high value of @xmath33 is due to the higher gas temperatures and the typically smaller @xmath34 values in the circumplanetary disc case . the value of the disc aspect ratio from equation ( [ hr ] ) is similar to the values obtained in the simulations by @xcite . they found no evidence for circumplanetary discs in their simulations of low mass planets ( less than 100 earth masses ) . they attributed this result to the weakness of the planet s vertical gravity compared to pressure , as can be seen in equation ( [ hr ] ) . for a sufficiently low mass planet , @xmath35 can be large enough that the concept of a disc breaks down and the gas around the planet is essentially spherical . as is well known , the circumplanetary disc temperatures at these accretion rates are too high to explain the existence of the icy satellites of jupiter and saturn @xcite . however , in a slower accretion phase as proposed by @xcite and further investigated by @xcite , the temperature would be lower and @xmath36 would be smaller , @xmath37 . we discuss this further in section [ concs ] . we compare the surface density in the circumplanetary disc , @xmath38 , to the surface density in the circumstellar disc , @xmath39 . we assume that most of the mass being accreted through the circumstellar disc outside the orbit of the planet is accreted by it . as discussed in the introduction , this efficient gas capture has been found in simulations of jupiter mass planets that orbit solar mass stars . the mass transfer rate through a steady state disc is given by @xmath40 @xcite , where the viscosity is parametrised with the @xmath41 prescription so that @xmath42 @xcite where @xmath43 . it then follows that @xmath44 where @xmath45 and @xmath46 are the viscosities in the circumstellar and circumplanetary discs respectively , and @xmath47 and @xmath48 similarly refer the @xmath41 values in the discs . consider a solar mass star and a jupiter mass planet , @xmath49 , and circumstellar and circumplanetary disc aspect ratios of @xmath50 and 0.3 respectively , as discussed above . for @xmath51 and equal @xmath41 values ( @xmath52 ) in the above equation , the ratio of the surface densities is @xmath53 . so the circumplanetary disc surface density is somewhat higher that the local circumstellar disc density . circumstellar discs could under some circumstances contain dead zones , regions where the disc ionization is too low for the magneto - rotational instability to operate as a source of turbulence ( gammie 1996 ) . dead zones operate where the temperature is sufficiently low that thermal ionization is weak ( less than about @xmath54 degrees ) and where the surface densities are high enough that external sources of ionization such as cosmic rays do not penetrate far below the disc surface . for expected conditions in a circumstellar disc , dead zones may extend from a few tenths of an au to several au , in some cases beyond jupiter s orbital radius ( e.g. * ? ? ? the model here suggests that the conditions for dead zone formation within a circumplanetary disc are perhaps more favorable than in the nearby circumstellar disc gas . the reason is that the circumstellar disc surface densities are higher than in the nearby circumstellar disc gas , while the temperatures are low enough to avoid sufficient ionisation in much of the circumplanetary disc . a difference in the properties of turbulence in the two discs violates our assumption of @xmath52 . but , it does so in a way that may further enforce the importance of dead zones in the circumplanetary disc . that is , @xmath55 would be expected to increase , leading to a decrease in @xmath56 we do not pursue the possibility of dead zones further in our disc models described later and assume a simple viscous disc . the viscous timescale in the disc is @xmath57 where @xmath58 is the orbital period of the planet . so we expect that such discs should be viscously relaxed for jupiter and saturn . this result has the implication that the viscous disc flow can be regarded to be in a steady state . a crude estimate of the steady state luminosity ratio of the circumplanetary to circumstellar discs in a system with a jupiter mass planet is @xmath59 where @xmath60 and @xmath61 are the planet and star radii , respectively . for @xmath62 and @xmath63 , we estimate @xmath64 to be of order a few percent . we see that circumplanetary discs are not very bright . we discuss this further in section [ concs ] . in this section we consider ballistic particle orbits in a planet - star system in order to find their nonkeplerian angular velocity and where the orbits begin to cross . as a simple two - dimensional disc model , we consider ballistic particles orbiting a planet of mass @xmath11 in the corotating frame of the star - planet system with a star of mass @xmath12 . the particles lie in the star - planet orbit plane . for such a model to represent a low pressure ( cold ) steady state fluid disc , each orbit must be periodic in the corotating frame and be nonintersecting , either with itself or neighboring orbits , in order that the velocity be single - valued in space . in addition , the orbit must be stable . the equation of motion of a ballistic particle at position @xmath65 for potential @xmath66 in the corotating frame is @xmath67 where @xmath68 is the angular velocity of the planet . the potential due to the rotation of the frame and the gravity of the planet and the star is given by @xmath69 where @xmath70 , @xmath71 and @xmath72 are the position vectors of center of mass , the planet , and the star respectively . the angular velocity of the star planet binary system is @xmath73 we adopt units where @xmath74 . the equations of motion for the particle with cartesian coordinates @xmath75 , whose origin is at the center of mass , are then @xmath76}{\left(\left[x-(1-\mu)\right]^2+y^2\right)^\frac{3}{2 } } + \frac{(1-\mu)(x+\mu)}{\left[(x+\mu)^2+y^2\right]^\frac{3}{2}}=0 \label{xeqtest}\ ] ] and @xmath77 ^ 2+y^2\right)^\frac{3}{2 } } + \frac{(1-\mu)y}{\left[(x+\mu)^2+y^2\right]^\frac{3}{2}}=0 . \label{yeqtest}\ ] ] following the standard procedure to obtain the hill equations , we transform these equations to a coordinate system centered on the planet and rescale the coordinates . the scaling is chosen so that the new coordinates scale with the hill radius , see equation ( [ rh ] ) . we make the change of variables to rescaled radius @xmath78 defined by @xmath79 where @xmath80 is the displacement of the planet from the center of mass . the equation of motion ( [ eqmot ] ) is then @xmath81 where @xmath82 is the potential in this new frame that we determine in section [ potential ] . we let @xmath83 and @xmath84 . since @xmath85 is small , we consider here only the terms to lowest order in @xmath85 . the equations to order @xmath86 become @xmath87 and @xmath88 in section [ cross ] we consider the higher order terms in @xmath85 that have been neglected here . the terms in equations ( [ xeq ] ) and ( [ yeq ] ) that are functions of @xmath89 and @xmath90 are expressed as the potential gradients @xmath91 and @xmath92 integrating these , we find the potential in the hill approximation to be @xmath93 in polar coordinates centered on the planet , so that @xmath94 and @xmath95 , the potential is given by @xmath96 the first term is a point mass potential of the planet . the other two terms are due to the rotation of the frame and the gravitational effects of the star . in dimensional form , the potential is given by @xmath97 which is valid for @xmath98 . from the @xmath22-component of the equation of motion ( [ eqmot ] ) with @xmath99 , we have that @xmath100 we substitute the axisymmetric terms in the potential of equation ( [ phih ] ) and the velocity in the @xmath101-direction , @xmath102 , to find @xmath103 where @xmath22 is the ( unscaled ) distance from the planet . there are corrections to this result due to nonlinear effects of nonaxisymmetric terms . the angular velocity about the planet , @xmath104 , is less than keplerian because of the effects of the star . we numerically solved equations ( [ xeq ] ) and ( [ yeq ] ) to find the trajectories of particles in this potential . we found closed periodic particle orbits in the frame of the planet , along the lines of @xcite , but in the hill approximation . we chose the initial position for the particle to be on the @xmath105-axis . the initial @xmath106-velocity is determined so that the particle has zero velocity in the @xmath105 direction when it again crosses the @xmath105-axis on the other side of the central planet . we find that the neighboring particle orbits first intersect for the initial position on the @xmath105-axis of @xmath107 ( or @xmath108 ) , as shown in the left panel of fig . [ orbitcross ] . this maximum radius of nonintersecting orbits , @xmath109 , is considered to represent a maximum disc radius for a cold disc , as discussed by @xcite . the orbits in in the left panel fig . [ orbitcross ] are asymmetric ( lopsided ) with respect to the y - axis in that the intersections occur for @xmath110 . on the other hand , the @xmath111 hill potential is symmetric about the y - axis . a more detailed analysis shows that there is another set of orbits for which the intersections occur for @xmath112 . in addition , there is a third set of orbits that remain symmetric about the y - axis and nonintersecting for @xmath113 . however , this set of orbits is dynamically unstable for @xmath113 . that is , there is a bifurcation that occurs at @xmath114 , as seen in the right panel of fig . [ orbitcross ] . beyond this radius , there continue to exist y - symmetric nonintersecting orbits , but they are unstable . there are stable orbits beyond this radius , but they intersect with neighboring orbits . in any case , this critical radius of @xmath115 is the limiting radius for orbits that represent a cold , steady - state disc . to test the validity the hill approximation , we also solved the equations of motion for the full gravitational potential in the frame of the planet . we determined the particle orbits in the same way as described above . we found that the orbits first intersect at a radius of @xmath116 with a mass ratio @xmath117 . the hill approximation is then reasonably accurate for these purposes , even for relatively large mass planets . and @xmath106 coordinates are in units of the hill radius @xmath1 . at a distance of about @xmath118 from the planet , the orbits begin to cross . right : stability parameter ( henon 1965 ) @xmath119 plotted against @xmath120 , the @xmath105 coordinate in units of @xmath1 where the orbit crosses the positive @xmath105-axis . for @xmath119 negative ( positive ) the orbits are dynamically stable ( unstable ) . the dashed line is for periodic orbits that eventually intersect each other at larger @xmath120 , as occurs in the left panel . the solid line tracks the stability of periodic orbits that remain symmetric about the @xmath106-axis for larger @xmath120 . notice that for @xmath121 , where orbits begin to cross in the left panel , there is a bifurcation in stability properties . at larger radii , @xmath106-symmetric orbits remain nonintersecting , but are unstable , while the intersecting orbits are marginally stable . , title="fig:",width=241 ] and @xmath106 coordinates are in units of the hill radius @xmath1 . at a distance of about @xmath118 from the planet , the orbits begin to cross . right : stability parameter ( henon 1965 ) @xmath119 plotted against @xmath120 , the @xmath105 coordinate in units of @xmath1 where the orbit crosses the positive @xmath105-axis . for @xmath119 negative ( positive ) the orbits are dynamically stable ( unstable ) . the dashed line is for periodic orbits that eventually intersect each other at larger @xmath120 , as occurs in the left panel . the solid line tracks the stability of periodic orbits that remain symmetric about the @xmath106-axis for larger @xmath120 . notice that for @xmath121 , where orbits begin to cross in the left panel , there is a bifurcation in stability properties . at larger radii , @xmath106-symmetric orbits remain nonintersecting , but are unstable , while the intersecting orbits are marginally stable . , title="fig:",width=241 ] by including terms of o(@xmath122 ) and o(@xmath85 ) in equations ( [ xeq ] ) and ( [ yeq ] ) , we obtain the dimensionless potential @xmath123 these higher order terms in @xmath124 introduce additional angular dependencies . the steady - state two - dimensional equations of motion for a fluid disc in the corotating frame are @xmath125 and @xmath126 where @xmath127 is the fluid velocity and @xmath128 is the force per unit disc mass that represents the effects of turbulent viscosity . quantity @xmath129 is the two - dimensional ( vertically integrated ) pressure . we consider a cylindrical coordinate system @xmath130 centered on the planet . the azimuthal viscous force due to shear is given by @xmath131 where @xmath132 is the surface density and @xmath133 is the kinematic turbulent viscosity . the potential @xmath66 is given by equation ( [ phif ] ) . we rescale the variables by @xmath124 as we did for the ballistic case in equation ( [ rscaled ] ) . we rescale the velocity and radius as @xmath134 and @xmath135 , respectively , and obtain @xmath136 and @xmath137 where @xmath138 @xmath139 @xmath140 @xmath141 and @xmath142 the potential , @xmath82 , is given by equation ( [ phiex ] ) and is independent of @xmath85 to lowest order , the hill approximation . the rescaled viscous shear force is @xmath143 where we used the @xmath41-prescription for the viscosity so that @xmath144 @xcite . we have left the disc aspect ratio about the planet in the unscaled form @xmath36 because it is invariant to scaling . we have not rescaled the surface density because doing so would have no effect on the velocity field , since we neglect disc self - gravity . notice that @xmath145 is independent of the mass ratio @xmath85 . this is also true of all other components of the viscous force . the rescaled pressure is given by @xmath146 which is independent of @xmath85 . thus , the transformed equations ( [ origas ] ) and ( [ origs ] ) are of the same form as the original equations ( [ origa ] ) and ( [ orig ] ) and are independent of @xmath85 in the hill approximation . this result also holds in an obvious extension to three dimensions . the rescaled circumplanetary disc equations are then very similar to the circumstellar disc equations previously analysed for close binary stars with order unity mass ratio . one difference is the form of the potential , which has a single nonaxisymmetric azimuthal number @xmath111 in the hill approximation . however , this is the dominant tidal term in close binaries . consider a disc in binary that orbits star 1 and is tidally perturbed by star 2 . at a small distance @xmath147 from star 1 , the potential terms due to the star 1 and the @xmath111 tidal component due to star 2 is given by @xmath148 . this potential is similar ( within a factor of 2 ) to the rescaled potential of equation ( [ phi ] ) with @xmath149 , and masses @xmath150 . in particular , the ratio of the @xmath111 tidal to central potentials is the same for both cases . therefore , the disc flow equations for a planet in the hill approximation are similar to those for a binary star system with unity mass ratio , for a given value of @xmath36 and @xmath41 . resonances occur in the disc where the forcing frequency matches a natural frequency in the disc . angular momentum can be transferred to the star from the disc by the tidal torques that are exerted at the lindblad resonances within the disc @xcite . at such resonances , rotationally modified pressure waves are launched . torques are exerted on the disc at radii where the waves damp . resonance torques could play a role in truncating the disc . in this section we consider whether circular , eccentric , or vertical resonances could lie within the disc . we only consider nearly circular orbits in this analysis . we find this approximation holds well out to the radius where orbits begin to cross ( section [ cross ] ) . since we expect the disc to be truncated inside or at this radius , we should find all of the resonances present in the disc in this approximation . however , for a disc as warm as suggested during the t tauri accretion phase , @xmath151 ( equation ( [ hr ] ) ) , some gas might extend beyond the orbit crossing radius and off - resonant forcing could play a role . off - resonant forcing is possible because for @xmath152 of order unity , the resonance width scales as @xmath153 . with such a large width , the resonance could overlap with the disc , even though the exact resonance location does not lie within the main body of the disc , as has been investigated in the binary star case by @xcite . we apply the angular velocity of the disc in the hill approximation given in equation ( [ angvel ] ) to the lindblad resonance condition @xcite . the circular lindblad resonances then occur where @xmath154 in the hill approximation we have only the @xmath111 term of the potential . we find that the only positions inside the orbit crossing radius is for the @xmath155 term which is at @xmath156 . for a cold disc , the resonance width is very small and the tidal forcing is very weak near the disc center . for a warm disc , stronger resonant excitation is possible . at higher order in @xmath124 , other @xmath152-values are present ( equation [ phiex ] ) . however , these higher order resonances also fail to lie inside the orbit crossing radius . if the orbit of the planet is slightly eccentric with eccentricity @xmath157 , the tidal forcing can be decomposed into a series or rigidly rotating potentials at various frequencies @xmath158 for integer @xmath159 @xcite . the eccentric lindblad resonances then occur where @xmath160 for @xmath111 , the lowest order resonance that lies within the orbit crossing radius has @xmath161 . the resonant torque scales as @xmath162 , where @xmath152 is the azimuthal wavenumber of the tidal potential . consequently , the torque would scale as @xmath163 , which suggests that the torque would be quite weak for modest eccentricities . for such a resonance to be able to overcome the effects of disc turbulent viscosity and truncate the disc , we roughly require @xmath164 ( see , * ? ? ? * ) . consequently , the disc viscosity would need to be very small . there is also a set of resonances associated with the vertical disc motions . for each @xmath152 , the vertical disc resonance lies closer to the planet than the corresponding horizontal ( coplanar ) lindblad resonance , although the torque it produces is a factor of @xmath165 smaller than the corresponding horizontal resonance . the vertical resonance generates horizontally propagating waves which travel towards the planet at the centre of the disc . the wave generation can transport angular momentum to the orbital motion of the planet - star system . the vertical resonances occur where @xmath166 where @xmath167 is the effective adiabatic index @xcite . we solve this equation with @xmath168 and find the @xmath111 resonance occurs at @xmath169 , which is somewhat outside the orbit crossing radius . resonances at higher @xmath152-values occur further away from the planet . the analysis in subsequent sections of this paper considers only one- and two - dimensional models that are not capable of finding vertical resonances . we consider the effect of the tidal perturbation on a circumplanetary accretion disc due to the presence of the star by following the work of @xcite . the tidal torque transfers angular momentum from the disc to the binary orbit , allowing material to accrete on to the central object @xcite . the disc attempts to expand by viscous forces , but tidal torques dominate the transport process only in the outer parts of the disc . in this approach , the disc pressure forces are ignored . consequently , this approach does not include the effects of off - resonant waves that could be present . it is more accurate for cool discs , @xmath170 , where the effects of waves are likely less important . such cool circumplanetary discs could arise at late stages of disc evolution where the accretion rates are lower as the disc disperses . we consider linearised equations for the tidal disturbances about the circular motions around the planet and initially include only gravitational and centrifugal forces . the unperturbed gas orbits with velocity @xmath171 in the corotating frame centered on the planet , where @xmath104 is given by equation ( [ angvel ] ) . we consider perturbations to this flow by the presence of the star . we denote the perturbed velocity as @xmath172 , where @xmath173 and @xmath174 . we follow the approach of @xcite , but adopt the corotating frame with the hill approximation . the linearised equations of motion based on equation ( [ orig ] ) are @xmath175 and @xmath176 where @xmath177 the force components are obtained from the negative gradient of potential @xmath178 in equation ( [ phih ] ) . since only @xmath111 disturbances are involved in the hill approximation , we represent @xmath179 and obtain @xmath180 } \label{ur}\ ] ] and @xmath181}. \label{utheta}\ ] ] we now consider the additional dissipation in the disc due to tidal perturbations when we introduce a small amount of dissipation into the flow . the two modes of dissipation in a newtonian fluid are the shear viscosity , @xmath133 , and the bulk viscosity , @xmath182 . the dissipation due to the bulk viscosity per unit radius is @xmath183 which we can evaluate using equations ( [ ur ] ) and ( [ utheta ] ) . the dissipation due to the shear viscosity is @xmath184 where @xmath185 and @xmath186 is the rate of strain tensor which is given by @xmath187 \\ \frac{1}{2}\left [ r \frac{\partial ( u_\theta / r)}{\partial r}+\frac{1}{r}\frac{\partial u_\theta}{\partial \theta}\right ] & \frac{1}{r}\frac{\partial u_\theta}{\partial \theta}+\frac{u_r}{r}\\ \end{array}\right).\end{aligned}\ ] ] from the dissipation we can find the torque on the disc due to the tidal perturbation . the rate of working per unit surface area of the disc by a torque , @xmath18 , is @xmath188,\end{aligned}\ ] ] where @xmath189 . the first term is the rate of convection of energy over the whole disc . its value depends only on the boundary conditions . we find the torque on the disc to be @xmath190 where @xmath191 , @xmath192 , @xmath193 @xcite . the torque @xmath194 is defined as the torque on an accretion disc without a companion . this has a viscous torque of @xmath195 and so the dissipation is @xmath196 in the left panel of fig . [ diss ] we plot the three scaled dissipations , @xmath197 , @xmath198 and @xmath199 as functions of the radius in the disc . this is similar to fig . 1 in @xcite , but we use the hill approximation . we see that the magnitude of the dissipation from the internal viscous torques , @xmath197 , and the tidal dissipation , @xmath199 , are equal very close to , but just inside , the radius in the disc where the particle orbits begin to cross . in the right panel of fig . [ diss ] we plot the scaled torques on the disc . the tidal torque in the disc starts to dominate the viscous torque just inside of the radius where the particle orbits cross . however , outside of the radius where the particle orbits cross , pressure and nonlinear effects can not be neglected . therefore , this linear solution is not valid beyond the orbit crossing radius . we use a one - dimensional ( radius only ) model of an accretion disc subject to a tidal torque in order to determine the surface density evolution . the governing equation for the surface density of a flat accretion disc centered on the planet is @xmath200 ' } \frac{\partial } { \partial r}\left(\nu \sigma r^3 ( -\omega')\right ) -\frac{\sigma r } { [ r^2 \omega ] ' } \frac{d t_{\rm gr}}{d m } \right ] + s(r ) \label{sigmaev}\ ] ] @xcite , where @xmath201 and the angular velocity is given in equation ( [ angvel ] ) . in the previous section , we found the tidal torque on the disc . however , this is only valid up to the position in the disc where the orbits cross . we adopt a torque function per unit disc mass on the disc of the form @xmath202 where @xmath203 is a constant . to model the tidal effects of orbit crossings , we select the torque parameters so that the torque acts in the region where the particle orbits begin to cross . we choose @xmath204 so that the disc is truncated quickly . the form of @xmath205 controls where the disc is truncated . we choose @xmath206 to model the effects of the inflowing circumstellar gas , we model gas injection at some radius @xmath207 over a narrow region of radial width @xmath208 at a steady rate @xmath209 with the local keplerian speed . function @xmath210 describes the mass injection that we take to be @xmath211 where @xmath212 is unity for @xmath213 and zero otherwise . we adopt a width @xmath214 . we solve this equation numerically on a fixed mesh that is uniform @xmath215 with 200 grid points ( like * ? ? ? we choose zero torque boundary conditions at the inner boundary @xmath216 and the outer boundary @xmath217 . the inner boundary allows the material there to be accreted by the planet . its position has been chosen so that it is about equal to the radius of a jupiter planet at a distance of @xmath218 from a central solar mass star . with the tidal torque acting , the outer boundary is far enough out that it does not affect the evolution because the tidal torque prevents the mass from reaching the outer regions . however , if there is no tidal torque , the disc extends as far as it can and the outer boundary will affect the mass in the disc because mass is removed there . the viscous torque is given in equation ( [ visct ] ) and we take @xmath219 at the boundaries to have zero torques there . we initially take the surface density to be a constant but very small value and allow it to build up by mass accretion at a radius @xmath207 . we set the accretion rate on to the disc to be @xmath220 . for a jupiter mass planet orbiting a solar mass star at a radius of @xmath218 this corresponds an accretion rate of @xmath221 . we start with a disc of nearly zero mass . we note that changing the accretion rate does not change the results in fig . [ evolution ] , only the amount of mass in the disc . we parametrise the viscosity with the @xmath41-prescription in equation ( [ alphavisc ] ) . for a keplerian disc with @xmath222 and a constant disc aspect ratio @xmath223 , we have gas kinematic turbulent viscosity @xmath224 disc mass builds up as mass is injected . we ran the numerical code until it reached a steady state . in fig . [ evolution ] we plot the surface density evolution of the disc for different injection radii . the surface density in each of the plots increases as time goes on . the top left plot shows the evolution without the tidal torque and an injection radius @xmath225 . we see that the disc spreads out as far as it can out to the outer boundary , where mass is removed . all the other three plots in fig . [ evolution ] include the tidal torque and we vary the position that mass is added from @xmath226 ( top right ) , @xmath227 ( bottom left ) and @xmath228 ( bottom right ) . with the tidal torque acting , the disc can not expand out to the outer boundary , it becomes truncated well inside that boundary . the position of the injection of the mass does not affect the outer boundary of the disc , it only mildly modifies the surface density profile of the steady state disc . in this section we find steady - state analytical solutions for the surface density of a circumplanetary accretion disc . our approach is similar to that of @xcite and @xcite , but we include the strong tidal torques near the orbit crossing radius . there are three regions in the disc , as shown in fig . [ sketch ] . inside of the radius where mass is added , @xmath207 , the disc acts as a normal accretion disc . outside of this radius , the disc acts as a decretion disc in terms of the density profile , although gas does not actually flow outward , as shown below . instead , this region acts as mass reservoir . however , material in this region is subject to exchange with the interior region @xmath229 by means of turbulent diffusion . a power law torque is applied only in the outer parts of the disc in @xmath230 . we assume for simplicity that the disc is in exact keplerian rotation about the planet with angular speed @xmath104 in the inertial frame . there is an accretion disc and in @xmath231 there is a decretion disc structure that acts as a mass reservoir . in this model , for @xmath232 there is no imposed gravitational torque and for @xmath230 a power law gravitational torque is applied.,width=302 ] the accretion rate through the disc is @xmath233 notice that @xmath234 is negative for accretion ( @xmath235 ) . mass is injected at a radius @xmath236 with rate @xmath237 , that is @xmath238 and the angular momentum equation is @xmath239 where the viscous torque per unit radius ( with equation [ visct ] ) is given by @xmath240 for a keplerian angular velocity where @xmath241 and @xmath242 is the gravitational torque . equation ( [ angularmom ] ) is equivalent to equation ( [ sigmaev ] ) with the time derivative set to zero and injection width @xmath243 set to ( nearly ) zero . in accordance with equation ( [ tgr ] ) , the gravitational torque per unit radius on the disc is taken to be @xmath244 where @xmath205 is defined by equation ( [ t0 ] ) . integrating equation ( [ mass ] ) we find @xmath245 where @xmath246 is a constant . with the boundary condition that @xmath247 at the disc outer edge , we see that @xmath248 . so there is no accretion flow outside of the injection radius , all of the material that is injected at @xmath207 flows inwards to be accreted on to the planet . we now integrate equation ( [ angularmom ] ) to find @xmath249 where @xmath250 is a constant to be determined by the inner boundary condition . as in section [ oned ] , we have chosen the gravitational torque to be zero in @xmath251 . in evaluating equation ( [ acc ] ) , we apply the form of kinematic viscosity given by equation ( [ nuv ] ) , @xmath252 , that holds for a constant @xmath41 and constant @xmath36 circumplanetary disc . with the zero viscous torque inner boundary condition at an inner radius @xmath253 , equation ( [ acc ] ) requires @xmath254 . we then obtain the standard viscous disc surface density in the inner parts of the disc @xmath255 , designated by subscript 1 , @xmath256\ ] ] @xcite . therefore , @xmath257 , for @xmath258 . in the other two regions of space where @xmath259 , we assume that @xmath260 , so that @xmath250 is very small compared to the other torques . for simplicity , we take @xmath261 in these regions . in the region outside of where the mass is injected , but where no gravitational torque is applied , @xmath262 , the surface density is @xmath263 in the outermost region , @xmath264 , the surface density is @xmath265 , \label{sig3}\ ] ] where @xmath266 and @xmath267 . we now have the surface density for the whole disc with the equations for @xmath268 , @xmath269 and @xmath270 . we compared these solutions to the one - dimensional cases of section [ oned ] with the same values for the input parameters . these steady state analytic solutions agree well ( better than 6% ) with the numerical solutions obtained at late times , plotted as the highest lines in the plots in fig . [ evolution ] . @xcite have also considered the flow outside some radius where mass is injected , but did not include the effects of the tidal field . they have a normal accretion disc in the inner parts , as we have . but in the outer parts of the disc in their model , the material flows outwards and is removed from the disc at a rate @xmath271 where @xmath272 is the outer edge of disc . for their standard model with @xmath273 and @xmath274 ( for jupiter radius @xmath275 ) , we find the mass loss rate is @xmath276 . if we take the disc radius to be the orbit crossing radius , the mass loss rate is about @xmath277 . they suggest that the removal of matter at the outer edge of the disc occurs through solar torques or collisions with highly shocked regions . however , in our model the tidal torque dominates the viscous torques in the outermost regions and prevents the outward flow of matter . hence , in our picture of a steady state , the rate that mass is accreted on to the central planet is the same rate that it is injected into the disc . in addition , there is a difference in the density profiles between the models outside the mass injection radius . in that region , @xcite determined a gradual tapering that varies in radius as @xmath278 for disk radius @xmath272 . in our model , the density falls off more slowly in radius in the main body of the disc and is abruptly terminated near the orbit crossing radius . the difference in the density structures can be traced to outer boundary conditions . @xcite employed a zero viscous stress outer boundary condition , @xmath279 . in our model , the effects of the tidal field can be approximately represented by an outer boundary condition at the tidal truncation radius ( orbit crossing radius ) that is given by @xmath280 as follows from the second relation of equation ( [ acc ] ) with @xmath261 . [ dcomp ] compares the density distributions resulting from the two boundary conditions . the density distribution we obtain approaches the @xcite distribution in the limit that the injection radius goes to zero . there are further modifications to the density profile due to pressure effects that are described in section [ concs ] . from equation ( [ sig3 ] ) , the width of the disc outer edge ( ignoring pressure effects ) is estimated as @xmath281 where the right - hand side is evaluated near the outer edge . if the disc flow is smooth and there are no resonances , the phase lag that produces a tidal torque due to the turbulent viscosity is estimated as @xmath282 , the inverse reynolds number of the flow . the resulting torque per unit disc mass is then estimated as @xmath283 , where @xmath284 is the @xmath111 component of the tidal potential . using equation ( [ w ] ) and assuming @xmath285 , it follows that @xmath286 . therefore , this torque is not capable of truncating the disc . instead , the much stronger tidal torque , due to orbit crossings and certain resonance effects discussed in section [ concs ] , truncates the disc in a small region of space . torques from the region well inside of where orbit crossings occur are then too weak to truncate the disc . the analytic solutions can also be considered to apply when the matter being injected into the disc has a nonkeplerian velocity , that is @xmath287 at @xmath207 is nonkeplerian . the gas in such a case adds angular momentum per unit mass to the system at a rate that differs from the keplerian rate . in reality , such gas would undergo a strong kelvin helzholtz instability with the disc . since the gas being added to the disc has a much lower density than the disc , we would expect this gas to be rapidly entrained by the disc that remains in nearly keplerian rotation . an example of such a situation occurs when coplanar inflowing gas meets the circumplanetary disc at the disc outer edge . we expect the inflowing gas in this case to be subkeplerian . in such a case , the decretion disc or mass reservoir region , as depicted in fig . [ sketch ] is very small . instead , nearly entire disc behaves an accretion disc . the subkeperian injected gas can be considered to impart a negative torque on an otherwise keplerian disc , in addition to the tidal torque . on the other hand , if much of the inflowing gas flows over the disc before becoming entrained by the disc , then the injection radius will be smaller than the disc outer radius , as in the cases plotted in fig . [ evolution ] . beyond the details of the flow properties in particular models , the results suggest that the disc structure is generally largely independent of the angular momentum per unit mass of the inflowing gas , as seen in figs . [ evolution ] and [ jc ] . the disc density distribution is mildly influenced by this quantity . the radial derivative of the surface density undergoes order unity changes at @xmath207 . in the model considered this section , the logarithmic surface density gradient changes from -1/2 to -1 across @xmath207 . also , for a fixed mass injection rate , the density in the outer parts of the disc scales as @xmath288 . but such variations do not lead to strong changes in disc structure by typically plausible values of @xmath289 . this is particularly true if the injection radius is near the disc outer edge . instead , the disc structure interior to the disc tidal truncation region is dominated by the effects of disc turbulent viscosity . to include the effects of gas pressure , we modeled the disc using a two - dimensional sph code with @xmath290 particles . the sph parameter @xmath291 was set to unity in these simulations , while @xmath292 was set to zero . the disc sound speed was crudely modeled as a constant equal to @xmath293 throughout . the disc aspect ratio was @xmath294 at @xmath295 ( in dimensionless units of these equations @xmath296 ) , where free particle orbit crossings occur . the fluid equations were taken in the hill approximation , following the force equation ( [ origs ] ) . the particles were initially distributed so that surface density was initially @xmath297 in an annulus @xmath298 . we chose @xmath299 and have two models , one with @xmath300 and the second with @xmath301 which we plot in fig . [ initial ] . the initial velocities were taken to be circular rotation having angular speed @xmath302 in the corotating frame , with @xmath104 given by equation ( [ angvel ] ) . particles are removed from the simulation if they reach the inner boundary at @xmath303 or the outer boundary located at @xmath304 . for each particle removed , a particle is injected at a random angle and random radius between @xmath305 and @xmath306 . therefore , the number of particles is fixed at @xmath290 at all times . fig . [ jevolution ] plots the angular momentum evolution as a function of time for the two initial disc sizes . the smaller disc starts with much less angular momentum than the larger one . but after only about 3 planetary orbits both discs have similar values of angular momentum and approach a steady state . in fig . [ final ] we plot the particles at a time of 6 planet orbital periods . we see that particle distributions look very similar and the disc has reached a near steady state . the viscous timescale of the disc can be estimated as @xmath307 . in sph , we have that @xmath308 ( e.g. , * ? ? ? the value of the kinematic viscosity @xmath133 is estimated as @xmath309 and the viscous timescale evaluates to about 8 orbits at @xmath310 . the simulated disc has therefore settled to a near equilibrium state on a timescale that is of order the estimated viscous timescale . this kinematic viscosity is a factor of ten or more larger than what is typically taken taken in circumplanetary disc simulations ( e.g. , * ? ? ? * ; * ? ? ? consequently , in those simulations , the timescales to reach a steady state are longer than the case here by similar factor . ( in addition , previous simulations have generally not started with a circumplanetary disc . some time is required for its formation from inflowing gas . ) for the simulations in @xcite , the timescale would be determined by the inherent viscosity in the code due to the finite differencing , since no viscous terms were included . the phasing of the arms determines the sign of the gravitational torque in the different regions of space as shown in fig . [ phase ] . the left plot shows the cumulative gravitational torque as a function of radius in a steady state disc after 6 planetary orbits . on the right is a plot of the shape of the spiral in the disc . the sign of the gravitational torque is equal to sign of @xmath311 and so is negative in the first and third quadrants the torque and positive in the second and fourth . as seen in the left plot , the gravitational torque is negative in the outermost region of the disc where it is strongest . the spiral there lies in the first and third quadrants , as seen in fig . [ final ] , in order the provide a net negative gravitational torque that removes angular momentum from the disc as gas accretes . the left plot in fig . [ phase ] also shows the cumulative disc mass distribution . more than 90% of the disc mass is located inside a radius of @xmath312 , where orbits crossings of free particles occur . the local gravitational torque is negative for @xmath313 . on the other hand , the cumulative gravitational torque on the disc inside the orbit crossing radius is actually positive . this does not mean that the gas gains angular momentum in this region . the viscous torques compensate by transporting the angular momentum out of this region so that the overall torque is negative . the negative cumulative gravitational disc torque is achieved somewhat outside the orbit crossing radius . a small amount of disc mass , less than 10% , that is located outside that radius provides the net negative torque on the disc . in that region of space , @xmath314 , the spiral arms are very prominent in figure [ final ] and gas response is highly nonlinear . a similar spiral structure was found in the case of a warm disc with @xmath315 within a binary star system by @xcite that we have argued should be similar to the circumplanetary disc case ( see section [ eqs ] ) . the sph code is less accurate in these outer low density regions because the interparticle pressure force calculation is more approximate . also , we have simplified the inflow on to the disc to be occurring within an annulus in the disc . but , we expect that the general properties of gravitational torque to generally hold . that is , the gravitational torque involves a relatively small amount of gas in the outer parts of disc . we have analysed the dynamics of a circumplanetary disc as an accretion disc subject to the tidal forces of the central star . we applied several techniques : ballistic particle orbits , one - dimensional simulations and analytic models , and two - dimensional sph simulations . we have shown that the gas dynamical equations for a circumplanetary disc can be rescaled to a form that is similar to those previously analysed for a disc in a binary star system with mass ratio of unity , but with a modified ( hill ) potential . the size of a circumplanetary disc is determined by the requirement that tidal torques from the central star remove angular momentum at a rate required by steady inflow on to the planet . in the crudest approximation that is accurate for a cold disc , disc streamlines consist of a set of stable , nested periodic ballistic particle orbits ( see fig . [ orbitcross ] ) . for a cold disk , the disc radius is determined to be the location of orbit intersections or instability that occurs at @xmath312 , for hill radius @xmath1 . this radius is close to the effective disc radius ( inside of which @xmath316 90% of the mass is located ) found in the sph simulations for a warm disc ( see figs . [ final ] and [ phase ] ) . in principle , the observational determination of the size of this disc then provides a constraint on the planet mass . in some previous disc models , the angular momentum of the inflowing gas was thought to control the disc size . according to these models , the compactness of the giant planet satellite systems , within a region of radius much less than @xmath1 , is a consequence of a low angular momentum inflow that penetrates well inside the hill sphere and deposits presatellite material there ( e.g. * ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? we suggest a different physical model for the processes that control the circumplanetary disc size and structure . we find that although the angular momentum per unit disc mass ( specific angular momentum ) of the inflowing gas plays a role in determining whether a circumplanetary disc can form , it has a minor influence on the disc structure . a steady state accretion disc does not achieve as compact a form as desired by these previous models . these results imply that the common scale of the disc seen in the simulations by @xcite and @xcite is not due to the specific angular momentum of the inflow , but instead to the tidal truncation effects . the angular momentum within the disc is redistributed by viscous torques . the disc structure inside the truncation radius is mainly determined by the properties of the disc turbulent viscosity , rather than the specific angular momentum of the inflowing gas ( see uppermost curves in fig . [ evolution ] ) . the average specific angular momentum in the disc is nearly independent of the specific angular momentum of the accreting gas ( see fig . tidal effects have a large - scale influence on the disc density distribution , as discussed in section [ analytic ] . we have concentrated on the case that the planet opens a gap in the disc , as is expected for a jupiter mass planet . some previous studies suggested that a small disc of size @xmath317 could result prior to and during early stages of gap opening ( e.g. * ? ? ? * ) . our results suggest that a turbulent accretion disc would be truncated at the orbit crossing radius , even if the inflowing gas has low specific angular momentum . there are are few qualifications in this statement . the simulations of @xcite found that circumplanetary discs do not exist prior to gap opening ( planet masses less than 100 earth masses ) . if low angular momentum gas is accreted in early stages of gap opening , then pressure forces from within the hill sphere could prevent a disc from extending out to the orbit crossing radius , if the planet s bondi radius is sufficiently smaller than its hill radius . under such conditions , we would expect the disc to expand to the bondi radius by viscous torques , since tidal torques would not be adequate to truncate the disc inside the bondi radius . the disc would then need to lose its angular momentum to the surrounding gas in the circumstellar disc . whether such a configuration is possible is unclear . in addition , gap opening or planet envelope contraction ( below its hill sphere ) might not begin when the bondi radius is smaller than the hill radius and so a disc may not form under such conditions . during the t tauri accretion phase , circumplanetary disc aspect ratios are expected to be large , @xmath151 , as a consequence of the weak gravitational forces due to the planet and relatively high temperatures ( see equation ( [ hr ] ) ) . there are some implications of these high disc aspect ratios . one is that the disc edge tapering is more gradual than in the thin disc case . the tapering must be on a scale @xmath318 in order to avoid instabilities that occur for sharper edges ( e.g. * ? ? ? the tapering can then have an influence on a substantial portion of the circumplanetary disc . @xcite estimated the disc outer radius based on where the angular momentum per unit disc mass departs from keplerian and begins to decline in radius . they obtained a disc outer radius of @xmath319 ( see the upper right panel of their fig . this radius is somewhat inside the orbit crossing radius of @xmath320 . the effect of the disc tapering on the distribution of disc angular momentum in our sph simulations can be seen in fig . it leads to subkeplerian velocities inside the orbit crossing radius , similar to the results of @xcite . the results of these two studies can be further reconciled by considering the differences in the models . in their simulations , the disc is hotter than in our simulations . consequently the pressure forces and their effects on departures from keplerian rotation in the outer parts of the disc should be stronger than in fig . [ vphi ] , as is consistent with their results . also the work of @xcite includes the inflowing gas from larger radii that is omitted in our simulations . this inflowing subkeplerian gas and its interaction with the disc outer edge as it becomes entrained could further modify the results plotted in fig . it is also possible that the disc radius is affected by the location of wave damping , as discussed below . such effects should be explored further . another implication of the high values of @xmath36 is that a small fraction of the disc mass can reside at larger disc radii than the orbit - crossing radius and so be subject to the effects of resonances that lie outside this radius , as discussed in section [ res ] ( see fig . [ phase ] ) . in addition , off - resonant tidal forcing of the outer parts of the disc that lie inside the orbit crossing radius can play an important role . the resonance width depends on @xmath321 and is not small . the substantial width can allow the resonant region to overlap with the denser parts of disc . as a result of these two effects , two - armed spiral waves can be launched in such discs , even though exact resonances do not lie within the main body of the discs . we have found evidence of such waves in our sph simulations ( see fig . [ final ] ) . for these waves to play a role in extracting angular momentum from the disc , they must damp in order to introduce irreversibility . otherwise , a standing wave is produced that results in little or no torque on the disc . some damping can be produced by the disc turbulent viscosity in the fractional amount that is roughly the ratio of the viscous wave damping rate to the wave group propagation rate , about @xmath322 . this ratio can be quite small for @xmath323 and the torque is reduced to a similarly small fraction of its potentially maximum value . for a thin disc , this damping could occur as launched waves propagate , steepen and shock , since their wavelengths are short compared to the disc radius . but for warm circumplanetary discs , this process is less important because the wavelengths are not short compared to the disc radius . in the sph simulations of section 7 , the damping seems to occur from the strongly nonlinear forcing in the outer disc , similar to the case of mildly warm discs ( @xmath324 previously investigated for binary star systems @xcite . the sph simulations suggest that the negative torque is produced in the outer parts of the disc , somewhat inside and beyond the radius where nested particle orbits cross or become unstable . this torque involves a relatively small amount of disc mass ( see fig . [ phase ] ) . however , the discs in our sph simulations were two - dimensional and not as warm as could occur , where @xmath151 . the wave damping issue should be explored in future work . the accretion disc model does not provide an obvious explanation for the locations of the regular satellites of jupiter and saturn that occur within @xmath4 of the planet . as discussed above , the disc structure is insensitive to the angular momentum of the inflowing gas and satellites lie well inside the tidal truncation radius of the disc . this conclusion is consistent with the results of recent three - dimensional simulations of @xcite and @xcite . we have considered circumplanetary discs with a smoothly varying turbulent viscosity with radius . but such discs could also harbor dead zones , as discussed in section [ properties ] . such situations could result in rapid disc density variations at their boundaries . the existence of satellites requires survival against the effects of migration . such density variations could affect satellite migration and possibly trapping satellites , since migration rates depend on density gradients ( e.g. , * ? ? ? but , if the inner dead zone boundary is due to thermal ionisation at a temperature of @xmath325 , then the disc temperatures there would be too hot for the survival of an icy satellite . of course , other sources of ionisation could change the locations of dead zone edges . in addition , the nonlinear feedback from waves in a low viscosity disc could slow migration @xcite . however , this slowing is less effective for the warmer circumplanetary discs . if the satellite formation occurs after partial disc depletion , as suggested by canup & ward ( 2002 ) , then the ionisation through the vertical extent of the disc becomes easier and the dead zones are less likely to occur at that stage . circumplanetary discs are not very bright as we see in equation ( [ lum ] ) . time dependent accretion and outbursts could occur in dead zones , as has been suggested to explain fu orionis outbursts in young stellar systems @xcite . during a circumplanetary disc outburst , the ratio of the circumplanetary to circumstellar luminosities could be much higher . we thank matthew bate , gennaro dangelo , and jim pringle for helpful discussions . rgm thanks the space telescope science institute for a giacconi fellowship . shl acknowledges support from nasa grant nnx07ai72 g .

weak lensing by large scale structure induces correlated ellipticities in the images of distant galaxies . the two - point correlation is determined by the matter power spectrum along the line of sight . we use the fully nonlinear evolution of the power spectrum to compute the predicted ellipticity correlation . we present results for different measures of the second moment for angular scales , and for alternative normalizations of the power spectrum , in order to explore the best strategy for constraining the cosmological parameters . normalizing to observed cluster abundance the rms amplitude of ellipticity within a radius is , almost independent of the cosmological model , with being the median redshift of background galaxies . nonlinear effects in the evolution of the power spectrum significantly enhance the ellipticity for for the rms ellipticity is , which is nearly twice as large as the linear prediction . this enhancement means that the signal to noise for the ellipticity is only weakly increasing with angle for , unlike the expectation from linear theory that signal to noise is strongly peaked on degree scales . the scaling with cosmological parameters also changes due to nonlinear effects . by measuring the correlations on small ( nonlinear ) and large ( linear ) angular scales , different cosmological parameters can be independently constrained to obtain a model independent estimate of both power spectrum amplitude and matter density . nonlinear effects also modify the probability distribution of the ellipticity . using second order perturbation theory we find that over most of the range of interest there are significant deviations from a normal distribution . # 1 # 1

astro-ph9611077

mapping the large - scale structure ( lss ) of the universe is one of the major goals of observational cosmology . traditionally this is performed using large surveys of galaxies , either the projected 2-dimensional distributions or the 3-dimensional surveys in redshift space . the main shortcoming of the galaxy surveys is that they trace light , while most of the matter appears to be dark . one therefore needs to translate the galaxy power spectrum into the matter power spectrum and in order to do so one has to make some assumptions on the nature of galaxy biasing . even in the simplest model this can be achieved only up to an unknown biasing parameter @xmath9 , which at present can not be theoretically estimated . while sophisticated n - body and hydro - dynamical simulations will eventually provide some answers to this question , at present the biasing relation between the light and matter remains rather poorly understood and prevents one from drawing definitive conclusions on the amplitude and distribution of mass fluctuations in the universe from galaxy survey data . it is clearly important to seek ways to estimate the large - scale structure that are insensitive to biasing . several observable tracers have been proposed that probe directly the underlying mass distribution : cosmic microwave background ( cmb ) anisotropies , gravitational lensing , peculiar velocity flows and abundances of massive nonlinear objects . this paper focuses on tracing the dark matter with gravitational lensing , in particular the effects of weak lensing by large scale structure on background galaxies . weak lensing magnifies and shears the images of distant galaxies . the shear induces an ellipticity in the image of an intrinsically circular galaxy . background galaxies are of course not circular , but by averaging over the observed ellipticities of a large number of galaxies , the induced ellipticity can be measured , and related to the mass fluctuations along the line of sight and to the spatial geometry of the universe . ellipticities of distant background galaxies averaged over several arcminute windows are sensitive to the mass power spectrum on scales of @xmath10 mpc . for a given spectrum of mass fluctuations , it is sensitive to the cosmological parameters @xmath8 and @xmath11 . thus while strong lensing which leads to multiple images probes non - typical regions of the universe which contain massive halos , weak lensing provides a different and a more direct measure of the mass fluctuations on large scales . the first calculations of the shear signal due to weak lensing that used modern models for the large scale structure power spectrum were those of @xcite and @xcite , based on the pioneering work by gunn ( 1967 ) . our work generalizes the results of these authors to include the effects of nonlinear evolution of the matter fluctuations for flat as well as open and @xmath12dominated cosmologies . villumsen ( 1996 ) has considered some aspects of the linear calculation for open models . very recently , bernardeau et al . ( 1996 ) , kaiser ( 1996 ) and stebbins ( 1996 ) have also made the linear calculation for different cosmologies . our work extends the results of the above authors by using the nonlinear power spectrum , which makes a significant difference on small angular scales of @xmath13 . for the nonlinear calculation we have used the prescription of @xcite as implemented by @xcite and @xcite which provide accurate power spectra for different models , valid from the linear to the strongly nonlinear regime . one can take two different approaches in interpreting a possible measurement of the shear signal . the first is to work within the framework of a physical model for the dark matter and background cosmology . such a model is best normalized to cobe and then its small scale predictions can be compared to the observational constraints of e.g. cluster or damped ly-@xmath14 system abundances , strong lensing statistics or peculiar velocity flows , or in our case the shear amplitude on a given angular scale . another possibility , which is less model dependent , is to compare the constraints from different tracers on the same physical scale . this way one can test the gravitational instability assumption that both tracers probe the same underlying power spectrum and place constraints on the cosmological parameters which scale differently with the tracers . at the present the cmb data constrain the power spectrum only on very large scales , where there are no available data from other tracers . while several current cmb experiments are approaching the scales probed by other tracers , cosmological parameters such as baryon density , reionization epoch and hubble constant become important and complicate the power spectrum reconstruction . the test mentioned above is possible between the cluster abundances ( or strong lensing statistics of multiple images at large separation , both of which trace essentially the same property ) and the peculiar velocities . unfortunately this method can not give model independent constraints on cosmological parameters , because the two tracers scale roughly equally with the matter density @xmath8 ( peebles 1980 , white , efstathiou & frenk 1993 ) . a similar comparison can also be made between weak lensing observations and the tracers discussed above . for example , the shear amplitude on @xmath1 angular scale and rich cluster abundance both probe power spectrum scales around @xmath15 mpc , so the ratio of the two is roughly independent of the shape of the power spectrum . one could hope that using such comparisons would lead to an estimate of the mean density of the universe , because the two tracers scale differently with it . unfortunately as shown in this paper this is not the case so that this is not a promising method of obtaining the mean density . nevertheless , weak lensing can also give important constraints on the power spectrum both on smaller and larger scales , where other tracers give less stringent constraints . on the smaller angular scales , our calculations provide the enhancement in the amplitude and the change in the scalings with @xmath16 and @xmath8 that arise due to nonlinear evolution . once reliable detections of the shear at different angular scales are available , these results can be used to constrain cosmological parameters and the matter power spectrum . the principal advantage of working with weak lensing is that no assumptions about the formation of observable structures need to be made to connect the mass power spectrum to the observed shear . the only unknown parameter is the redshift distribution of source galaxies , and in principle this can also be determined observationally . in 2 and the appendix , the formalism for the weak lensing calculation is presented following seljak ( 1995 , 1996 ) , who generalized the work by @xcite and @xcite to a non - flat universe and non - linear regime . the derivation presented here is complementary to recent derivations by bernardeau et al . ( 1996 ) and kaiser ( 1996 ) . the linear and nonlinear evolution of the power spectrum is introduced in 2.1 , and the resulting dependence on cosmological parameters for power law spectra is obtained in 2.2 . in 3 realistic cdm - like spectra are used to predict the rms shear . we provide accurate power law fits for the dependence of the shear on the parameters @xmath17 , @xmath3 , @xmath16 and @xmath8 for different cosmological models . 3.1 provides predictions of the shear for cobe and cluster - abundance normalized spectra for the range of angular scales which are being probed by current and forthcoming observations . the effects of non - gaussianity in the distribution of shear are considered in 4 where we compute the skewness of the distribution . this is followed by a discussion and conclusions in 5 . gravitational lensing shears and magnifies the images of distant galaxies . the relation of the shear to perturbations in the gravitational potential along the line of sight is developed in the appendix . here we shall use the formalism presented in the appendix to derive expressions for 3 different measures of the second moment of the shear . the observable mean ellipticity of galaxy images can be simply defined in the approximation that the images are ellipses with complex eccentricity @xmath18 given by the axes lengths @xmath9 and @xmath19 as : @xmath20 , where @xmath21 is the position angle of the major axis . in the limit of weak distortions , the eccentricity is the same as ellipticity @xmath22 . while individual galaxies have intrinsic ellipticities which can not be separated from gravitational stretching , we have assumed here that averaging over several galaxies leaves only the gravitational component which can in principle be measured . this mean ellipticity @xmath23 will be denoted as @xmath24 and is defined by equation ( [ shear3 ] ) in the appendix . we shall use @xmath25 to denote its amplitude in what follows . information on the ellipticity is contained in the trace - free part of the shear tensor , which is obtained by integrating over all the deflectors between us and the galaxy and over the distribution of background galaxies @xmath26 ( see appendix ) , @xmath27 here @xmath28 is the gravitational potential and @xmath29 is the radial comoving distance with @xmath30 being the horizon distance . we also introduced the comoving angular distance @xmath31 defined in equation ( [ rchi ] ) . the radial distribution of background galaxies is described with the window @xmath32 . if all the galaxies are assumed to lie at the same redshift @xmath3 , corresponding to @xmath33 , then @xmath34 , and @xmath35 takes the simple form , @xmath36 there are several two - point statistics that can quantify the induced ellipticites in galaxy images . the easiest to obtain from observations is the rms ellipticity @xmath37 , where the overbar indicates the average within a circular aperture of radius @xmath17 . equation ( [ shear ] ) can be used to express @xmath38 in terms of the power spectrum of density perturbations @xmath39 . following the derivation presented in the appendix we obtain for the ellipticity variance @xmath40d\chi , \label{p}\end{aligned}\ ] ] where @xmath41 with @xmath42 being the bessel function of first order and @xmath19 is the expansion factor normalized to unity today . a related quantity is the two - point polarization correlation function @xmath43 , which is given by a similar expression to the one above , replacing the term @xmath44 with @xmath45 : @xmath46\ , d\chi . \label{cpp}\ ] ] the fourier transform of @xmath47 is the angular ellipticity power spectrum @xmath48 , which is often the optimal statistic to use ( kaiser 1992 ) . here we use the quantity @xmath49 , which gives the contribution to the variance per log - interval in @xmath50 and is given by @xmath51 this gives an estimate of correlations at an angle @xmath52 . no assumption on the matter power spectrum has been made in the above expressions and one can use them both in the linear and nonlinear regime . the nonlinear spectrum is in general a non - separable function of wavenumber and redshift , thus leading to a complicated coupling of the dependences on the distance factors and the growth of perturbations . it also leads to a nonlinear dependence of the predicted ellipticity correlations on the shape and amplitude of the initial power spectrum . thus the amplitude of the ellipticity correlations as well as their dependence on cosmological parameters can change in the nonlinear regime ( i.e. , on small angular scales ) . in the rest of the paper we shall explore in detail the outcome of these effects of nonlinear evolution . if the gravitational potential changes in time its power spectrum will depend on the radial distance @xmath29 . in linear theory this dependence is independent of @xmath53 and can be written in terms of the potential growth factor @xmath54 as @xmath55 , which gives for the density power spectrum @xmath56(\chi)p_\delta(k)$ ] , where @xmath57 is the linear growth factor for the density . this can be approximated as ( lahav et al . 1991 ) @xmath58 we have ignored the weak @xmath11 dependence of the logarithmic growth factor . for @xmath59 , this expression simplifies to @xmath60 for a flat @xmath61 model the gravitational potential does not change in time in the linear regime and the region where @xmath35 peaks dominates the radial integral in equations ( [ p]-[ps ] ) . typically this is at half the mean comoving distance to the galaxies , so if background galaxies lie at @xmath62 typical deflectors lie at @xmath63 . in low @xmath8 models the gravitational potential increases with @xmath29 , differently for open and for cosmological constant models . in the small @xmath64 limit the growth factor only depends on @xmath8 ( villumsen 1996 ) . on scales where @xmath65 approaches or exceeds unity nonlinear evolution of the power spectrum becomes important . this is particularly important if one is discussing low @xmath8 models , where both cluster abundances and peculiar velocity normalizations give higher density normalization on scales of interest . in the quasilinear regime the nonlinear spectrum can be computed using perturbation theory . an alternative semi - analytic approach which is accurate from the linear to the strongly nonlinear regime ( up to density contrasts @xmath66 ) was proposed by hamilton et al . ( 1991 ) and further developed in subsequent work ( e.g. , nityananda & padmanabhan 1994 ; peacock & dodds 1996 ; jain , mo & white 1995 ; padmanabhan et al . 1996 ) . this prescription involves mapping the nonlinear spectrum at a wavenumber @xmath53 to the linear spectrum at a unique wavenumber @xmath67 which is given by a spherical collapse model . this leads to a mapping of the form @xmath68 $ ] , where the dimensionless power @xmath69 is defined above , and @xmath70 is a function which varies in a simple way for different initial spectra . the linear wavenumber @xmath71 is given by @xmath72^{-1/3}$ ] . the functional form of the mapping is thus specified by the function @xmath70 , which has been calibrated using high resolution n - body simulations and is accurate for a wide range of initial spectra . for the predictions of the rms ellipticity in 3 we shall use the fitting formulae of @xcite for @xmath61 and of @xcite for the open and @xmath12models to describe the nonlinear power spectrum as a function of wavenumber and redshift . we refer the reader to the above references for details of the formulae and their implementation . while the formulae can differ by a few tens of percent for certain spectra , we find that our results for the ellipticity are not affected by more than a few percent as the ellipticity integrates the power spectrum over a range of wavenumbers and redshifts . in particular the results for the cdm - like models shown in figures 4 - 9 would be completely unaffected by our choice of fitting formulae . in the next section we shall use the nonlinear evolution of cdm - like spectra to predict the rms ellipticity signal . here we consider the simplified case of power law initial spectra @xmath73 to compute the scalings of @xmath47 and @xmath74 with source redshift @xmath3 , and the cosmological model parameters @xmath16 and @xmath8 . the parameter @xmath16 is the rms fluctuation in the mass on scales of @xmath15mpc , and serves to normalize the power spectrum . we use the linear evolution of the power spectrum for which some of the scalings can be obtained analytically . we also consider the qualitative modification due to nonlinear evolution by using the stable clustering regime which gives the maximal effect of nonlinearities . the length scales that contribute to the ellipticity are of order @xmath15 mpc for @xmath75 . on these scales most realistic spectra have a spectral index @xmath76 . on smaller angular scales , the contribution to the ellipticity comes from higher wavenumbers which sample a steeper part of the spectrum ( lower @xmath77 ) . we therefore choose power law spectra with @xmath78 and @xmath79 for comparison with realistic spectra on angular scales of interest . for a power law spectrum the @xmath80integral for @xmath81 is analytic , and gives : @xmath82 valid for @xmath83 . the constant @xmath84 gives the normalization of the power spectrum . aside from the numerical prefactors , the result for @xmath74 is the same and involves the same @xmath29 integral . for simplicity we shall take all the source galaxies to be at the same redshift this is a good approximation as the result is insensitive to the source distribution for a given median source redshift . the window function @xmath35 then takes the form given in equation ( [ delta ] ) . for @xmath61 , @xmath85 , and the @xmath29 integral is analytic as well giving the following power law dependence : @xmath86 using @xmath87 , the dependence of @xmath81 on the source redshift @xmath3 can be obtained from the above equation . for @xmath88 and @xmath89 , the result for @xmath81 is not analytic except in the limit of @xmath90 which is not of interest for realistic situations . it is simple to understand qualitatively the two physical effects that enter . @xmath91 for a given redshift , the distance @xmath29 increases as @xmath8 decreases and @xmath11 increases . therefore there is more path length in the line of sight integral as @xmath33 is larger for a given @xmath3 . the factor @xmath35 involves factors of the comoving angular diameter distance @xmath92 and @xmath93 ; it also increases with decreasing @xmath8 and increasing @xmath11 . ( ii ) the second factor is the linear growth factor @xmath54 . growth of structure slows down at low redshift in an open universe , and to a lesser extent in a @xmath12 dominated universe . therefore , at a given redshift the linear growth factor , normalized to unity today , increases as @xmath8 decreases , and for a given @xmath8 it decreases as @xmath11 increases . the effect of both the distance and growth factors is to increase the contribution from the @xmath29 integral in equation ( [ powerlaw ] ) relative to the einstein - de sitter case . the net contribution increases as @xmath8 decreases , and for a given @xmath8 , as @xmath11 increases . in the case of @xmath94 , the distance factors dominate over the linear growth factor leading to a net increase over the case with the same @xmath8 . however the enhancement due to the integral over @xmath29 has to contend with the factor of @xmath95 outside the integral which is considerably larger . the net result is shown in figure [ fig1 ] for source galaxies at @xmath96 ( upper panel ) and @xmath97 ( lower panel ) . the y - axis shows @xmath95 times the @xmath29-integrand of equation ( [ powerlaw ] ) . thus @xmath81 or @xmath98 is proportional to the area under the curves shown , aside from numerical factors which depend on the shape of the spectrum . the figure shows that the peak contribution for a reasonable choice of @xmath3 comes from the range @xmath99 . the scaling of @xmath38 ( the same as that of the square root of @xmath81 ) with @xmath8 is dependent both on @xmath3 and the shape of the power spectrum . for @xmath100 the dependence is well fit by the power law , @xmath101 for @xmath102 and @xmath103 , the above equation is quite accurate . for @xmath97 the scaling depends on the spectrum , and can be approximated as , @xmath104 the result for @xmath79 is valid provided a low-@xmath53 cutoff is imposed to keep the integral finite . as expected , at higher source redshifts the enhancement due to the distance and growth factors is larger and leads to a weaker @xmath8 dependence . the same occurs as @xmath77 decreases , as then there is more power on large scales , and therefore more weight to the high-@xmath64 part of the integral . the results of equations ( [ scaling1 ] ) and ( [ scaling2 ] ) are valid for @xmath105 , and are intended to cover the range of source redshift and spectra spanned by realistic models . a simple way to understand the dependence on @xmath8 follows from the fact that most of the lensing contribution comes from about half the distance to the source galaxies . thus the @xmath8 dependence of equations ( [ scaling1 ] ) and ( [ scaling2 ] ) can be compared to that of @xmath8 at the redshift @xmath106 , corresponding to @xmath107 . @xmath108 is well approximated by @xmath109 for @xmath96 and by @xmath110 for @xmath97 if @xmath79 . for a shallower slope there is more power on small scales , which are closer for a given angle . hence the results are closer to those of equations ( [ scaling1 ] ) and ( [ scaling2 ] ) for @xmath78 if one uses @xmath111 ( it makes very little difference if one uses @xmath31 instead of @xmath29 for this purpose ) . thus as a first approximation the @xmath8 dependence can be obtained by replacing the @xmath8 term in @xmath81 for the flat case , by the value of @xmath8 at @xmath112 to @xmath113 the distance to the source galaxies . the above scalings are derived using linear evolution of the power spectrum . on smaller angular scales nonlinear evolution could play an important role in the next section we compute the change in the above scalings due to nonlinear effects . it is complicated to estimate the angular scale where nonlinear effects are important because each angle involves the projection of a range of length scales at different redshifts . the answer will therefore depend on the slope of the power spectrum . a rough estimate of the length scales that make the dominant contribution to @xmath38 is given by figure [ dist ] . this figure shows the transverse distance for a given @xmath17 at the redshifts estimated from figures [ fig1]a and [ fig1]b as providing the dominant contribution to @xmath114 for source galaxies at @xmath115 . for @xmath13 this length scale is less than @xmath116mpc , which is the correlation length of galaxies and therefore a good demarcation of the nonlinear regime . we may thus expect that the linear results for @xmath114 will be modified for @xmath13 . an analytic estimate of the change in the above scalings due to strongly nonlinear effects can be made using the stable clustering solution . if one assumes that the lensing is dominated by clustering on small scales ( valid in the limit of small angles and low source redshift ) , which stabilized at high redshift , then the dimensionless power is a function of the physical wavenumber times @xmath117 . this property extends to the open models as well , because at high enough redshift , @xmath118 . it can then be shown that , if the linear spectrum is normalized to the same @xmath16 today , then the enhancement factor of the nonlinear spectrum relative to the linear one is proportional to @xmath119 . this follows from a reversed version of the argument of peacock & dodds ( 1996 ) , which was made for spectra normalized to the same initial value . a particularly simple example is provided by the @xmath79 spectrum , for which the nonlinear slope and growth in the einstein - de sitter case are the same as in the linear regime . hence the nonlinear spectrum is given at all @xmath53 by a constant enhancement factor relative to the linear spectrum . for @xmath120 one can approximate @xmath121 for @xmath102 . therefore the @xmath8 dependence in @xmath74 due to strongly nonlinear evolution of the power spectrum is @xmath122 . in this regime then , @xmath123 times the contribution of the distance factors , which further lowers the power of @xmath8 . this estimate is meant to provide an upper bound on the enhancement for open models due to the nonlinear contribution . for realistic redshifts the contribution from such small scales does not dominate , and the exact , numerical estimates of the nonlinear spectrum must be integrated over @xmath53 and @xmath29 . still , it is clear that the @xmath8 dependence of weak lensing amplitude will be significantly weaker than the linear relation predicted from the scalings in previous subsection . the detailed predictions for the rms ellipticity depend on the shape and amplitude of the power spectrum , the distribution of source galaxies , and on the cosmological parameters @xmath8 and @xmath11 . in the following subsection we shall consider two alternative normalizations of realistic cdm - like power spectra . a useful , model independent description of the scalings can also be obtained by fitting the dependence on the different parameters to power laws . this provides a reasonably accurate approximation to the result , and shows the qualitative behavior more clearly . table 1 provides such power law fits to the scalings of @xmath114 with @xmath17 , @xmath3 , @xmath16 , and @xmath8 for einstein - de sitter , open and @xmath12dominated cosmologies . the three specific models we have chosen are : einstein - de sitter ( @xmath61 , @xmath100 ) , open ( @xmath124 , @xmath100 ) , and @xmath12dominated ( @xmath125 ) . the latter two models are representative of the class of open and flat @xmath12dominated cosmologies as the results do not change significantly for @xmath8 in the range @xmath126 . the matter power spectrum used in the calculations is the nonlinear @xmath127 cdm spectrum , where @xmath128 is the shape parameter defined in bardeen et al . ( 1986 ) and roughly corresponds to @xmath129 in cdm models . the scalings given in table 1 are not strongly sensitive to the shape of the power spectrum as shown in figure [ shape ] and discussed in the next subsection . .scaling of the polarization @xmath38 with @xmath17 , @xmath3 and with cosmological model parameters [ cols="^,^,^,^,^",options="header " , ] [ table ] on scales @xmath13 nonlinear evolution makes a significant difference to all the scalings . figure [ ratio ] shows the ratio of @xmath38 computed using the nonlinear / linear power spectrum for the above three models . the ratio is significantly larger than unity for @xmath13 , and reaches a factor of two for @xmath130 . it is the largest for the open model , where the nonlinear effects are more important because of a slower linear growth of perturbations . since the scalings differ in the linear and nonlinear regimes , table 1 gives power law fits for three different angles , @xmath131 , to represent the nonlinear , weakly nonlinear and linear regimes , respectively . these give @xmath38 to good accuracy over most of the range : @xmath132 ; @xmath133 ; @xmath134 ; @xmath135 . the results can be approximated by the following equations for the three cosmological models given above . the angle @xmath17 below is in arcminutes , and where the scalings differ on @xmath136 and @xmath1 , the latter scaling is given in brackets in italics . since the scaling with @xmath3 is nearly the same for the two values of @xmath17 , we have used an intermediate value in the following equations . @xmath137\ & = & \ 0.08\ \ \theta^{-0.37 } \ z_s^{0.58}\ \sigma_8^{1.25 ( { \it 1})}\ \omega_m^{0.66 ( { \it 0.81 } ) } \ \ \ \ \ \ \ \ \ : \ \omega_m=1.0\ , \nonumber \\ \bar{p}\left[2'({\it 15'})\right]\ & = & \ 0.04\ \ \theta^{-0.47}\ z_s^{0.66}\ \sigma_8^{1.38 ( { \it 1})}\ ( \omega_m/0.3)^{0.66 ( { \it 0.81 } ) } \ : \ \omega_m=0.3\ , \nonumber \\ \bar{p}\left[2'({\it 15'})\right]\ & = & \ 0.04\ \ \theta^{-0.42}\ z_s^{0.76}\ \sigma_8^{1.29 ( { \it 1})}\ ( \omega_m/0.3)^{0.60 ( { \it 0.68 } ) } \ : \ \omega_\lambda=0.7\ , . \label{scalings - lambda}\end{aligned}\ ] ] these power law fits can be used to obtain predictions for the rms ellipticity for any desired choice of cosmological parameters . for @xmath138 , the results are consistent with the linear theory estimates for power law spectra ( with @xmath77 between @xmath139 and @xmath140 ) obtained in the previous section . the @xmath38 curve is enhanced at small angles due to nonlinear evolution . this enhancement causes the dependence on @xmath17 to be much closer to a power law over the range @xmath141 than the linear prediction , which has much greater curvature due to the curvature of the power spectrum . this power law is steeper for the open and @xmath94 models due to stronger nonlinear evolution . on @xmath142 , the dependence on @xmath16 is significantly stronger than linear , especially for the open model . nonlinear evolution also weakens the dependence on @xmath8 as anticipated in the previous section . combining the scaling with @xmath16 and @xmath8 ( relevant for comparison with other tracers ) , @xmath38 for @xmath142 scales as @xmath143 , whereas on large scales of @xmath144 , it scales as @xmath145 . thus on small scales @xmath38 measures @xmath146 ( the power of @xmath8 is even lower for @xmath147 or @xmath148 ) , whereas on large scales it measures @xmath145 . using the above differences a comparison of @xmath38 at small and large angular scales can constrain @xmath16 and @xmath8 separately . thus nonlinear evolution , at the expense of a more complex dependence on various parameters , can help break the degeneracy between these parameters . to distinguish an open from a flat @xmath12model with the same value of @xmath8 one needs to use the difference in the scaling with @xmath3 , as the other scalings are very similar . the @xmath12model predicts a faster increase with @xmath3 due to a faster increase of the distance factors with redshift than the open case . once it becomes feasible to get estimates of the redshifts of source galaxies , this difference can become a useful test of @xmath94 as it is nearly independent of the shape and normalization of the power spectrum . in choosing the most realistic power spectrum to compute the polarization signal , the approach closest to other tracers of the mass density field would be to use the abundance of galaxy clusters or the peculiar velocity field of galaxies to estimate the amplitude @xmath16 , and to then use the galaxy power spectrum to constrain the shape ( peacock & dodds 1996 ; peacock 1996 ) . the merit of this approach is that it agrees with the observational constraints on the scales which are most relevant for weak lensing observations . an alternative is to work with physical models for the dark matter and cosmology which give the shape of the power spectrum , and use cobe - normalization to fix the spectrum on very large scales and thus its overall amplitude . the merit of this approach is that for a given model both the spectrum of perturbations and their evolution is fully determined . however , the final spectrum may or may not agree with the present day observations ( which interpretation is still somewhat uncertain ) . we shall present results using both approaches . the first of the two reliable tracers of mass fluctuations on large scales is the abundance of large galaxy clusters with a typical mass of @xmath149 , which corresponds to a linear scale of about @xmath15mpc . these objects are very rare and according to the press - schechter formalism ( press & schechter 1974 ) their number density depends exponentially on the amplitude of mass fluctuation on the corresponding linear scale . this allows an accurate estimate of the amplitude , which typically gives @xmath150 ( white , efstathiou & frenk 1993 , viana & liddle 1995 , eke , cole & frenk 1996 , pen 1996 ) in an @xmath61 universe , almost independent of the shape of the power spectrum . at present this gives the best constraint on the linear density power spectrum , with an error of about 10 - 20% . the dependence on @xmath8 is approximately @xmath151 , and differs only slightly for open and spatially flat models . the scaling with @xmath8 and @xmath11 is still somewhat uncertain as it requires calibration with large n - body simulations . peculiar velocities are also tracers of underlying mass fluctuations . in the linear regime there is a simple relation between velocity and density fields ( peebles 1980):@xmath152 , where @xmath153 . the @xmath8 dependence on mass fluctuation amplitude from velocity data is @xmath154 , which is almost the same as for the cluster abundance in a flat universe and only slightly less so in an open universe . comparison of the two constraints therefore can not significantly constrain @xmath8 . given a set of velocity measurements one can reconstruct the particular combination @xmath155 and estimate its power spectrum . recent comparisons between mark iii tully - fisher catalogs and iras survey tend to favor @xmath156 ( @xcite , @xcite ) , where @xmath157 is the linear bias parameter for iras galaxies . this combined with @xmath158 leads to @xmath159 , smaller than the value derived from the cluster abundance data . this would therefore reduce the weak lensing predictions presented in this paper . however , analysis of mark iii catalog with potent gives a higher value of @xmath160 ( kolatt & dekel 1996 ) , so it seems prudent to adopt the cluster abundance normalization until the discrepancy in the velocity data is resolved . based on the above discussion , a useful functional form for the density power spectrum is given by a cdm type transfer function with two free parameters , the amplitude @xmath161 and the shape parameter @xmath162 ( peacock & dodds 1996 ) . this is the model that will be adopted in computing predictions of ellipticity polarizations for cluster - abundance normalization . we shall use the fitting formulae in equations ( 47 - 49 ) of @xcite to compute @xmath16 as a function of @xmath8 their normalization is very close to the results of @xcite and pen ( 1996 ) , though a bit higher than that of @xcite . before proceeding with the results for @xmath38 we consider the contribution to the integral from different ranges in @xmath53 . figure [ fig11 ] compares the contribution per logarithmic interval in wavenumber to @xmath163 and @xmath164 at @xmath138 , @xmath96 with the logarithmic contribution to @xmath165 , which is also an integral of the power spectrum over @xmath53 with a real space top - hat filter . all the distributions are normalized so that they peak at unity . one can see that the distribution of @xmath98 is broader and that of @xmath164 narrower than the corresponding @xmath165 distribution , but in general they sample very much the same scales . thus the optimal statistic to use in order to compare the normalization @xmath16 with other tracers is @xmath164 , @xmath74 at @xmath166 , @xmath75 . the comparison can then be made nearly independent of the shape of the power spectrum . we need to choose a redshift distribution for the source galaxies in order to compute @xmath38 . at present this is spectroscopically known only for galaxies with magnitude in i below 23 ( lilly et al . 1996 ; cowie et al . 1996 ) , which have median redshift @xmath167 , while typical lensing observations reach several magnitudes deeper ( e.g. mould et al . hst observations of abell 2218 coupled with cluster mass reconstruction indicate that the median redshift increases from @xmath168 at r=22 to @xmath169 at r=25 ( kneib et al . there could however be a population even at redshift beyond 1 , as indicated by the detection of the shear around a @xmath170 cluster ( luppino & kaiser 1996 ) . most of the results we present are for @xmath96 , but they can be adapted to any desired @xmath3 using the scalings of table 1 . for simplicity we have assumed that all the galaxies are at the same distance . this is the best possible case in the sense that it makes the distributions in @xmath53 space more narrow than for more realistic cases . fortunately the difference between our results and those using a different , more realistic distribution of source galaxies with the same median redshift is usually very small ( kaiser 1992 , villumsen 1996 ) . figure [ pvstheta1 ] shows the dependence of @xmath38 on @xmath17 for the three cosmological models , all with shape parameter @xmath127 , except for the @xmath61 cobe - normalized model . the thick solid curve is for @xmath61 , the dashed curve for @xmath124 and the dotted curve for @xmath124 , @xmath171 . the normalization chosen is ( a ) @xmath172 , ( b ) cluster - abundance , and ( c ) cobe . for the open and flat cobe models , we have used @xmath173 so that the value of @xmath174 and @xmath16 are reasonable . the three cobe models are : flat , @xmath61 , tilted with @xmath175 , @xmath176 , @xmath177 ; open with @xmath173 , @xmath178 , @xmath179 ; flat , @xmath12model with @xmath173 , @xmath178 , @xmath180 ( bunn & white 1996 ) . since the @xmath12model chosen has a high @xmath16 , it turns out to have nearly the same @xmath38 prediction as the tilted @xmath61 model . the thin solid curve is computed using the linear spectrum for the @xmath61 model . as expected from the results of figure [ ratio ] , it is only for @xmath144 that the linear and nonlinear curves start to coincide . for cluster - abundance normalization , the amplitude of ellipticity correlations very weakly depends on the mean density for a reasonable range of @xmath8 . the enhancement for low @xmath8 models from the growth and distance factors discussed in 2 nearly cancels out the @xmath181 dependence one might have expected from the term outside the integral in equation [ p ] . therefore one has to make subtler comparisons of the signal at small and large @xmath17 to break the degeneracy between @xmath16 and @xmath8 . this was discussed above following equation ( [ scalings - lambda ] ) . it indicates that comparing weak lensing measurements over different scales can more powerfully constrain @xmath8 than comparing the weak lensing amplitude with other tracers . for cobe normalization on the other hand , a detection of the correlated ellipticity with reasonable error bars can immediately constrain the cosmological model in question , but of course there might still remain several different models which equally well fit the data . the effect of choice of statistic , shape of the power spectrum , and the redshift of source galaxies is shown in figures [ pvssig ] and [ shape ] . figure [ pvssig ] shows @xmath182 for the three cosmological models with cluster - abundance and cobe normalization as in figure [ pvstheta1 ] . the curves for @xmath183 flatten at small @xmath17 more than the @xmath114 curves . this is to be expected because @xmath114 integrates over all scales larger than @xmath184 ( where @xmath185 is the typical distance ) and so can only increase toward small scales . because there is little power on very small scales it eventually saturates ( this typically happens on subarcminute scales , keeton , kochanek & seljak 1996 ) . the fourier space quantity @xmath164 on the other hand receives contributions only from scales smaller than @xmath186 and therefore decreases on very small scales for realistic power spectra . the shape of the @xmath81 curve is very similar to that of @xmath98 , and for @xmath187 , @xmath81 is well approximated by @xmath188 . figure [ shape ] compares @xmath38 for the @xmath189 cdm spectrum to the @xmath127 spectrum used in the rest of the figures . the upper and lower set of curves are for @xmath190 and @xmath191 respectively . since the @xmath189 spectrum has more power at small scales , it leads to a larger @xmath114 at small @xmath17 . the curves cross around @xmath192 , but the difference is not large for @xmath193 . figure [ zs ] shows the variation of @xmath194 with @xmath3 for @xmath138 and @xmath195 . the models shown are the same as in the cluster - abundance normalized panel of figure [ pvstheta1 ] . as shown in table 1 , @xmath25 increases with @xmath3 fastest for the @xmath12models . so far we have only discussed the importance of nonlinear effects on the second order statistic , which as we have seen increases the signal above the values predicted by linear calculations . nonlinear evolution however also changes the distribution function of correlated ellipticity or magnification . in particular , the tails of distribution grow and become asymmetric , reflecting the nonlinear growth of perturbations into small and overdense objects . this leads to a distribution of magnification which is skewed toward positive values . such skewness is of interest for determining cosmological parameters by itself ( bernardeau et al . 1996 ) , but also affects the inference of cosmological parameters from the measurements of 2-point statistics . if the distribution of polarization at a given angular scale @xmath17 can be assumed to be normal distributed then the usual results from gaussian statistics apply . for example a single measurement of polarization amplitude within a circular aperture has a rayleigh probability distribution @xmath196 , which has exponentially suppressed tail and from which the limits on cosmological parameters can be derived analytically . several independent measurements can be easily combined and only a few observations suffice to obtain strong limits on cosmological parameters . if , on the other hand , the distribution is strongly non - gaussian so that the tails of distribution are important then one can not use these simple arguments to derive the limits and a larger set of measurements ( a `` fair sample '' ) is needed to sufficiently sample the tails so that the variance converges to a true value . at the same time , higher order statistics become easier to measure and can provide further tests of cosmological parameters . for measurements of density perturbations a simple criterion of where the nonlinear effects become important is @xmath197 , so for scales above @xmath198 mpc the universe is in the linear regime and gaussian statistic can be used . for smaller scales nonlinear effects become important which leads to deviations from a normal distribution even if the initial field was gaussian . for projected quantities such as shear and magnification the answer is not so simple . the projection integrates over all the scales , so there will always be contribution from very small and thus very nonlinear scales . on the other hand , projection will reduce the deviations from a gaussian distribution because of the central limit theorem : if we sum over sufficiently large number of deflectors the final distribution will be a gaussian regardless of the probability distribution of individual deflectors . we thus expect that the nonlinear effects will be more important for small angles and for bright ( nearby ) background galaxies . to obtain a more quantitative answer we calculated the skewness of local convergence @xmath199 using second order perturbation theory . skewness for local convergence does not vanish as in the case of ellipticity ( bernardeau et al . 1996 ) . the second moments of @xmath200 and @xmath114 are the same and so one may expect the higher order moments of convergence to be indicative of the nonlinear effects on the ellipticity distribution as well . note that the quantity @xmath201 differs from the quantity @xmath202 used by bernardeau et al . the latter is defined so that it is independent of the amplitude of power spectrum in the second order perturbation theory , while @xmath201 is a more useful variable to quantify the deviations from the gaussianity . as discussed in the appendix , second order perturbation theory should be approximately valid in the regime of interest here , between 1 to 1@xmath203 . second order perturbation theory accurately estimates the skewness over a much wider range of scales than it does the variance . only on the smallest scales it underestimates the skewness , but typically not more than a factor of two , as shown by n - body simulations ( colombi , bouchet & hernquist 1996 ) . using the expressions derived in appendix ( see also bernardeau et al . 1996 ) we can compute the value of @xmath201 for any of the models discussed in this paper . figure [ skew1 ] presents the distribution of @xmath201 as a function of @xmath17 for 3 different redshifts in 3 different models : flat model with @xmath204 , curvature and cosmological constant models , both with @xmath172 and @xmath205 ( this is very close to the cluster - abundance normalization ) . in all the models we use @xmath127 . for simplicity we will choose @xmath206 as the value where the distribution becomes non - gaussian . for background galaxies at @xmath207 we find that the distribution is non - gaussian even at @xmath208 in the flat model , where the polarization signal is only 0.25% . at @xmath96 and @xmath190 the corresponding values are @xmath209 and @xmath210 . for curvature and cosmological constant dominated models the non - gaussian effects are even more important because of higher normalization at @xmath15mpc , although the effect is offset by the longer radial distance to a given redshift . the conclusion is that in most of the observationally interesting regime the nongaussian effects will be quite important . while this complicates the issue of how large a fair sample should be , it also implies that skewness will be fairly easy to measure with a reasonably small sample . because @xmath211 depends sensitively on @xmath8 but not on the amplitude of the power spectrum ( bernardeau et al . 1996 ) this test may become a promising way to determine @xmath8 . the analytical approach presented above provides only a limited view of the properties of ellipticity distribution . for example , such an analysis can not answer whether skewness is produced by a few outlying points generated by rare clusters , in which case a rather large sample would be needed to sample these properly . for a quantitative estimate of what constitutes a fair sample for measuring the variance requires therefore knowledge of at least the fourth moment . such estimates of a fair sample for measuring the second and third moment can be made more directly by examining the distribution of correlated ellipticities in n - body simulations ( wambsganss , cen & ostriker 1996 ) . weak gravitational lensing has several advantages over other probes of large scale structure . it is directly sensitive to the underlying mass distribution and is therefore independent of biasing , which has plagued many of the cosmological tests based on the distribution of galaxies . while peculiar velocities in principle similarly probe the dark matter distribution directly , in practice they suffer because of large observational errors and various biases associated with these . this is the main reason why higher order statistics of velocity fields have not provided strong constraints on cosmological models . gravitational lensing could provide a much cleaner way of performing power spectrum and higher order statistical tests , provided that one can overcome observational difficulties and reach the 1% level of polarization , quite feasible with the new generation of composite ccd cameras . for a filled survey with an area @xmath212 and @xmath213 measured galaxies the noise variance in @xmath74 is given by @xmath214 , where @xmath215 is the typical intrinsic shear for a single galaxy . although one has to average over sufficient number of galaxies to reach this level of signal , this is not a severe limitation for sufficiently deep exposures with several hundred thousand galaxies per square degree , so on degree scales the signal to noise is of the order of 100 ( kaiser 1996 ; fig . [ sn ] ) and in fact sampling variance ( finite number of independent areas of size @xmath216 ) is more important than noise on large angles . on smaller scales the noise increases ( proportional to @xmath217 ) , but the signal also increases nearly as rapidly , so that the overall signal to noise remains approximately flat down to @xmath136 scales in open and cosmological constant models and slowly decreases in flat models ( fig . [ sn ] ) . based on this it appears there should be sufficient power present in a survey of this size to measure the signal even on arcminute scales with a high statistical accuracy , so a comparison between small and large scales to constrain @xmath8 would indeed be feasible , provided that the systematic effects can be kept under control . the other limitation at present is that the redshift distribution of background galaxies is unknown beyond @xmath218 ( lilly et al . 1996 ; cowie et al . 1996 ) . new observations with keck will be able to extend this spectroscopically up to @xmath219 ( e.g. lowenthal et al . 1996 ) and even beyond , if spectra of magnified arcs are measured . in addition , weak lensing on high redshift clusters can be used to constrain the redshift distribution of faint galaxies without measuring their redshifts , although at present there appears to be some controversy over whether the faintest galaxies are predominantly around or below @xmath62 ( kneib et al . 1995 ) or above that ( luppino & kaiser 1996 ) . finally , the best hope to constrain their redshift distribution may lie in the photometric redshifts technique ( conolly et al . 1995 ; sawicki , lin & yee 1996 ) . this would allow one to perform such precision tests as the growth of power spectrum with redshift , needed to estimate the value of @xmath11 . for example , based on hubble deep field observations , sawicki et al . 1996 find that the redshift distribution of @xmath220 galaxies is bimodal , with a large contribution of ( predominantly blue ) galaxies between redshifts 2 - 3 , while at @xmath221 most of the galaxies are below @xmath62 . in both cases the weak lensing signal to noise with these limiting magnitudes would be comparable to the one in fig . [ sn ] , so in principle by comparing the signals in the two magnitude bins it would be straightforward to deduce the value of @xmath11 . the caveat is that the faintest galaxies in hdf appear to be very irregular and small , so it is not clear whether weak lensing analysis from the ground could be performed on these objects . while more investigation is needed along these lines , there appears to be a real promise of measuring cosmological parameters with future weak lensing observations . we have analyzed theoretical predictions for weak lensing ellipticity correlations over a broad range of angles and for a variety of cosmological models . our principal conclusions are the following : \a ) the amplitude of the rms ellipticity varies between 4 - 6% on arcminute scales to 0.5 - 1% on degree scales for background galaxies at @xmath96 and grows as @xmath222 . in this range the amplitude only weakly depends on the value of cosmological parameters if cluster abundance ( or velocity flows ) normalization is adopted . a single measurement of weak lensing amplitude on @xmath223 will provide a determination of @xmath224 which can be compared with the cluster - abundance or velocity flow predictions . because these tracers all scale similarly with @xmath8 this will not provide an independent determination of @xmath8 , for which a more complicated analysis will be needed . \b ) nonlinear effects are important for second order statistics on scales below @xmath225 and enhance the signal expected from linear theory . this enhancement can be up to a factor of two on arcminute scales . from this it follows that the signal to noise does not decrease as rapidly on small scales as expected from linear theory and is in fact approximately flat in certain models from arcminute to degree scales ( see figure [ sn ] ) . therefore measurements over this whole angular range should be pursued , especially since they provide complementary information . the interpretation of cosmological models is more cleanly addressed on large angular scales , where nonlinear effects are negligible and the distribution function is primordial . small angular scales can provide insight on the nonlinear evolution of perturbations . if our current understanding of the latter is correct then combining measurements on large and small angular scales will help break the degeneracy between the cosmological parameters , because of different scalings in the linear and nonlinear regimes . the ellipticity on small angles measures @xmath226 , on large angles it measures @xmath227 , while the variation with @xmath3 can constrain @xmath11 . thus , to obtain model independent measurements of the cosmological parameters it is necessary to measure the signal over a range of angles and source redshifts . \c ) nonlinear effects on the distribution of ellipticity are significant on scales below @xmath209 and imply that gaussian statistics can not be applied to the data . this means that a larger sample will be required to measure second order statistics than expected from a gaussian distribution . conversely , skewness and other higher moments will be easy to measure and may provide a strong test of cosmological parameters independent of the tests above . in conclusion , weak lensing has the promise to become the next testing ground for cosmological theories . the expected signal to noise is above unity over a large range of angles and limiting magnitudes . this should allow one to test sensitively the spectrum of density perturbations and its evolution in time , in both the linear and the nonlinear regimes . is very grateful to jens villumsen for initiating his interest in weak lensing , and to him and richhild moessner for useful discussions . we thank bernard fort , nick kaiser , chris kochanek , yannick mellier , stella seitz , simon white , and especially peter schneider for stimulating discussions . we thank the referee for suggestions which helped improve the paper and led to the addition of figure 8 . in this appendix we present a general description of lensing in a weakly perturbed universe . the derivation presented here follows seljak ( 1995 , 1996 ) and differs from recent derivations in bernardeau et al . ( 1996 ) and kaiser ( 1996 ) in that it employs a global description of photon trajectories . the geometrical interpretation requires only knowledge of spherical trigonometry and it is thus particularly simple to derive the general weak lensing expressions . moreover , it can easily be generalized beyond weak lensing approximation or small deflection angles . the framework is a perturbed robertson - walker with small - amplitude metric fluctuations . we will only consider scalar perturbations here , in which case the metric in the longitudinal gauge can be written using conformal time @xmath228 and comoving coordinates @xmath229 as @xmath230 here @xmath231 is the scale factor expressed in terms of conformal time . we will adopt units such that @xmath232 . we set the two scalar potentials to be equal , which is a good approximation in the matter dominated epoch . the space part of the background metric can be written as @xmath233 where @xmath234 is the curvature term which can be expressed using the present density parameter @xmath235 ( only the present day density parameter will be used in this paper , hence we may drop the subscript 0 ) and the present hubble parameter @xmath236 as @xmath237 . the relation between the radial distance @xmath29 and the redshift @xmath64 can be obtained from the friedmann equation , @xmath238 , where @xmath239 is the expansion factor and the densities of matter , cosmological constant and curvature are expressed in terms of the critical density . the density parameter @xmath235 can have contributions from mass density @xmath240 or vacuum energy density @xmath241 , @xmath242 . the advantage of using the conformal time @xmath228 is that the metric becomes conformally euclidean ( @xmath243 ) , 3-sphere ( @xmath244 ) or 3-hyperboloid ( @xmath245 ) and leads to a simple geometrical description of light propagation . in an unperturbed universe a photon emitted from a source toward the observer will travel along a null geodesic in the radial direction with a radial position given by @xmath246 . adding a perturbation changes the photon trajectory . the change in photon direction is governed by the space part of the geodesic equation , which applied to the metric ( equation [ metric ] ) gives @xmath247 where the symbol @xmath248 denotes the transverse derivative of potential . null geodesics obey @xmath249 , from which follows the relation @xmath250 . here @xmath28 can be interpreted as the newtonian potential , since on scales smaller than the horizon it obeys the cosmological poisson equation . it can be viewed as providing a force deflecting the photons and affecting their travel time while they propagate through the unperturbed space - time . equation ( [ dndl ] ) is a generalization of einstein s deflection angle formula and includes the well - known factor of 2 difference compared to newtonian gravity . even when metric perturbations are present , one can continue to parameterize the geodesic with the unperturbed comoving radial distance @xmath29 . the deflection angle at a given position @xmath29 can be calculated using a locally flat coordinate system , which allows a plane wave expansion for the potential @xmath28 , provided that the longest correlation length is small compared to the curvature length ( this condition is well satisfied for the power spectrum in our universe ) . the effect of the deflection angle on the photon transverse position must however include the curvature effects . in practice this means that one only needs to know how to solve triangles using the spherical , euclidean or hyperboloid trigonometry ( figure [ fig_appb ] ) . because the only observable photon direction is that at the observer s position we will propagate photons relative to their final direction ( i.e. backwards in time ) . we will also adopt a small deflection angle approximation , because one does not expect large deflection angles due to the lensing , but the expressions can be generalized to remove this restriction . in this approximation the transverse derivatives in equation ( [ dndl ] ) can be approximated with the transverse derivatives with respect to the observed direction of the photon or with respect to any other fiducial direction that has a small angular separation with the photon ( for example the unperturbed direction to the source ) . in this plane approximation the observed photon direction @xmath251 can be described with a two - dimensional angle @xmath252 with respect to the fiducial direction , @xmath253 . suppose a photon is observed at an angle @xmath252 relative to some fiducial position . as it propagates through the universe the photon is additionally deflected according to equation ( [ dndl ] ) ( see figure [ fig_appb ] ) . this leads to the transverse photon excursion @xmath254 relative to the unperturbed line . from euclidean , spherical or hyperboloid trigonometry one finds that an individual deflection by @xmath255 at @xmath256 leads to an excursion at @xmath29 given by @xmath257 . the total excursion is given by an integral over individual deflections , @xmath258 because of gravitational lensing the `` true '' surface brightness ( i.e. the surface brightness one would see in the absence of any lensing ) at position @xmath33 is mapped into the observed one , @xmath259 , where @xmath260 is the angular deflection of a photon caused by intervening mass , which follows from equation ( [ xperp ] ) @xmath261 while the deflection @xmath260 is not directly observable , its gradient is , through the stretching and magnification of distant galaxies . this is described by the two - dimensional shear tensor defined in equation ( [ shear ] ) . to compute it we expand the potential in equation ( [ dtheta ] ) across two neighboring rays separated at the observer by @xmath262 using the _ unperturbed _ separation @xmath263 . this approximation assumes that the components of shear tensor @xmath264 are small ( kaiser 1992 ) , or , similarly , that the relative deflection between the neighboring rays is small compared to the unperturbed separation . this is the so - called weak lensing approximation . the shear tensor can be diagonalized and decomposed into its trace @xmath265 and ellipticity ( or polarization ) @xmath266 given by @xmath267 while trace @xmath268 magnifies ( or demagnifies ) the images of galaxies , ellipticity @xmath24 stretches the images and can be observed by averaging over a sufficient number of galaxies so that the noise caused by the intrinsic ellipticity of galaxies becomes smaller than the signal . the simplest average is the one within a circular window of radius @xmath17 , which can be written in fourier space as @xmath269 we decomposed the wavevector into the radial component @xmath270 and the 2-dimensional transverse component @xmath271 with the amplitude @xmath272 . we introduced the variables @xmath273 , where @xmath274 is the azimuthal angle of @xmath271 . the integral over @xmath275 can be readily performed to give @xmath276 where @xmath41 . we may now employ the small angle approximation ( limber 1956 ) , which is valid in the limit where the radial window function is broad compared to the typical wavelength that contributes to the integral . in this limit only the modes perpendicular to the radial direction will contribute to the integral , all the others being suppressed because of cancellation of positive and negative fluctuations along the line of sight . this allows one to put @xmath277 . with this approximation one obtains for rms ellipticity amplitude @xmath278 @xmath279 ensemble average can be expressed in terms of power spectrum of potential @xmath280 or density @xmath281 , @xmath282 where power spectrum @xmath283 and expansion factor @xmath19 depend on time . integrating over the dirac @xmath284 function and over the radial component of wavevector we finally obtain equation ( [ p ] ) , which gives the rms amplitude of polarization . the same expression can also be used for rms amplitude of magnification @xmath200 . the rms amplitude for each of the two ellipticity components is simply @xmath285 of the total . because skewness of ellipticity vanishes for symmetry reasons we will use skewness of local convergence as an estimate of the nongaussian nature of ellipticity distribution . calculating it proceeds along the same lines as for the variance ( bernardeau 1995 ) . it is defined as an ensemble average of the third moment of the mean magnification , @xmath286 ^ 3 \right\rangle.\ ] ] in linear theory the 3-point correlation function vanishes , so one needs to go beyond the linear approximation to obtain a nonvanishing value . in second order perturbation theory the density field is given by @xmath287.\end{aligned}\ ] ] note that the time dependence is all in the growth factors of @xmath288 , which is calculated using linear theory ( neglecting the extremely weak dependence of @xmath289 on @xmath235 , bouchet et al . ensemble averaging of skewness gives 6 identical terms , which after integrating over the radial wavevectors leads to the following expression , @xmath290 the integrals can be further simplified using similar manipulations as in bernardeau ( 1995 ) , which leads to the expression ( see also bernardeau et al . 1996 ) @xmath291 . \label{skew}\end{aligned}\ ] ] this expression was used in 4 to evaluate the importance of nonlinear evolution on the distribution function of polarization . several approximations have been employed in obtaining the result above , which limit its validity . second order perturbation theory results are only valid in the domain where density perturbations are not much larger than unity , so on very small scales a more involved calculation would be needed . however , these corrections are typically not more than a factor of two . because our purpose is to estimate at which scale the nongaussian effects become important second order perturbation theory suffices . another approximation is the use of small angle approximation , which leads to corrections on angular scales above @xmath209 ( bernardeau 1995 ) . in the regime of main interest for us , which is between @xmath292 this is not an important correction . finally , expression ( [ skew ] ) neglects the deviation of the photon trajectory from the unperturbed path , which also contributes at the second order . this correction is less important than the main term used in equation ( [ skew ] ) and can be neglected ( bernardeau et al . 1996 ) . bardeen , j. m. , bond , j. r. , kaiser , n. , & szalay , a. s. 1986 , , 304 , 15 bernardeau , f. 1995 , a&a , 301 , 309 bernardeau , f. , van waerbeke , l. , & mellier , y. 1996 , astro - ph/9609122 blandford , r. d. , saust , a. b. , brainerd , t. g. , & villumsen , j. v. 1991 , , 251 , 600 bouchet , f. r. , juszkiewicz , r. , colombi , s. & pellat , r. 1992 , , 394 , l5 bunn , e. f. & white , m. 1996 , astro - ph/9607060 colombi , s. , bouchet f. r. , & hernquist l. 1996 , , 465 , 14 connolly , a. j. , scabai , i. , szalay , a. s. , koo , d. c. , kron , r. c. , & munn , j. a. , , 110 , 2655 cowie l. , l. , songaila , a. , hu , e.i m. , & cohen , j. g. 1996 , astro - ph/9606079 davis , m. , nusser , a. , & willick , j. a. 1996 , astro - ph/9604101 eke , v. r. , cole , s. , & frenk , c. s. 1996 , astro - ph/9601088 fort , b. , mellier , y. , dantel - fort , m. , bonnet , h. & kneib , j .- p . 1996 , a&a , 310 , 705 gunn , j. e. 1967 , , 147 , 61 hamilton , a. j. s. , kumar , p. , lu , e. , & matthews , a. 1991 , , 374 , l1 jain , b. , mo , h. j. , & white , s. d. m. 1995 , , 276 , l25 kaiser , n. 1992 , , 388 , 272 kaiser , n. 1996 , astro - ph/9610120 keeton , c. r. , kochanek , c. s. , seljak , u. 1996 , astro - ph/9610163 kneib , j .- p . , ellis , r. s. , smail , i. , couch , w.j . , sharples , r. m. 1995 , astro - ph/9511015 kolatt , t. & dekel , a. 1995 , astro - ph/9512132 lahav , o. , lilje , p. b. , primack , j. r. , & rees , m. j. 1991 , , 251 , 128 lilly , s. j. , le fevre , o. , hammer , f. & crampton , d. 1996 , astro - ph/9601050 lowenthal , j. d. et al . 1996 , astro - ph/9612239 luppino , g. , & kaiser , n. 1996 , astro - ph/9601194 miralda - escude , j. 1991 , , 380 , 1 mould , j. , blandford , r. , villumsen , j. , brainerd , t. , smail , i. , small , t. , & kells , w. 1994 , , 271 , 31 nityananda , r. & padmanabhan , t. 1994 , , 271 , 976 padmanabhan , t. , cen , r. , ostriker , j. p. & summers f. j. 1996 , , 466 , 604 peacock , j. a. , & dodds , s. j. 1996 , , astro - ph/9603031 peacock , j. a. 1996 , astro - ph/9608151 peebles , p. j. e. 1980 , the large - scale structure of the universe ( princeton : princeton university press ) pen , u. 1996 , astro - ph/9610147 press , w. h. & schechter , p. 1974 , , 187 , 425 sawicki , m. j. , lin , h. , yee , h. k. c. 1996 , astro - ph/9610027 seljak , u. 1995 , ph.d . thesis , mit , unpublished seljak , u. 1996 , , 463 , 1 stebbins . a. 1996 , preprint astro - ph/9609149 viana , p. t. p. , & liddle , a. r. 1995 , astro - ph/9511007 villumsen , j. v. 1996 , , 281 , 369 wambsganss , j. , cen , r. , ostriker , j. p. 1996 , astro - ph/9610096 willick , j. a. , strauss , m. a. , dekel , a. , & kolatt , t. 1996 , astro - ph/9612240 white , s. d. m. , efstathiou , g. , & frenk c. s. 1993 , , 262 , 1023

starting from the qcd sum rules with nonlocal condensates for the pion distribution amplitude , we derive another sum rule for its derivative and its `` integral '' derivatives defined in this work . we use this new sum rule to analyze the fine details of the pion distribution amplitude in the endpoint region . the results for endpoint - suppressed and flat - top ( or flat - like ) pion distribution amplitudes are compared with those we obtained with differential sum rules by employing two different models for the distribution of vacuum - quark virtualities . we determine the range of values of the derivatives of the pion distribution amplitude and show that endpoint - suppressed distribution amplitudes lie within this range , while those with endpoint enhancement flat - type or cz - like yield values outside this range .

1006.2936

many of the theoretical and phenomenological analyses of qcd processes rely upon the factorization of the underlying dynamics into a short - distance dominated part , amenable to qcd perturbation theory , and a large - distance part that has to be taken from experiment or be determined by nonperturbative methods . among such processes , the pion form factors ( electromagnetic and transition ) play the role of a theoretical laboratory to test various ideas and techniques . the key ingredient in these descriptions is the pion distribution amplitude ( da ) @xmath1 which represents the pion bound state . at leading twist two it is defined in terms of a nonlocal axial current and reads @xcite @xmath2 u(0 ) | \pi ( p ) \rangle |_{z^{2}=0 } = i f_{\pi}p^{\mu } \int_{0}^{1 } dx\ , { \rm e}^{ix(z\cdot p ) } \varphi_{\pi}^{\rm ( t=2)}(x,\mu_{0}^{2 } ) , \label{eq : pi - da}\ ] ] where is the longitudinal momentum fraction carried by the valence quark ( antiquark ) in the pion and the path - ordered exponential , i.e. , the lightlike gauge link @xmath3 = { \cal p } \exp \left [ -ig \int_{0}^{z } dy^{\mu } t^{a}a_{\mu}^{a}(y ) \right ] \ , , \label{eq : con}\ ] ] ensures gauge invariance . the pion da has an expansion in the basis of the gegenbauer polynomials which constitute the eigenfunctions of the one - loop meson evolution equation . at a typical hadronic scale @xmath4 , which serves as a normalization scale , one then has @xcite @xmath5 + \ldots \ , \label{eq : pion - da}\ ] ] where @xmath6 is the asymptotic pion da . by virtue of the leptonic decay @xmath7 , one obtains the normalization @xmath8 , which fixes @xmath9 . rather than try to derive the pion da as a whole , one attempts to reconstruct it from its first few moments @xmath10 where @xmath11 . the values of the moments may be determined by means of qcd sum rules ( sr)s with local @xcite or nonlocal condensates @xcite , or be computed by numerical simulations on the lattice @xcite . once they are known , one can use them to reverse engineer the pion da , with a precision depending upon the influence of the magnitude of the discarded higher - order moments . it was shown in ref . @xcite that , using qcd sum rules with nonlocal condensates , one can de facto resort to the first two gegenbauer coefficients @xmath12 , while @xmath13 with @xmath14 turn out to be negligible . once the shape of the pion da has been determined at some ( low ) normalization scale @xmath15 around 1 gev@xmath16 , one can evolve the gegenbauer coefficients to higher values of the momentum scale using the efremov - radyushkin - brodsky - lepage @xcite evolution equation which is determined by means of qcd perturbation theory . it turns out that another quantity which is intertwined with the form factors of the pion is its inverse moment @xmath17 this quantity is one of the key ingredients of the pion - photon transition form factor , a process that has attracted the continuous attention of theorists @xcite and experimentalists @xcite . actually , the most recent measurement of this observable by the babar collaboration @xcite has provided controversial results . at moderate values of the momentum transfer , up to 10 gev@xmath16 , the new high - precision babar data agree well with the previous cleo data @xcite and can be best described by pion das that have their endpoints strongly suppressed @xcite as the bakulev - mikhailov - stefanis ( bms ) model @xcite derived from qcd sum rules with nonlocal condensates . by contrast , the high-@xmath18 babar data show an unexpected growth with @xmath18 which can not be understood on the basis of collinear factorization and calls for pion das that have their endpoints strongly enhanced @xcite . this intriguing behavior consists the basic motivation for the present investigation , though we will not attempt to describe any data . we shall employ in this work the method of qcd srs with nonlocal condensates ( nlc)s with the aim to estimate the slope of the pion da in the region @xmath19 , trying to understand the fine structure of the pion da in this region vs. the ansatz for the quark - virtuality distribution in the nonperturbative qcd vacuum . our main interest will be in the behavior of pion das with distinct endpoint characteristics . qcd srs were mainly proposed with the purpose of studying the _ integral _ characteristics of the pion da . to overcome this restriction , we shall design an operator for defining _ integral derivatives _ of the pion da . these will supplement the results obtained with srs which employ the _ standard derivative _ of the pion da . in this latter case , we will use in our analysis not only a delta - function ansatz for the vacuum quark - virtuality distribution , but also a refined model which describes the large - distance regime more accurately . the paper is organized as follows . in sec . [ sec : qcd - vac ] , we briefly discuss the main features of the nonperturbative qcd vacuum in terms of the scalar quark condensate . we also give the corresponding srs and recall their main ingredients . section [ sec : pi - slope ] is devoted to the calculation of the slope of the pion da in the endpoint region employing two different techniques : integral srs and differential srs . finally , sec . [ sec : concl ] contains our conclusions , while some important technical details are given in four appendices . the basic idea underlying the nlc approach is that the vacuum condensates possess a correlation length which endows the vacuum quarks with a non - zero average virtuality @xmath20 ( see , for instance , @xcite ) . to analyze the nonlocality of the vacuum condensate , it is useful to parameterize the lowest one . therefore , one has @xmath21=1 $ ] . ] @xmath22q(z)}\rangle\equiv m_s(z^2)$ ] with the help of the vacuum distribution function @xmath23 : @xmath24 that describes the distribution of the vacuum - quark virtuality @xmath25 @xcite . assuming @xmath26 , that takes into account only a fixed virtuality @xmath27 of the vacuum quarks , leads to the simplest gaussian model @xmath28 for the scalar quark condensate @xcite . the parameter @xmath29 represents the typical quark momentum in the vacuum and is given by @xmath30 in this work , we use the value @xmath31 gev@xmath16 , which is supported by several analyses , though values within the interval @xmath32 $ ] gev@xmath16 are still acceptable ( see @xcite and references cited therein ) . the qcd srs with nonlocal condensates for the pion da were first proposed in @xcite and were significantly improved in @xcite from which we quote @xmath33_{\rm q } \ . \label{eq : piondaqcdsr}\end{aligned}\ ] ] here , @xmath34 is the @xmath35-meson da , whereas @xmath36 and @xmath37 are , respectively , the decay constants of the @xmath35 and the @xmath38-meson . the @xmath35-meson state is an effective state that collects the @xmath39 and the @xmath40 meson . the nonperturbative ingredients in the theoretical part of the sr are the gluon - condensate term @xmath41 and the quark - condensate contribution @xmath42_{\rm q}$ ] . this latter contribution contains the vector - condensate term ( v ) , the mixed quark - gluon condensate term ( t ) , and the scalar condensate term ( s ) . the explicit expressions for the nonperturbative contributions and the nlo spectral density @xmath43 are given in appendices [ app : a.1 ] and [ app : a.2 ] , respectively . it turns out that in the endpoint region , the first radiative correction in the spectral density , which is of @xmath44 , is too large , thus overshadowing the zeroth order perturbative contribution and the nonperturbative contribution . for that reason , we use in this work the leading - order ( lo ) approximation @xmath45 . in order to include radiative corrections into the spectral density when analyzing the endpoint region one would have to resum all radiative corrections , a formidable task outside the scope of the present investigation . as already mentioned in the introduction , the endpoint region of the pion da turns out to be of particular importance for a variety of pion observables see @xcite for an in - depth discussion of this issue . for example , in the case of the pion - photon transition form factor , one finds that in order to comply with the cleo data and those babar data close to them one needs endpoint - suppressed pion das , while the high-@xmath18 babar data can only be described with flat - type pion das . therefore , it is crucial to explore the fine details of the pion da in a region around the origin @xmath19 . one way to study the endpoint regime of the pion da is provided by the inverse moment @xmath46 @xcite . in fig . [ fig : inv.mom.y ] we show the values of @xmath47 with @xmath48 for some pion da models with a characteristic behavior in the endpoint region . from this figure one sees that the bms pion da ( dashed green line ) , derived with the aid of nonlocal condensates @xcite , exhibits an evident endpoint suppression , while all other models have more ( flat - top dashed - dotted - dotted red line and cz dashed - dotted green line ) or less ( asymptotic da ) endpoint enhancement . the considered models of the pion da , and their parameters , are given in appendix [ app : damodels ] . in the present work we will probe the endpoint region of the pion da also by another means , namely , the `` integral '' derivative which will be defined next . , integrated in the region @xmath49 to @xmath50 for different pion das . left panel : solid blue line mptotic ; dashed green line bms @xcite . right panel : dashed - dotted green line cz @xcite ; dashed - dotted - dotted red line flat - top da [ eq . ( [ eq : flat - da ] ) with @xmath51 . [ fig : inv.mom.y],title="fig:",scaledwidth=45.0% ] , integrated in the region @xmath49 to @xmath50 for different pion das . left panel : solid blue line mptotic ; dashed green line bms @xcite . right panel : dashed - dotted green line cz @xcite ; dashed - dotted - dotted red line flat - top da [ eq . ( [ eq : flat - da ] ) with @xmath51 . [ fig : inv.mom.y],title="fig:",scaledwidth=45.0% ] as usual , we study the srs in the fiducial interval of the borel parameter @xmath78 $ ] , where both terms , the continuum contribution and the nonperturbative one , each contributes about @xmath79 to the whole sr ( [ eq : piondaqcdsr ] ) . this induces an uncertainty of the order of @xmath80 . moreover , the quantities to be calculated with the sr ( the integral derivatives ) should not ( crucially ) depend on the borel parameter . therefore , we should take care that this dependence is minimized . to achieve this goal , we attempt to minimize the root - mean - square deviation by varying the @xmath81(x)$ ] contribution in the fiducial borel interval . on this account , we can average the @xmath82-dependence of the pion da contribution [ first term in eq . ( [ eq : piondaqcdsr : dk ] ) ) ] in order to get a more reliable form of the sr . therefore , we write @xmath83(x ) & = & \langle[d^{(\nu+2)}\varphi_\pi](x , m^2)\rangle \\ & \equiv & \frac{1}{m^2_\text{max}-m^2_\text{min } } \int_{m^2_\text{min}}^{m^2_\text{max}}\!\!dm^2\ , [ d^{(\nu+2)}\varphi_\pi](x , m^2 ) \ . \label{eq : mean - da}\end{aligned}\ ] ] in fig . [ fig : d.k1.x05.m2 ] we show the @xmath82-dependence of @xmath84(x , m^2)$ ] , obtained from sr ( [ eq : piondaqcdsr : dk ] ) , for different values of @xmath85(x)$ ] and evaluating it for @xmath86 . the average value in the fiducial interval is @xmath87(0.5)=4.8 \pm 0.5 $ ] . ( x , m^2)$ ] on the auxiliary borel parameter @xmath82 for @xmath86 . the solid blue line corresponds to @xmath85(x)=6.4 $ ] , whereas the dashed lines refer to @xmath85(x)=6.7 $ ] ( upper curve ) and @xmath85(x)=7.1 $ ] ( lower curve ) , using in all cases @xmath86 . the two vertical lines delimit the fidelity region @xmath78 $ ] . [ fig : d.k1.x05.m2],scaledwidth=50.0% ] we discuss now the integral sum rules and their applications . the main contribution from the singularities in the sr for @xmath88(x , m^2)$ ] stems from the @xmath56-region around @xmath89 . this is because we used a delta - ansatz model [ cf . ( [ eq : ansv ] ) , ( [ eq : anstril ] ) ] for the condensates , which implies that the nonperturbative contributions have delta - function and heaviside - function behaved terms ( [ eq : phi_s])([eq : phi.g ] ) . therefore , in order to take into account all nlc contributions , we should analyze the region @xmath90 and use @xmath91 gev@xmath16 which corresponds to @xmath92 . moreover , the image of the operator @xmath66 for @xmath93 is numerically very close to the result obtained with the differentiation method ( see next subsection)for any @xmath56 . thus , the integral sr ( [ eq : piondaqcdsr : dk ] ) becomes close to the differential sr which we will consider in the next section . for these reasons , we analyze the constructed sr ( [ eq : piondaqcdsr : dk ] ) for @xmath94 and @xmath95 and present the results in fig . [ fig : ddda012 ] by the solid line that is inside the light gray strip bounded by the short - dashed lines . for the sake of comparison , the predictions for the asymptotic da ( dashed - dotted line ) and the bms da bunch obtained in the nlc sr analysis of ref . @xcite(shaded band limited by long - dashed lines ) are also shown . from this figure we see that our sr estimates for @xmath96(x)$ ] agree fairly well with the bms model see also table [ tab : all - derivatives ] . this table shows estimates for the third - order integral derivative of the pion da for @xmath86 , using ( i ) the sum rule given by eq.([eq : piondaqcdsr : dk ] ) and ( ii ) the pion da models we discussed above , and also flat - type das which we consider below . -dependence of @xmath88(x)$ ] for the bms bunch of pion das @xcite ( shaded green band within long - dashed lines ) in comparison with the sr result ( [ eq : mean - da ] ) ( narrow gray strip ) in all three panels . the left panel shows the predictions for @xmath97 , whereas those for @xmath98 and @xmath99 are shown in the middle and the right panel , respectively . the dashed - dotted line represents the asymptotic result @xmath100(x)=6 - 3x/2^\nu$ ] . [ fig : ddda012],title="fig:",scaledwidth=35.0% ] -dependence of @xmath88(x)$ ] for the bms bunch of pion das @xcite ( shaded green band within long - dashed lines ) in comparison with the sr result ( [ eq : mean - da ] ) ( narrow gray strip ) in all three panels . the left panel shows the predictions for @xmath97 , whereas those for @xmath98 and @xmath99 are shown in the middle and the right panel , respectively . the dashed - dotted line represents the asymptotic result @xmath100(x)=6 - 3x/2^\nu$ ] . [ fig : ddda012],title="fig:",scaledwidth=35.0% ] -dependence of @xmath88(x)$ ] for the bms bunch of pion das @xcite ( shaded green band within long - dashed lines ) in comparison with the sr result ( [ eq : mean - da ] ) ( narrow gray strip ) in all three panels . the left panel shows the predictions for @xmath97 , whereas those for @xmath98 and @xmath99 are shown in the middle and the right panel , respectively . the dashed - dotted line represents the asymptotic result @xmath100(x)=6 - 3x/2^\nu$ ] . [ fig : ddda012],title="fig:",scaledwidth=35.0% ] second , we consider a particular flat - type pion da which is provided by the ads / qcd correspondence in the holographic approach see , for instance , refs . @xcite . in that case one has @xmath107 yielding @xmath108(0.5 ) = 14 $ ] . third , we study an alternative flat - like pion da which results from the gegenbauer expansion of unity by retaining only the first few harmonics . one obtains @xmath109 with a profile shown in fig . [ fig : piondamodels ] in comparison with the models already mentioned : bms solid line ; cz long - dashed blue line ; flat - top da given by eq . ( [ eq : flat - da])dotted red line ; flat - like da given by eq . ( [ eq:3gegeda])dashed - dotted green line . ) , dotted red line flat - top da from eq . ( [ eq : flat - da ] ) with @xmath105 . all das are normalized at the same scale @xmath110 gev@xmath16 with or without evolution @xcite . [ fig : piondamodels],scaledwidth=45.0% ] it is interesting to notice that using the pion da given by expression ( [ eq:3gegeda ] ) to calculate the pion - photon transition form factor according to radyushkin s expression ( 24 ) in ref . @xcite , one actually reproduces the gross features of his results . this proves that the inclusion of the first few gegenbauer polynomials in ( [ eq:3gegeda ] ) does not affect the result obtained with @xmath111 in the momentum range @xmath112 gev@xmath16 in a crucial way . this type of da , i.e. , ( [ eq:3gegeda ] ) , yields for the integral derivative the value @xmath113(0.5)=22.5 $ ] , which is much larger than the range of values determined via our sr . this holds true also for the other two flat - type das considered above . recalling that the leading - order qcd sum rules with the minimal gaussian model for the nonlocal condensates provide much smaller values of the integral derivative of the pion da , one may conclude that it is very difficult to reconcile flat - type pion das with sr ( [ eq : piondaqcdsr : dk ] ) . on the other hand , also the cz model yields third - order integral derivatives which are incompatible with the values derived from our sr ( [ eq : piondaqcdsr : dk])see table [ tab : all - derivatives ] . note that a similar statement also applies to the pion da proposed in @xcite , which employs a brodsky - huang - lepage ansatz for the @xmath114-dependence of the pion wave function see table [ tab : all - derivatives ] . the upshot of this table is that the sr for the integral derivative of the pion da is fulfilled by the bms bunch , whereas flat - type das have no overlap with the estimated range of values . the same is true for the cz model . @xmath115 & approach / model & `` integral '' derivative @xmath84(0.5)$ ] & derivative @xmath116 + @xmath115 1 & integral sr ( [ eq : piondaqcdsr : dk ] ) & @xmath117 & @xmath118 + @xmath115 2 & differential sr ( [ eq : dpiondaqcdsr ] ) & & @xmath119 + @xmath115 3 & sr ( [ eq : dpiondaqcdsr ] ) with smooth nlc ( [ eq : smoothmodel ] ) & & @xmath120 + @xmath115 4 & bms bunch @xcite & @xmath121 & @xmath122 + @xmath115 5 & asymptotic da & @xmath123 & @xmath124 + @xmath115 6 & cz da @xcite & @xmath125 & @xmath126 + @xmath115 7 & da from @xcite & @xmath127 & @xmath128 + @xmath115 8 & flat - type da , eq . ( [ eq:3gegeda ] ) & @xmath129 & @xmath130 + @xmath115 9 & flat - top da ( eq . ( [ eq : flat - da ] ) , @xmath105 ) & @xmath131 & @xmath132 we close this subsection by considering the usual derivative @xmath133 of the pion da which encapsulates the key characteristics of the pion da at small @xmath56 . this quantity can be extracted from fig . [ fig : ddda012 ] , where we have plotted the results for @xmath134(x)$ ] we obtained with the sr ( [ eq : piondaqcdsr : dk ] ) for different values of @xmath61 . to determine @xmath133 one can use expansion ( [ eq : d.k.x.series ] ) @xmath135(x ) & \approx & \varphi'_\pi(0 ) + \varphi''_\pi(0)\frac{x}{2!2^{\nu+1 } } \label{eq : dda00}\end{aligned}\ ] ] and subtract the second derivative for which the asymptotic value @xmath136 is used . the involved error @xmath137 was estimated by varying ( [ eq : dda00 ] ) within the narrow grey strip in fig . [ fig : ddda012 ] . to be more specific , one obtains @xmath138(x ) - \varphi''_\pi(0)\frac{x}{2!2^{\nu+1 } } \approx [ d^{(\nu+2)}\varphi_\pi^\text{sr}](x ) + \frac{3x}{2^{\nu}}=5.5 \pm 1.5 \label{eq : dda0}\end{aligned}\ ] ] for any @xmath139 and @xmath140 . the error in ( [ eq : dda0 ] ) is a combination of the uncertainties originating from sr ( [ eq : piondaqcdsr : dk ] ) and the error in the determination of @xmath141 . the above finding can be compared with what one obtains for the bms and the cz model ( displayed in table [ tab : all - derivatives ] ) @xmath142 = 1.07^{+5.87}_{-4.68}=-3.61 \div 6.95 \label{eq : der.da.bms}\end{aligned}\ ] ] and @xmath143\simeq 26.2 \ , \label{eq : der.da.cz}\end{aligned}\ ] ] respectively , using in both cases the normalization scale @xmath144 gev@xmath16 . quite analogously to the integral derivative , the cz model yields also for the standard derivative much larger values than those estimated in ( [ eq : dda0 ] ) . by contrast , the pion da model proposed in @xcite though it provides a similarly large integral derivative like the cz da has a usual derivative at the origin which is zero due to the strong exponential suppression of this da in the vicinity of the origin . another way to study the behavior of the pion da in the small-@xmath56 region is provided by the differentiation of the sr ( [ eq : piondaqcdsr ] ) , which yields @xmath145 we shall evaluate this sr for the threshold value @xmath146 gev@xmath16 , recalling that we employing a lo expression for the spectral density . using the simplest delta - ansatz model for the condensates ( cf.([eq : anstril ] ) ) , only one nonperturbative term survives , namely , the four - quark - condensate @xmath147 where @xmath148 is represented by eq . ( [ eq : delta.phi.s])see fig . [ fig:4q ] in appendix [ app : a.1 ] . the vector - quark condensate ( v ) and the gluon condensate ( g)recall eq . ( [ eq : piondaqcdsr])give zero contributions in the region @xmath149 , where @xmath150 with @xmath151 gev@xmath16 . on the other hand , the antiquark - gluon - quark condensate ( t ) amounts to a vanishing contribution in the region @xmath152 . for @xmath153 gev@xmath16 ( or equivalently @xmath154 ) this contribution can also be neglected . even if we assume a behavior of the various condensates differing from the delta - ansatz model , using , for instance , a smooth model like ( [ eq : smoothmodel ] ) for the scalar quark condensate ( which implies a decay at large distances not slower than the exponential decay see further below ) , the ( v ) , ( g ) , and ( t ) terms give only a small non - zero contribution in the small-@xmath56 region . therefore , these terms can be neglected in first approximation in both mentioned models . in this section , we shall study the differential sr ( [ eq : dpiondaqcdsr ] ) using the mentioned models for the scalar condensate . the calculation of the nonlocal version of the four - quark - condensate contribution involves only the scalar quark condensate . the four - quark contribution @xmath155 was obtained in @xcite and can be used with various types of vacuum - quark distributions @xmath23 . ) for a particular choice of the intrinsic parameters as explained in the text . [ fig : fs],scaledwidth=40.0% ] accordingly , the probability for finding vacuum quarks with very large virtualities is very small , as one can see from fig . [ fig : fs ] . therefore , the distribution function @xmath156 diminishes in the region @xmath157 , the latter region corresponding to the values @xmath158 gev@xmath16 of the borel parameter . on the basis of this result , one @xcite can use the method of ref . @xcite to obtain the model - independent expression ( [ eq : delta.phi.s ] ) . then , one finds for the four - quark - condensate contribution to the sr ( [ eq : dpiondaqcdsr ] ) the following expression @xmath159 we see from this equation that the nonperturbative contribution to the sr is mainly due to the scalar - quark condensate at large and moderate distances @xmath160 . for the concrete evaluation of the differential sr we again employ the same criteria as already used in the integral sr for both the continuum and the nonperturbative terms . applying these criteria , the low boundary for the borel parameter turns out to be very small , viz . , @xmath161 gev@xmath16 . though this low value is consistent with the applied criteria , it is not in good agreement with standard qcd sum - rule approaches in which a higher minimum borel - parameter value is used . therefore , we use @xmath162 gev@xmath16 , a value also employed in the qcd sr for the pion da in ref.@xcite in connection with the moments of the pion da . keep also in mind that using a lower value of the borel parameter would cause the decrease of the first derivative of the pion da at the endpoints . to continue , we define the derivative @xmath163 in terms of the mean value in the fiducial interval @xmath78 $ ] following the definition on the rhs of ( [ eq : mean - da ] ) . this helps minimizing the sensitivity of @xmath164 on the choice of the borel parameter by means of the variation of the @xmath35 contribution @xmath165 . the delta - ansatz @xmath26 leads to a simple expression for the nonperturbative contribution to sr ( [ eq : dpiondaqcdsr ] ) , notably , @xmath166 note that we defined the @xmath35-meson contribution @xmath167 by means of the minimum of the root mean square deviation . on the other hand , the dependence of @xmath168 on the borel parameter @xmath82 for the delta - ansatz model is controlled by eq . ( [ eq : dpiondaqcdsr ] ) and is shown in the left panel of fig.[fig : srm2 ] . thus , the average value of the pion da derivative in the fiducial borel interval is @xmath169see table [ tab : all - derivatives ] . -dependence of @xmath168 in the sr with the @xmath35-meson contribution @xmath167 . the broken lines represent @xmath168 for @xmath170 ( lower line ) and @xmath171 ( upper line ) . right panel : the solid curve illustrates the @xmath82-dependence of @xmath168 in the sr which employs the smooth quark condensate model with a @xmath35-meson contribution given by @xmath172 . the broken lines denote @xmath168 at @xmath170 ( lower line ) and @xmath171 ( upper line ) . [ fig : srm2],title="fig:",scaledwidth=45.0% ] -dependence of @xmath168 in the sr with the @xmath35-meson contribution @xmath167 . the broken lines represent @xmath168 for @xmath170 ( lower line ) and @xmath171 ( upper line ) . right panel : the solid curve illustrates the @xmath82-dependence of @xmath168 in the sr which employs the smooth quark condensate model with a @xmath35-meson contribution given by @xmath172 . the broken lines denote @xmath168 at @xmath170 ( lower line ) and @xmath171 ( upper line ) . [ fig : srm2],title="fig:",scaledwidth=45.0% ] we go forward and discuss the consequences of the smooth model for the quark - virtuality distribution in the differential sr . though the delta - ansatz model is useful , because of its simplicity , there is an indication from the heavy - quark effective theory @xcite that in reality the quark - virtuality distribution @xmath173 should be parameterized in a different way as to ensure that the scalar condensate decreases exponentially at large distances . moreover , in order that the vacuum matrix element @xmath174 exists , the quark - virtuality distribution @xmath23 should decrease faster than any power @xmath175 as @xmath176 @xcite . for this reason , the authors of @xcite suggested a two - tier model for @xmath173 which has a smooth dependence on the quark virtuality @xmath25 , namely , @xmath177 where @xmath178 is the modified bessel function . this model , the so - called `` smooth model '' , depends on two parameters @xmath179 and @xmath180 that parameterize , respectively , the long- and short - distance behavior of the nonlocal condensates @xcite . for large distances @xmath181 this model leads to the asymptotic form @xmath182 it is instructive to consider a purely exponential decay of the quark - virtuality distribution and study its influence on the quark condensate . this can be realized in the model of @xcite by choosing @xmath183 , whereas the second parameter @xmath184 gev can be taken from the qcd srs for the heavy - light meson in heavy quark effective theory see @xcite . the two parameters @xmath185 and @xmath179 are responsible for the large-@xmath186 behavior of the scalar - quark condensate , cf . ( [ eq : smooth - x ] ) . the third parameter @xmath187 gev@xmath188 is defined in terms of the parameters @xmath189 , and @xmath27 via the following equation : @xmath190 which we evaluate for the value of the nonlocality parameter @xmath151 gev@xmath16 . the main effect of using a smooth model for the quark - virtuality distribution relative to the gaussian form , @xmath191 , is the induced increase of the nonperturbative contribution to the sr , so that @xmath192 we analyzed the sr ( [ eq : dpiondaqcdsr ] ) for this model using a particular choice of its parameters , notably , @xmath193 and determined the dependence of @xmath168 on the borel parameter @xmath82 . the result is shown graphically in the right panel of fig.[fig : srm2 ] . the average value of the derivative @xmath168 in the fiducial borel interval is @xmath194 . thus , the nonperturbative contribution @xmath195 , obtained from the smooth model , is approximately two times larger xthan the analogous contribution @xmath196 in the delta ansatz : @xmath197 . in addition to this result , marked by a black dot , we show in fig.[fig : smoothsr.l.n ] the dependence of @xmath133 on the choice parameters @xmath185 and @xmath179 of the smooth model . from this picture and the relation ( [ eq:4qterm ] ) , we may come to the conclusion that , choosing a model for the condensate that has a slower decay at large distances ( small @xmath185 or @xmath179 ) , may cause an increase of the nonperturbative contribution to the sr ( [ eq : piondaqcdsr ] ) and entail also an increase of the value @xmath133 . on the other hand , choosing a model for the condensate with a faster decay at large distances ( large @xmath185 or @xmath179 ) , may lead to a decrease of the nonperturbative contribution to the sr ( [ eq : piondaqcdsr ] ) and therefore to a decrease of the value @xmath133 . on the parameter @xmath185 of the smooth scalar condensate model ( [ eq : smoothmodel ] ) for @xmath198 ( dashed - dotted red line ) , @xmath199 ( solid blue line ) , and @xmath200 ( dashed green line ) . here , we show only the central value of the derivative @xmath201 we obtained from the sr analysis ( [ eq : dpiondaqcdsr ] ) . the black dot symbol marks the position which corresponds to the model parameters @xmath183 , @xmath199 , and @xmath194 , that corresponds to the third row in table [ tab : all - derivatives ] . [ fig : smoothsr.l.n],scaledwidth=45.0% ] in the last column of table [ tab : all - derivatives ] we collect the values of the ( usual ) pion da derivative at @xmath202 , using different sr approaches ( first three rows ) and pion da models ( last six rows ) . the current investigation was partly motivated by the recent results obtained by the babar collaboration @xcite on the pion - photon transition form factor which indicate an unexpected growth of this quantity with @xmath18 above @xmath203 gev@xmath16 up to the highest momentum value of 40 gev@xmath16 measured . as it was pointed out in @xcite ( see also @xcite ) , such behavior is impossible for endpoint - suppressed pion das and brings into play a flat - type profile for the pion da , as proposed in refs . therefore , it appears to be of crucial importance to have a theoretical tool in our hands able to reveal the particular characteristics of pion das precisely in the kinematic endpoint region . the method mostly used in the past to extract such information employs the inverse moment @xmath204extensively discussed in @xcite . in the present work we proposed another , more direct way , to access the endpoint characteristics of the pion da which makes use of its derivatives . the first step was to define the notion of `` integral '' derivatives @xmath205(x)$ ] by means of an appropriate operator @xmath62 [ cf . ( [ eq : d.nu.x ] ) ] . next , we formulated an `` integral '' sum rule , eq.([eq : piondaqcdsr : dk ] ) , for these quantities by taking into account an effective @xmath35-meson contribution . using this sum rule , we determined the range of values of the integral derivatives and displayed it in the first entry of the second column of table [ tab : all - derivatives ] showing it also graphically in fig . [ fig : ddda012 ] . we also calculated the integral derivative for different characteristic pion da models and listed its value in the same table . the list of pion das includes the bms , the cz , the asymptotic , and two options for flat - type da models , one given by eq . ( [ eq : flat - da ] ) with @xmath105``flat - top '' model the other being a flat - like model parameterized in terms of eq . ( [ eq:3gegeda ] ) . we also analyzed the usual derivative of the pion da in the vicinity of the origin . this procedure is helpful in revealing the fine details of the pion distribution amplitude in the endpoint region see table [ tab : all - derivatives ] for the results . our findings can be summarized as follows . first , at the end of subsec . [ subsec : int - sr ] , we applied the result for the integral derivatives @xmath88(x)$ ] in order to reproduce the usual derivative of @xmath206 at the origin @xmath59 . the result @xmath207 is shown in the third column of table [ tab : all - derivatives ] in comparison with some characteristic models for the pion da . second , in subsec . ( [ subsec : dif - sr ] ) , we studied the derivative of the pion da @xmath133 in terms of the differentiation of the sr ( [ eq : piondaqcdsr ] ) that leads to the differential sr ( [ eq : dpiondaqcdsr ] ) . it turns out that the only nonperturbative content in this sr is mainly defined by the scalar quark condensate . the nonperturbative contribution is proportional to the second inverse moment ( [ eq:4qterm ] ) of the distribution @xmath23 of the vacuum - quark virtualities and is defined by the behavior of the quark condensate at large and moderate distances between the vacuum quarks . the results for the derivative of the pion da @xmath133 are shown in table [ tab : all - derivatives ] for the delta - ansatz model ( [ eq : ansv ] ) and also for the two - tier smooth model ( [ eq : smoothmodel ] ) with the model parameters @xmath183 , @xmath199 . the dependence of the derivative of the pion da @xmath133 on the choice of the model parameters @xmath185 and @xmath179 is illustrated in fig . [ fig : smoothsr.l.n ] . as we see from table [ tab : all - derivatives ] , the results from the differential ( [ eq : dpiondaqcdsr ] ) and the integral srs ( [ eq : piondaqcdsr : dk ] ) agree with each other . the integral derivative of the pion da , based on a new sr derived in this work , remains smaller than the asymptotic value and overlaps with the range of values determined with the bms bunch of pion das , while there is no agreement with the cz da and the flat - type models considered . the same conclusions can be drawn also for the usual derivative of the pion da , which follows from the differential sr ( [ eq : dpiondaqcdsr ] ) . it is worth mentioning that employing the integral and the differential sum rules , we found that the leading - order qcd sum rules ( [ eq : piondaqcdsr ] ) , which employ the minimal gaussian model for the nonlocal condensates , can not be satisfied by flat - type pion distribution amplitudes . given that an increasing behavior of the scaled pion - photon transition form factor can only be achieved with flat - type pion das , the independent experimental confirmation of this effect , e.g. , by the belle collaboration , becomes extremely crucial for our theoretical understanding of basic qcd exclusive processes . we would like to thank alexander bakulev for stimulating discussions and useful remarks . a.p . is indebted to prof . maxim polyakov for the warm hospitality at bochum university , where most of this work was carried out . a.p . and s.m . acknowledge financial support from nikolay rybakov . this work received partially support from the heisenberg landau program under grants 2009 , and 2010 , the russian foundation for fundamental research ( grants no . 07 - 02 - 91557 , no . 08 - 01 - 00686 , and no . 09 - 02 - 01149 ) , and the brfbr - jinr cooperation program , contract no . a.p . wishes to thank the ministry of education and science of the russian federation ( `` development of scientific potential in higher schools '' projects : no . 2.2.1.1/1483 and no . 2.1.1/1539 ) , and the daad foundation for a research scholarship . [ app : a.1 ] in sections [ sec : qcd - vac ] and [ sec : pi - slope ] , we used the following expressions for the vacuum distribution functions @xmath208 the meaning of these expressions and their connection to the initial nlcs has been discussed in detail in refs . the contributions to the qcd sr ( [ eq : piondaqcdsr ] ) with nonlocal condensates , @xmath209 , associated with these expressions , are shown below . here , and in what follows , we use @xmath210 , @xmath211 . then we obtain @xmath212 + { \left(\bar x\rightarrow x\right)}+ \right . \nonumber \\ & & \qquad\qquad \left . + \theta(1>\delta)\theta\left(\delta > x>\bar\delta\right ) \left[\bar\delta + \left(\delta-2\bar xx\right)\ln(\delta)\right ] \right\ } \ , \label{eq : phi_s } \\ \delta\phi_v\left(x , m^2\right ) & = & \frac{a_s}{m^4 } \left(x\delta'\left(\bar x-\delta\right)+{\left(\bar x\rightarrow x\right)}\right)\ , \\ \delta\phi_{t}\left(x , m^2\right ) & = & \delta\phi_{t_1}\left(x , m^2\right ) + \delta\phi_{t_2}\left(x , m^2\right ) + \delta\phi_{t_3}\left(x , m^2\right)\ , \nonumber\end{aligned}\ ] ] @xmath213 \left(\frac1{\delta } - 2\right)\theta(1>2\delta ) + \theta(2\delta > x ) { \nonumber}\right.\\ & & \left . \times \theta(x>\delta)\theta(x>3\delta-1 ) \frac{\bar x}{\bar\delta } \left[\frac{3x}{\delta } -6 - \frac{1+\bar x}{\bar\delta}\right ] \right\ } + { \left(\bar x\rightarrow x\right)}\ , \label{eq : phi_t1 } \\ \delta\phi_{t_2}\left(x , m^2\right ) & = & \frac{4 a_s}{m^4}\bar x \left\{\frac{\delta(x-2\delta)}{\delta}\theta(1>2\delta ) -\theta(2\delta > x)\theta(x>\delta)\theta(x>3\delta-1 ) \right . \nonumber \\ & & \left . \times \frac{1 + 2x-4\delta}{\bar\delta\delta^2}\right\ } + { \left(\bar x\rightarrow x\right)}\ , \label{eq : phi_t2 } \\ \delta\phi_{t_3}\left(x , m^2\right ) & = & \frac{3 a_s\bar x}{m^4\bar\delta\delta } \left\{\theta(2\delta > x)\theta(x>\delta)\theta(x>3\delta-1 ) \left[2-\frac{\bar x}{\bar\delta}-\frac{\delta}{\bar\delta } \right]\right\ } { \nonumber}\\ & & + { \left(\bar x\rightarrow x\right)}\ , \\ \delta\phi_g\left(x , m^2\right ) & = & \frac{\langle \alpha_s gg \rangle}{24\pi m^2 } \left(\delta\left(x-\delta\right)+ { \left(\bar x\rightarrow x\right)}\right)\ . \label{eq : phi.g}\end{aligned}\ ] ] in the above equations , we used the abbreviation @xmath214 , while for the quark and the gluon condensates the standard estimates @xcite @xmath215 gev@xmath216 , @xmath217 gev@xmath218 and @xmath219 gev@xmath16 , normalized at @xmath220 gev@xmath16 , have been adopted . in sec . [ subsec : dif - sr ] , we applied not only the delta ansatz , discussed above , but also the smooth model , proposed in refs.@xcite , which is based on an exponential decay of the condensate ( [ eq : smooth - x ] ) . for this reason , we use the model - independent expression for the four - quark contribution @xcite to obtain eq . ( [ eq:4qterm ] ) @xmath221 \\\nonumber & & + { \left(\bar x\rightarrow x\right)}\ , \end{aligned}\ ] ] where @xmath222 , @xmath223 , and @xmath224 . encoding nonperturbative input in ( [ eq : piondaqcdsr ] ) . [ fig:4q],scaledwidth=40.0% ] [ app : damodels ] in our work we compared the results following from the differential sr ( [ eq : dpiondaqcdsr ] ) for the usual derivative of the pion da , @xmath133 , with those obtained with the help of the integral derivative , @xmath205(x)$ ] , i.e. , ( [ eq : piondaqcdsr : dk ] ) , evaluating them for the bms da bunch , the cz model , and a couple of flat - type pion das . the bms bunch of pion das was obtained @xcite as the sum of the first three terms in the gegenbauer - polynomial expansion ( [ eq : pion - da ] ) . their gegenbauer coefficients were obtained from an analysis of the sr ( [ eq : piondaqcdsr ] ) for the @xmath225-moments ( @xmath226 ) with the nonlocal condensate contributions being presented in appendix [ app : a.1 ] . it was found that only the two first gegenbauer coefficients @xmath227 and @xmath228 give sizeable contributions to the @xmath225-moments , whereas the higher gegenbauer terms give merely tiny contributions . for this reason , the bms bunch das are two - parameter models . the central values of @xmath227 and @xmath228 for the whole bms bunch in the @xmath229 space are @xmath230 and @xmath231 using as a normalization scale @xmath144 gev@xmath16 @xcite . on the other hand , the cz pion da contains only the first nontrivial gegenbauer polynomial corresponding to the coefficient @xmath227 @xcite . it was derived from qcd srs for the @xmath232-moments ( @xmath233 ) , using local condensates . for the sake of consistency , we use here a value of @xmath234 obtained after evolution to the normalization scale @xmath144 gev@xmath16 , see for more details @xcite . [ app : a.2 ] the spectral density @xmath237 with nlo accuracy was obtained in @xcite and was found to be @xmath238 \frac{1}{2\pi^2 } \ . \label{eq : rho - nlo}\end{aligned}\ ] ] for our numerical calculations , which take into account radiative corrections , we use the following function -dependence of the function @xmath239 for @xmath97 ( solid blue line ) , @xmath98 ( dashed green line ) , and @xmath99 ( dashed - dotted red line ) . [ fig : deltanlovslo],scaledwidth=45.0% ] @xmath240\left(x\right ) } { [ d^{(\nu+2)}\rho_\text{pert}^\text{lo } ] \left(x\right ) } -100 \ . \label{eq : dnx}\end{aligned}\ ] ] the @xmath56-dependence of the function @xmath239 is illustrated in fig . [ fig : deltanlovslo ] for different values of @xmath61 : solid blue line@xmath97 , dashed green line@xmath98 , and dashed - dotted red line@xmath99 . in order to ensure a weak dependence of the results on the particular model for the condensates adopted , and in order to include all condensate contributions , one has to construct the sr by integrating the pion da sr ( [ eq : piondaqcdsr ] ) over a large enough interval of @xmath56 . for this reason , we introduced in sec . [ subsec : int - sr ] the integral derivatives @xmath241(x)$ ] , with the two lowest - order ones being given in terms of eq . ( [ eq : int - dev012 ] ) . the next higher derivative reads @xmath242(x ) = \frac{1}{x}\int\limits_0^x\frac{dy}{y } \int\limits_0^y\frac{\varphi(t)}{t}\,dt = \frac{1}{x}\int\limits_0^x\frac{\varphi(t)}{t}dt \int\limits_t^x\frac{dy}{y } = \frac{1}{x}\int\limits_0^x\frac{\varphi(t)}{t } \ln\left(\frac{x}{t}\right)dt \ . \label{eq : d-3}\end{aligned}\ ] ] assuming @xmath54 , we find @xmath55(x ) = \frac{1}{x}\int\limits_0^x\!dy\,[d^{(n)}\varphi](x ) \ . \label{eq : d - n}\end{aligned}\ ] ] to obtain an expression for @xmath243(x)$ ] for any @xmath244 and @xmath245 , one has to rearrange the integration order to get @xmath246(x ) & = & \frac{1}{x}\int\limits_0^x\!\!\frac{dy_n}{y_n } \int\limits_0^{y_n}\!\!\frac{dy_{n-1}}{y_{n-1 } } \ldots \int\limits_0^{y_2}\!\!\frac{dy_{1}}{y_{1 } } \int\limits_0^{y_1}\!\!\frac{\varphi(t)}{t}dt = \frac{1}{x}\int\limits_0^{x}\!\!\frac{\varphi(t)}{t } \frac{1}{n!}\left(\!\int\limits_t^x\!\!\frac{dy}{y}\right)^n\!\!dt \ . \label{eq : d - n-2}\end{aligned}\ ] ] this can be readily generalized to the expression ( [ eq : d.nu.x ] ) using ( [ eq : f.n.x ] ) that allows one to establish this transformation for any real @xmath61 , @xmath64 . the dependence on @xmath49 of the function @xmath70 is shown in fig . [ fig : fykx ] for @xmath247 and @xmath248 . as one sees from eq . ( [ eq : d.nu.x ] ) , the function @xmath70 acts as a smooth projector to the area around the origin of @xmath49 . if a taylor expansion exists for @xmath58 at @xmath59 , then by applying ( [ eq : d.nu.x ] ) one finds @xmath250(x ) = \varphi'(0)+\varphi''(0)\frac{x}{2!2^{k+1 } } + \sum\limits_{n=2}^\infty\varphi^{(n+1)}(0 ) \frac{x^n}{(n+1)!(n+1)^{k+1 } } \ . \label{eq : d.k.x.seriesa}\end{aligned}\ ] ] using ( [ eq : d.nu.x ] ) in combination with ( [ eq : d.k.x.seriesa ] ) , one can obtain the following properties of the operator @xmath62 : @xmath251(x ) = \lim\limits_{\nu\to \infty } [ d^{(\nu)}\varphi](x ) = \varphi'(0 ) $ ] and @xmath252 . moreover , from ( [ eq : d.k.x.seriesa ] ) it follows that the introduced operator @xmath253(x)$ ] reproduces at small @xmath56 and large @xmath61 the derivative of @xmath58 at @xmath59 . a. v. radyushkin , ubna preprint p2 - 10717 , 1977 [ hep - ph/0410276 ] ( unpublished ) . g. p. lepage and s. j. brodsky , phys . lett . * b87 * , 359 ( 1979 ) ; phys . d22 * , 2157 ( 1980 ) ; + a. v. efremov and a. v. radyushkin , phys . lett . * b94 * , 245 ( 1980 ) ; theor . phys . * 42 * , 97 ( 1980 ) . v. l. chernyak and a. r. zhitnitsky , phys . rept . * 112 * , 173 ( 1984 ) . s. v. mikhailov and a. v. radyushkin , jetp lett . * 43 * , 712 ( 1986 ) ; sov . . phys . * 49 * , 494 ( 1989 ) . s. v. mikhailov and a. v. radyushkin , sov . * 52 * , 697 ( 1990 ) . a. p. bakulev and a. v. radyushkin , phys . lett . * b271 * , 223 ( 1991 ) . s. v. mikhailov and a. v. radyushkin , phys . rev . * d45 * , 1754 ( 1992 ) . l. del debbio , few body syst . * 36 * , 77 ( 2005 ) . v. et al . _ , phys . rev . * d74 * , 074501 ( 2006 ) . m. a. donnellan _ et al . _ , pos * lat2007 * , 369 ( 2007 ) . a. p. bakulev , s. v. mikhailov , and n. g. stefanis , phys . lett . * b508 * , 279 ( 2001 ) . * b590 * , 309(e ) ( 2004 ) . f. del aguila and m. k. chase , nucl . phys . * b193 * , 517 ( 1981 ) . + e. p. kadantseva , s. v. mikhailov , and a. v. radyushkin , sov . 44 * , 326 ( 1986 ) . p. kroll and m. raulfs , phys b387 * , 848 ( 1996 ) . a. v. radyushkin and r. ruskov , nucl . phys . * b481 * , 625 ( 1996 ) . i. v. musatov and a. v. radyushkin , phys . rev . * d56 * , 2713 ( 1997 ) . n. g. stefanis , w. schroers , and h .- c . kim , phys . b449 * , 299 ( 1999 ) . a. schmedding and o. yakovlev , phys . rev . * d62 * , 116002 ( 2000 ) . n. g. stefanis , w. schroers , and h .- c . kim , eur . j. * c18 * , 137 ( 2000 ) . b. meli , b. nii , and k. passek , phys . d65 * , 053020 ( 2002 ) . m. diehl , p. kroll , and c. vogt , eur . j. * c22 * , 439 ( 2001 ) . b. meli , d. mller , and k. passek - kumeriki , phys . rev . * d68 * , 014013 ( 2003 ) . a. p. bakulev , s. v. mikhailov , and n. g. stefanis , phys . rev . * d67 * , 074012 ( 2003 ) . a. p. bakulev , s. v. mikhailov , and n. g. stefanis , phys . lett . * b578 * , 91 ( 2004 ) . s. s. agaev , phys . * d72 * , 114010 ( 2005 ) ; _ ibid . _ * d73 * , 059902(e ) ( 2006 ) . a. p. bakulev , s. v. mikhailov , and n. g. stefanis , phys . rev . * d73 * , 056002 ( 2006 ) . n. g. stefanis , nucl . * 181182 * , 199 ( 2008 ) . s. v. mikhailov and n. g. stefanis , nucl . * b821 * , 291 ( 2009 ) . h. j. behrend _ _ , z. phys . * c49 * , 401 ( 1991 ) . j. gronberg _ _ , phys . rev . * d57 * , 33 ( 1998 ) . b. aubert _ et al . _ , phys . rev . * d80 * , 052002 ( 2009 ) . s. v. mikhailov and n. g. stefanis , mod . a24 * , 2858 ( 2009 ) . invited talk presented by the first author at workshop `` recent advances in perturbative qcd and hadronic physics '' , 2025 july 2009 , ect * , trento , italy , in honor of prof . anatoly efremov s 75th birthday . s. v. mikhailov and n. g. stefanis , nucl . * b198 * , 199 ( 2010 ) . invited talk given by the first author at the 3rd joint international hadron structure09 conference , tatranska strba ( slovak republic ) , aug . 30sept . 3 , 2009 . a. v. radyushkin , phys . d * 80 * ( 2009 ) 094009 . m. v. polyakov , jetp lett . * 90 * , 228 ( 2009 ) . a. v. radyushkin , in _ perspectives in hadronic physics : proceedings of the ictp conference , 1216 may 1997 , trieste , italy _ , edited by s. boffi , c. c. d. atti , and m. giannini ( world scientific , singapore , 1997 ) , pp . a. p. bakulev and s. v. mikhailov , phys . rev . * d65 * , 114511 ( 2002 ) . a. p. bakulev , k. passek - kumeriki , w. schroers , and n. g. stefanis , phys . rev . * d70 * , 033014 ( 2004 ) ; _ ibid . _ * d70 * , 079906(e ) ( 2004 ) . a. p. bakulev , s. v. mikhailov , and n. g. stefanis , annalen phys . * 13 * , 629 ( 2004 ) . a. p. bakulev , s. v. mikhailov , and n. g. stefanis , in _ proceedings of the international workshop on light cone physics : hadrons and beyond , durham , uk , august 5th9th , 2003 _ , edited by s. dalley ( institute for particle physics phenomenology , university of durham , 2003 ) , pp . 172177 . s. j. brodsky and g. f. de teramond , phys . rev . * d77 * , 056007 ( 2008 ) . h. r. grigoryan and a. v. radyushkin , phys . rev . * d78 * , 115008 ( 2008 ) . h. j. kwee and r. f. lebed , phys . rev . * d77 * , 115007 ( 2008 ) . s. s. agaev and m. a. gomshi nobary , phys . rev . * d77 * , 074014 ( 2008 ) . g . wu and t. huang , arxiv:1005.3359 [ hep - ph ] . v. l. chernyak and a. r. zhitnitsky , nucl . * b201 * , 492 ( 1982 ) . a. p. bakulev and a. v. pimikov , phys . . lett . * 4 * , 377 ( 2007 ) . a. v. radyushkin , phys . lett . * b271 * , 218 ( 1991 ) . a. p. bakulev and s. v. mikhailov , mod . a11 * , 1611 ( 1996 ) . a. v. radyushkin , in _ continuous advances in qcd : proceedings of the workshop , 1820 feb 1994 , minneapolis , minnesota _ , edited by a. v. smilga ( world scientific , river edge , n.j , 1994 ) , pp . . m. a. shifman , a. i. vainshtein , and v. i. zakharov , nucl . phys . * b147 * , 385 ( 1979 ) . a. p. bakulev and s. v. mikhailov , z. phys . * c68 * , 451 ( 1995 ) .

motivated by applications in quantum information and quantum control , a new type of `` numerical range , the relative '' numerical range denoted , is introduced . it arises upon replacing the unitary group in the definition of the classical " numerical range by any of its compact and connected subgroups . the geometric properties of the relative `` numerical range are analysed in detail . counterexamples prove its geometry is more intricate than in the classical case : e.g. is neither star - shaped nor simply - connected . yet , a well - known result on the rotational symmetry of the classical '' numerical range extends to , as shown by a new approach based on lie theory . furthermore , we concentrate on the subgroup , i.e. the-fold tensor product of , which is of particular interest in applications . in this case , sufficient conditions are derived for being a circular disc centered at origin of the complex plane . finally , the previous results are illustrated in detail for .

math-ph0702005

@xmath0 `` numerical ranges and @xmath0 '' numerical radii @xcite naturally relate to optimization problems on the unitary orbit of a given matrix . a particularly interesting class of applications concerns quantum systems as shown in more detail in the accompanying paper @xcite . this is because quantum dynamics and quantum control of closed hamiltonian systems are governed by the unitary group of the underlying hilbert space and its subgroups . for instance , in quantum information and quantum computation , where spin-@xmath9 two - level systems are exploited as elementary units called quantum bits or qubits for short the full dynamics of @xmath5 qubits take place on the unitary group @xmath10 . this includes two types of coherent time evolutions , ( i ) _ within _ each individual qubit independently as well as ( ii ) _ between _ coupled qubits . the former is brought about by so - called _ local actions_. hence , we refer to the corresponding subgroup as the subgroup of local time evolution @xmath11 . it takes the form of an @xmath5-fold tensor product of @xmath6 with itself . clearly , matrices ( states ) that are themselves @xmath5-fold tensor - products remain in this product form under conjugation with elements of @xmath11 . therefore , quantum optimization problems which can be expressed via local actions motivate in a most natural way the study of a subset of the classical @xmath0 " numerical range , namely the _ local @xmath0 " numerical range_. it is but a special instance of the more general structure which we introduce here as the _ relative @xmath0 " numerical range _ , where the full unitary group is restricted to some arbitrary compact connected subgroup . to be mathematically precise , let @xmath2 denote the compact and connected lie group of all unitary matrices of size @xmath12 , i.e. @xmath13 if and only if @xmath14 with @xmath15 , where @xmath16 stands for the conjugate transpose . moreover , let @xmath17 be the subgroup of all special unitary matrices , i.e. @xmath18 if and only if @xmath13 with @xmath19 . finally , let @xmath20 be the @xmath5-fold tensor product of @xmath6 , which consists of all @xmath5-fold kronecker products of the from @xmath21 for all @xmath22 . following terminology from quantum information as mentioned above we refer to @xmath11 as the subgroup of _ local _ unitary transformations . now , for arbitrary complex matrices @xmath23 the _ local @xmath0 " numerical range _ of @xmath24 is defined as @xmath25 obviously , @xmath26 is a subset of the classical @xmath0 " numerical range of @xmath24 , denoted @xmath27 . however , its geometry is significantly more intricate than in the classical case and has yet not been studied in any systematic way , cf . analysing its basic properties leads naturally to the following slightly more general object @xmath28which we call the _ relative @xmath0 " numerical range _ of @xmath24 . here , @xmath29 may be any compact and connected subgroup of @xmath2 . clearly , @xmath1 is invariant under unitary similarities of @xmath24 and @xmath0 by elements of @xmath29 . moreover , with @xmath1 being a continuous image of @xmath29 , it is compact and connected . however , in contrast to the classical @xmath0 " numerical range , cf . @xcite , counterexamples show that @xmath1 is neither star - shaped nor simply connected . in general , the actual size and shape of @xmath1 are mostly unknown . the purpose of the present paper is to initiate the study of the relative @xmath0 `` numerical range and in particular its geometry . starting points for our investigations are well - known results on the classical @xmath0 '' numerical range . therefore , we have summarized the fundamental geometric properties of @xmath27 at the beginning of section [ sec:2 ] . subsequently , the above mentioned counterexamples and straightforward consequences of ( [ def : relnumrange ] ) are collected in subsection [ subsec:2.1 ] . the main result , a generalisation on rotationally symmetric @xmath0-umerical ranges , is presented in subsection [ subsec:2.2 ] . particularly , the original work by li and tsing @xcite is recovered as a immediate corollary . in section [ sec:3 ] we return to local @xmath0 `` numerical ranges . here , the central conclusion is that rotationally symmetric local @xmath0 '' numerical ranges are essentially circular discs . finally , we illustrate the previous results by determining all complex @xmath30-matrices , the local @xmath0 " numerical range of which is a circular disc centered at the origin of the complex plane . these computations lead to a conjecture on unitray similarity to block - shift form of such matrices . some concluding remarks on numerical issuse and perspectives for future work can be found in section [ sec:4 ] . in section [ intro ] , we introduced the @xmath0 " numerical range of @xmath31 relative to a compact and connected subgroup @xmath29 of @xmath2 as follows @xmath32 for simplicity , we call @xmath33 _ the relative @xmath0 " numerical range _ whenever any misinterpretation is excluded . below , we aim at a better understanding of the geometry of the relative @xmath0 `` numerical range . in particular , we focus on geometric properties of the classical @xmath0 '' numerical range which are preserved by passing to the relative one . for this reason , we first recall some basic facts about the classical @xmath0 " numerical range . let @xmath34 be any complex matrix . the classical numerical and @xmath0 " numerical range of @xmath35 are defined by @xmath36 and @xmath37 respectively . since any unitary transformation @xmath38 can be factored in the form @xmath39 with @xmath40 , the set @xmath27 does not change , if we restrict @xmath38 to @xmath17 . moreover , if @xmath41 for some @xmath42 with @xmath43 we have @xmath44 hence @xmath27 coincides with @xmath45 , if @xmath0 is of the above form . here , we used the fact , that @xmath2 acts transitively on the unit sphere of @xmath46 . this will be of interest later on , if one wants to define a numerical range of @xmath24 relative to @xmath29 , see also section [ sec:3 ] , remark [ rem:3.1 ] . probably the most basic geometric properties of the numerical and @xmath0 `` numerical range are compactness and connectedness , both following from the fact that @xmath45 and @xmath27 are continuous images of compact and connected sets . the first deep result on the geometry of the numerical range , its convexity , was obtained independently by hausdorff @xcite and toeplitz @xcite . for a proof in a modern textbook we refer to @xcite . later , the result was extended by westwick @xcite and poon @xcite to @xmath0 '' numerical ranges for one of the operators @xmath24 or @xmath0 being hermitian . yet , in general @xmath0 " numerical ranges are non - convex except for @xmath47 , and even for normal matrices convexity may fail for @xmath48 , cf . however , by a result of cheung and tsing @xcite , they are always star - shaped with respect to the star - center @xmath49 . the above mentioned theorems describe features which are in common to all @xmath0 `` numerical ranges . another type of results aims at characterising @xmath0 '' numerical ranges @xmath27 of particularly simple form such as an interval or a circular disc . here , we mention only two results in this direction . [ thm : hermitian_case ] if @xmath50 are hermitian , then @xmath27 degenerates to a compact interval @xmath51 $ ] on the real line with @xmath52 where @xmath53 and @xmath54 are the eigenvalues of @xmath24 and @xmath0 , respectively . [ thm : disc ] let @xmath31 and let @xmath55 denote the unitary orbit of @xmath24 . the following statements are equivalent : 1 . the unitary orbit of @xmath24 satisfies @xmath56 for all @xmath57 . the @xmath0 " numerical range of @xmath24 satisfies @xmath58 for all @xmath34 and all @xmath57 , i.e. @xmath27 is rotationally symmetric for all @xmath34 . the @xmath24-numerical range of @xmath24 satisfies @xmath59 for all @xmath57 , i.e. @xmath60 is rotationally symmetric . the matrix @xmath24 is unitarily similar to a block matrix @xmath61 such that all @xmath62 are square blocks and @xmath63 if @xmath64 . @xmath46 can be factored into a direct sum of mutually orthogonal subspaces @xmath65 such that @xmath66 for @xmath67 and @xmath68 . the @xmath0 " numerical range @xmath27 is a circular disc in the complex plane centered at the origin for all @xmath34 . @xmath60 is a circular disc in the complex plane centered at the origin . theorem [ thm : hermitian_case ] can be traced back to von neumann . for a proof we refer to @xcite . theorem [ thm : disc ] is a resume of theorem 2.1 and corollary 2.2 in @xcite . more on @xmath0 `` numerical ranges and the @xmath0 '' numerical radii can be found in a special issue of _ linear and multilinear algbra _ by ando and li @xcite . [ rem:2.1 ] it is essential in theorem [ thm : disc ] to require the symmetry condition in statement ( b ) and ( f ) _ for all _ @xmath34 . in fact , @xmath27 can be a circular disc , although neither @xmath24 nor @xmath0 satisfy any of the above conditions , for instance , if @xmath24 is hermitian and @xmath0 skew - hermitian . our starting points for analysing the geometry of the relative @xmath0 " numerical range are the above mentioned facts on @xmath27 . obviously , @xmath1 is compact and connected by the same argument as before . however , the previous results on convexity and star - shapedness fail as the following three counterexamples show . in particular , the first and third one prove that @xmath1 is in general not even simply connected . here , the first example exploits the fact that the subgroup @xmath29 itself need not be simply connected , while the second and third one are based on the fact that the product of convex or star - shaped subsets of @xmath69 is in general neither convex nor star - shaped . [ ex1 ] let @xmath70 then we have @xmath71 obviously , @xmath1 is not simply connected and therefore neither star - shaped nor convex . however , this is not really surprising , since @xmath72 is a closed one - parameter subgroup and hence itself not simply connected . [ ex2 ] let @xmath73 and let @xmath74 by the trace identity @xmath75 for all @xmath31 and @xmath76 it is easy to see that @xmath77 . moreover , as @xmath78 is normal and @xmath79 is hermitian for @xmath80 , a straightforward calculation yields @xmath81 & \quad\mbox{and}\quad & w(c_2,a_2 ) = \{1 + { { \rm i}}b \;|\ ; -1 \leq b \leq 1\}.\end{aligned}\ ] ] hence we obtain @xmath82 which is obviously not convex . the relative @xmath0 " numerical ranges of examples [ ex2 ] and [ ex3].,scaledwidth=45.0% ] [ ex3 ] let @xmath83 and let @xmath84 as in example [ ex2 ] , we have @xmath85 and @xmath86\ } \quad\mbox{for}\quad j=1,2,3.\ ] ] thus , it is easy to see that the equilateral triangle @xmath87 forms the outer boundary of @xmath1 , while the origin and a large part of the interior of @xmath88 does not belong to @xmath1 . therefore @xmath1 is not simply connected and hence not star - shaped either . in both figures , the outer boundary of @xmath1 has been computed analytically and drawn in for a better visualization . in some sense , example [ ex3 ] is more significant than [ ex2 ] . nevertheless , we presented example [ ex2 ] , because it seems to be impossible to imitate [ ex3 ] in dimension @xmath89 . this raises the question which is the lowest dimension @xmath12 for which a compact , simply connected subgroup @xmath3 and matrices @xmath24 , @xmath0 exist such that @xmath1 is not simply connected . moreover , the reader should note that both can be interpreted as relative numerical ranges , cf . remark [ rem:3.1 ] of section [ sec:3 ] . having seen that most of the geometric properties of the classical @xmath0 " numerical range are not preserved , we present a series of results which guarantee that some geometric features of @xmath27 are passed to @xmath1 under additional assumptions on @xmath24 , @xmath0 , and @xmath29 . the first one is a trival fact based on the properties of the hermitian form @xmath90 . [ hermitianlem ] let @xmath29 be any compact and connected subgroup of @xmath2 and let @xmath91 be hermitian . then the @xmath0 " numerical range of @xmath24 is a compact subinterval of @xmath92 . the identity @xmath93 for all hermitian matrices @xmath91 implies immediately the above assertion . the next two propositions focus on special types of subgroups direct sums and tensor products which are of particular interest for applications , cf . both results show that the properties of the classical @xmath0 " numerical range are maintained , if the matrices @xmath91 are compatible with the corresponding subgroup structure . [ prop : directsum ] let @xmath94 and @xmath95 be compact and connected subgroups and let @xmath29 be the direct sum of @xmath96 and @xmath97 , i.e. @xmath98 , where @xmath99 is defined by @xmath100 moreover , let @xmath101 and @xmath102 with @xmath103 for @xmath104 . then the @xmath0 " numerical range of @xmath24 is given by @xmath105 in particular , @xmath1 is star - shaped ( convex ) , if @xmath106 and @xmath107 are star - shaped ( convex ) . the result follows from @xmath108 and the fact that the sum of two star - shaped ( convex ) sets is again star - shaped ( convex ) . [ prop : tensorproduct ] let @xmath94 and @xmath95 be compact and connected subgroups and let @xmath29 be the tensor product of @xmath96 and @xmath97 , i.e. @xmath109 , where @xmath110 denotes the kronecker product of @xmath111 and @xmath112 . moreover , let @xmath113 and @xmath114 with @xmath103 for @xmath104 . then the @xmath0 " numerical range of @xmath24 is given by @xmath115 in particular , @xmath1 is star - shaped ( convex ) , if either @xmath106 is star - shaped ( convex ) and @xmath116 is contained in a ray @xmath117 of the complex plane or vice versa . ( [ eq : product ] ) is obtained by the trace identity @xmath118 . moreover , @xmath7 is star - shaped ( convex ) , if @xmath106 is , as the following two operations preserve star - shapedness and convexity : ( a ) rotation by a fixed complex number , ( b ) multiplication by a subinterval of @xmath119 , cf . appendix , lemma [ lem : starshaped ] . the following corollary is an immediate consequence of the previous results . let @xmath29 , @xmath24 and @xmath0 either be defined as in proposition [ prop : directsum ] or as in proposition [ prop : tensorproduct ] and furthermore , let @xmath120 be hermitian for @xmath104 . then @xmath121 \vspace{-4mm}\ ] ] or @xmath122,\ ] ] respectively , where @xmath123 $ ] and @xmath124 $ ] . this subsection contains the main result of the paper a generalisation of the circular - disc theorem by li and tsing @xcite , cf . theorem [ thm : disc ] . more precisely , we characterize all matrices @xmath31 , the relative @xmath0 `` numerical range of which is rotationally symmetric for all @xmath0 . but first , we illustrate by a counterexample that the _ relative _ @xmath0 '' numerical range of a block - shift matrix as li and tsing s result might suggest is in general not rotationally symmetric . [ ex4 ] let @xmath73 and let @xmath125 we claim that @xmath1 is not circular , although its classical @xmath0 " numerical range is a circular for all @xmath126 . a rigorous proof of this fact will be given after corollary [ corlocrange ] . moreover , we note that easier examples can be constructed via the subgroup of example [ ex1 ] . however , we favoured the above one as it will be of interest also in section [ sec:3 ] . yet , for the main result and its proof , we need some further notation . we call the @xmath29-orbit @xmath127 of @xmath31 _ weakly rotationally symmetric _ , if @xmath128 holds for all @xmath57 . moreover , let @xmath129 denote the lie algebra of @xmath29 and let @xmath130 for @xmath131 be the operator defined by @xmath132 : = \omega a - a \omega.\ ] ] finally , a maximal abelian subalgebra of @xmath129 is called a torus algebra of @xmath29 . [ prop : rotsym ] let @xmath31 and let @xmath133 . then the following statements are equivalent : 1 . the orbit @xmath134 satisfies the relation @xmath135 . 2 . the relative @xmath0 " numerical range of @xmath24 satisfies @xmath136 for all @xmath34 . 3 . the relative @xmath24-numerical range of @xmath24 satisfies @xmath137 . if @xmath138 is an irrational multiple of @xmath139 and if one of the above statements ( a ) , ( b ) , or ( c ) holds for @xmath138 , then all of them are satisfied for all @xmath57 . the implications ( a ) @xmath140 ( b ) and ( b ) @xmath140 ( c ) are obvious . for proving ( c ) @xmath140 ( a ) we assume without loss of generality @xmath141 . now , there is a @xmath142 such that @xmath143 moreover , using the cauchy - schwarz inequality and the unitary invariance of the frobenius norm we obtain @xmath144 thus , in fact equality holds in eq . ( [ eq : cauchyschwarz ] ) and therefore we have @xmath145 for some @xmath146 with @xmath147 . substituting @xmath145 in eq . ( [ eq2.1 ] ) yields @xmath148 and hence @xmath149 . this finally implies ( a ) . if one of the above statements ( a ) , ( b ) or ( c ) holds for some @xmath150 , we know that all of them are satisfied for @xmath138 . in particular , by ( a ) we have @xmath151 for some @xmath142 . thus , by induction we obtain @xmath152 for all @xmath153 . now , compactness of @xmath29 and eq . ( [ eq2.2 ] ) imply that for all @xmath57 there exists a @xmath154 such that @xmath155 and thus @xmath156 for all @xmath57 . hence ( a ) and therefore also ( b ) and ( c ) are satisfied for all @xmath57 . [ rem:2 ] proposition [ prop : rotsym ] is a slight generalization of the first part of theorem 2.1 in @xcite . the above proof , which is almost the same in @xcite , is included only for completeness . as an immediate consequence of the above proposition we obtain . [ cor : rotsym ] let @xmath157 be compact and connected subgroups of @xmath2 . 1 . the relative @xmath0 " numerical range of @xmath24 is rotationally symmetric for all @xmath34 if and only if the orbit @xmath134 is weakly rotationally symmetric . 2 . if the relative @xmath0 `` numerical range @xmath1 is rotationally symmetric for all @xmath34 , then the relative @xmath0 '' numerical range @xmath158 is as well . \(a ) @xmath159 \(b ) this follows from ( a ) and the fact that weak rotational symmetry of @xmath134 implies weak rotational symmetry of @xmath160 . now , by the previous definitions our main theorem reads as follows . [ thm : rotsym ] let @xmath29 be a compact and connected subgroup of @xmath2 with lie algebra @xmath129 and let @xmath161 be a torus algebra of @xmath129 . then the orbit @xmath134 of @xmath31 , @xmath141 is weakly rotationally symmetric if and only if 1 . there exists @xmath131 such that @xmath24 is an eigenvector of the operator @xmath162 to a non - zero eigenvalue , or equivalently , 1 . there exist @xmath154 and @xmath163 such that @xmath164 is an eigenvector of the operator @xmath165 to a non - zero eigenvalue . before proving theorem [ thm : rotsym ] , the subsequent lemma will clarify the relation between the operator @xmath162 and the weak rotational symmetry of the orbit @xmath134 . [ lemrotsym ] for @xmath166 and @xmath167 let @xmath168 then @xmath169 is equal to the eigenspace of the operator @xmath162 to the eigenvalue @xmath170 , i.e. @xmath171 `` @xmath172 '' : let @xmath173 . then @xmath174 holds for all @xmath175 . hence differentiating eq . ( [ eq2.3 ] ) on both sides , yields @xmath176 for @xmath177 . therefore , @xmath24 is an eigenvector of @xmath162 to the eigenvalue @xmath170 . `` @xmath178 '' : now , let @xmath24 be an eigenvector of @xmath162 to the eigenvalue @xmath170 . then we consider the curve @xmath179 . an easy calculation shows that @xmath180 satisfies the linear differential equation @xmath181 by uniqness of solutions of eq . ( [ eq2.4 ] ) we have @xmath182 and thus @xmath173 . \(a ) `` @xmath183 '' : let @xmath131 and assume that @xmath24 is an eigenvector of the operator @xmath162 to some eigenvalue @xmath184 . then @xmath185 is purely imaginary , i.e. @xmath186 for some @xmath187 , since @xmath162 is skew - hermitian with respect to the scalar product @xmath188 . therefore , lemma [ lemrotsym ] implies @xmath189 and this shows that @xmath134 is weakly rotationally symmetric . `` @xmath190 '' : now , let @xmath134 be weakly rotationally symmetric and choose @xmath191 . thus there is a @xmath142 such that @xmath192 and hence @xmath193 now , we consider the closed abelian subgroup @xmath194 generated by @xmath195 . since @xmath29 is compact , @xmath194 is the direct product of a torus @xmath196 and a finite subgroup , c.f . @xcite . continuity and eq . ( [ eq2.6 ] ) imply , that for all @xmath197 there exists a unique @xmath198 depending on @xmath38 such that @xmath199 furthermore , let @xmath200 be a basis of the lie algebra of @xmath196 such that the following conditions hold : @xmath201 for all @xmath202 . since @xmath203 for some @xmath153 , we can assume @xmath204 . hence , there are @xmath205 such that @xmath206 . moreover , by eq . ( [ eq2.7 ] ) there exist @xmath207 for @xmath202 such that @xmath208 now , from eq . ( [ eq2.5 ] ) it follows that at least one @xmath209 is an irrational multiple of @xmath139 . without loss of generality let @xmath210 . hence , we obtain @xmath211 and thus @xmath212 for all @xmath153 . finally , let @xmath175 and @xmath213 be a sequence such that @xmath214 . since the map @xmath215 is a diffeomorphism of @xmath216 onto the one - parameter subgroup @xmath217 by eq . ( [ eq:2.7a ] ) , we conclude @xmath218 and so @xmath219 therefore , lemma [ lemrotsym ] yields the desired result . \(b ) `` @xmath190 '' : if @xmath164 is an eigenvector of @xmath220 to a non - zero eigenvalue part ( a ) implies that the @xmath29-orbit of @xmath164 is weakly rotationally symmetric . however , the @xmath29-orbit of @xmath24 and @xmath164 are equal . this proves the first part of ( b ) . `` @xmath183 '' : let @xmath134 be weakly rotationally symmetric . then by part ( a ) there exist @xmath131 and @xmath187 such that @xmath221 moreover , by a well - known fact from lie theory , @xmath222 is @xmath29-conjugate to some element @xmath163 , i.e. @xmath223 , cf . @xcite and thus @xmath224 now , eqs . ( [ eq:2.8a ] ) and ( [ eq:2.8b ] ) imply @xmath225 . hence , @xmath164 is an eigenvector of @xmath220 and thus the proof of ( b ) is complete . [ rem:2.3 ] equation ( [ eq2.6 ] ) in the above proof which can be rewritten as @xmath226 for all @xmath153 and some @xmath227does in general not imply @xmath228 for all @xmath175 as the following example shows . [ ex5 ] for @xmath229 let @xmath230 then a straightforward computation yields @xmath231 and thus @xmath232 for all @xmath153 . however , @xmath24 and @xmath233 do not satisfy eq . ( [ eq2.9 ] ) , as the necessary condition of lemma [ lemrotsym ] is not met , i.e. , @xmath24 is not an eigenvector of @xmath234 to the eigenvalue @xmath235 . here , we have @xmath236 , which is obvious a rational multiple of @xmath139 and thus does not completely fit the previous requirements . actually , we have the following conjecture . it is impossible to find a counterexample also satisfing @xmath191 . to see this , one probably has to use deeper results on the structure of the torus algebra and its root space decomposition of @xmath29 . theorem [ thm : rotsym ] suggests that the set @xmath237 where @xmath161 is an arbitray torus algebra of @xmath129 , plays a crucial role in the further analysis of matrices with rotationally symmetric relative @xmath0 " numerical range . more precisely , we have the following corollary . [ cor1rotsym ] let @xmath161 be a torus algebra of @xmath129 . the relative @xmath0 " numerical range of @xmath31 , @xmath141 is rotationally symmetric for all @xmath34 if and only if there exists @xmath154 such that @xmath164 is in @xmath238 . this follows immediately from theorem [ thm : rotsym ] and proposition [ prop : rotsym ] . next , we show that part ( d ) and ( e ) of theorem [ thm : disc ] can easily be derived from theorem [ thm : rotsym ] by choosing @xmath239 . [ cor2rotsym ] the classical @xmath0 " numerical range of @xmath24 is rotationally symmetric for all @xmath34 if and only if @xmath24 is unitarily similar to a block - shift matrix @xmath240 , i.e. @xmath61 is of block form such that all the @xmath62 are square blocks and @xmath63 if @xmath64 . we may assume without loss of generality that @xmath141 . applying theorem [ thm : rotsym](b ) to @xmath241 and the torus algebra @xmath242 yields that the unitary orbit of @xmath24 is weakly rotationally symmetric if and only if @xmath24 is unitarily similar to a matrix @xmath240 such that @xmath243 for some @xmath163 and @xmath187 . now , @xmath88 can be arranged such that @xmath244 and @xmath245 for all @xmath246 . choosing a block partition of @xmath240 corresponding to the one of @xmath88 , eqs . ( [ eq2.10 ] ) and ( [ eq2.11 ] ) yield @xmath247 for all @xmath246 . thus condition ( [ eq2.12 ] ) implies @xmath63 if @xmath248 and hence @xmath240 has the required form . in contrast to part ( f ) of theorem [ thm : disc ] on the classical @xmath0 " numerical range of @xmath24 , the relative one need not be a circular disc in order to be rotationally symmetric , a counterexample being provided by example [ ex1 ] . finally , we want to obtain some information on the lie - algebraic structure of the set of all matrices with rotationally symmetric relative @xmath0 " numerical range . [ lemliestruc ] let @xmath222 be skew - hermitian and let @xmath141 be an eigenvector of @xmath130 to the non - trivial eigenvalue @xmath170 , @xmath57 . then we have 1 . @xmath249 is an eigenvector of @xmath130 to a non - trivial eigenvalue @xmath250 . 2 . @xmath24 is nilpotent . @xmath251 \neq 0 $ ] and @xmath252 = 0 $ ] . \(a ) let @xmath222 be skew - hermitian and @xmath253 such that @xmath254 . then @xmath255 \(b ) from the identity @xmath256 = { { \rm i}}\varphi a$ ] we obtain @xmath257 for all @xmath258 by induction . therefore , we have @xmath259 for all @xmath258 , where @xmath260 denotes the frobenius norm . this implies @xmath261 for some @xmath258 , otherwise eq . ( [ eq : nil ] ) would contradict the fact @xmath262 . \(c ) again , let @xmath222 be skew - hermitian and @xmath253 such that @xmath254 . then by the jacobi - identity for the double commutator we obtain @xmath263 & = & - \big([a,[a^{\dagger},\omega ] ] + [ a^{\dagger},[\omega , a ] ] \big)\\ & = & [ a,\mathrm{ad}_{\omega}(a^{\dagger } ) ] - [ a^{\dagger},\mathrm{ad}_{\omega}(a)]\\ & = & -{{\rm i}}\varphi [ a , a^{\dagger } ] - { { \rm i}}\varphi [ a^{\dagger},a ] = 0,\end{aligned}\ ] ] i.e. , @xmath222 and @xmath251 $ ] commute . to prove that @xmath251 $ ] does not vanish , we assume the converse . hence , @xmath24 is normal and thus part ( b ) implies @xmath264 . this , however , contradicts our assumptions on @xmath24 and therefore @xmath251 \neq 0 $ ] . [ corliestruc ] 1 . the @xmath29-orbit of @xmath24 is weakly rotationally symmetric if and only if the @xmath29-orbit of @xmath249 is as well . 2 . if the @xmath29-orbit of @xmath24 is weakly rotationally symmetric then also the @xmath29-orbit of @xmath265,a]$ ] . \(a ) this follows immediately by lemma [ lemliestruc](a ) and theorem [ thm : rotsym](a ) . \(b ) lemma [ lemliestruc ] and the identity @xmath266 = \mathrm{ad}_{[x , y]}$ ] imply that @xmath130 and @xmath267}$ ] commute and thus they admit a simultaneous eigenspace decomposition . hence @xmath169 is invariant under @xmath267}$ ] . in particular , @xmath267}(a ) = [ [ a , a^{\dagger}],a]$ ] is again an eigenvector of @xmath130 to the eigenvalue @xmath170 . therefore , theorem [ thm : rotsym](a ) yields the desired result . so far we have seen that @xmath24 and @xmath251 $ ] are contained in @xmath169 . this , however , does not imply that @xmath251 = \lambda a$ ] for some @xmath146 . therefore , we introduce the following notion to analyse the situation in more detail . for @xmath31 let the _ separation index _ @xmath268 of @xmath24 be defined by @xmath269 if @xmath24 is not contained in any eigenspace @xmath169 , then we set @xmath270 . [ prop : liestruc ] if the separation index of @xmath24 is equal to @xmath271 then the lie algebra generated by @xmath272 , @xmath273 and @xmath274 $ ] is isomorphic to @xmath275 . by assumption there exist @xmath166 and @xmath57 , @xmath253 such that @xmath173 with @xmath276 . as in corollary [ corliestruc](b ) , we obtain the invariance of @xmath169 under @xmath267}$ ] and thus @xmath277}(a ) = \lambda a\ ] ] for some @xmath278 . here , @xmath185 has to be real , since the operator @xmath267}$ ] is hermitian with respect to the scalar product @xmath188 . moreover , as in lemma [ lemliestruc ] we obtain @xmath277}(a^{\dagger } ) = -\lambda a^{\dagger}.\ ] ] hence , we have the following commutator relations : @xmath279 & = & 2{{\rm i}}[a , a^{\dagger}],\\[2 mm ] [ { { \rm i}}(a + a^{\dagger}),{{\rm i}}[a , a^{\dagger } ] ] & = & \lambda ( a - a^{\dagger}),\\[2 mm ] [ { { \rm i}}[a , a^{\dagger}],(a - a^{\dagger } ) ] & = & \lambda { { \rm i}}(a + a^{\dagger}).\end{aligned}\ ] ] therefore , @xmath280 $ ] , @xmath281 and @xmath282 generate a @xmath283-dimensional lie subalgebra of @xmath284 . if @xmath184 then a straightforward rescaling of @xmath285 and @xmath286 shows that the generated lie subalgebra is isomorphic to @xmath275 . if @xmath287 then @xmath288 and @xmath289 as well as @xmath288 and @xmath286 commute . hence we can assume that @xmath288 and @xmath289 are diagonal . this , however , contradicts @xmath290 = x$ ] and thus we are done . an alternative approach to proposition [ prop : liestruc ] is given by the following lemma , which explicitly determines all @xmath24 with @xmath291 . it is a straightforward consequence of corollary [ cor2rotsym ] and therefore stated without proof . [ lem : liestruc ] the separation index of @xmath24 is @xmath271 if and only if @xmath24 is unitarily similar to @xmath292 for some @xmath146 , @xmath184 , where all entries of @xmath293 are zero except the one in the @xmath294-th row and @xmath295-th column with @xmath296 , @xmath297 . [ rem:4 ] 1 . proposition [ prop : liestruc ] is independent of the subgroup @xmath3 . in particular , @xmath24 can be the eigenvector of some @xmath131 with @xmath291 , while @xmath274 $ ] is not contained in the lie algebra @xmath129 of @xmath29 . for instance , let @xmath298 @xmath299 = \begin{bmatrix } -1 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0 \end{bmatrix}.\ ] ] thus the subgroup corresponding to the subalgebra generated by @xmath272 , @xmath273 and @xmath274 $ ] is in general not contained in @xmath29 , even if the @xmath29-orbit of @xmath24 is weakly rotationally symmetric . on the other hand , characterizing all subgroups @xmath29 having the inclusion property @xmath300\ , \in \mathfrak{k}\ ] ] for all @xmath24 with weakly rotationally symmetric @xmath29-orbit seems to be an open problem . if the separation index of @xmath24 is greater than @xmath271 , then proposition [ prop : liestruc ] is in general not true , as the following example shows . @xmath301 : = \begin{bmatrix } -1 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & -4 & 0\\ 0 & 0 & 0 & 4 \end{bmatrix}.\ ] ] by corollary [ cor2rotsym ] , we have @xmath302 . moreover , the lie algebra generated by @xmath272 , @xmath273 and @xmath274 $ ] is not isomorphic to @xmath275 , as claimed above . in this subsection we specify the previous results to the @xmath5-fold tensor product of @xmath6 , i.e. @xmath303 in quantum mechanics and , in particular , in quantum information , @xmath11 is called the _ subgroup _ of _ local action_. therefore , we call the corresponding relative @xmath0 `` numerical range the _ local _ @xmath0__-numerical range of _ _ @xmath24 and introduce the short - hand notation @xmath304 note that replacing @xmath11 by @xmath305 in definition ( [ eq:2.20 ] ) would yield the same local @xmath0 '' numerical range , which can easily be seen by the identity @xmath306 [ rem:3.1 ] following eq . ( [ num - cnum ] ) one might naively assign a relative numerical range @xmath307 to an operator @xmath31 by a definition like @xmath308 with @xmath309 , @xmath310 . however , such a concept is inappropriate as it would depend on the particular choice of @xmath309 . yet , for the local case or more general , if @xmath311 is a tensor product of special unitary groups , there is a canonical subset of the unit sphere , to wit the set of all @xmath312 with @xmath313 , @xmath314 , on which @xmath29 acts transitively . this allows for properly defining the _ local numerical range of _ @xmath24 as the set @xmath315 which in turn immediately yields the local analogue of eq . ( [ num - cnum ] ) . moreover , note that in physical terms , the local numerical range is nothing else than the classical numerical range restriced to the set of all _ pure product states_. some of its implications are analysed in the accompanying paper @xcite . now , for applying theorem [ thm : rotsym ] , we have to choose a torus algebra in the lie algebra of @xmath316 . a straighforward way of doing so is presented in the following . let @xmath317 and @xmath318 be lie subgroups with lie algebras @xmath319 and @xmath320 , respectively . then the lie algebra of the tensor product @xmath321 is denoted by @xmath322 . it is given by @xmath323 moreover , let @xmath324 denote the set of all @xmath31 with @xmath325 . [ lemlocrange ] let @xmath326 and @xmath327 be torus algebras of the subalgebras @xmath328 and @xmath329 , respectively . then @xmath330 is a torus algebra of @xmath322 . if , moreover , @xmath331 and @xmath332 , then the converse is also true . let @xmath333 such that @xmath334 = 0 $ ] for all @xmath335 , i.e. @xmath336 = } \\ & = & [ \omega_1,\omega'_1 ] \otimes i_{n_2 } + i_{n_1 } \otimes [ \omega_2,\omega'_2 ] = 0\end{aligned}\ ] ] for all @xmath337 and @xmath338 . by the fact that @xmath339 \otimes i_{n_2}$ ] and @xmath340 $ ] are orthogonal with respect to the scalar product @xmath188 , we obtain the equivalence @xmath341 = 0 \quad\mbox{for all $ \omega ' \in \mathfrak{t}_1 \widehat{\oplus } \mathfrak{t}_2 $ } & \\ & \longleftrightarrow & \\ & [ \omega_1,\omega'_1 ] = 0 \quad\mbox{and}\quad [ \omega_2,\omega'_2 ] = 0 \quad\mbox{for all $ \omega'_1 \in \mathfrak{t}_1 $ , $ \omega'_2 \in \mathfrak{t}_2 $ } & \end{aligned}\ ] ] now , if @xmath326 and @xmath327 are maximal abelian , then @xmath342 and @xmath343 are contained in @xmath326 and @xmath327 , respectively , and thus @xmath330 is maximal abelian , too . on the other hand , let @xmath344 be abelian and suppose maximality of @xmath330 , the above equivalence shows that @xmath345 belongs to @xmath330 , if @xmath346 = 0 $ ] for all @xmath347 and @xmath104 . hence , it follows @xmath348 and @xmath349 , if the map @xmath350 is one - to - one . this , however , is guaranteed under the additional assumption @xmath331 and @xmath332 . therefore , @xmath326 and @xmath327 are maximal abelian , too . note that the additional assumption for the converse in lemma [ lemlocrange ] is necessary as the following example shows . let @xmath351 and define @xmath352 then @xmath330 is a torus algebra in @xmath353 . however , @xmath326 is not maximal abelian in @xmath354 . now , let @xmath355 be the lie algebra of @xmath11 and let @xmath356 be the subset of all diagonal matrices . obviously , @xmath357 is a torus algebra of @xmath355 by lemma [ lemlocrange ] which yields the following corollary . [ corlocrange ] the local @xmath0 " numerical range @xmath358 of @xmath359 is rotationally symmetric for all @xmath360 if and only if there exists @xmath361 such that @xmath362 , i.e. @xmath363 = { { \rm i}}\varphi uau^{\dagger}.\ ] ] for some @xmath364 and @xmath57 , @xmath253 . this follows immediately from corollary [ cor1rotsym ] and lemma [ lemlocrange ] . finally , we are prepared to present the main result of this section , which roughly speaking excludes the possibility of an annulus for rotationlly symmetric local @xmath0 " numerical ranges . [ thmlocrange ] the local @xmath0 " numerical range @xmath358 of @xmath365 is rotationally symmetric for all @xmath360 if and only if it is a circular disc in the complex plane centered at the origin for all @xmath360 . before approaching theorem [ thmlocrange ] we provide the following technical lemma . [ lem3locrange ] let @xmath366 and @xmath367 satisfy the relation @xmath368 = { { \rm i}}\varphi a \quad \mbox{for some $ \varphi \in { \mathbb{q}}$}.\ ] ] then there exists a rational @xmath369 such that eq . ( [ eq2.45 ] ) holds . let @xmath366 , i.e. @xmath370 and let @xmath371 denote the diagonal entries of @xmath88 , i.e. @xmath372 then one can find a matrix @xmath373 such that @xmath374 cf . lemma [ lemlocrange ] . moreover , a straightforward calculation shows that @xmath375 = { { \rm i}}\varphi a \quad\longleftrightarrow\quad \left\ { \begin{array}{l } a_{ii } = 0 \quad \mbox{for all $ i = 1 , \dots , 2^n$}\\ x_{\mathrm{ad } } \ , \mu = ( \varphi,\dots,\varphi)^{\top } , \end{array } \right.\ ] ] where @xmath376 is a matrix of appropriate size depending on @xmath24 with entries equal to @xmath377 or @xmath378 . in particular , @xmath379 . hence , we have @xmath380 = { { \rm i}}\varphi a , \quad \delta \in \mathfrak{t}_\mathrm{loc } \quad\longleftrightarrow\quad \left\ { \begin{array}{l } a_{ii } = 0 \quad \mbox{for all $ i = 1 , \dots , 2^n$}\\ x_{\mathrm{loc}}\ , \mu = 0 \\ x_{\mathrm{ad}}\ , \mu = ( \varphi,\dots,\varphi)^{\top } \end{array } \right.\ ] ] with @xmath381 . now , by assumption there exists a @xmath382 such that eq . ( [ eq2.49 ] ) is satisfied for some @xmath383 . this , however , implies that eq . ( [ eq2.49 ] ) has in particular rational solutions , i.e. solutions in @xmath384 . `` @xmath385 '' : @xmath159 `` @xmath140 '' : suppose that @xmath358 is rotationally symmetric for all @xmath360 . it is sufficient to show that zero is contained in @xmath358 . therefore , we can assume @xmath386 without loss of generality . thus , by proposition [ prop : rotsym ] and theorem [ thm : rotsym](a ) there exists @xmath387 such that @xmath388 is a circle around the origin in the complex plane . by theorem [ thm : rotsym](b ) and lemma [ lemlocrange ] we can assume that @xmath222 is of diagonal form @xmath389 and satisfies the relation @xmath256 = { { \rm i}}\varphi a$ ] . by rescaling @xmath222 such that @xmath198 is rational and by invoking lemma [ lem3locrange ] we further suppose that for @xmath390 all @xmath391 are rational . now , let @xmath392 be the least common multiple of the denominaters of all @xmath391 for @xmath393 . then @xmath394 and thus @xmath395}$ ] is a circle in the complex plane surrounding the origin @xmath396-times . moreover , we have @xmath397 therefore , the homotopy @xmath398 \times [ 0,{\textstyle}\frac{\pi}{2 } ] \to su_{\rm loc}(2^n)$ ] of the form @xmath399 with @xmath400 satisfies @xmath401 for all @xmath402 \times [ 0,{\textstyle}\frac{\pi}{2}]$ ] . it follows that @xmath395}$ ] is homotopic to its inverse by the homotopy @xmath403 hence , @xmath404 has to cross the origin , cf . appendix [ appendix ] , lemma [ lem : homotopy ] , and thus , the origin is contained in @xmath358 . therefore , @xmath358 is a circular disc . in the remainder of this section , we exemplify the previous results by determining the set of all matrices @xmath405 , the local @xmath0 `` numerical range of which is a cirular disc centered at the origin . these investigations will lead to a conjecture about `` local '' similarity to block - shift form . but first as promised we tackle the problem of computing all @xmath405 with circular local @xmath0 '' numerical range . by corollary [ corlocrange ] , we can focus on the set of all @xmath406 which satisfy @xmath407 = { { \rm i}}\varphi \widehat{a}.\ ] ] for some @xmath364 and @xmath253 . let @xmath408 and @xmath409 . ( [ eq2.35 ] ) can be rewritten as @xmath410 for all @xmath411 . moreover , a straightforward calculation shows @xmath412 now , considering all possibilities for @xmath413 to be equal to @xmath253 and taking into account eq . ( [ eq2.37 ] ) leads to 32 different cases . however , by the symmetry of eq . ( [ eq2.36 ] ) , these can be reduced to the following 16 ones , while the remaining ones can be obtained by transposition . [ cols="^,^,^,^ " , ] for example , case 1 in the above table means that @xmath414 and all other differences @xmath413 do not equal @xmath198 . hence , eq . ( [ eq2.36 ] ) reads @xmath415 , @xmath416 , and @xmath417 otherwise . thus @xmath418 has the form @xmath419 where the symbol @xmath420 denotes an arbitrary complex number . in the same way one can compute @xmath418 in all other cases . here , we only list the solutions for the above table . the remaining ones as mentioned before can be obtained by transposition . @xmath421 finally , we are prepared to provide a rigorous proof of what we claimed in example [ ex4 ] : the local @xmath0 " numerical range of the block - shift matrix @xmath422 is not rotationally symmetric for all @xmath126 . according to the above classification , we have to show that @xmath24 is not locally unitarily similar , i.e. similar via an element in @xmath8 , to one of the above cases . this , however , can be checked by `` brute force '' and is left to the reader . the above computations reveal an interesting contrast to corollary [ cor2rotsym ] : on the one hand , we have seen by example [ ex4 ] that not every matrix which is similar to block - shift from via a local unitary transformation has a circular local @xmath0 `` numerical range . on the other hand , there are matrices with circular local @xmath0 '' numerical range , which are not locally unitarily similar to block - shift form , cf . case 16 in the above table . however , combining corollary [ cor : rotsym ] and [ cor2rotsym ] , every matrix with circular local @xmath0 `` numerical range has to be `` globally '' unitarily similar to block - shift form , i.e. via a transformation in @xmath423 . this raises the question : what is a smallest subgroup @xmath424 of @xmath423 containing @xmath11 such that every matrix with circular local @xmath0 '' numerical range is similar to block - shift form via a unitary transformation in @xmath424 ? by corollary [ corlocrange ] , we can reduce the problem to studying the smallest subgroup @xmath425 of ( signed ) permutations , such that any element in @xmath426 is similar to block - shift form via a permutation in @xmath425 . one should note that in the above table all matrices are either in block - shift form or similar to block - shift form via a permutation of the following type : @xmath427 this leads to the following conjecture : [ conlocrange ] every element of @xmath428 is similar to a block - shift matrix via an element of @xmath429 . here , @xmath430 denotes the subgroup of all ( signed ) _ local _ permutations , i.e. @xmath430 consists of all matrices of the form @xmath431 with @xmath432 for @xmath22 , and @xmath433 stands for the subgroup generated by @xmath434 where the factor @xmath435 appears in all possible positions . note that the two subgroups @xmath430 and @xmath436 operate via similarity on the set of tensor products @xmath437 with @xmath438 in a completely different way . while @xmath430 acts on each factor @xmath439 separately , @xmath436 does not effect the factors themselves but interchanges their order . moreover , @xmath430 and @xmath436 commute and hence @xmath440 is isomorphic to the direct product of @xmath430 and @xmath436 . we call it the _ extended local permutation group_. it is easy to check that @xmath428 is invariant under conjugation with elements of @xmath441 . therefore , we expect @xmath441 to be the smallest subgroup of permutations satisfying the above conjecture . 2 . the group @xmath441 is closely related to the _ weyl group _ of @xmath11 . more precisely , the action of @xmath442 on @xmath443 coincides with the weyl group action of @xmath11 on the torus algebra @xmath443 . however , conjugation by elements of @xmath433 can not be achieved by elements of the weyl group . these ideas suggest how to generalise the above result to the settings of arbitrary compact lie groups . we introduced a new mathematical object , the _ relative _ @xmath0 `` _ numerical range _ @xmath1 of an operator @xmath24 . in particular , we studied its geometry by comparing its properties with the classical @xmath0 '' _ numerical range_. we showed that although the _ relative _ @xmath0 `` _ numerical range _ is compact and connected as in the classical case , it is neither star - shaped nor simply connected . moreover , necessary and sufficient conditions for circular symmetry of @xmath1 have been derived . these results generalise a former theorem by li and tsing @xcite and lead also to a deeper understanding of the classical case in lie theoretical terms . moreover , in view of applications in quantum information , we introduced the _ @xmath0 '' _ numerical range _ @xmath26 as a special case of the relative @xmath0 " numerical range thus inheriting its mathematical structure . in particular detail , we analysed circular symmetry of local @xmath0 `` numerical ranges , which are of special interest in quantum control and quantum information @xcite . here , we proved that local @xmath0 '' numerical ranges with circular symmetry have to be circular discs centered at the origin of the complex plane . this is not evident as relative @xmath0 `` numerical ranges are in general not simply connected . finally , we applied our results to characterise all @xmath444-matrices with circular local @xmath0 '' numerical range . however , explicit formulas for the radius of a circular relative @xmath0 `` numerical range or , more general , for the _ relative @xmath0 " numerical radius _ of @xmath24 @xmath445lead to open research problems . therefore , general numerical algorithms for finding sharp bounds on the size of @xmath1 are highly desirable . geometric optimisation methods for the classical as well as the local @xmath0 '' numerical radius can be found in @xcite . yet another interesting open problem is to find out conditions which guarantee convergence of these methods , such as intrinsic gradient flows , to the relative @xmath0 `` numerical radius or at least to boundary points of the respective relative @xmath0 '' numerical range . problems of this kind are anticipated to be illuminating both for mathematical structure and for quantum applications . this work has been supported in parts by the german research foundation ( dfg ) , grant he 1858/10 - 1 konnew as well as the integrated eu programme qap . parts of it were done while the third author was affiliated to national ict australia and the department of information sciences and engineering , the australian national university , canberra act 0200 . national ict australia is funded by the australian government s department of communications , information technology and the arts and the australian research council through _ backing australia s ability _ and the ict research centre of excellence programs . 9 t. ando and c .- k . li , editors . , volume 37 , 13 of _ linear and multilinear algebra _ , pages 1238 . gordon and breach , 1994 . dynamical systems that sort lists , diagonalize matrices , and solve linear programming problems . in _ proc . ieee of the 27th conference on decision and control _ , 799803 , austin , texas , 1988 . . t. brcker and t. tom dieck . . graduate texts in mathematics . springer , new york , 1985 . cheung and n .- k . the @xmath0-numerical range of matrices is star - shaped . , 41:245250 , 1996 . j. dazord . on the @xmath0-numerical range of a matrix . , 212/213:2129 , 1994 . g. dirr and u. helmke and m. kleinsteuber and s. glaser and th . the local numerical range : examples , conjectures and numerical algorithms , kyoto , 2006 . g. dirr and u. helmke and m. kleinsteuber and th . a new type of @xmath0-numerical range arising in quantum computing , berlin , 2006 . s.j . glaser , t. schulte - herbrggen , m. sieveking , o. schedletzky , n.c . nielsen , o.w . srensen , and c. griesinger . unitary control in quantum ensembles : maximizing signal intensity in coherent spectroscopy . , 280:421424 , 1998 . k. gustafson and d. rao . . springer , new york , 1997 . m. goldberg and e.g. straus . elementary inclusion relations for generalized numerical ranges . , 18:124 , 1977 . d. hausdorff . der wertevorrat einer bilinearform . , 3:314316 , 1919 . u. helmke and j.b . springer , london , 1994 . u. helmke , k. hper , j. b. moore , and th . schulte - herbrggen . gradient flows computing the c - numerical range with applications in nmr spectroscopy . 23 : 283308 , 2002 . -numerical ranges and @xmath0-numerical radii . , 37:5182 , 1994 . li and n .- k . matrices with circular symmetry on their unitary orbits and @xmath0-numerical ranges . , 111(1):1928 , 1991 . s. ojasiewicz . sur les trajectoires du gradient dune fonction analytique . seminari di geometria 1982 - 1983 , universit di bologna , istituto di geometria , dipartimento di matematica , 1984 . j. von neumann . . princeton university press , princeton , 1955 . j. von neumann . . , 1:286300 , 1937 . y. poon . another proof of a result of westwick . , 9:3537 , 1980 . t. schulte - herbrggen . . phd thesis , eth zrich , 1998 . 12752 . t. schulte - herbrggen , g. dirr , u. helmke and s. glaser . accompanying paper in wonra proceedings . e " print : http://arxiv.org/pdf/math-ph/0701035 , 2007 . r. toeplitz . das algebraische analogon zu einem satze von fejr . , 2:187197 , 1918 . r. westwick . a theorem on numerical range . , 2:311315 , 1975 . the following appendix contains two technical lemmas which we referred to in the previous sections . although most readers will be familiar with these results , we incude them for completeness . [ lem : starshaped ] let @xmath446 be a convex ( or star - shaped ) subset of a complex or real vector space and let @xmath447 $ ] be a non - negative , real interval , i.e. @xmath448 . then the set @xmath447 \cdot w : = \ { \lambda \cdot w \;|\ ; \lambda \in [ r , s ] , w \in w\}$ ] is also convex ( star - shaped ) . for @xmath449 there is nothing to prove thus we may assume without loss of generality @xmath450 , as for @xmath451 we can use the identity @xmath447 \cdot w = [ r / s,1 ] \cdot sw$ ] and the fact that @xmath452 is convex or star - shaped if @xmath446 is convex or star - shaped . firstly , we consider the case that @xmath446 is convex . let @xmath453 \cdot w$ ] , i.e. there are @xmath454 and @xmath455 $ ] such that @xmath456 and @xmath457 . we have to show @xmath458 \cdot w\ ] ] for all @xmath459 $ ] . without loss of generality let @xmath460 and define @xmath461 and @xmath462 by @xmath463 as @xmath459 $ ] it holds that @xmath464 and @xmath465 . furthermore , we have @xmath466 hence @xmath467 is in @xmath468 \cdot w$ ] , as convexity of @xmath446 implies that the left side of eq . ( [ eq2.2a ] ) is obviously in @xmath468 \cdot w$ ] . now , we assume that @xmath446 is star - shaped with star center @xmath469 . let @xmath470 \cdot w$ ] , i.e. there is @xmath471 and @xmath472 $ ] such that @xmath473 . we have to show that there exists a star center @xmath474 \cdot w$ ] such that @xmath475 \cdot w\ ] ] for all @xmath459 $ ] . let @xmath476 and define @xmath461 and @xmath462 in the same way as in eq . ( [ eq2.2a ] ) with @xmath477 and @xmath478 . then we have @xmath479 as @xmath480 does not depend on @xmath481 , it follows that @xmath468 \cdot w$ ] is star - shaped with star center @xmath480 . [ lem : homotopy ] let @xmath482 \to { \mathbb{c}}$ ] be a closed curve and let @xmath483 be its inverse , i.e. @xmath484 . moreover , let @xmath485 be any point in the interior of @xmath486 , i.e. the winding number of @xmath486 with respect to @xmath487 is not equal to zero . then any homotopy from @xmath486 to its inverse @xmath483 has to cross @xmath487 . let @xmath488 denote the winding number of a closed curve with respect to @xmath487 and assume that @xmath489 \times [ c , d ] \to { \mathbb{c}}$ ] is a homotopy from @xmath486 to its inverse @xmath483 such that @xmath490 for all @xmath491 \times [ c , d]$ ] . as well - known , the winding number assumes only integer values and satisfies the equality @xmath492 . therefore , by continuity of @xmath404 , the winding number of @xmath486 with respect to @xmath487 has to be zero . this , in turn , contradicts our hypothesis and thus @xmath404 has to cross @xmath487 , i.e. @xmath493 for some @xmath491 \times [ c , d]$ ] .

let be a compact , connected symplectic manifold with a hamiltonian action of a compact-dimensional torus . suppose that is equipped with an anti - symplectic involution compatible with the-action . the real locus of is the fixed point set of . duistermaat introduced real loci , and extended several theorems of symplectic geometry to real loci . in this paper , we extend another classical result of symplectic geometry to real loci : the kirwan surjectivity theorem . in addition , we compute the kernel of the real kirwan map . these results are direct consequences of techniques introduced by tolman and weitsman . in some examples , these results allow us to show that a symplectic reduction has the same ordinary cohomology as its real locus , with degrees halved . this extends duistermaat s original result on real loci to a case in which there is not a natural hamiltonian torus action .

math0209111

let @xmath0 be a compact symplectic manifold , and suppose an @xmath1-dimensional torus @xmath2 acts on @xmath0 in a hamiltonian fashion . let @xmath7 be a moment map . the components of the moment map are ( equivariant ) morse - bott functions on @xmath0 , and hence may be used to determine the ( equivariant ) topology of @xmath0 . now suppose that @xmath3 is an anti - symplectic involution on @xmath0 that anti - commutes with the @xmath2 action : @xmath8 for all @xmath9 and @xmath10 . we assume that @xmath11 is nonempty . we call the fixed point set @xmath12 of @xmath3 the _ real locus _ of @xmath0 . the real locus is a lagrangian submanifold of @xmath0 . the motivating example of a real locus is a complex variety @xmath0 under complex conjugation @xmath3 . in this case , the real locus is the set of real points on the variety . by ( [ eq : anticommute ] ) the subgroup of @xmath2 of elements of order 2 acts on @xmath13 . we call this subgroup @xmath14 , the _ real torus _ in @xmath2 . it is immediate that @xmath15 . duistermaat @xcite proved that the real locus has full moment image , that is , that @xmath16 . moreover , he showed that the components of the moment map are morse - bott functions for the real locus , when using @xmath17 coefficients , and so we can understand the topology of @xmath13 via the moment map . biss , guillemin and the second author @xcite proved that these moment map components can also be used to understand the equivariant topology of the real locus with respect to the restricted @xmath14 action . in addition , sjamaar and oshea have generalized the results of duistermaat for hamiltonian actions of nonabelian lie groups @xcite . the principle behind these results is that real loci should behave in a fashion similar to the symplectic manifolds of which they are submanifolds . we show that this philosophy applies in the context of symplectic reductions . given a compact hamiltonian @xmath2-space @xmath0 , kirwan @xcite proved that when @xmath2 acts freely on @xmath18 , the inclusion map @xmath19 induces a surjection in equivariant cohomology with rational coefficients : @xmath20 the map @xmath21 is called the kirwan map . in @xcite the authors note that under reasonable assumptions about the torsion of the fixed point sets and the group action , this map is surjective over the integers as well . for @xmath22 one can state the result as follows . for any prime @xmath23 , assume that _ either _ there is no @xmath23-torsion in the @xmath24-cohomology of the fixed point set @xmath25 , _ or _ every point not in @xmath25 has a free @xmath26 action . then the map ( [ eq : surjection ] ) is surjective with integer coefficients . for higher dimensional @xmath2 , the authors assume the action is quasi - free to prove kirwan s surjectivity over @xmath24 , although this is stronger than necessary . one natural question is , what is the kernel of @xmath21 ? this question was answered in @xcite and refined by the first author in the case of rational coefficients @xcite . suppose that @xmath0 has the additional structure of an anti - symplectic involution , compatible with torus action as specified above . suppose also that @xmath27 is nonempty . in this article we prove a surjectivity result for real loci analogous to kirwan s result ( theorem 1 ) . we also use tolman and weitsman s equivariant morse theoretic methods to compute the kernel of this real version of the kirwan map ( theorem 2 ) . thus the main contribution of this article is to complete the program begun by duistermaat in showing that the real locus of a hamiltonian @xmath2-space has a @xmath28-action which behaves as if it were itself a hamiltonian @xmath2-space . not only is the equivariant cohomology ring of @xmath13 described by restrictions to fixed points , but there is a well - defined notion of real reduction " and the major theorems about the cohomology rings of reduced spaces go through in the real case . the proofs we present are straightforward extensions of the work of atiyah - bott , kirwan , and tolman - weitsman to the real case . we then explore the ramifications of real reduction and these theorems in a series of examples . the key to kirwan s proof of surjectivity is the analysis of the function @xmath29 in the case of @xmath30 , this is not a morse function , but it is morse - bott except at @xmath31 , the function s minimum value . for @xmath32 , there are a finite number of critical values of @xmath33 where @xmath34 is not morse - bott , in addition to the minimum . moreover , @xmath35 kirwan uses the critical sets of @xmath36 to argue inductively ( on the critical sets ) that the equivariant cohomology of @xmath0 surjects onto the ordinary cohomology of @xmath37 . it is clear that surjectivity for general values @xmath38 of @xmath33 follows by shifting the moment map by an appropriate constant . these arguments easily go through when the function is restricted to the real locus , as we will see . a fundamental step in the proof of surjectivity and the computation of the kernel is a lemma due to atiyah and bott @xcite . this lemma describes the local topology , in and around a fixed point component of @xmath25 . under the hypothesis of no 2-torsion ( definition [ def:2torsion ] ) we show that this lemma still holds for @xmath28-cohomology with @xmath17 coefficients . we then apply the lemma and an inductive argument to both the question of surjectivity and to the computation of the kernel for real loci . for the rest of this article , let @xmath0 be a compact , connected symplectic manifold , endowed with a hamiltonian action of a compact torus @xmath2 and moment map @xmath33 . let @xmath3 be an anti - symplectic involution , anti - commuting with the torus action , and let @xmath39 be the set of order 2 elements , plus the identity . recall that @xmath14 acts on @xmath13 . we begin with a definition . [ def:2torsion ] let @xmath40 be a critical point of @xmath33 . let @xmath41 be the maximal connected subtorus of @xmath2 fixing @xmath23 , @xmath42 the weight lattice of @xmath2 , and @xmath43 the weight lattice of @xmath44 . let @xmath45 be the connected component of @xmath46 containing @xmath23 . we denote by @xmath47 the weights of the @xmath44 action on @xmath48 , the fiber over @xmath23 of the normal bundle to @xmath45 in @xmath0 , where @xmath49 . we say that @xmath23 is a _ 2-torsion point _ if @xmath50 for some @xmath51 . suppose @xmath0 has a symplectic involution @xmath3 with real locus @xmath13 . suppose @xmath23 is critical for @xmath33 , fixed by @xmath44 , and @xmath45 is the connected component of @xmath46 containing @xmath23 . if @xmath23 lies in @xmath13 , the @xmath44 action on the normal bundle to @xmath45 in @xmath0 restricts to an @xmath52 action on the normal bundle @xmath53 to @xmath54 in @xmath13 . let @xmath55 be the weights of the irreducible representations obtained by this action on the fiber over @xmath23 . then @xmath23 is a 2-torsion point if and only if @xmath56 is trivial for some @xmath57 . the functions @xmath58 and a family of perturbations ( see section [ se : kernel ] ) have critical sets fixed by various subtori of @xmath2 . tolman and weitsman prove @xmath59 is a surjection provided that @xmath2 acts quasi - freely . as they note , one needs to ensure that the negative normal bundles to all critical sets of these functions have nontrivial equivariant euler classes . for the real locus case , no 2-torsion points " plays the same role as quasi - free " does in the symplectic case . at the critical sets of this family of functions , the negative normal bundles have top stiefel - whitney classes which are nontrivial in the @xmath60-component of @xmath61 for each connected component @xmath45 of the critical set . this amounts to the requirement that , over any point in @xmath45 , the negative normal bundle splits into one - dimensional _ nontrivial _ representations of @xmath14 . the condition that @xmath13 contain no @xmath62-torsion points ensures this for the family of perturbations of @xmath58 . the first main theorem is the surjectivity analogue of for real loci . [ th : surjectivity ] suppose @xmath0 is a compact symplectic manifold with a hamiltonian @xmath63-action , and that @xmath13 is the real locus of @xmath0 . let @xmath33 be a moment map on @xmath0 and @xmath38 a regular value of @xmath33 . suppose further that @xmath63 acts freely on @xmath18 and that @xmath13 contains no @xmath62-torsion points . then the _ real kirwan map _ in @xmath14-equivariant cohomology with @xmath17 coefficients @xmath64 induced by inclusion , is a surjection . we omit the superscript @xmath1 in @xmath63 when the dimension is clear . the hypothesis that the real locus have no @xmath62-torsion points is reasonably strong . real loci of toric varieties and coadjoint orbits in type @xmath65 satisfy this hypothesis , for example , but the real loci of maximal coadjoint orbits in type @xmath66 do not . the next task we complete is to specify the kernel of @xmath67 . the kernel is precisely the real analog of the kernel found in @xcite . for every @xmath68 , let @xmath69 let @xmath70 denote the fixed point set , and let @xmath71 finally , define the ideal @xmath72 [ th : kernel ] suppose @xmath0 is a compact symplectic manifold with a hamiltonian @xmath2-action , an anti - symplectic involution @xmath3 anti - commuting with @xmath2 , and real locus @xmath13 . let @xmath33 be a moment map on @xmath0 and @xmath38 a regular value of @xmath33 such that @xmath2 acts freely on @xmath18 and that @xmath13 contains no @xmath62-torsion points . then the kernel of the real kirwan map is the ideal @xmath73 , so there is a short exact sequence @xmath74 where the cohomology is taken with @xmath17 coefficients . the remainder of this paper is organized as follows . in section [ se : top ] , we review @xmath75-equivariant cohomology with specific attention to the case of @xmath17 coefficients and @xmath76 . here we make explicit the thom isomorphism theorem and the @xmath17 version of the atiyah - bott lemma . in section [ se : morsekirwan ] , we study morse - kirwan theory on real loci . in section [ se : surjectivity ] , we discuss reduction , an induced anti - symplectic involution on the symplectic reduction , and we prove the surjectivity theorem ( theorem [ th : surjectivity ] ) . in section [ se : kernel ] , we prove theorem [ th : kernel ] , the description of the kernel of the real kirwan map . finally , in section [ se : egs ] , we work out several pertinent examples . in particular , we present an example of a symplectic reduction which has the same cohomology as its real locus , with degrees divided in half . as @xmath77 does not in general have a torus action , this example generalizes duistermaat s original work on real loci , in which he uses the @xmath2 action to make an analogous statement for @xmath0 and its real locus . the authors would like to thank victor guillemin , robert kleinberg , dan dugger and reyer sjamaar for useful comments during the preparation of this paper . in this section we discuss equivariant cohomology and the thom isomorphism in @xmath17 coefficients . we then state a topological lemma ( lemma 2.1 ) that plays an important role in the proofs of theorems 1 and 2 . given a bundle @xmath78 with a @xmath17 action exactly fixing @xmath79 , we show the top stiefel - whitney class @xmath80 is not a zero - divisor in @xmath81 . similarly , given a @xmath14 action on a space @xmath82 that has no 2-torsion , we show that @xmath83 is not a 0-divisor . this result implies that a certain long exact sequence splits into short exact sequences ( lemma 2.3 ) allowing inductive arguments to be applied . this is not surprising given the analogous result due to tolman and weitsman @xcite for the @xmath2-equivariant cohomology of @xmath79 , where @xmath2 is a compact torus acting on @xmath82 and fixing @xmath79 , and the cohomology is taken with integer coefficients . let @xmath75 be a compact lie group acting on @xmath0 . then the equivariant cohomology of @xmath0 is defined to be the ordinary cohomology of the _ borel construction_. let @xmath84 be an equivariantly contractible space on which @xmath75 acts freely . the borel construction is @xmath85 where the right - hand side is the set of equivalence classes of pairs @xmath86 such that @xmath87 and @xmath88 for all @xmath89 . by definition , the @xmath75-equivariant cohomology of @xmath0 is @xmath90 . all @xmath91 are homotopic for different choices of @xmath84 , and so any choice of @xmath84 will induce the same equivariant cohomology ring . for any compact oriented @xmath75-manifolds @xmath45 and @xmath0 and any @xmath75-equivariant map @xmath92 , there is an associated map @xmath93 defined by @xmath94 on equivalence classes . this induces the pullback and the pushforward in cohomology : @xmath95 where @xmath96 , and may be negative . in particular , equivariant cohomology is functorial . note that the pushforward does not preserve degree , whereas the pullback is a ring map . in the case of the map @xmath97 which sends all of @xmath0 to a point @xmath98 , the induced map in equivariant cohomology @xmath99 turns @xmath100 into a module over @xmath101 . we note that @xmath100 is not always a _ free _ module over @xmath102 : in particular , if @xmath75 acts freely on a finite dimensional manifold @xmath0 , then @xmath103 ( as can be seen by noticing that the fiber of @xmath104 is @xmath84 , which is contractible ) has finite - dimensional cohomology . however , @xmath102 is not finite for nontrivial @xmath75 , and so the module is not free . when @xmath105 as vector spaces , we say that @xmath0 is _ equivariantly formal_. clearly , the module structure is free in this case . hamiltonian @xmath75-spaces are examples of equivariantly formal spaces . an important fact is that for @xmath106 , one choice of @xmath107 is @xmath108 . thus , with @xmath17 coefficients , @xmath109\ ] ] where @xmath110 for all @xmath57 . let @xmath111 be a submanifold of real codimension @xmath112 . in ordinary cohomology ( with any ring of coefficients ) , the thom isomorphism is a map between the cohomology of @xmath45 and the compactly supported cohomology of the normal bundle @xmath113 of @xmath45 in @xmath0 . the thom isomorphism does not preserve degree ; it increases degree by @xmath112 : @xmath114 this latter ring is identified using excision to the relative cohomology as follows : @xmath115 where @xmath116 and @xmath117 are the disk and sphere bundles of @xmath45 in @xmath0 , respectively . the class @xmath118 is called the _ thom class _ of @xmath119 . then @xmath120 , where @xmath121 is the map induced by the projection @xmath122 . the class @xmath123 has the identifying property that , @xmath124 , where @xmath123 is restricted to the fiber of @xmath125 . in equivariant cohomology we avoid the issue of compact support by remaining in the relative cohomology ring . the following discussion is an equivariant version of well - known results found in @xcite and @xcite . we now restrict ourselves to the coefficient ring @xmath17 . let @xmath75 be a compact lie group acting on @xmath0 . let @xmath45 be a @xmath75-invariant submanifold of @xmath0 with codimension @xmath112 with normal bundle @xmath119 . then @xmath126 is a co - dimension @xmath112 submanifold of @xmath91 . let @xmath127 be its normal bundle and @xmath128 the projection . note that @xmath129 if @xmath119 is identified with a @xmath75-invariant tubular neighborhood of @xmath45 . choose a @xmath75-invariant riemannian metric on @xmath127 , and let @xmath130 and @xmath131 be the unit disk and sphere bundles , respectively . as @xmath45 is an oriented submanifold of @xmath0 , @xmath131 is an oriented sphere bundle over @xmath126 . consider the bundle @xmath132^{i } & \ar[d]^{p } ( dn_g , sn_g ) \\ & n_g } \ ] ] where @xmath133 is the fiber over any point @xmath134 in the relative bundle @xmath135 . the cohomology @xmath136 is isomorphic to that of the relative normal bundle @xmath137 . there is a unique _ equivariant thom class _ @xmath138 ( in @xmath17 coefficients ) such that @xmath139 restricted to the ( relative ) fiber over any point is the unique nonzero class in the _ ordinary _ cohomology @xmath140 , where @xmath141 and @xmath142 are the intersection of the fiber with the disk and sphere bundles , respectively . in particular , the _ equivariant thom isomorphism _ in @xmath17 coefficients is the map @xmath143 obtained by the composition of @xmath144 induced by the projection , and the map @xmath145 which multiplies any class in @xmath146 by @xmath139 using the cup product . in other words , @xmath147 . [ le : restrictthom ] let @xmath45 be a submanifold of @xmath0 , where @xmath45 and @xmath0 are compact , oriented manifolds . let @xmath119 be the normal bundle of @xmath45 in @xmath0 . let @xmath75 act on @xmath0 fixing @xmath45 , such that for every point @xmath148 , there is a @xmath149 acting nontrivially on @xmath23 . in cohomology with @xmath17 coefficients , the restriction of the equivariant thom class @xmath139 to the submanifold @xmath45 is the top equivariant stiefel - whitney class of the bundle @xmath119 . [ re:2.2 ] the reason this theorem is so powerful is that it has the following interpretation . let @xmath150 be the inclusion . the composition @xmath151^{\hspace{-1in}{\mathbf{t } } } & h^{*+k}_g(\nu n , \nu n - n ) \cong h^{*+k}_g(m , m - n ) \ar[r]^{\hspace{1 in } j^ * } & h^{*+k}_g(m ) } \ ] ] is the pushforward @xmath152 . here @xmath153 is the map induced by the inclusion @xmath154 . lemma [ le : restrictthom ] states that , in @xmath17-cohomology , @xmath155 , where @xmath156 is the pullback . in @xmath24-cohomology , @xmath157 , the equivariant euler class of the normal bundle . the top stiefel - whitney class is defined using the squaring operation " . for @xmath158 , let @xmath159 . the @xmath160 _ equivariant stiefel - whitney class _ is the unique class @xmath161 such that @xmath162 . thus @xmath163 since @xmath164 , and @xmath165 is an isomorphism , it follows that @xmath166 . we show in the proof of lemma [ le : atiyahbott ] that , under appropriate assumptions regarding the @xmath75 action on @xmath119 ( an assumption implied by no 2-torsion ) , multiplication by the equivariant stiefel - whitney class is injective . suppose @xmath0 is compact and symplectic , equipped with an anti - symplectic involution @xmath3 . we assume that there is a compact torus @xmath2 acting on @xmath0 in a hamiltonian fashion , and a real torus @xmath167 acting on the real locus @xmath168 . one of the critical properties of hamiltonian actions on compact manifolds is the richness of the structure of the fixed point sets . we use the behavior of the group action on the normal bundles of these fixed sets to obtain cohomological information . the following lemma describes the cohomology locally near a fixed point set . it is crucial in making subsequent inductive arguments . [ le : atiyahbott ] let @xmath169 be a rank @xmath112 bundle over a compact , connected , oriented manifold @xmath45 , and @xmath75 either a compact torus or @xmath106 acting on @xmath82 whose fixed point set is exactly @xmath45 . suppose also that the cohomology of @xmath45 has no torsion over @xmath24 . choose a @xmath75-invariant riemannian metric and let @xmath170 and @xmath171 be the disk and sphere bundles , respectively , of @xmath82 . then the long exact sequence induced by the inclusions @xmath172 splits into short exact sequences @xmath173 where the coefficient ring is taken to be @xmath24 if @xmath75 is a compact torus , and @xmath17 for @xmath106 . in the case that @xmath106 , the requirement that @xmath75 act on the fibers of @xmath82 nontrivially is weaker than the condition of no @xmath62-torsion for @xmath174 . the condition that @xmath45 have no torsion is unnecessary if the coefficient ring is @xmath17 . the proof of [ re:2.2 ] can be found in @xcite in the case that @xmath175 , a compact torus . suppose that @xmath106 and the coefficient field is @xmath17 . since @xmath45 is fixed by @xmath75 we have @xmath176\otimes h^*(n).\ ] ] we show that the leading term of @xmath177 in ( [ eq : ncohom ] ) is nonzero , where @xmath178 and @xmath112 is the rank of @xmath82 . as we have assumed that @xmath45 has no torsion , if the leading term in @xmath179 is not zero , then the equivariant stiefel - whitney class is not a zero - divisor . consider the inclusion of a point @xmath180 and the induced projection @xmath181 we need only show that the projection of @xmath177 is nonzero . the fiber @xmath182 over @xmath23 is a representation of @xmath75 which , since @xmath75 is simply a product of @xmath17 s , splits into real one - dimensional representations @xmath183 of @xmath75 . each of these representations is nontrivial , by our assumption that @xmath75 fixes exactly @xmath45 in the bundle @xmath82 . the bundle @xmath184 is the pullback of the canonical line bundle over @xmath185 under the projection @xmath186 to one component . this bundle has nontrivial first stiefel - whitney class by the axiomatic definition of these classes . thus the projection of @xmath177 is @xmath187 which is nonzero . we identify @xmath188 with @xmath189 using the thom isomorphism , and @xmath190 with @xmath191 by contraction . then the map @xmath192 is identified with the injective map given by multiplication by the euler class ( in @xmath24 coefficients ) or the stiefel - whitney class ( in @xmath17 coefficients ) . it follows that the long exact sequence @xmath193 & h^{*-1}_g(se ) \ar[r ] & h^*_g(de , se ) \ar[r ] & h^*_g(de)\ar[r ] & h^*_g(se)\ar[r ] & \dots } \ ] ] splits into short exact sequences . another proof of this lemma can be found in @xcite . we return to case where @xmath0 is a compact connected symplectic manifold with a hamiltonian @xmath2 action with moment map @xmath194 . suppose also that @xmath0 is equipped with an anti - symplectic involution that anti - commutes with @xmath2 . let @xmath13 be the real locus of @xmath0 . we fix an inner product on @xmath195 . in this section , we will prove three important lemmas . in the first , we determine the critical sets of @xmath196 in @xmath13 . duistermaat showed @xcite that for any component of the moment map @xmath197 , if @xmath198 is the set of critical points for @xmath197 on @xmath0 , the set of critical points of @xmath199 is @xmath200 . we prove an analogous result for @xmath196 . [ le : critsets ] let @xmath198 be the set of critical points of the function @xmath201 , and let @xmath202 be the set of critical points of @xmath203 . then @xmath198 is preserved by @xmath3 , and @xmath204 . recall that @xmath13 is the fixed point set of the involution @xmath3 , and that @xmath3 commutes with the moment map @xmath33 . for any point @xmath205 we have @xmath206 . suppose @xmath23 is critical for @xmath58 . the involution @xmath3 induces an isomorphism of tangent spaces . so @xmath207 implies that @xmath208 is critical for @xmath58 . thus @xmath198 is preserved by @xmath3 . suppose that a @xmath205 is critical for @xmath209 . then clearly @xmath23 is critical for @xmath210 , so @xmath211 . now suppose that @xmath205 is critical for @xmath210 . then for all @xmath212 , @xmath213\\ & = 2\phi(p)\cdot [ d(\phi)_p(v)+d(\phi\circ\sigma)_p(v)]\\ & = 2\phi(p)\cdot [ d(\phi)_p(v)+d(\phi)_p(v)]\\ & = 4\phi(p)\cdot d\phi_p(v ) = 2d|\!|\phi|\!|^2_p(v),\end{aligned}\ ] ] so @xmath23 is critical for @xmath58 as well . in the next two lemmas , we show that the functions @xmath199 and @xmath196 are morse - kirwan functions on @xmath13 , when @xmath13 contains no @xmath62-torsion points . as before , @xmath0 is compact , connected , and symplectic with a hamiltonian torus action and involution @xmath3 fixing the real locus @xmath13 . we suppose also that @xmath13 contains no @xmath62-torsion points . the following two lemmas are real locus versions of results of kirwan @xcite . the proof of the first lemma is nearly identical to that given by tolman and weitsman in @xcite . the only differences that occur are a result of the possibility of @xmath62-torsion points . the proof of the second lemma contains some additional analysis concerning critical sets . we include the proofs here for completeness . [ le : morsekirwan1 ] choose an invariant riemannian metric on @xmath0 . given any @xmath68 , define @xmath214 . let @xmath215 be a critical set of index @xmath216 for @xmath217 , and assume that @xmath215 is the only critical set in the preimage of an @xmath218-neighborhood around @xmath219 , for some @xmath220 . define @xmath221 then the long exact sequence of the pair @xmath222 splits into short exact sequences in equivariant cohomology , with @xmath17 coefficients : @xmath223 the function @xmath217 is morse - bott at every connected component of the critical set . when @xmath224 is generic , the critical sets are precisely the connected components of the fixed points . when @xmath224 is not generic , then the critical sets are connected components of @xmath225 , where @xmath226 is some closed subtorus of @xmath2 . since the action is smooth , these sets are submanifolds . the negative normal bundle to @xmath215 is oriented , except at the minimum , where the negative normal bundle is zero . consider the long exact sequence in equivariant cohomology with @xmath17 coefficients , @xmath227 let @xmath228 be a nonminimal critical value corresponding to @xmath215 . by assumption , @xmath229 is the only critical value in the interval @xmath230 $ ] . denote the negative disc and sphere bundles to @xmath215 in @xmath13 by @xmath231 and @xmath232 respectively . we let @xmath216 denote the morse index of @xmath215 in @xmath13 . following an identical argument to ( * proof of proposition 2.1 ) , we obtain a commutative diagram , with @xmath17 coefficients , @xmath233 & h^{*}_{{t_{{\mathbb r}}}}(q^+,q^-)\ar[r]^{\gamma_c}\ar[d]_{\iso } & h^*_{{t_{{\mathbb r}}}}(q^+ ) \ar[r]^{\beta_c}\ar[d ] & h^*_{{t_{{\mathbb r}}}}(q^-)\ar[r ] & \\ & h_{{t_{{\mathbb r}}}}^*(d_c , s_c)\ar[r]^{\delta_c}\ar[d]_{\iso } & h_{{t_{{\mathbb r}}}}^*(d_c ) & & \\ & h^{*-\lambda}_{{t_{{\mathbb r}}}}(d_c ) \ar[ur]_{\cup w_c } & & & } \end{array}\ ] ] where @xmath234 is the equivariant top stiefel - whitney class of the bundle @xmath235 . the left - most vertical arrows in the diagram are excision and the thom isomorphism with @xmath17 coefficients . remark [ re:2.2 ] ensures that the diagonal arrow is indeed the cup product with @xmath236 . because there are no @xmath62-torsion points in @xmath237 , @xmath14 fixes no sub - bundle of the negative normal bundle . lemma [ le : atiyahbott ] ensures that the cup product map @xmath238 is injective , and so @xmath239 and @xmath240 must also be injective . thus , the long exact sequence splits . if @xmath241 is the minimum critical value , the spaces @xmath242 and @xmath215 are equivariantly homotopic , and @xmath243 is empty . thus , @xmath244 and @xmath245 . therefore , the sequence splits in this case as well . this completes the proof of the lemma . [ le : morsekirwan2 ] choose an invariant riemannian metric on @xmath0 . given @xmath246 , let @xmath247 . let @xmath248 be a critical set for @xmath249 , and assume that @xmath250 is the only critical set in the preimage of an @xmath218-neighborhood around @xmath251 , for some @xmath220 . define @xmath252 then the long exact sequences of the pair @xmath253 split into short exact sequences in equivariant cohomology , with @xmath17 coefficients : @xmath254 suppose @xmath250 is not the critical set at the @xmath31-level set . while the function @xmath249 may not be morse - bott at @xmath250 , kirwan proves @xcite that the critical sets behave on a cohomological level as if they were nondegenerate . there is a smooth stratification @xmath255 of @xmath0 , where @xmath79 is an index set of vectors in @xmath256 , such that , for some @xmath257 , the stratum @xmath258 contains precisely those @xmath259 such that the limit of the path of steepest descent for @xmath249 from @xmath260 is in @xmath250 . kirwan proves that the inclusion @xmath261 is an equivalence of @xmath2-equivariant cohomology . furthermore , close to @xmath250 , @xmath258 coincides with a @xmath2-invariant submanifold @xmath262 of a neighborhood of @xmath250 , and @xmath262 has a well - defined orientable normal bundle . let @xmath263 be the fixed point set of the involution @xmath3 of @xmath250 . all critical sets of @xmath249 restricted to @xmath13 will be of the form @xmath263 , by lemma [ le : critsets ] . then @xmath263 is the only critical set of @xmath264 in @xmath265 and not in @xmath266 . we assume @xmath263 is connected , for each component can be treated individually by a direct sum argument in the following discussion . @xmath263 lies in @xmath267 , the submanifold of a neighborhood of @xmath263 in @xmath13 formed by the fixed point set of @xmath3 on @xmath262 ( @xmath267 is smooth since @xmath3 is a smooth action on @xmath262 ) . furthermore , @xmath267 has a well - defined oriented normal bundle in @xmath13 , @xmath267 coincides with @xmath268 close to @xmath263 , and @xmath269 is an equivalence in @xmath2-equivariant cohomology by @xcite . consider the long exact sequence in equivariant cohomology with @xmath17 coefficients , @xmath270 except at the minimum , there is a commutative diagram of the form . the @xmath62-torsion hypothesis guarantees that @xmath14 acts nontrivially on the negative normal bundle to @xmath267 , and that the cup product with the equivariant stiefel - whitney class is an injection . by kirwan s work ( extended to the real case as above ) the negative normal bundle to @xmath267 plays the role ( cohomologically ) that the negative normal bundle to the critical set plays in the case that the critical set is itself a smooth manifold . finally , at the minimum @xmath271 , the spaces @xmath265 and @xmath263 are equivariantly homotopic , and @xmath266 is empty . thus , @xmath272 and @xmath273 . the lemma now follows . let @xmath0 be a compact symplectic manifold with hamiltonian @xmath2-action with moment map @xmath33 . let @xmath25 denote the set of fixed points of @xmath0 . kirwan s injectivity theorem @xcite states that the inclusion @xmath274 induces an injection in equivariant cohomology @xmath275 the real locus analogue is proved in @xcite using algebraic techniques . schmid hints that there is a morse - kirwan theoretic proof in @xcite , but does not provide the details . [ th : qinjection ] let @xmath0 be a compact symplectic manifold with hamiltonian @xmath2-action with moment map @xmath33 . let @xmath13 be the real locus of @xmath0 and let @xmath276 be the fixed points of @xmath14 in @xmath13 , with natural inclusion map @xmath277 . assume that @xmath278 contains no @xmath62-torsion points . then the pullback in equivariant @xmath17-cohomology @xmath279 is injective . choose an element @xmath68 such that the critical set of @xmath197 is precisely @xmath278 . order the critical values of @xmath280 as @xmath281 , and let @xmath282 be the critical points with @xmath283 . denote @xmath284 the theorem holds for @xmath285 because this set is empty . we proceed by induction . suppose that @xmath286 is an injection . we need to apply lemma [ le : morsekirwan1 ] , at the critical sets of @xmath197 , which are precisely the fixed point components of @xmath278 . thus , since @xmath278 contains no @xmath62-torsion points , we conclude that there is a short exact sequence @xmath287 let @xmath288 be the inclusion @xmath289 . then we have a commutative diagram , with @xmath17 coefficients , @xmath290 & h^{*}_{{t_{{\mathbb r}}}}(q_i^+,q_i^-)\ar[r]\ar[d]^{\iso } & h^*_{{t_{{\mathbb r}}}}(q_i^+ ) \ar[r]\ar[d]^{i^*_+ } & h^*_{{t_{{\mathbb r}}}}(q_i^- ) \ar[r]\ar[d]^{i^*_- } & 0 \\ 0\ar[r ] & h^{*}_{{t_{{\mathbb r}}}}(f_i)\ar[r]^= & h^*_{{t_{{\mathbb r}}}}(q_i^+\cap f ) \ar[r ] & h^*_{{t_{{\mathbb r}}}}(q_i^-\cap f ) \ar[r ] & 0 . } \ ] ] the map @xmath291 is an injection by the inductive hypothesis . a diagram chase shows that @xmath292 is an injection . finally , notice that @xmath293 is equivariantly homotopy equivalent to @xmath294 . this completes the proof . suppose that @xmath38 is a regular value of @xmath33 , and that @xmath2 acts freely on @xmath18 . then @xmath295 is a symplectic manifold with a canonical symplectic form @xmath296 . notice that @xmath33 is @xmath3-invariant : @xmath297 . thus , @xmath3 acts on the level sets @xmath18 , taking @xmath2-orbits to @xmath2-orbits . therefore , there is an induced involution @xmath298 on @xmath299 . there are natural maps @xmath300^{i } \ar[d]^{\pi } & m \\ m/\!/t(\mu ) & } \ ] ] the symplectic form @xmath296 on @xmath301 satisfies the property that @xmath302 . it is immediate that @xmath3 also satisfies @xmath303 . the involution @xmath298 on @xmath301 is anti - symplectic with respect to @xmath296 . notice first that @xmath304 but now we also notice that @xmath305 thus @xmath306 . finally , we notice that @xmath307 is an injection , completing the proof . the real locus @xmath308 of @xmath301 is a lagrangian submanifold . this follows immediately from the proposition , as duistermaat has shown in @xcite that the real locus of any symplectic manifold with any anti - symplectic involution is lagrangian . the torus @xmath2 acts freely on @xmath18 , so the subset @xmath14 of order two elements acts freely on @xmath309 . thus the _ reduction _ of the real locus is well defined : @xmath310 [ pr : real ] let @xmath0 be a compact symplectic manifold , and suppose @xmath2 acts on @xmath0 in a hamiltonian fashion . suppose @xmath0 has real locus @xmath13 . let @xmath311 be a regular value of the moment map such that @xmath2 acts freely on @xmath18 . then @xmath312 the symplectic reduction @xmath301 consists of @xmath2-orbits with moment image @xmath38 . the points of @xmath308 are those orbits which are fixed by @xmath3 . on the other hand , the reduction @xmath313 consists of @xmath14-orbits in @xmath13 with moment image @xmath38 . given a @xmath2-orbit @xmath314 fixed by @xmath3 , there is a @xmath14-orbit in @xmath314 that is fixed pointwise by @xmath3 , for suppose that @xmath315 for some @xmath316 . then @xmath317 is fixed by @xmath3 , and moreover , @xmath318 is the orbit we seek . in particular , we can view @xmath318 as a @xmath14-orbit in @xmath13 with moment image @xmath38 . finally , given a @xmath14-orbit @xmath318 in @xmath13 with moment image @xmath38 , this orbit extends to @xmath319 , a @xmath2-orbit fixed by @xmath3 in @xmath0 . we claim that this is the only @xmath14-orbit extending to @xmath319 . suppose that @xmath320 for two distinct @xmath14-orbits @xmath321 and @xmath322 . then there is a @xmath316 such that @xmath323 . but then , @xmath324 . thus , @xmath325 . but now , because @xmath2 acts freely on @xmath18 , the element @xmath326 must be in @xmath14 , so @xmath327 , completing the proof . in this section , we will prove that there is a surjection from the @xmath14-equivariant cohomology of @xmath13 onto the ordinary cohomology of @xmath313 . we will first show this for @xmath328 , and we can then deduce the theorem for all other values @xmath38 by shifting the moment map @xmath33 appropriately . we will use the function @xmath329 as a morse - kirwan function on @xmath13 and apply lemma [ le : atiyahbott ] inductively to show that @xmath330 with @xmath17 coefficients is a surjective map . suppose a torus @xmath2 acts on @xmath0 in a hamiltonian fashion . let @xmath31 be a regular value of the moment map @xmath33 , and assume that @xmath2 acts freely on @xmath331 . suppose further that @xmath13 is the real locus of @xmath0 and that @xmath13 contains no @xmath62-torsion points . then theorem [ th : surjectivity ] states that the map with @xmath17 coefficients @xmath332 induced by inclusion is a surjection . we prove this now . proof of theorem [ th : surjectivity ] let @xmath333 , and let @xmath334 be the finitely many critical values of @xmath335 , where @xmath336 . choose @xmath220 so that @xmath337 and @xmath338 for all @xmath57 . let @xmath339 the set @xmath340 is empty , while @xmath341 is equivariantly homotopy equivalent to @xmath342 . the free @xmath14 action implies @xmath343 in @xmath17 coefficients . we note that @xmath293 is equivariantly homotopy equivalent to @xmath294 and @xmath344 . as @xmath13 has no 2-torsion points , we apply lemma [ le : morsekirwan2 ] to @xmath293 and @xmath345 to obtain a short exact sequence with @xmath17 coefficients @xmath346 thus , we have a surjection @xmath347 over @xmath17 , because @xmath345 is equivariantly homotopy equivalent to @xmath348 . over all @xmath57 we obtain a sequence of surjections in equivariant cohomology with @xmath17 coefficients @xmath349 @xmath350 induced by inclusions @xmath351 the composition is the kirwan map ( [ eq : realkirwan ] ) . we have only proved the theorem for the regular value @xmath328 . suppose @xmath38 is some other regular value of @xmath33 such that @xmath2 acts freely on @xmath18 . since @xmath33 is determined only up to its constant term , we may choose a new moment map @xmath352 . this new moment map will have @xmath353 , and so we may apply the above argument to @xmath354 to prove that there is a surjection @xmath355 we now compute the kernel of the real kirwan map , @xmath67 ( theorem [ th : kernel ] ) . the technique for real loci is identical to the symplectic analogue . however , we must impose some additional requirements on our real locus . first , we require injectivity of the equivariant cohomology of the real locus to the equivariant cohomology of its fixed point set ( theorem [ th : qinjection ] ) . second , we require that the multiplication by the top stiefel - whitney class of the negative normal bundles to certain functions be nonzero . this follows from the proof of lemma [ le : atiyahbott ] . here the @xmath28-equivariant stiefel - whitney class plays the role that the equivariant euler class does in @xcite . following @xcite but restricting our attention to the real locus , we define @xmath69 for any @xmath68 , and @xmath356 where @xmath276 . finally , define the ideal @xmath357 theorem [ th : kernel ] states that @xmath73 is the kernel of @xmath67 . proof of theorem [ th : kernel ] first , we show that @xmath358 . let @xmath359 be a class in @xmath73 for some @xmath68 . the real - valued function @xmath360 is a morse - kirwan function on @xmath13 , in the sense of lemma [ le : morsekirwan1 ] . let @xmath361 be the critical values of @xmath217 , and let @xmath57 be such that @xmath362 . applying lemma [ le : morsekirwan1 ] inductively to the function @xmath217 , we see that @xmath363 . but @xmath364 is homotopy equivalent to @xmath294 , where @xmath365 for sufficiently small @xmath366 . then @xmath367 implies @xmath368 that is , @xmath369 . next , following methods introduced in @xcite , we show that @xmath370 . a class @xmath371 implies that @xmath372 . by the injectivity on the real locus ( theorem [ th : qinjection ] ) , it suffices to find @xmath373 such that @xmath374 for all components @xmath282 of the fixed point set @xmath375 . let @xmath333 , and order the critical sets @xmath376 of @xmath335 with @xmath377 so that the critical values are @xmath378 . note that @xmath379 and @xmath380 we have @xmath381 in particular , critical points consist of @xmath382 , as well as any points @xmath383 such that @xmath384 for all @xmath385 , or @xmath386 is perpendicular to @xmath387 for all @xmath385 . since the differential of @xmath33 is 0 at fixed points , the fixed point set @xmath388 is critical for @xmath335 . note that @xmath389 . now assume that @xmath390 for all @xmath391 and @xmath392 . applying lemma [ le : morsekirwan2 ] to @xmath335 , we see that @xmath393 , where @xmath394 is some multiple of the equivariant top stiefel - whitney class @xmath395 of the negative normal bundle @xmath396 to @xmath397 for @xmath335 . suppose @xmath390 , @xmath398 . we find @xmath399 such that @xmath400 , @xmath401 . then @xmath402 for @xmath401 and we apply the argument inductively to find @xmath403 . let @xmath404 where @xmath405 denotes the connected components of the fixed point set . note that if @xmath406 and @xmath407 , then @xmath399 . the class @xmath408 is constructed by finding a new morse - bott - kirwan function @xmath409 on @xmath13 such that : 1 . @xmath388 is critical for @xmath409 . @xmath397 is critical for @xmath409 , and the negative normal bundle @xmath410 equals @xmath396 . @xmath411 for all @xmath398 . 4 . @xmath412 implies @xmath413 for any component @xmath278 of @xmath388 . provided such a function @xmath409 exists , we construct @xmath408 as follows . let @xmath414 be critical sets of @xmath409 such that @xmath415 . since @xmath397 is critical , we let @xmath416 for all @xmath417 with @xmath418 . by theorem [ le : morsekirwan2 ] we may choose @xmath419 by condition ( 1 ) we have @xmath420 for all connected components @xmath278 in @xmath388 such that @xmath421 it then follows by ( 3 ) that @xmath422 for all @xmath398 by the argument at the beginning of this proof . lastly , we note that by condition ( 4 ) , @xmath399 . it is left to prove such a function exists . this is achieved by perturbing @xmath335 into a family of functions parametrized by @xmath423 , all of which are equivariantly perfect , and then we apply lemma 3.3 to an appropriate perturbation to find @xmath424 . define @xmath425 where @xmath426 . note that for all @xmath383 and @xmath380 , @xmath427 thus @xmath428 is singular on @xmath13 exactly when @xmath429 , when @xmath430 or when @xmath431 is perpendicular to @xmath432 . in particular , @xmath428 is singular on @xmath388 , satisfying condition ( 1 ) . we note also that @xmath397 is critical for @xmath428 ( condition ( 2 ) ) , since for @xmath433 and @xmath380 , @xmath434 this argument also shows that condition ( 3 ) holds : the normal bundles @xmath435 and @xmath436 are equal , since @xmath437 if and only if @xmath438 ( when @xmath433 ) . thus @xmath395 is the equivariant stiefel whitney class of the normal bundle to @xmath397 for @xmath428 . note that each of the functions in this perturbed family is still morse - bott - kirwan , in the sense of lemma [ le : morsekirwan2 ] , since in that lemma , we make no assumption on the value @xmath439 . since @xmath440 , we note that if @xmath413 , then for large enough @xmath216 , @xmath441 . as @xmath13 is compact , there is some @xmath442 such that @xmath443 implies @xmath444 for all connected components @xmath278 of @xmath388 with @xmath445 . thus conditions ( 1)(4 ) are satisfied by @xmath446 . thus there exists @xmath447 and @xmath448 for all @xmath449 , as desired . let @xmath450 be the standard symplectic form on @xmath451 , and let @xmath30 act on @xmath451 by rotation . let @xmath452 with symplectic form @xmath453 . consider the hamiltonian @xmath22 action on @xmath0 which sends @xmath454 then the involution @xmath3 which switches the factors of @xmath0 , @xmath455 is anti - symplectic and anti - commutes with the @xmath30 action . moreover , the real locus of this involution is @xmath456 , the diagonal copy of @xmath451 . the @xmath14 action on @xmath13 is an action by @xmath17 , where the nontrivial element acts by rotation by @xmath457 . let @xmath33 be a moment map for the @xmath30 action on @xmath0 , and let @xmath38 be a regular value . then @xmath458 and @xmath459 . theorem [ th : surjectivity ] implies that there is a surjection @xmath460 the real locus @xmath451 is equivariantly formal , since @xmath461 has classes only in even degrees . thus , as a vector space , @xmath462 the only class in degree @xmath463 is the equivariant class in @xmath464 , and so the only element of @xmath465 which is not in the kernel of @xmath466 is this equivariant class . suppose @xmath0 is a compact hamiltonian @xmath2 space . suppose further that @xmath0 has only finitely many fixed points , and that for @xmath467 , the weights of the isotropy representation of @xmath2 on @xmath468 are pairwise independent over @xmath24 and over @xmath17 . then we say that @xmath0 is a _ @xmath469 gkm space_. such examples include toric varieties , all coadjoint orbits in type @xmath65 , and some ( nonmaximal ) coadjoint orbits in other types . in this case , a result of @xcite states that there is an isomorphism @xmath470 dividing degrees in half . moreover , if we write down each of these rings as subrings of the equivariant @xmath17-cohomology of the appropriate fixed point sets , then the subrings are equal with a change in grading inherited from the module structure . both the left- and right - hand sides of equation ( [ eq : bghequality ] ) are modules over the corresponding cohomology of a point . in both cases , this coefficient ring is the polynomial ring @xmath471 $ ] but the generators are regarded as degree 2 classes on the left , and as degree 1 classes on the right . for example , suppose @xmath472 is a coadjoint orbit of type @xmath473 . in figure 1 , we show an equivariant class @xmath474 , represented by a polynomial associated to each fixed point , pictured here on the moment map image . as a class in @xmath475 , @xmath474 is a degree @xmath62 class . as a class in @xmath476 , it is a degree @xmath463 class . now let @xmath194 be the moment map , and suppose @xmath311 is a regular value such that @xmath2 acts freely on @xmath18 . as a result of kirwan s surjectivity for symplectic reductions and theorem [ th : surjectivity ] , we have two short exact sequences in equivariant cohomology , with @xmath17 coefficients : @xmath290 & k \ar[r]^<<<<<{\kappa } & h_t^{2*}(m;{\mathbb z}/2{\mathbb z})\ar[r]\ar@{=}[d ] & h^{2*}(m/\!/t(\mu);{\mathbb z}/2{\mathbb z})\ar[r ] & 0 \\ 0\ar[r ] & k_{{\mathbb r } } \ar[r]^<<<<<{\kappa_{{\mathbb r } } } & h_{{t_{{\mathbb r}}}}^{*}(q;{\mathbb z}/2{\mathbb z})\ar[r ] & h^{*}(q/\!/{t_{{\mathbb r}}}(\mu);{\mathbb z}/2{\mathbb z})\ar[r ] & 0 . } \ ] ] the tolman - weitsman theorem and theorem [ th : kernel ] imply that the generators of @xmath477 and @xmath73 are identified under the equality . for example , figure 2 demonstrates the class @xmath474 which is a degree @xmath62 class in @xmath478 and is a degree 1 class in @xmath479 , for the @xmath38 and @xmath224 shown . thus , we have a commutative diagram @xmath290 & k \ar[r]^<<<<<{\kappa}\ar@{=}[d ] & h_t^{2*}(m;{\mathbb z}/2{\mathbb z})\ar[r]\ar@{=}[d ] & h^{2*}(m/\!/t(\mu);{\mathbb z}/2{\mathbb z})\ar[r ] & 0 \\ 0\ar[r ] & k_{{\mathbb r } } \ar[r]^<<<<<{\kappa_{{\mathbb r } } } & h_{{t_{{\mathbb r}}}}^{*}(q;{\mathbb z}/2{\mathbb z})\ar[r ] & h^{*}(q/\!/{t_{{\mathbb r}}}(\mu);{\mathbb z}/2{\mathbb z})\ar[r ] & 0 . } \ ] ] by the five lemma , @xmath480 as rings . this result is analogous to duistermaat s original result on the topology of real loci , ; however @xmath77 may not have a hamiltonian torus action . thus this example extends duistermaat s result to symplectic reductions . in the symplectic setting , a toric variety is a compact symplectic @xmath481-dimensional manifold with an effective hamiltonian @xmath63-action . let the moment map be @xmath482 . then @xmath483 is a simple rational convex polytope . delzant @xcite proved that there is a bijection between simple rational convex polytopes in @xmath484 and toric varieties . for every simple rational convex polytope , delzant constructed a toric variety @xmath485 , which is a symplectic reduction of affine space . namely , @xmath486 . danilov @xcite computed the ordinary cohomology of a toric variety in terms of combinatorial data encoded in its moment polytope . he showed that @xmath487/i,\ ] ] where @xmath488 is an ideal of classes depending on how faces of @xmath489 intersect . in this case , it is natural to think of the map @xmath490\to h^*(m_\delta;{\mathbb c})\ ] ] as the kirwan map , and that it is a surjection follows by an argument similar to kirwan s . while @xmath491 is not compact , the compactness assumption can be replaced by the assumption that the moment map is proper and bounded in some direction . the action of @xmath492 on @xmath491 certainly satisfies these assumptions . indeed , an equivariant extension of kirwan s surjectivity , due to the first author @xcite , can also be applied , giving a map @xmath493 where @xmath494 . the kernel computation does not go through in the same way , but we have an alternative description of @xmath495 which allows us to compute the kernel for real loci of toric varieties . we can compute the equivariant cohomology of toric varieties using a combinatorial description due to goresky , kottwitz and macpherson @xcite . provided @xmath485 has no 2-torsion , this combinatorial description also holds for real loci @xcite , and so we know that as rings , @xmath496 moreover , this equality commutes with the two surjections , giving a commutative diagram @xmath290 & k = i\ar[r ] & h^{2*}_{t^d}({\mathbb c}^d;{\mathbb z}/2{\mathbb z } ) \ar@{=}[d]\ar[r ] & h^{2*}_{t^n}({\mathbb c}^d/\!/t^k;{\mathbb z}/2{\mathbb z } ) \ar@{=}[d]\ar[r ] & 0 \\ 0\ar[r ] & k_{{\mathbb r}}\ar[r ] & h^{*}_{{t_{{\mathbb r}}}^d}({\mathbb r}^d;{\mathbb z}/2{\mathbb z } ) \ar[r ] & h^{*}_{{t_{{\mathbb r}}}^n}({\mathbb r}^d/\!/{t_{{\mathbb r}}}^k;{\mathbb z}/2{\mathbb z } ) \ar[r ] & 0 } \ ] ] thus , by the five lemma , @xmath497 . this description of the cohomology of @xmath498 as a quotient of a polynomial ring is identical to danilov s description of the cohomology of @xmath485 . it was proved using other methods by davis and januszkievicz @xcite . t. delzant , hamiltoniens priodiques et images convexes de lapplication moment . [ periodic hamiltonians and convex images of the momentum mapping ] _ bull . soc . math . france _ * 116 * ( 1988 ) , no . 3 , 315339 . s. tolman and j. weitsman , on the cohomology rings of hamiltonian @xmath2-spaces . _ northern ca symplectic geometry seminar _ , 251258 , amer . 2 , vol 196 , amer . math . soc . , providence , ri , 1999 .

biological and cellular systems are often modeled as graphs in which vertices represent objects of interest ( genes , proteins , drugs ) and edges represent relational ties among these objects ( binds - to , interacts - with , regulates ) . this approach has been highly successful owing to the theory , methodology and software that support analysis and learning on graphs . graphs , however , often suffer from information loss when modeling physical systems due to their inability to accurately represent multiobject relationships . hypergraphs , a generalization of graphs , provide a framework to mitigate information loss and unify disparate graph - based methodologies . in this paper , we present a hypergraph - based approach for modeling physical systems and formulate vertex classification , edge classification and link prediction problems on ( hyper)graphs as instances of vertex classification on ( extended , dual ) hypergraphs in a semi - supervised setting . we introduce a novel kernel method on vertex- and edge - labeled ( colored ) hypergraphs for analysis and learning . the method is based on exact and inexact ( via hypergraph edit distances ) enumeration of small simple hypergraphs , referred to as hypergraphlets , rooted at a vertex of interest . we extensively evaluate this method and show its potential use in a positive - unlabeled setting to estimate the number of missing and false positive links in protein - protein interaction networks .

1703.04823

graphs provide a mathematical structure for describing relationships between objects in a system . owing to their intuitive representation , well - understood theoretical properties , the wealth of the algorithmic methodology and available code base , graphs have also become a major framework for modeling biological systems . protein - protein interaction networks , protein 3d structures , drug - target interaction networks , metabolic networks and gene regulatory networks are some of the major representations of biological systems . unfortunately , molecular and cellular systems are only partially observable and may contain significant amount of noise due to their inherent stochastic nature as well as the limitations of both low - throughput and high - throughput experimental techniques . this highlights the need for the development and application of computational approaches for predictive modeling ( e.g. , inferring novel interactions ) and identifying interesting patterns in such data . learning on graphs can be generally seen as supervised or unsupervised . under a supervised setting , typical tasks involve _ graph classification _ ; i.e. , the assignment of class labels to entire graphs @xcite , _ vertex or edge classification _ ; i.e. , the assignment class labels to vertices or edges in a single graph @xcite , or _ link prediction _ ; i.e. , the prediction of the existence of edges in graphs @xcite . alternatively , frequent subgraph mining @xcite , motif finding @xcite , clustering @xcite , and community detection @xcite are traditional unsupervised approaches . regardless of the category , the development of techniques that capture local / global network structure , measure graph similarity and incorporate domain - specific knowledge in a principled manner lie at the core of all these problems . the focus of this study is on classification problems across various biological networks . a straightforward approach to this problem is the use of topological and other descriptors ( e.g. , vertex degree , clustering coefficient , betweenness centrality ) that summarize graph neighborhoods . these descriptors straightforwardly lead to vector - space representations of vertices or edges in the graph , after which standard machine learning algorithms can be applied to learn a target function @xcite . another approach involves the use of kernel functions on graphs @xcite . kernels are mappings of pairs of objects from an input space @xmath0 to an output space @xmath1 with special properties , such as symmetry and positive semi - definiteness , that lead to efficient learning . graph kernels often exploit similar ideas as traditional vector - space approaches . finally , classification on graphs can be approached using probabilistic graphical models such as markov random fields @xcite and related label - propagation @xcite or flow - based @xcite methods . these `` global '' formulations are generally well adjusted to learning smooth functions over neighboring nodes . despite the success and wide adoption of these methods in machine learning and computational biology , it is well - understood that graph representations suffer from information loss since every edge can only encode pairwise relationships @xcite . a protein complex , for instance , can not be distinguished from a set of proteins that interact only pairwise . such disambiguation , however , is important in order to understand the biological activity of these molecules . hypergraphs , a generalization of graphs , naturally capture these higher - order relationships @xcite . as we show later , they also provide a representation that can be used to unify several conventional classification problems on ( hyper)graphs as a single vertex classification approach on hypergraphs . in this paper , we present and evaluate a kernel - based framework for the problems of vertex classification , edge classification and link prediction in graphs and hypergraphs . we first use the concepts of hypergraph duality to demonstrate that all such classification problems can be unified through the use of hypergraphs . we then describe the development of edit - distance hypergraphlet kernels for vertex classification in hypergraphs and combine them with support vector machines into a semi - supervised predictive methodology . finally , we use sixteen biological network data sets , eleven assembled specifically for this work , to provide evidence that the proposed approaches compare favorably to the previously established methods . * graphs . * a graph @xmath2 is a pair @xmath3 , where @xmath4 is a set of vertices ( nodes ) and @xmath5 is a set of edges . in a vertex - labeled graph , a labeling function @xmath6 is defined as @xmath7 , where @xmath8 is a finite alphabet . similarly , in an edge - labeled graph , another labeling function @xmath9 is given as @xmath10 , where @xmath11 is also a finite set . a rooted graph @xmath2 is a graph together with one distinguished vertex called the root . we denote such graphs as @xmath12 , where @xmath13 is the root . a neighborhood graph @xmath14 of a vertex @xmath13 is a rooted graph constructed from @xmath2 such that all nodes at distance at least @xmath15 from @xmath16 ( and corresponding edges ) are removed . in this work we focus on undirected ( the order of the vertices in each edge can be ignored ) , simple graphs ( graphs without self - loops ) . additionally , for the simplicity of presentation , we ignore weighted graphs ; i.e. , graphs where a non - negative number is associated with each vertex . generalization of our approach and terminology to directed and weighted graphs is straightforward . a walk @xmath17 of length @xmath18 in a graph @xmath2 is a sequence of nodes @xmath19 such that @xmath20 , for @xmath21 . if @xmath22 , @xmath17 is called a cycle of length @xmath23 . a path @xmath24 in @xmath2 is a walk in which all nodes are distinct . a connected graph is a graph where there is a path between any two nodes . * hypergraphs . * a hypergraph @xmath2 is a pair @xmath3 , where @xmath4 is the vertex set as previously defined and @xmath25 is a family of non - empty subsets of @xmath4 called hyperedges . as in the case of graphs , one can define a vertex - labeled , edge - labeled , rooted , and neighborhood hypergraphs . a hyperedge @xmath26 is said to be incident with a vertex @xmath16 if @xmath27 and two vertices are called adjacent if there is an edge that contains both vertices . the neighbors of a vertex @xmath16 in a hypergraph are the vertices adjacent to @xmath16 . two hyperedges are said to be adjacent if their intersection is non - empty . finally , the degree @xmath28 of a vertex @xmath16 in a hypergraph is given by @xmath29 , whereas the degree @xmath30 of a hyperedge @xmath26 is defined as its cardinality ; that is , @xmath31 . a walk @xmath17 of length @xmath18 in a hypergraph @xmath32 is a sequence of vertices and hyperedges @xmath33 such that @xmath34 for each @xmath21 and @xmath35 . if @xmath22 , @xmath17 is called a cycle of length @xmath23 . a path @xmath24 in a hypergraph is a walk in which all nodes and edges are distinct . a connected hypergraph is a hypergraph where there exists a path between any two nodes . * isomorphism . * consider two graphs , @xmath36 and @xmath37 . we say that @xmath2 and @xmath38 are isomorphic , denoted as @xmath39 , if there exists a bijection @xmath40 such that @xmath41 if and only if @xmath42 for all @xmath43 . if @xmath2 and @xmath38 are hypergraphs , an isomorphism is defined as interrelated bijections @xmath40 and @xmath44 such that @xmath45 if and only if @xmath46 for all hyperedges @xmath47 . isomorphic graphs ( hypergraphs ) are structurally identical . an automorphism is an isomorphism of a graph ( hypergraph ) to itself . * edit distance . * consider two vertex- and hyperedge - labeled hypergraphs @xmath2 and @xmath38 . the edit distance between these hypergraphs corresponds to the minimum number of edit operations necessary to transform @xmath2 into @xmath38 , where edit operations are defined as insertion / deletion of vertices / hyperedges and substitutions of vertex and hyperedge labels . any sequence of edit operations that transforms @xmath2 into @xmath38 is referred to as an edit path ; hence , the hypergraph edit distance between @xmath2 and @xmath38 corresponds to the length of the shortest edit path between them . this concept can be generalized to the case where each edit operation is assigned a cost . hypergraph edit distance then corresponds to the edit path of minimum cost . , where @xmath48 and @xmath49 @xmath50 with its dual hypergraph @xmath51 , where @xmath52 and @xmath53 such that @xmath54 , @xmath55 , @xmath56 and @xmath57 . panel ( b ) shows an example of graph @xmath2 with two degree - one vertices that lead to the dual hypergraph @xmath58 with self - loops ; @xmath59 and @xmath60 . panel ( c ) shows an extended dual hypergraph that is proposed to formulate link prediction as an instance of vertex classification in hypergraphs . to make a prediction regarding the existence of edge @xmath61 , shown as a dashed line on the left side , an extended dual hypergraph is created in which @xmath61 is added to the set of vertices @xmath62 . updates are made to hyperedges @xmath63 and @xmath60 ( dashed ) that correspond to those vertices in @xmath2 that are incident with the edge @xmath61 . ] let @xmath32 be a hypergraph , where @xmath64 and @xmath65 . the dual hypergraph of @xmath2 , denoted as @xmath51 , is obtained by constructing the set of vertices as @xmath66 and the set of hyperedges as @xmath67 such that @xmath68 . figure [ duality : example]a - b shows two examples of a hypergraph @xmath2 and its dual hypergraph representation @xmath69 . observe that the hyperedges of the original hypergraph @xmath2 are the vertices of the dual hypergraph @xmath69 , whereas the hyperedges of @xmath69 are constructed using the hyperedges of @xmath2 that are incident with the respective vertices . we are interested in binary classification on hypergraphs . the following paragraphs briefly define three distinct classification problems , formulated here so as to naturally lead to the methodology proposed in the next section . * vertex classification . * given a set of rooted hypergraphs @xmath70 , where each @xmath71 corresponds to the same , possibly disconnected , hypergraph @xmath32 rooted at a different vertex of interest @xmath72 . here , one aims to learn some classifier function @xmath73 using a labeled training set @xmath74 , where @xmath75 , as a means of assigning class labels to each unlabeled vertex in @xmath76 . a number of classical problems in computational biology map straightforwardly to vertex classification ; e.g. , protein function prediction @xcite , disease gene prioritization @xcite , and so on . * hyperedge classification . * given a possibly disconnected hypergraph @xmath32 , the objective is to learn a discriminant function @xmath77 from a labeled training set @xmath78 , where @xmath79 , and infer class annotations for every unlabeled hyperedge in @xmath25 . an example of edge classification is the prediction of types of macromolecular interactions such as positive vs. negative regulation . * link prediction . * let @xmath32 be a hypergraph with some missing hyperedges and let @xmath80 be all non - existent hyperedges in @xmath2 ; i.e. , @xmath81 , where @xmath82 represents all possible hyperedges over @xmath4 . the goal is to learn a target function @xmath83 and infer the existence of all missing hyperedges . examples of link prediction include predicting protein - protein interactions , predicting drug - target interactions , and so on . a number of prediction problems in computational biology can be considered within a semi - supervised framework , where a set of labeled and a set of unlabeled examples are used to construct classifiers that discriminate between positive and negative examples . a special category of semi - supervised learning occurs when labeled data contain only positive examples ; i.e. , where the negative examples are either unavailable or ignored ; say , if the set of available negatives is small or biased . such problems are generally referred to as learning from positive and unlabeled data or positive - unlabeled learning @xcite . many prediction problems in molecular biology belong to the open world category ; i.e. , due to various experimental reasons , the absence of evidence of class labels is not the evidence of absence . such problems lend themselves naturally to the positive - unlabeled setting . research in machine learning has recently established tight connections between traditional supervised learning and ( non - traditional ) positive - unlabeled learning . under mild conditions , a classifier that optimizes the ranking performance ; e.g. , area under the roc curve @xcite , in the non - traditional setting has been shown to also optimize the performance in the traditional setting @xcite . similar relationships have been established in approximating posterior distributions @xcite as well as in recovering the true performance accuracy in the traditional setting for a classifier evaluated in a non - traditional setting @xcite . the latter two problems require estimation of class priors ; i.e. , the fractions of positive and negative examples in ( representative ) unlabeled data @xcite . we consider binary classification problems on graphs and hypergraphs and propose to unify all such learning problems through semi - supervised vertex classification on hypergraphs . first , vertex classification falls trivially into this framework . second , the problems of edge classification in graphs and hyperedge classification in hypergraphs are equivalent to the problem of vertex classification on dual hypergraphs . as discussed in section [ duality ] , both graphs and hypergraphs give rise to dual hypergraph representations and , thus , ( hyper)edge classification on a graph @xmath2 straightforwardly translates into vertex classification on its dual hypergraph @xmath58 . we note here that vertices with the degree of one in @xmath2 give rise to self - loops in the dual hypergraph @xmath58 . to account for them , we add one dummy node per self - loop with the same vertex label as the original vertex and connect them with an appropriately labeled edge . third , one can similarly see link prediction as vertex classification on dual hypergraphs , where the set of existing links is treated as positive data , the set of known non - existing links is treated as negative data , and the remaining set of missing links is treated as unlabeled data . this formulation further requires an extension of dual hypergraph representations as follows . consider a particular negative or missing link @xmath84 in the original graph @xmath2 with its dual hypergraph @xmath58 ( fig . [ duality : example]c ) . to make a prediction on this edge @xmath61 , we must first introduce a new vertex @xmath61 in the dual hypergraph as well as modify those hyperedges in @xmath58 that correspond to the vertices @xmath85 in @xmath2 ( fig . [ duality : example]c ) . we denote this extended hypergraph as @xmath86 . it now easily follows that the sets of negative and unlabeled examples can be created by considering a collection of extended graphs @xmath86 , one at a time , for all non - existing vertices @xmath84 or a subset thereof . since most graph data in biological networks lack large sets of representative negative examples , we approach vertex classification , ( hyper)edge classification and link prediction as instances of vertex classification on ( extended , dual ) hypergraphs in a positive - unlabeled setting . we believe this is a novel and useful attempt at generalizing three distinct graph classification problems in a common kernel - based semi - supervised setting . the following sections introduce hypergraphlet kernels that are the core of our classification approach . to @xmath87 vertices , with the root node of each hypergraphlet inscribed in a square . all hypergraphlets are presented in a compressed notation ; e.g. , the two non - isomorphic hypergraphlets @xmath88 and @xmath89 are shown in one drawing . ] * hypergraphlets . * inspired by graphlets @xcite , we define _ hypergraphlets _ as small , simple , connected , rooted hypergraphs . a hypergraphlet with @xmath15 vertices is called an @xmath15-hypergraphlet ; and the @xmath90-th hypergraphlet of order @xmath15 is denoted as @xmath91 . we consider hypergraphlets up to isomorphism and will refer to these isomorphisms as root- and label - preserving isomorphisms when hypergraphs are rooted and labeled . figure [ hypergraphlets ] displays all non - isomorphic unlabeled @xmath15-hypergraphlets with up to three vertices . there is only one hypergraphlet of order @xmath92 ( @xmath93 ; fig . [ hypergraphlets]a ) and one hypergraphlet of order @xmath94 ( @xmath95 ; fig . [ hypergraphlets]b ) . on the other hand , there are nine hypergraphlets of order @xmath87 ( @xmath96 ; fig . [ hypergraphlets]c ) and @xmath97 hypergraphlets of order @xmath98 ( not shown ) . we refer to all these hypergraphlets as _ base hypergraphlets _ since they correspond to the case when @xmath99 . consider now a vertex- and hyperedge - labeled ( or _ fully labeled _ for short ) hypergraphlet with @xmath15 vertices and @xmath100 hyperedges , where @xmath8 and @xmath11 denote the vertex - label and hyperedge - label alphabets , respectively . if @xmath101 and/or @xmath102 , then automorphic structures with respect to the same base hypergraphlet may exist ; hence , the number of fully labeled hypergraphlets per base structure is generally smaller than @xmath103 . for example , if one only considers vertex - labeled @xmath87-hypergraphlets , then there are @xmath104 vertex - labeled hypergraphlets corresponding to the asymmetric base hypergraphlets @xmath88 , @xmath105 and @xmath106 but only @xmath107 corresponding to the base hypergraphlets @xmath108 , @xmath89 , @xmath109 , @xmath110 , @xmath111 , @xmath112 ; see table [ hypergraphs : polya ] . this is a result of symmetries in the base hypergraphlets that give rise to automorphisms among vertex - labeled structures . similarly , if @xmath102 , then new symmetries may exist with respect to the base hypergraphlets that give rise to different automorphisms among hyperedge - labeled structures . in section [ growth : pattern ] , we provide a more detailed discussion on these symmetries . the relevance of these symmetries and enumeration steps relates to the dimensionality of the hilbert space in which the prediction is carried out . motivated by the case for graphs @xcite , we introduce _ hypergraphlet kernels_. let @xmath113 , be a fully labeled hypergraph where @xmath6 is a vertex - labeling function @xmath7 , @xmath9 is a hyperedge - labeling function @xmath10 , and @xmath114 . the vertex- and hyperedge - labeled @xmath15-hypergraphlet count vector for any vertex @xmath13 is defined as @xmath115 where @xmath116 is the count of the @xmath90-th fully labeled @xmath15-hypergraphlet and @xmath117 is the total number of vertex- and hyperedge - labeled @xmath15-hypergraphlets . a kernel function between the @xmath15-hypergraphlet counts for vertices @xmath118 and @xmath16 is defined as an inner product between @xmath119 and @xmath120 ; i.e. , @xmath121 the hypergraphlet kernel function incorporating all hypergraphlets up to the size @xmath122 is given by @xmath123 where @xmath122 is a small integer . in this work we use @xmath124 due to the exponential growth of the number of base hypergraphlets . consider a fully labeled hypergraph @xmath113 . given a vertex @xmath13 , we define the vector of counts for a @xmath125-generalized edit - distance hypergraphlet representation as @xmath126 where @xmath127 here , @xmath128 is the set of all @xmath15-hypergraphlets such that for each @xmath129 there exists an edit path of total cost at most @xmath125 that transforms @xmath130 into @xmath131 and @xmath132 is a user - defined constant . in words , the counts for each hypergraphlet @xmath91 are updated by also counting all other hypergraphlets @xmath133 that are in the @xmath125 vicinity of @xmath91 . the function @xmath134 can be used to adjust the weights of these pseudocounts . we set @xmath135 for all @xmath90 and @xmath136 and the cost of all edit operations was also set to @xmath92 . this restricts @xmath125 to nonnegative integers . the length-@xmath125 edit - distance @xmath15-hypergraphlet kernel @xmath137 between vertices @xmath118 and @xmath16 can be computed as an inner product between the respective count vectors @xmath138 and @xmath139 ; i.e. , @xmath140 finally , the length-@xmath125 edit - distance hypergraphlet kernel function is given as @xmath141 the edit operations considered here incorporate substitutions of vertex labels , substitutions of hyperedge labels , and insertions / deletions ( indels ) of hyperedges . given these edit operations , we also define three subclasses of edit - distance hypergraphlet kernels referred to as vertex label - substitution @xmath142 , hyperedge label - substitution @xmath143 and hyperedge - indel kernels @xmath144 . although the functions from equations ( [ eq : kernel ] ) and ( [ eq : kernel2 ] ) are defined as inner products , other formulations such as radial basis functions can be similarly considered @xcite . we also note that the combined kernels from equations ( [ eq : combined : kernel ] ) and ( [ eq : combined : kernel2 ] ) can be generalized beyond linear combinations @xcite . for the simplicity of this work , however , we only explore equal - weight linear combinations and normalize the functions from equations ( [ eq : kernel ] ) and ( [ eq : kernel2 ] ) using a cosine transformation . the implementation and the analysis of hypergraphlet kernels is an extension of the available solutions for string kernels @xcite . let @xmath145 be a neighborhood hypergraph , as defined in section [ sec : graphs ] and suppose it is significantly smaller than the original hypergraph @xmath2 . the hypergraphlet counting algorithm takes @xmath146 steps , where @xmath147 is the maximum degree of a vertex . similarly , the generation of the minimum cost edit path takes @xmath148 per single hypergraphlet edit operation . therefore , for each vertex @xmath16 an order of @xmath149 operations are necessary , where the @xmath150 term enumerates possible @xmath15-hypergraphlets in @xmath151 . note that the possible number of edges in a hypergraph @xmath152 can be significantly larger than the possible number of edges in a standard graph . hence , in a practical setting , the edit distance hypergraphlet kernels could greatly benefit from effective sampling techniques or exploitation of special types of hypergraphlets . the proposed implementation for computing hypergraphlet kernel functions is computed in time linear in the number of non - zero elements . in this section we summarize classification problems , data sets , and evaluation methodology . the hypergraphlet kernels were evaluated on the problems of edge classification and link prediction , both of which require generation of dual hypergraphs followed by the subsequent vertex classification approach . * protein - protein interaction data . * the protein - protein interaction ( ppi ) data was used for both edge classification and link prediction . in the context of edge classification , we are given a ppi network where each interaction is annotated as either direct physical interaction or a co - membership in a complex . the objective is to predict the type of each interacting protein pair as physical vs. complex ( pc ) . for this task , we used the budding yeast _ s. cerevisiae _ ppi network assembled by ben - hur and noble @xcite . another important task in ppi networks is discovering whether two proteins interact . despite the existence of high - throughput experimental methods for determining interactions between proteins , the ppi network data of all organisms is incomplete @xcite . furthermore , high - throughput ppi data contains a potentially large fraction of false positive interactions @xcite . therefore , there is a continued need for computational methods to help guide experiments for identifying novel interactions . under this scenario , there are two classes of link prediction algorithms : ( 1 ) prediction of direct physical interactions @xcite and ( 2 ) prediction of co - membership in a protein complex @xcite . in this paper , we focused on the former task and assembled nine species - specific data sets comprised solely of direct protein - protein interaction data derived from public databases ( bind , biogrid , dip , hprd , and intact ) as of january 2017 . we considered only one protein isoform per gene and used experimental evidence types described by lewis _ et al . _ specifically , we constructed link prediction tasks for : ( 1 ) bacterium _ e. coli _ ( ec ) , ( 2 ) budding yeast _ s. cerevisiae _ ( sc ) , ( 3 ) nematode worm _ c. elegans _ ( ce ) , ( 4 ) thale cress _ a. thaliana _ ( at ) , ( 5 ) fruit fly _ d. melanogaster _ ( dm ) , ( 6 ) human _ h. sapiens _ ( hs ) , ( 7 ) fission yeast _ s. pombe _ ( sp ) , ( 8) brown rat _ r. norvegicus _ ( rn ) , and ( 9 ) house mouse _ m. musculus _ ( mm ) . * drug - target interaction data . * identification of interactions between drugs and target proteins is an area of growing interest in drug design and therapy @xcite . in a drug - target interaction ( dti ) network , nodes correspond to either drugs or proteins and edges indicate that a protein is a known target of the drug . here we used dti data for both edge classification and link prediction . in the context of edge labeling , we are given a dti network where each interaction is annotated as direct ( binding ) or indirect , as well as assigned modes of action as activating or inhibiting . the objective is to predict the type of each interaction between proteins and drug compounds . for this task , we derived two data sets : ( 1 ) indirect vs. direct ( i d ) binding derived from matador , and ( 2 ) activation vs. inhibition ( ai ) assembled from stitch . under link prediction setting , the learning task is to predict drug - target protein interactions . in particular , we focus on four drug - target classes : ( 1 ) enzymes ( ez ) , ( 2 ) ion channels ( ic ) , ( 3 ) g protein - coupled receptors ( gr ) , and ( 4 ) nuclear receptors ( nr ) ; originally assembled by yamanishi _ et al . _ table [ datasets ] summarizes all data sets used in this work . llllll * type * & * dataset * & & & & + + & & @xmath153 & @xmath154 & @xmath155 & @xmath156 + ppi & pc & @xmath157 & @xmath158 & @xmath159 & @xmath160 + & & @xmath161 drugs & & & + & & @xmath162 targets & & & + & & @xmath163 drugs & & & + & & @xmath164 targets & & & + + & & @xmath153 & @xmath154 & @xmath165 & @xmath166 + & ec & @xmath167 & @xmath168 & @xmath169 & @xmath170 + & sc & @xmath171 & @xmath172 & @xmath173 & @xmath174 + & ce & @xmath175 & @xmath176 & @xmath177 & @xmath178 + & at & @xmath179 & @xmath180 & @xmath181 & @xmath182 + & dm & @xmath183 & @xmath184 & @xmath185 & @xmath186 + & hs & @xmath187 & @xmath188 & @xmath189 & @xmath190 + & sp & @xmath191 & @xmath192 & @xmath193 & @xmath194 + & rn & @xmath195 & @xmath196 & @xmath197 & @xmath198 + & mm & @xmath199 & @xmath200 & @xmath201 & @xmath202 + & & @xmath203 drugs & & & + & & @xmath204 targets & & & + & & @xmath205 drugs & & & + & & @xmath206 targets & & & + & & @xmath207 drugs & & & + & & @xmath208 targets & & & + & & @xmath209 drugs & & & + & & @xmath210 targets & & & + to incorporate domain knowledge into the ppi networks , we exploited the fact that each vertex ( protein ) in the graph is associated with its amino acid sequence . two methods were used to develop vertex alphabet . first , we mapped each protein into a vector of @xmath18-mer ( @xmath211 ) counts and then applied hierarchical clustering on these count vectors . a result of the clustering step assigned one of the @xmath212 vertex labels for each node . second , we used protein sequences to predict their molecular and biological function ( gene ontology terms ) using the fann - go algorithm @xcite . hierarchical clustering was subsequently used on the predicted term scores to group proteins into @xmath213 broad functional categories . in the case of dti data , target proteins were annotated in a similar manner . for labeling drug compounds , we used the chemical structure similarity matrix computed from simcomp @xcite , transformed it into a dissimilarity matrix and then applied hierarchical clustering to group compounds into @xmath214 structural categories . for each data set , we evaluated all hypergraphlet kernels by comparing them to two in - house implementations of random walk kernels on hypergraphs . the random walk kernels were implemented as follows : given a hypergraph @xmath2 and two vertices @xmath118 and @xmath16 , simultaneous random walks @xmath215 and @xmath216 were generated from @xmath118 and @xmath16 using random restarts . however , in contrast to random walks on standard graphs , a random walk in a hypergraph is a two - step process such that at each step one must simultaneously ( 1 ) pick hyperedges @xmath217 and @xmath218 incident with current vertices @xmath118 and @xmath16 respectively , and ( 2 ) pick destination vertices @xmath219 and @xmath220 . this process is repeated until a pre - defined number of steps is reached . in the conventional random walk implementation on hypergraphs , a walk was scored as 1 if the entire sequences of vertex and hyperedge labels between @xmath215 and @xmath216 matched ; otherwise , a walk was scored as 0 . after 10,000 steps , the scores over all walks were summed to produce a kernel value between @xmath118 and @xmath16 . in order to construct a random walk similar to the hypergraphlet edit distance approach , a cumulative random walk kernel was also implemented . here , any match between the labels of vertices @xmath221 and @xmath222 , or hyperedges @xmath223 and @xmath224 in the @xmath90-th step of each walk was scored as 1 , while a mismatch was scored as 0 . thus , a walk of length @xmath225 could contribute between @xmath226 and @xmath225 to the total count . in each of the random walks , the probability of restart was selected from a set @xmath227 and the result with the highest accuracy is reported . on the ppi data sets we also evaluated the performance of pairwise spectrum kernels @xcite . the @xmath18-mer size was varied from @xmath228 and the result with the highest accuracy is reported . finally , in the case of the edit distance kernels , we computed the set of normalized hypergraphlet kernel matrices @xmath229 using @xmath230 , @xmath231 , @xmath232 , and @xmath233 for all pairs @xmath234 obtained from a grid search over @xmath235 , @xmath236 and @xmath237 . the result with the highest accuracy is reported . the performance of each method was evaluated through a 10-fold cross - validation . in each iteration , 10% of nodes in the network are selected for the test set , whereas the remaining 90% are used for training . support vector machine ( svm ) classifiers were used to construct all predictors and perform comparative evaluation . we used svm@xmath238 with the default value for the capacity parameter @xcite . once each predictor was trained , we used platt s correction to adjust the outputs of the predictor to the 0 - 1 range @xcite . finally , we estimated the area under the roc curve ( auc ) , which plots the true positive rate ( sensitivity , @xmath239 ) as a function of false positive rate ( 1 - specificity , @xmath240 ) . llll * dataset / method * & * pc * & * i d * & * ai * + + hypergraphlet kernel ( @xmath241 ) & @xmath242 & @xmath243 & @xmath244 + hypergraphlet kernel ( @xmath245 ) & @xmath246 & @xmath247 & @xmath248 + + random walk & @xmath249 & @xmath250 & @xmath251 + cumulative random walk & @xmath252 & * 0.834 * & @xmath253 + hypergraphlet kernel ( @xmath241 ) & @xmath254 & @xmath255 & @xmath256 + hypergraphlet kernel ( @xmath245 ) & * 0.781 * & @xmath257 & * 0.845 * + we first evaluated the performance of hypergraphlet kernels in the task of predicting the types of interactions between pairs of proteins in a ppi network , as well as interaction types and modes of action between proteins and chemicals in dti data . as described in section [ sec : problem ] we first converted the input hypergraph to its dual hypergraph and then used the dual hypergraph for vertex classification . table [ edge : classification : results ] lists the auc estimates for each method and data set . figure [ auc : curves ] shows roc curves for one representative data set from each classification task and network type . observe that the edit distance kernel ( @xmath245 ) outperformed the traditional hypergraphlet kernel ( @xmath241 ) on all data sets . edit distance kernels achieved the highest aucs on two of the three data sets over random walk kernels . therefore , these results provide evidence of the feasibility of this alternative approach to edge classification via exploiting hypergraph duality . llllllllll * method / dataset * & * ec * & * sc * & * ce * & * at * & * dm * & * hs * & * sp * & * rn * & * mm * + + hypergraphlet kernel ( @xmath241 ) & @xmath258 & @xmath259 & * 0.900 * & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath251 & @xmath264 + hypergraphlet kernel ( @xmath245 ) & @xmath265 & @xmath266 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 & @xmath273 + + random walk & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath258 & @xmath280 & @xmath281 + cumulative random walk & @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath250 & @xmath286 & @xmath287 & @xmath288 & @xmath289 + pairwise spectrum kernel ( @xmath290 ) & @xmath291 & @xmath292 & @xmath293 & * 0.911 * & @xmath294 & @xmath266 & @xmath295 & @xmath296 & @xmath297 + hypergraphlet kernel ( @xmath298 ) & @xmath299 & @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath264 & @xmath304 & @xmath305 & @xmath273 + hypergraphlet kernel ( @xmath306 ) & * 0.742 * & @xmath266 & @xmath307 & @xmath308 & @xmath309 & @xmath310 & @xmath311 & @xmath305 & @xmath312 + hypergraphlet kernel ( @xmath313 ) & @xmath314 & @xmath315 & @xmath316 & @xmath302 & @xmath317 & @xmath318 & @xmath305 & @xmath319 & @xmath320 + hypergraphlet kernel ( @xmath321 ) & @xmath322 & @xmath323 & @xmath307 & @xmath324 & @xmath309 & @xmath309 & * 0.830 * & @xmath325 & @xmath270 + hypergraphlet + pairwise spectrum & @xmath322 & * 0.883 * & @xmath326 & @xmath327 & * 0.858 * & * 0.878 * & @xmath328 & * 0.818 * & * 0.847 * + llllll * method / dataset * & * ez * & * ic * & * gr * & * nr * + + hypergraphlet kernel ( @xmath241 ) & @xmath329 & @xmath318 & @xmath330 & @xmath267 + hypergraphlet kernel ( @xmath245 ) & @xmath329 & @xmath270 & @xmath331 & @xmath332 + + random walk & @xmath333 & @xmath334 & @xmath335 & @xmath320 + cumulative random walk & @xmath336 & @xmath337 & @xmath338 & @xmath305 + hypergraphlet kernel ( @xmath298 ) & @xmath339 & @xmath270 & @xmath340 & @xmath341 + hypergraphlet kernel ( @xmath306 ) & @xmath342 & @xmath330 & @xmath340 & @xmath343 + hypergraphlet kernel ( @xmath313 ) & @xmath344 & @xmath330 & @xmath309 & @xmath345 + hypergraphlet kernel ( @xmath321 ) & * 0.922 * & * 0.863 * & * 0.858 * & * 0.941 * + the performance of hypergraphlet kernels was further evaluated on the problem of link prediction on multiple ppi and dti network data sets . tables [ ppi : lp : results ] and [ dti : lp : results ] show the performance accuracies for each hypergraph - based method across all link prediction data sets . these results demonstrate good performance of our methods , with edit - distance kernels generally having the best performance . the primary objective of our study was to present a new approach whose value will increase as biological data becomes more frequently modeled by hypergraphs . at this time , such data sets are not readily available . we used the alphamax algorithm @xcite for estimating class priors in positive - unlabeled learning to estimate the number of missing links and misannotated ( false positives ) interactions on each ppi network . for example , if we assume a tissue and cellular component agnostic model ( i.e. , any two proteins can interact ) , we obtained that the number of missing interactions on the largest component of the human ppi network ( see table [ datasets ] ) is about 5% ( i.e. , approximately 2.5 million interactions ) , while the number of misannotated interactions is close to 11% which translates to about 4,985 interactions . in the case of yeast , we computed that less than 1% of the potential protein interactions are missing which is close to 95,000 . the number of misannotated interactions is close to 13% , which is about 3,400 misannotated protein pairs . some of these numbers fall within previous studies that suggest that the size of the yeast interactome is between 13,500 @xcite and 137,000 @xcite ; however , the size of the human interactome is estimated to be within 130,000 @xcite and 650,000 @xcite interactions . a recent paper by lewis _ et al . _ @xcite presents a scenario where yeast and human interactome size could reach 400,000 and over two million interactions , respectively . in any case , we note that these estimates were made as a proof of concept for the proposed methodology under the assumption of representative positive data . they however can serve as further validation of the usefulness of our problem formulation and underlying methodology . additional tests and experiments , potentially involving exhaustive classifier and parameter optimization , will be necessary for more accurate and reliable estimates , especially for understanding the influence of potential biases within the ppi network data . the literature on the similarity - based measures for learning on hypergraphs is relatively scarce . most studies revolve around the use of random walks for clustering that were first used in the field of circuit design @xcite . historically , typical hypergraph - based learning approaches can be divided into ( 1 ) tensor - based approaches , which extend traditional matrix ( spectral ) methods on graphs to higher - order relations for hypergraph clustering @xcite , and ( 2 ) approximation - based approaches that convert hypergraphs into standard weighted graphs and then exploit conventional graph clustering and ( semi- ) supervised learning @xcite . the methods from the first category provide a direct and mathematically rigorous treatment of hypergraph learning , although most tensor problems are np - hard . as a consequence , this line of research remains largely unexplored despite a renewed interest in tensor decomposition approaches @xcite . regarding the second category , there are two commonly used transformations for graph - based hypergraph approximation : ( 1 ) the star expansion and ( 2 ) the clique expansion . these methods are reviewed and compared by agarwal _ et al . _ @xcite . under a supervised learning framework , wachman and khardon @xcite propose random walk - based hypergraph kernels on ordered hypergraphs , while sun _ et al . _ @xcite present a hypergraph spectral learning formulation for multi - label classification . more recently , bai _ et al . _ @xcite introduced a hypergraph kernel that transforms a hypergraph into a directed line graph and computes a weisfeiler - lehman isomorphism test between directed graphs . a major drawback of most such approaches is that no graph representation fully captures the hypergraph structure . for instance , ihler _ et al . _ @xcite have shown that it is impossible to have an exact representation of a hypergraph via a graph while still retaining its cut properties . therefore , there is a need for a robust hypergraph - based methodology for learning directly on hypergraph data . this paper presents a learning framework for the problems of vertex classification , ( hyper)edge classification , and link prediction in graphs and hypergraphs . the key to our approach is the use of hypergraph duality in order to cast each classification problem as an instance of vertex classification . this work also presents a new family of kernel functions defined directly on hypergraphs . using the terminology of bleakey _ et al . _ @xcite , our method belongs to the category of `` local '' techniques . that is , it captures the structure of local neighborhoods , rooted at the vertex of interest , and should be distinguished from `` global '' models such as markov random fields or diffusion kernels @xcite . the body of literature on graph learning is vast . we therefore selected to perform extensive comparisons against a limited set of methods that are most relevant to ours . the development of hypergraphlet kernels derives from the graph reconstruction conjecture , an idea of using small graphs to probe large graphs @xcite . hypergraphlet kernels prioritize accuracy over run time and , it may be argued , do not follow some recent trends in machine learning that generally trade off accuracy for improved scalability and real - time performance . we therefore propose that hypergraphlet kernel approaches , in particular those based on edit distances , be predominantly used on sparse graphs of moderate size . fortunately , all graphs used in this work fall into that category . increased accuracy , in general , benefits experimental biologists who typically use prediction to prioritize targets for experimental validation . the proposed methodology was evaluated on multiple data sets for edge classification and link prediction in biological networks . the results show that hypergraphlet kernels are competitive with other approaches and readily deployable in practice . through limited tests , we also find that combining hypergraphlet kernels with pairwise spectrum kernels achieves better accuracy than either of the methods does individually . we thank matthew carey for his help in implementing hyperedge - indel kernels . this work was partially supported by the national science foundation ( nsf ) grant dbi-1458477 , national institutes of health ( nih ) grant r01 mh105524 , and the indiana university precision health initiative . 10 s. agarwal , k. branson , and s. belongie . higher order learning with graphs . in _ proc . 23rd international conference on machine learning _ , icml 06 , pp . 1724 , 2006 . s. agarwal , j. lim , l. zelnik - manor , p. perona , d. kriegman , and s. belongie . beyond pairwise clustering . in _ proc . 18th conference on computer vision and pattern recognition _ , cvpr 05 , pp . 838845 , 2005 . l. bai , p. ren , and e. r. hancock . a hypergraph kernel from isomorphism tests . in _ proc . 22nd international conference on pattern recognition _ , icpr 14 , pp . 38803885 , 2014 . a. ben - hur and w. s. noble . kernel methods for predicting protein - protein interactions . , 21(suppl 1):i38i46 , 2005 . c. berge . . north - holland , 1973 . g. blanchard , g. lee , and c. scott . semi - supervised novelty detection . , 11:29733009 , 2010 . k. bleakley , g. biau , and j - p . vert . supervised reconstruction of biological networks with local models . , 23(13):i57 , 2007 . j. a. bondy and r. l. hemminger . graph reconstruction - a survey . , 1(3):227268 , 1977 . c. borgs , j. chayes , l. lovsz , v. t. ss , and k. vesztergombi . counting graph homomorphisms . in _ topics discrete math _ , algorithms and combinatorics , pp . springer berlin heidelberg , 2006 . s. r. bul and m. pelillo . a game - theoretic approach to hypergraph clustering . in _ proc . 22nd advances in neural information processing systems _ , nips 09 , pp . 15711579 , 2009 . f. chung - graham . . cbms regional conference series in mathematics , 1997 . w. t. clark and p. radivojac . analysis of protein function and its prediction from amino acid sequence . , 79(7):20862096 , 2011 . j. cong , l. hagen , and a. kahng . random walks for circuit clustering . in _ proc . 4th international asic conference _ , asic 91 , pp . p142.1p142.4 , 1991 . f. denis , r. gilleron , and f. letouzey . learning from positive and unlabeled examples . , 348(1):7083 , 2005 . c. elkan and k. noto . learning classifiers from only positive and unlabeled data . in _ proc . 14th international conference on knowledge discovery and data mining _ , kdd 08 , pp . 213220 , 2008 . t. fawcett . an introduction to roc analysis . , 27:861874 , 2006 . s. fortunato . community detection in graphs . , 486(3 - 5):75174 , 2010 . s. m. gomez , w. s. noble , and a. rzhetsky . learning to predict protein - protein interactions from protein sequences . , 19(15):18751881 , 2003 . g. t. hart , a. k. ramani , and e. m. marcotte . how complete are current yeast and human protein - interaction networks ? , 7(11):120 , 2006 . m. hattori , y. okuno , s. goto , and m. kanehisa . development of a chemical structure comparison method for integrated analysis of chemical and genomic information in the metabolic pathways . , 125(39):1185311865 , 2003 . m. hein , s. setzer , l. jost , and s. s. rangapuram . the total variation on hypergraphs - learning on hypergraphs revisited . in _ proc . 26th advances in neural information processing systems _ , nips 13 , pp . 24272435 , 2013 . h. huang , b. m jedynak , and j. s. bader . where have all the interactions gone ? estimating the coverage of two - hybrid protein interaction maps . , 3(11):120 , 2007 . e. ihler , d. wagner , and f. wagner . modeling hypergraphs by graphs with the same mincut properties . , 45(4):171175 , 1993 . s. jain , m. white , and p. radivojac . estimating the class prior and posterior from noisy positives and unlabeled data . in _ proc . 30th advances in neural information processing systems _ , nips 16 , pp . 26932701 , 2016 . s. jain , m. white , and p. radivojac . recovering true classifier performance in positive - unlabeled learning . in _ proc . 31st aaai conference on artificial intelligence _ , aaai 17 , 2017 . s. jain , m. white , m. w. trosset , and p. radivojac . nonparametric semi - supervised learning of class proportions . , 2016 . c. jiang , f. coenen , and m. zito . a survey of frequent subgraph mining algorithms . , 28(01):75105 , 2013 . t. joachims . . kluwer academic publishers , 2002 . s. klamt , u - u . haus , and f. theis . hypergraphs and cellular networks . , 5(5):16 , 2009 . d. koller and n. friedman . . mit press , 2009 . r. i. kondor and j. d. lafferty . diffusion kernels on graphs and other discrete structures . in _ proc . 19th international conference on machine learning _ , icml 02 , pp . 315322 , 2002 . m. leordeanu and c. sminchisescu . efficient hypergraph clustering . in _ proc . 15th international conference on artificial intelligence and statistics _ , volume 22 of _ aistats 12 _ , pp . 676684 , 2012 . a. c. f. lewis , n. s. jones , m. a. porter , and c. m. deane . what evidence is there for the homology of protein - protein interactions ? , 8:114 , 9 2012 . d. liben - nowell and j. kleinberg . the link - prediction problem for social networks . , 58(7):10191031 , 2007 . j. lugo - martinez and p. radivojac . generalized graphlet kernels for probabilistic inference in sparse graphs . , 2(2):254276 , 2014 . s. martin , d. roe , and j - l . faulon . predicting protein - protein interactions using signature products . , 21(2):218226 , 2005 . a. k. menon , b. van rooyen , c. s. ong , and r. c. williamson . learning from corrupted binary labels via class - probability estimation . in _ proc . 32nd international conference on machine learning _ , icml 15 , pp . 125134 , 2015 . r. milo , s. shen - orr , s. itzkovitz , n. kashtan , d. chklovskii , and u. alon . network motifs : simple building blocks of complex networks . , 298:824827 , 2002 . y. moreau and l. c. tranchevent . computational tools for prioritizing candidate genes : boosting disease gene discovery . , 13(8):523536 , 2012 . e. nabieva , k. jim , a. agarwal , b. chazelle , and m. singh . whole - proteome prediction of protein function via graph - theoretic analysis of interaction maps . , 21(suppl 1):i302i310 , 2005 . j. platt . probabilistic outputs for support vector machines and comparisons to regularized likelihood methods . in _ advances in large margin classifiers _ , mit press , pp . 6174 , 2000 . g. plya . kombinatorische anzahlbestimmungen fr gruppen , graphen und chemische verbindungen . , 68:145254 , 1937 . n. przulj . biological network comparison using graphlet degree distribution . , 23(2):e177e183 , 2007 . n. przulj , d. g. corneil , and i. jurisica . modeling interactome : scale - free or geometric ? , 20(18):35083515 , 2004 . p. purkait , t - j . chin , h. ackermann , and d. suter . clustering with hypergraphs : the case for large hyperedges . in _ proc . 13th european conference on computer vision _ , eccv 14 , pp . 672687 , 2014 . j. qiu and w. s. noble . predicting co - complexed protein pairs from heterogeneous data . , 4(4):e1000054 , 2008 . a. k. ramani and e. m. marcotte . exploiting the co - evolution of interacting proteins to discover interaction specificity . , 327(1):273284 , 2003 . h. g. ramaswamy , c. scott , and a. tewari . mixture proportion estimation via kernel embedding of distributions . , 2016 . k. rieck and p. laskov . linear - time computation of similarity measures for sequential data . , 9:2348 , 2008 . r. sharan , i. ulitsky , and r. shamir . network - based prediction of protein function . , 3:88 , 2007 . j. shawe - taylor and n. cristianini . . cambridge university press , cambridge cb2 8ru , uk , 4th edition , 2001 . n. shervashidze , s. v. n. vishwanathan , t. h. petri , k. mehlhorn , and k. m. borgwardt . efficient graphlet kernels for large graph comparison . in _ proc . 12th international conference on artificial intelligence and statistics _ , aistats 09 , pp . 488495 , 2009 . m. p. h. stumpf , t. thorne , e. de silva , r. stewart , h. j. an , m. lappe , and c. wiuf . estimating the size of the human interactome . , 105(19):69596964 , 2008 . l. sun , s. ji , and j. ye . hypergraph spectral learning for multi - label classification . in _ proc . 14th international conference on knowledge discovery and data mining _ , kdd 08 , pp . 668676 , 2008 . k. tsuda and h. saigo . graph classification . in _ managing and mining graph data _ , volume 40 of _ advances in database systems _ , pp . 337363 , 2010 . v. vacic , l. m. iakoucheva , s. lonardi , and p. radivojac . graphlet kernels for prediction of functional residues in protein structures . , 17(1):5572 , 2010 . k. venkatesan et al . an empirical framework for binary interactome mapping . , 6:8390 , 2009 . s. v. n. vishwanathan , n. n. schraudolph , r. i. kondor , and k. m. borgwardt . graph kernels . , 11:12011242 , 2010 . c. von mering , r. krause , r. i. kondor , b. snel , m. cornell , s. g. oliver , s. fields , and p. bork . comparative assessment of large - scale data sets of protein - protein interactions . , 417(6887):399403 , 2002 . g. wachman and r. khardon . learning from interpretations : a rooted kernel for ordered hypergraphs . in _ proc . 24th international conference on machine learning _ , icml 07 , pp . 943950 , 2007 . y. wang and j. zeng . predicting drug - target interactions using restricted boltzmann machines . , 29(13):i126 , 2013 . j. xu and y. li . discovering disease - genes by topological features in human protein - protein interaction network . , 22(22):28002805 , 2006 . y. yamanishi , m. araki , a. gutteridge , w. honda , and m. kanehisa . prediction of drug - target interaction networks from the integration of chemical and genomic spaces . , 24(13):i232i240 , 2008 . l. v. zhang , s. l. wong , o. d. king , and f. p. roth . predicting co - complexed protein pairs using genomic and proteomic data integration . , 5(1):38 , 2004 . d. zhou , j. huang , and b. schlkopf . learning with hypergraphs : clustering , classification , and embedding . in _ proc . 19th advances in neural information processing systems _ , nips 06 , pp . 16011608 , 2006 . x. zhu and z. ghahramani . learning from labeled and unlabeled data with label propagation . in _ technical report cmu - cald-02 - 107 _ , carnegie mellon university , 2002 . here we characterize the feature space of fully labeled hypergraphlets by describing the dimensionality of count vectors @xmath346 . we are interested in the order of growth of @xmath117 as a function of @xmath15 , @xmath8 and @xmath11 . suppose that @xmath2 and @xmath38 are base hypergraphlets with @xmath15 vertices and @xmath100 hyperedges . we say that @xmath2 and @xmath38 belong to the same equivalence class if and only if the total number of ( non - isomorphic ) fully labeled hypergraphlets corresponding to the base cases @xmath2 and @xmath38 are equal for any @xmath8 and @xmath11 . the total counts of labeled hypergraphlets over all alphabet sizes induce a partition of base hypergraphlets into equivalence classes . we denote the set of all equivalence classes over the hypergraphlets of order @xmath15 as @xmath347 . for example , the set of vertex- and hyperedge - labeled @xmath87-hypergraphlets can be partitioned into either : two symmetry classes when @xmath348 : @xmath349 and @xmath350 , or seven symmetry classes when @xmath102 : @xmath351 , @xmath352 , @xmath353 , @xmath354 , @xmath355 , @xmath356 and @xmath357 . table [ hypergraphs : polya ] summarizes equivalence classes induced by partitioning base hypergraphlets up to the order of @xmath98 along with the cardinality of each set . overall , observe that the cardinality of @xmath358 can be significantly larger than those reported for graphlets @xcite because the possible number of hyperedges in a hypergraphlet is generally much larger than the possible number of edges in a graphlet . additionally , hyperedge - labels require base hypergraphlets @xmath2 and @xmath38 to have an equal number of hyperedges . where @xmath360 is the number of ( non - isomorphic ) fully labeled hypergraphlets corresponding to any base hypergraphlet from the equivalence class @xmath361 . we use this decomposition to compute the total dimensionality of the count vectors by first finding the equivalence classes corresponding to the base hypergraphlets and then counting the number of labeled hypergraphlets for any one member of the group . llc + @xmath362 & @xmath363 & @xmath364 + @xmath365 & @xmath92 & @xmath366 + @xmath367 & @xmath92 & @xmath368 + @xmath369 & @xmath87 & @xmath104 + @xmath370 & @xmath371 & @xmath107 + @xmath372 & @xmath373 & @xmath374 + @xmath375 & @xmath376 & @xmath377 ) + @xmath378 & @xmath379 & @xmath380 ) + + @xmath362 & @xmath363 & @xmath360 + @xmath365 & @xmath92 & @xmath366 + @xmath367 & @xmath92 & @xmath381 + @xmath369 & @xmath92 & @xmath382 + @xmath370 & @xmath94 & @xmath383 + @xmath384 & @xmath92 & @xmath385 + @xmath386 & @xmath94 & @xmath387 + @xmath388 & @xmath92 & @xmath389 + @xmath390 & @xmath92 & @xmath391 + @xmath392 & @xmath92 & @xmath393 + in the case of undirected fully labeled hypergraphlets , @xmath360 can also be computed by applying the theory of enumeration developed by plya @xcite . in order to get the derivation of the complete generating function for each equivalence class @xmath361 , we first define the automorphism group @xmath394 of a given vertex- and hyperedge - labeled hypergraph @xmath32 . that is , in the case of fully - labeled hypergraphs , set @xmath394 is a collection of permutations ( automorphisms ) of @xmath4 and @xmath25 . therefore , the counting problem can be re - formulated as follows : let @xmath2 be a base hypergraphlet of @xmath15 vertices and @xmath100 hyperedges , and @xmath394 be the automorphism group of @xmath2 over @xmath4 and @xmath25 . then , each permutation @xmath395 can be written uniquely as the product of disjoint cycles such that for each integer @xmath396 ( @xmath397 ) , we define @xmath398 ( @xmath399 ) as the number of cycles of length @xmath18 ( @xmath400 ) in the disjoint cycle expansion of @xmath401 . interestingly , the generalized formula for the cycle index of @xmath394 , denoted as @xmath402 , is a polynomial in @xmath403 given by @xmath404 by applying plya s theorem in the context of enumerating vertex- and hyperedge - labeled hypergraphlets corresponding to any base hypergraphlet in @xmath405 , we get that @xmath360 is determined by substituting @xmath366 for each variable @xmath406 and @xmath407 for each variable @xmath408 in @xmath402 . hence , @xmath409 where @xmath394 is the automorphism group of a base hypergraphlet from @xmath361 . as an example , consider the equivalence class @xmath410 with @xmath411 and @xmath412 ( figure [ hypergraphlets ] illustrates an unlabeled version of hypergraphlet @xmath112 ) . the automorphism group @xmath413 ; thus , @xmath414 . therefore , it follows that , @xmath415

in this work , we extend the non - orthogonal amplify - and - forward ( naf ) cooperative diversity scheme to the mimo channel . a family of space - time block codes for a half - duplex mimo naf fading cooperative channel with relays is constructed . the code construction is based on the non - vanishing determinant ( nvd ) criterion and is shown to achieve the optimal diversity - multiplexing tradeoff ( dmt ) of the channel . we provide a general explicit algebraic construction , followed by some examples . in particular , in the single - relay case , it is proved that the golden code and the perfect code are optimal for the single - antenna and two - antenna case , respectively . simulation results reveal that a significant gain ( up todb ) can be obtained with the proposed codes , especially in the single - antenna case . cooperative diversity , relay channel , amplify - and - forward ( af ) , multiple - input multiple - output ( mimo ) , space - time block code , non - vanishing determinant ( nvd ) , diversity - multiplexing tradeoff ( dmt ) .

cs0509006

in this paper , we use boldface lower case letters @xmath3 to denote vectors , boldface capital letters @xmath4 to denote matrices . @xmath5 represents the complex gaussian random variable . @xmath6 $ ] stands for the expectation operation and @xmath7}^{{\scriptscriptstyle \mathsf{t } \!}}},{{[\cdot]}^{{\scriptscriptstyle \dag \!}}}$ ] denote the matrix transposition and conjugated transposition operations . @xmath8 is the euclidean vector norm and @xmath9 is the frobenius matrix norm . @xmath10 is the cardinality of the set @xmath11 . @xmath12 means @xmath13 . @xmath14 , @xmath15 , @xmath16 and @xmath17 stand for the real field , complex field , rational field and the integer ring respectively . for any quantity @xmath18 , @xmath19 and similarly for @xmath20 and @xmath21 . on a wireless channel , diversity techniques are used to combat channel fadings . recently , there has been a growing interest in the so called _ cooperative diversity _ techniques , where multiple terminals in a network cooperate to form a virtual antenna array in order to exploit spatial diversity in a distributed fashion . in this manner , spatial diversity gain can be obtained even when a local antenna array is not available . since the work of @xcite , several cooperative transmission protocols have been proposed @xcite . these protocols can be categorized into two principal classes : the amplify - and - forward ( af ) scheme and the decode - and - forward ( df ) scheme . in practice , the af scheme is more attractive for its low complexity since the cooperative terminals ( relays ) simply forward the signal and do not decode it . actually , for most _ ad hoc _ wireless networks , it is not realistic for other terminals to decode the signal from a certain user , because the codebook is seldom available and the decoding complexity is unacceptable in most cases . the non - orthogonal amplify - and - forward ( naf ) scheme was proposed by nabar _ et al . _ @xcite for the single - relay channel and was then generalized to the multiple - relay case by azarian _ et al._@xcite . in @xcite , it is shown that the naf scheme outperforms all previously proposed af schemes in terms of the fundamental diversity - multiplexing tradeoff ( dmt)@xcite and that it is optimal within the class of af schemes in the single - relay case . the superiority of the naf scheme comes from the fact that the source terminal is allowed to transmit during all the time , which boosts up the multiplexing gain . however , even though they showed that the dmt of this scheme can be achieved using a gaussian random code of sufficiently large block length , no practical coding scheme that achieves the tradeoff has been proposed since then . the main contributions of our work are summarized as follows : 1 . we extend the single - antenna naf scheme proposed in @xcite to the multiple - antenna case . we establish a lower bound on the optimal dmt of the mimo naf channel . in particular , we show that the maximum diversity order of a single - relay mimo naf channel is lower - bounded by the sum of the maximum diversity order of the source - destination channel and the maximum diversity orders of the source - relay - destination product channels . this lower bound is tight when the source , relay and destination antenna number @xmath22 , @xmath23 and @xmath24 satisfy @xmath25 2 . we provide an explicit algebraic construction of _ short _ block codes that achieves the optimal dmt of the general multiple - antenna multiple - relay naf cooperative diversity scheme . our algebraic code construction is inspired by the non - vanishing determinant ( nvd ) space - time codes design for mimo rayleigh channels @xcite . first , we show that for any linear fading gaussian channel ( not only the rayleigh channel as in @xcite ) @xmath26 in the high snr regime , the error event of a `` good '' space - time code @xmath27 ( which will be properly defined later ) occurs only when the channel is in outage . therefore , the optimal dmt can always be achieved by @xmath27 . then , as in @xcite , we derive equivalent signal models of the af cooperative schemes in the form ( [ eq : linear_model ] ) , subject to certain input constraint ( _ e.g. _ , block diagonal for multiple - relay channel ) . since codes that achieve the optimal dmt of the equivalent channel ( [ eq : linear_model ] ) also achieve the optimal dmt of the corresponding cooperative channel , optimal codes for an af cooperative channel can be obtained from the nvd criterion . as a result , we show that for a single - relay af channel with @xmath22 antennas at the source terminal , a @xmath28 full rate nvd space - time code ( _ e.g. _ , the golden code @xcite for the single - antenna case and the @xmath1 perfect code @xcite for the two - antenna case ) can be directly applied to construct an optimal block code . in the @xmath0-relay case , the optimal code is constructed from a block - diagonal nvd space - time code with @xmath0 blocks . the performance of our construction is confirmed by simulation results . the rest of the paper is outlined as follows . section [ sec : model ] introduces the system model and recalls the single - antenna naf protocol as well as the equivalent channel model . in section [ sec : mimo_naf ] , we extend the naf scheme to the mimo channel and develop a lower bound on the optimal diversity - multiplexing tradeoff . the codes design criteria are derived in section [ sec : criteria ] and the explicit algebraic construction that satisfies the design criteria is provided in section [ sec : construction ] . section [ sec : some - examples ] shows some examples of channel configuration and the parameters of the corresponding optimal codes . simulation results on our construction are available in section [ sec : results ] . section [ sec : conclusion ] contains some concluding remarks . for continuity of demonstration , most proofs are left in the appendices . we consider a wireless network with @xmath29 sources ( users ) and only one destination . the channels are slow fading ( or delay - limited ) , _ i.e. _ , the channel coherence time is much larger than the maximum delay that can be tolerated by the application . for the moment , we assume that all the terminals are equipped with only one antenna . the multi - antenna case will be treated separately in section [ sec : mimo_naf ] . the channel is shared in a tdma manner , _ i.e. _ , each user is allocated a time slot for the transmission of its own data . within the same time slot , any of the other @xmath0 users can help the current user transmit its information . the extension to a more general orthogonal access scheme is straightforward . suppose that the network configuration is symmetric . without loss of generality , we consider only one time slot and the channel model becomes a single - user relay channel with one source , @xmath0 relays and one destination , as shown in fig . [ fig : relay_model ] . here , we exclude the multi - user case , where information of more than one user can circulate at the same time in the network ( _ e.g. _ , the cma - naf scheme proposed in @xcite ) . ) , one destination ( @xmath30 ) and @xmath0 relays ( @xmath31).,scaledwidth=40.0% ] in fig . [ fig : relay_model ] , variables @xmath32 and @xmath33 stand for the channel coefficients that remain constant during a block of length @xmath34 . as in the previous works we cite here , we assume that all the terminals work in half duplex mode , _ i.e. _ , they can not receive and transmit at the same time . the channel state information ( csi ) is supposed to be known to the receiver but not to the transmitter . in our work , we consider the naf protocol ( @xcite ) . in this scheme , the relays simply scale and forward the received signal . however , unlike the orthogonal af protocols , the source can keep transmitting during the transmission of the relays . is the normalization factor such that @xmath35 is subject to the power constraint.,scaledwidth=50.0% ] in the single - relay case , each frame is composed of two partitions of @xmath36 symbols . the frame length @xmath37 is supposed to be smaller than the channel coherence time @xmath34 , _ i.e. _ , the channel is static during the transmission of a frame . the half duplex constraint imposes that the relay can only transmit in the second partition . the frame structure is illustrated in fig . [ fig : frame_struct ] , from which we get the following signal model @xmath38where @xmath39 , @xmath40 are the transmitted signals from the source with normalized power and the received signals at the destination , respectively , in the @xmath41 partition ; @xmath42 is the received signal at the relay in the first partition ; @xmath43 are independent additive white gaussian noise ( awgn ) vectors with i.i.d . unit variance entries ; the channels between different nodes are independently rayleigh distributed , _ i.e. _ , @xmath44 ; @xmath45 is the geometric gain representing the ratio between the path loss of the source - relay link and the source - destination link ; @xmath46 is the normalization factor satisfying @xmath47 , _ i.e. _ , @xmath48we consider a short term power constraint , _ i.e. _ , the power allocation factors @xmath49 s do not depend on the instantaneous channel realization @xmath50 and @xmath51 , but can depend on @xmath45 and @xmath52 . we impose that @xmath53 so that @xmath52 denotes the average received snr at the destination . since the channel coefficients and the awgn are normalized , @xmath54 represents the average received snr per two partitions as well . ] . as shown in @xcite , the channel model ( [ eq : signal_model ] ) is equivalent to @xmath36 channel uses of a @xmath55 channel : @xmath56where @xmath57\ { { { \lowercase{{\pmb{u}}}}}}_2[i]\bigr]}^{{\scriptscriptstyle \mathsf{t } \!}}}$ ] for @xmath58 and @xmath59 $ ] denotes the @xmath41 symbol in the @xmath60 partition . in the following , we consider a more convenient normalized model @xmath61where @xmath62 is the equivalent awgn and @xmath63 in the multiple - relay case , a superframe of @xmath0 consecutive cooperation frames is defined , as shown in fig . [ fig : frame_struct_nrelays ] . it is assumed that the channel is static during the transmission of the whole superframe ( of @xmath64 symbols ) . the @xmath0 relays take turns to cooperate with the source . within each cooperation frame , the cooperation is in exactly the same manner as in the single - relay case . however , by allowing an encoding over the whole superframe , a diversity order of @xmath29 is achieved . a coding scheme @xmath65 is said to achieve _ multiplexing gain _ @xmath66 and _ diversity gain _ @xmath67 if @xmath68 where @xmath69 is the data rate measured by bits per channel use ( pcu ) and @xmath70 is the average error probability using the maximum likelihood ( ml ) decoder . the optimal dmt of the single - antenna @xmath0-relay naf channel @xmath71is found in @xcite , where the achievability is proved by using a gaussian random code with a sufficiently long block length . in this section , we generalize the naf protocol to the mimo case , where each terminal is equipped with multiple antennas . the notation @xmath72 will be used to denote a single - relay channel with @xmath73 and @xmath24 antennas at the source , relay and destination . all matrix variables defined for a single - relay channel apply to a multiple - relay channel with an index @xmath74 denoting the @xmath41 relay . for convenience of demonstration , we only present the signal model of the single - relay channel . the extension to the multiple - relay case is straightforward . in the case @xmath75 , a direct generalization of ( [ eq : signal_model ] ) is as follows : @xmath76where @xmath77 are @xmath78 and @xmath79 independent matrices , respectively , with zero mean unit variance i.i.d . gaussian entries ; @xmath80 s are @xmath81 matrices with i.i.d . zero mean unit variance entries , representing the space - time signal from the source ; @xmath82 and @xmath83 are independent awgn matrices with normalized i.i.d . entries ; the power allocation factors @xmath49 s satisfy @xmath84 so that @xmath52 denotes the received snr per receive antenna at the destination ; @xmath85 is an @xmath86 matrix equivalent to the `` normalization factor '' @xmath46 in the single - antenna case and is subject to the power constraint @xmath87 which can be simplified to @xmath88now , as in the single - antenna case , we obtain an equivalent single - user mimo channel @xmath89where @xmath90^{{{\scriptscriptstyle \mathsf{t } \!}}}\ { { { \uppercase{{\pmb{x}}}}}}_2[i]^{{{\scriptscriptstyle \mathsf{t } \!}}}\bigr]}^{{\scriptscriptstyle \mathsf{t } \!}}}$ ] and @xmath91^{{{\scriptscriptstyle \mathsf{t } \!}}}\ { { { \uppercase{{\pmb{y}}}}}}_2[i]^{{{\scriptscriptstyle \mathsf{t } \!}}}\bigr]}^{{\scriptscriptstyle \mathsf{t } \!}}}$ ] are the vectorized transmitted and received signals with @xmath92 $ ] denoting the @xmath41 column of the matrix @xmath93 ; @xmath62 is the equivalent awgn ; the equivalent channel matrix @xmath94 is @xmath95with @xmath96and @xmath97 being the whitening matrix satisfying @xmath98 . with more antennas at the relay than at the source , the relay can do better than simple forwarding . in this case , the received signal at the relay is in the @xmath22-dimensional subspace generated by the @xmath22 eigenmodes of @xmath100 , represented by @xmath101 from the singular value decomposition @xmath102 . since the relay - destination channel @xmath103 is isotropic , it is of no use to forward the received signal in more than @xmath23 antennas ( spatial directions ) . however , by using only a subset of the antennas at the relay , we can not obtain all the diversity gain provided by the channel @xmath103 . to exploit all the available diversity , we propose two schemes . the first scheme is the virtual @xmath0-relay scheme . since no csi is available at the transmitter and therefore no antenna combination is _ a priori _ better than the others , one solution is to use all the @xmath104 antenna combinations equally . intuitively , the typical outage event is that all the antenna combinations are in deep fade , which implies that the channel @xmath103 is also in outage . in this scheme , a superframe of @xmath104 cooperation frames is constructed . within each cooperation frame , a different combination of @xmath22 antennas is used . the second scheme is the antenna selection scheme , which is used only when limited feedback from the destination to the relay is available . in this case , the destination tells the relay which antenna combination is optimal according to a given criterion ( _ e.g. _ , maximization of mutual information ) . with this scheme , maximum diversity gain is obtained without a superframe structure , which means a significant reduction of coding - decoding complexity . as an example , let us consider a @xmath105 relay channel with @xmath106 . in this case , the received signal at the relay can be projected into a one - dimensional subspace . consider a superframe of @xmath23 cooperation frames . in each cooperation frame , a different relay antenna is used . this scheme is virtually an @xmath0-relay single - antenna channel . the only difference is that the equivalent source - relay link , which is @xmath107 after the matched filter operation , is the same for the @xmath0 virtual relays . in this scheme , the achievable diversity order of the source - relay - destination link is @xmath23 , since the channel is in outage only when @xmath107 is in deep fade or all the @xmath23 relay - destination links are in deep fade . when feedback is possible , the antenna selection scheme can be used . the difference from the first scheme is that only the antenna with maximum relay - destination channel gain ( say , @xmath108 ) is used in the relaying phase . therefore , only one cooperation frame is needed and the diversity order is also @xmath23 . with the discussion above , considering the case @xmath75 is without loss of generality . in addition , since the destination is usually equipped with more antennas than the relays are in practice , we will restrict ourselves to the case @xmath109 hereafter . in the rest of this section , we study the optimal dmt of the mimo naf cooperative channel . unlike the single - antenna case , a closed form expression of the dmt of the mimo naf channel is difficult to obtain , since the probability distribution ( in the high snr regime or not ) of the eigenvalues of @xmath94 defined in ( [ eq : channelh_mimo ] ) is unknown . in the following , we will derive a lower bound on the tradeoff , as a generalization of the dmt of the single - antenna naf channel provided in @xcite . to this end , we first study the dmt of a rayleigh product channel . [ prop : prod ] let @xmath110 be @xmath111 independent matrices with i.i.d . entries distributed as @xmath112 . assume that @xmath113 and define @xmath114 , then the optimal dmt curve @xmath115 of the rayleigh product channel , _ i.e. _ , the channel defined by @xmath116 with @xmath117 being the awgn is a piecewise - linear function connecting the points @xmath118 , where @xmath119 ^ 2}{2}\right\rfloor.\ ] ] see appendix [ app : proof_prop1 ] . a more general result is given by @xcite , where a rayleigh product channel is seen as a special case of the double scattering channels and the assumption @xmath120 is unnecessary . from proposition [ prop : prod ] , we note that a. @xmath115 only depends on @xmath121 and @xmath18 , which means that interchanging @xmath122 and @xmath123 does not change @xmath115 , which is obvious if we consider the fact that @xmath124 has the same eigenvalues as @xmath125 ; b. @xmath126 is upper - bounded by @xmath127 and coincides with it when @xmath128 ; c. when @xmath129 , @xmath130 . without loss of generality , assume that @xmath131 . intuitively , by increasing @xmath123 and keeping @xmath122 unchanged , the diversity gain of the channel @xmath103 is increasing and so is that of the product channel @xmath132 . when @xmath133 , the best possible dmt for @xmath134 is achieved . with this @xmath123 , the fading effect of @xmath103 vanishes , since @xmath135 . in other words , it is of no use to have @xmath136 , for @xmath122 and @xmath137 fixed . the optimal dmts of a rayleigh product channel for @xmath138 are illustrated in fig . [ fig : dmt_pr ] . as indicated in remark 1 , when @xmath139 for @xmath138 , the tradeoffs of the rayleigh product channels are the same as those of the corresponding rayleigh channels , _ i.e. _ , the @xmath55 and @xmath140 rayleigh channels . [ thm : dmt_mimonaf ] for a single - relay mimo naf channel ( [ eq : signal_model_mimo ] ) , we have @xmath141 where @xmath142 and @xmath143 are the optimal dmt of the fading channel @xmath144 and the fading product channel @xmath145 , respectively . see appendix [ app : proof_thm1 ] . the interpretation of ( [ eq : dnafmimo ] ) is as follows . first , since the transmitted signal passes through the source - destination link all the time , a diversity gain @xmath142 can be obtained . then , due to the half duplex constraint , only half of the transmitted signal is protected by the source - relay - destination link , _ i.e. _ , the channel defined by @xmath145 . fig . [ fig : naf_mimo_dmt ] shows the lower bound ( [ eq : dnafmimo ] ) for a rayleigh channel with @xmath146 and @xmath147 . this result of theorem [ thm : dmt_mimonaf ] can be generalized to the @xmath0-relay case . let @xmath148 and @xmath149 be channels related to the @xmath41 relay and be similarly defined as @xmath103 and @xmath100 . the following theorem gives a lower bound on the optimal dmt of the @xmath0-relay mimo naf channel . [ thm : dmt_mimonaf_n ] for an @xmath0-relay mimo naf channel , we have @xmath150 in particular , when all the relays have the same number of antennas , we have @xmath151 see appendix [ app : proof_thm2 ] . from ( [ eq : dmt ] ) , we see that the bound in ( [ eq : lb3 ] ) is actually the optimal tradeoff in the single - antenna case ( _ i.e. _ , @xmath152 ) . [ coro:1 ] let @xmath153 . then , we have @xmath154 if all the channels are rayleigh distributed and @xmath155 , then , we have @xmath156 see appendix [ app : proof_thm2 ] . in this section , we will derive design criteria for a family of _ short _ codes to achieve the optimal dmt of an @xmath0-relay mimo naf channel . let us first define the `` good '' code mentioned in section [ sec : intro ] . let @xmath157 be an alphabet that is scalably dense , _ i.e. _ , for @xmath158 @xmath159then , an @xmath160 space - time code @xmath27 is called a rate-@xmath123 nvd code if it 1 . is @xmath157-linear is @xmath157-linear means that each entry of any codeword @xmath161 is a linear combination of symbols from @xmath157 . ] ; 2 . transmits on average @xmath123 symbols pcu from the signal constellation @xmath157 ; 3 . has the non - vanishing determinant ( nvd ) property with @xmath162 a constant independent of the snr . ] . the following theorem is fundamental to our construction . [ thm : dmt ] for any linear block fading channel @xmath163where @xmath100 is an @xmath164 channel and @xmath165 is the awgn , the achievable dmt of a rate-@xmath123 nvd code @xmath27 satisfies @xmath166where @xmath167 and @xmath168 is the outage upper bound of the dmt for the channel @xmath100 . see appendix [ app : proof_thm3 ] . in particular , for a full rate code ( @xmath169 ) , the upper bound @xmath168 is achievable . this theorem implies that the nvd property is fundamental for @xmath27 to achieve all the diversity gain @xmath67 , for any linear fading channel . for a given diversity gain @xmath67 , the achievable multiplexing gain @xmath66 of such @xmath27 is a shrunk version of @xmath170 , the best that we can have for channel @xmath100 . one of the consequences of theorem [ thm : dmt ] is the possibility of constructing optimal codes ( in terms of the dmt ) based on the nvd criteria for some channels . for example , we can get an equivalent mimo space - time model for the single - antenna fast fading channel ( also called a gaussian parallel channel ) as @xmath171with @xmath172 and @xmath173 diagonal @xmath174 matrices . the best code that we can have is a rate-@xmath175 nvd code @xmath27 due to the diagonal constraint . according to theorem [ thm : dmt ] , we have @xmath176 with @xmath168 the dmt of ( [ eq : ff ] ) without the diagonal constraint . in fact , we can verify that @xmath177 coincides with the dmt of the fast fading channel . the nvd criterion includes the product distance criterion since the determinant of a diagonal matrix is the product of the diagonal entries . in addition , it implies that the product distance should be non - vanishing as the constellation size increases . note that another such general result as theorem [ thm : dmt ] , has been derived independently in @xcite . in @xcite , the nvd property is derived from the mismatched eigenvalue bound ( worst case rotation ) while the results in @xcite are derived using the worst case codeword error probability , which is effectively the same thing as the worst case rotation . theorem [ thm : dmt ] is a generalization of the result in @xcite ( for the full rate codes ) to a rate-@xmath123 code . this result is more adapted to the algebraic construction of explicit codes for the relay channel . with theorem [ thm : dmt ] , we are ready to give out the design criteria of the optimal codes for the naf cooperative channel . the following theorem states the main result of our work . [ thm : achievability ] let @xmath27 be a rate-(@xmath178 ) nvd block diagonal code , _ i.e. _ , @xmath179 where @xmath180 s are @xmath28 matrices . now consider an equivalent code @xmath181 whose codewords are in the form @xmath182 with @xmath183 then , @xmath181 achieves the optimal dmt of the @xmath0-relay mimo naf channel with @xmath22 transmit antennas at the source , by transmitting @xmath184 in the @xmath41 cooperation frame . the code @xmath181 is of length @xmath185 . see appendix [ app : proof_thm4 ] . in section [ sec : mimo_naf ] , a lower bound on the optimal dmt of a mimo naf channel is derived . here , theorem [ thm : achievability ] shows that the _ exact _ optimal tradeoff can always be achieved by a code @xmath181 , even though we can not obtain its closed form expression . we assume that the modulation used by the source is either a qam or a hex modulation . the fields representing the modulated symbols will be either @xmath186 or @xmath187 . we denote it as @xmath188 . for each algebraic number field @xmath189 , the ring of integers is denoted @xmath190 . we recall that a codeword @xmath191 is represented by a block diagonal matrix @xmath192 with @xmath193 being a square @xmath28 matrix . the criteria to fulfill are the following : 1 . _ full rate _ : the number of qam or hex independent symbols in a codeword is equal to @xmath194 corresponding to a multiplexing gain of @xmath22 symbols pcu ; 2 . _ full rank _ : @xmath195 3 . _ non - vanishing determinant _ : @xmath196 with @xmath162 being some strictly positive constant . we use the same methods as in @xcite . some particular cases can be found in @xcite . the main difference is in the choice of the base field @xmath197 . in @xcite , this base field was equal to @xmath188 . here , we choose a galois extension of @xmath188 with degree @xmath0 and denote @xmath198 the elements of its galois group @xmath199 . now , we construct a cyclic algebra whose center is @xmath197 . we need a cyclic extension over @xmath197 of degree @xmath178 . we denote it @xmath200 . the generator of its galois group is @xmath201 . the code construction needs two steps . 1 . construction of the cyclic algebra @xmath202 such that @xmath203 and @xmath204 . in the matrix representation , we have @xmath205 and @xmath206 . application of the embeddings of @xmath207 . in terms of matrices , we construct , in step 1 , the square @xmath28 matrix @xmath208 . then , by applying the embeddings of @xmath207 , the codeword is @xmath209 where we can identify @xmath210 from ( [ eq : codeword - structure ] ) . as usual , we restrict the information symbols to be in @xmath211 , that is , @xmath212 $ ] ( qam symbols ) or @xmath213 $ ] ( hex symbols ) . so , instead of being in @xmath197 , we will be in @xmath214 and in the same way , we will be in @xmath215 instead of @xmath200 . the infinite space - time code is defined as being the set of all matrices @xmath216 the code @xmath217 of ( [ eq : space - time - code ] ) is full rate . in the submatrix @xmath208 , there are @xmath178 independent elements of @xmath215 . each element in @xmath215 is a linear combination of @xmath178 elements of @xmath214 . finally , each element of @xmath214 is a linear combination of @xmath0 qam or hex symbols . so , each codeword @xmath191 is a linear combination of @xmath194 qam or hex symbols . if @xmath218 , then the code @xmath217 is full rank . in @xcite , it is proved that if @xmath218 , then the cyclic algebra @xmath219 is a division algebra ( each element has an inverse ) . if @xmath218 , then the code @xmath217 has a non - vanishing determinant , more precisely @xmath220 because of the structure of @xmath191 , its determinant is@xmath221 but , @xmath222 is the reduced norm of @xmath223 thus it belongs to @xmath214 . so , @xmath224 with @xmath225 $ ] or @xmath226 $ ] . since @xmath227 unless @xmath228 , we get @xmath229 . finally , the following result is derived . the code @xmath217 of ( [ eq : space - time - code ] ) with @xmath230 or a subspace of @xmath215 ( which will be in the following an ideal of @xmath215 ) achieves the dmt of the mimo naf cooperative channel when @xmath0 is the number of relays and @xmath22 is the number of antennas at the source . the proof is straightforward and uses the results of the @xmath147 above lemmas . as in @xcite , we may be interested in constructing codes that achieve the dmt and that behave well in terms of error probability even for small alphabets such as qpsk ( 4qam ) . in that case , we add another constraint to our codes design , the shaping factor . this new constraint implies that @xmath231 . moreover , as in @xcite , the linear transform that sends the vector composed by the @xmath232 qam or hex information symbols to @xmath233 has to be unitary . the following examples will illustrate this claim . we give some examples of the code construction . our code for an @xmath0-relay @xmath234-antenna channel is denoted @xmath235 . in the case of single - relay single - antenna channel , the codewords are @xmath55 matrices . because the golden code satisfies to all the criteria of subsection [ sub : criteria ] , it achieves the optimal dmt of the channel . optimal codes for the case @xmath236 relays can not be found in the literature . for the @xmath237-relay case , we propose the following code . codewords are block diagonal matrices with @xmath237 blocks . each block is a @xmath55 matrix . let @xmath238 and @xmath239 with @xmath240 be an extension of @xmath186 of degree @xmath237 . we choose @xmath241 . in fact , we try to construct the golden code on the base field @xmath242 instead of the base field @xmath186 . moreover , the number @xmath243 is no more equal to @xmath74 because @xmath74 is a norm in @xmath242 ( @xmath244 ) . we choose here , in order to preserve the shaping of the code , @xmath245 . we prove in appendix [ zeta8-not - norm ] that @xmath246 and thus that this code satisfies to the full rank and the nvd conditions . such a code uses @xmath247 qam symbols . let @xmath248 , @xmath245 and @xmath249 . the ring of integers of @xmath200 is @xmath250\right\ } $ ] . let @xmath251 and @xmath252 . codewords are given by @xmath253 with @xmath254 and @xmath255 changes @xmath256 into @xmath257 . the generalization to @xmath258 relays is straightforward . codewords are block diagonal matrices with @xmath259 blocks . each block is a @xmath55 matrix . let @xmath238 and @xmath260 with @xmath261 be an extension of @xmath186 of degree @xmath259 . we choose @xmath262 . we choose here , in order to preserve the shaping of the code , @xmath263 . we prove in appendix [ zeta16-not - norm ] that @xmath264 and thus that this code satisfies to the full rank and the nvd conditions . since @xmath265 and @xmath266 , we need a code whose codewords are represented by a @xmath1 nvd space - time code . the @xmath1 perfect code of @xcite satisfies to all criteria . we assume here that the source uses @xmath237 antennas and that there are @xmath237 relays . the idea is to construct a @xmath1 perfect code not on the base field @xmath267 as it is the case in @xcite , but on the base field @xmath268 . thus , a rate-@xmath237 nvd code can be constructed as follows : * take @xmath269 . * choose @xmath270 . * finally , take @xmath245 . we can show , in the same way as in appendix [ zeta8-not - norm ] , that if @xmath271 was a norm in @xmath272 , then @xmath74 must be a norm in @xmath273 which contradicts the results of @xcite . the case of @xmath274 is obvious since @xmath275 . now , in order to prove that @xmath276 is not a norm , it is enough to replace @xmath256 by @xmath277 and @xmath278 by @xmath279 in appendix [ zeta8-not - norm ] and show in the same way that if @xmath277 was a norm in @xmath280 , then @xmath74 must be a norm in @xmath273 . in this section , we provide the simulation results on the performance of some of the codes proposed in section [ sec : some - examples ] . the performance is measured by the frame error rate ( fer ) _ vs. _ receive snr per bit . for simplicity , we set the power allocation factors @xmath281 for all the scenarios that we considered in this section . an optimization on the @xmath49 s in function of @xmath45 and @xmath52 can improve the performance and @xmath52 are high enough to give a better performance over the non - cooperative case . ] . however , this kind of optimization is out of the scope of this paper and will not be considered here . the transmitted signal constellation is @xmath259- and @xmath282-qam . the geometric gain @xmath45 varies from @xmath283 to @xmath284db . [ fig : r1 ] shows the performance of the golden code on the single - relay single - antenna channel . the performance of the channel without relay is also shown in the figures . in this case , the frame length is @xmath259 symbols . compared to the non - cooperative case , the golden code achieves diversity @xmath237 . for @xmath259-qam , a gain of @xmath285db ( _ resp . _ @xmath286 and @xmath287db ) is observed for @xmath288db ( _ resp . _ @xmath289 and @xmath284db ) at @xmath290 . first of all , note that in the low snr regime , the non - cooperative channel is better than the cooperative channel . this is due to the error cumulation ( at the relay ) which is more significant than the diversity gain provided by the relay in this regime . then , we see that the difference between @xmath291db and @xmath292db is negligible , which means that a geometric gain of @xmath2db is enough to achieve the ( almost ) best performance of the golden code . in practice , it is often possible to find this kind of `` helping agent '' ( with a geometric gain of @xmath2db ) . when we increase the spectral efficiency ( @xmath282-qam ) , same phenomena can be observed except that the gain of the relay channel is reduced . still , a gain of @xmath293 and @xmath294db can be obtained at @xmath290 for @xmath288 and @xmath284db . the performance of @xmath295 on the four - relay single - antenna channel is illustrated in fig . [ fig : r4 ] . the frame length is @xmath296 . for @xmath292db , a gain of @xmath297db ( _ resp . _ @xmath298db ) at @xmath290 is obtained with @xmath259-qam ( _ resp . _ @xmath282-qam ) . as an example for the case @xmath300 , we consider the @xmath299 channel . here , we use the virtual @xmath0-relay scheme with @xmath301 . as discussed in section [ sec : n_r - geq - n_s ] , the diversity order of this scheme is comparable to the @xmath237-relay single - antenna channel . to compare these two channels , we use the same code @xmath302 . the performance is shown in fig . [ fig : comp_121_211 ] . as compared to the @xmath237-relay single - antenna channel , the @xmath299 channel has a gain of @xmath303db at @xmath290 with @xmath288db . with @xmath292db , the two channels have essentially the same performance . in fact , the inter - relay cooperation in the virtual two - relay channel improves the receive snr ( @xmath147db ) of the source - relay channel with antenna combining . thus , the geometric gain @xmath45 is increased effectively . however , as stated before , the global performace is not sensitive to @xmath45 for large @xmath45 s . this is why there is a gain only with small @xmath45 s . in the case of @xmath305 , we consider a single - relay channel with two antennas at each terminal . the code @xmath306 is actually the @xmath1 perfect code . for the non - cooperative scenario , we take the same code for fairness of comparison . more precisely , the non - cooperative channel we consider here is equivalent to a cooperative channel with @xmath307 and @xmath308 . as shown in fig . [ fig : r1_mimo ] , the gain of the cooperative channel over the non - cooperative channel is much less significant in the snrs of interest . this is because the diversity order of the perfect code in the two - antenna non - cooperative channel is already @xmath259 and a diversity gain does not play an important role in the scope of interest . note that at @xmath309 , the gain of the cooperative channel with @xmath288db over the non - cooperative channel is @xmath237db for @xmath259-qam and @xmath147db for @xmath282-qam . also note that the difference between different @xmath45 s is within @xmath175db . in this scenario , we consider the shadowing effect of a wireless channel . assume that each link between terminals is shadowed . mathematically , the channel matrix is multiplied by a random scalar variable , the shadowing coefficient . suppose that this variable is log - normal distributed of variance @xmath310db@xcite and that the shadowing is independent for different links . [ fig : mimoshadowed ] shows the performance of the cooperative channel with the use of @xmath306 ( @xmath259-qam ) at @xmath288db , as compared to the non - cooperative channel . fer is the averaged frame error rate on the channel fading and the shadowing . as shown in fig . [ fig : mimoshadowed ] , the slope of the fer - snr curve of the non - cooperative channel is reduced as compared to the non - shadowing case , in the scale of interest . since the shadowing is independent between different links , the cooperative channel mitigates the shadowing effect and we get a larger gain over the non - cooperative channel than in the non - shadowing case ( fig . [ fig : r1_mimo_4qam ] ) . at @xmath290 , this gain is @xmath247db , in contrast to @xmath311db in the non - shadowing case ( fig . [ fig : r1_mimo_4qam ] ) . in this paper , a half - duplex mimo amplify - and - forward cooperative diversity scheme is studied . we derived the optimal diversity - multiplexing tradeoff of a mimo rayleigh product channel , from which we obtain a lower bound on the optimal diversity - multiplexing tradeoff of a mimo naf cooperative channel . moreover , we established a lower and upper bound on the maximum diversity order of the proposed mimo naf channel and showed that they coincide when the numbers of antenna satisfy certain conditions . based on the non - vanishing determinant criterion , we constructed a family of short space - time block codes that achieve the dmt of our mimo naf model . our construction is systematic and applies to a system with arbitrary number of relays and arbitrary number of antennas . numerical results on some explicit example codes revealed that significant gain in terms of snr can be obtained even with some non - optimized parameters . this gain is much more important in the single - antenna case than in the mimo case . fortunately , in reality , it is also the case that we need cooperative diversity only when local antenna array is not available . nevertheless , it still remains two important open problems to solve : 1 . optimization of the power allocation factors : based on the statistical knowledge of the channel ( notably @xmath45 ) , how to choose the factors @xmath49 s in order to optimize the code performance according to certain criteria ? we set @xmath281 for simplicity . it is clear that the optimal dmt is independent of these parameters . however , in practice , for different @xmath45 and @xmath52 , the factor @xmath49 s are significant for the performance ( _ e.g. _ , the error rate performance ) . how to analyze the impact theoretically is an interesting future work ; 2 . optimization of the matrix @xmath85 : we set @xmath85 to be identity matrix and derived the lower bound ( [ eq : dnafmimo ] ) . based on the receiver csi at the relay , is there an optimal matrix @xmath85 that gives a better dmt than the lower bound ( [ eq : dnafmimo ] ) ? this problem is independent of the code we use . solving this problem may lead to solution for the exact dmt of the mimo naf channel . for sake of simplicity , we use the _ dot ( in)equalites _ throughout the proofs to describe the behavior of different quantities in the high snr regime . more precisely , * for probability related quantities , @xmath312 * for mutual information related quantities , @xmath313 * for sets , @xmath314 @xmath21 , @xmath20 , @xmath315 and @xmath316 are similarly defined . for any nonnegative random variable @xmath317 , the _ exponential order _ is defined as @xmath318 we denote @xmath319 . [ lemma : pdfexp ] let @xmath320 be a @xmath321-distribution random variable with @xmath322 degrees of freedom , the probability density function of its exponential order @xmath323 satisfies let @xmath11 be a certain set , @xmath324 s be independent random variables with @xmath325 , and @xmath326 , then we have @xmath327where @xmath328 . _ in the high snr regime , the outage probability is @xcite @xmath329 with @xmath330 . let us define @xmath331 and @xmath332 . the entries of @xmath103 and @xmath100 being i.i.d . rayleigh distributed , @xmath333 and @xmath334 are two central complex wishart matrices @xcite . let @xmath335 and @xmath336 be the ordered eigenvalues of @xmath333 and @xmath83 , then we have @xcite @xmath337 \end{aligned}\right.\]]with @xmath338 and @xmath339 being the normalization factors . hence , the joint pdf of @xmath340 is @xmath341,\end{aligned}\]]where @xmath342 is the normalization factor . define @xmath343 and @xmath344 for @xmath345 . then , we have @xmath346 . \label{eq : pdfjoint}\end{aligned}\]]first , we only consider @xmath347 , since otherwise , @xmath348 decays exponentially with @xmath52@xcite . then , we can show that @xmath349 { \doteq}{{\mathsf{snr } } } ^{-\sum_{k=1}^l\sum_{i < k}(\alpha_i-\beta_k)^+}.\]]to see this , let us rewrite @xmath350_{i , j=1}^l$ ] as @xmath351where the equations are obtained by iterating the identity @xmath352 for @xmath353 . since @xmath354 for @xmath317 close to @xmath355 , we have @xmath356 if @xmath357 and @xmath358 otherwise . as shown in the recursive relation above , we must have @xmath359 , so that @xmath360 does not decay exponentially . in this case , we have @xmath361 . thus , we have @xmath362 , and in a recursive manner , we get ( [ eq : detexp ] ) . finally , we can write the outage probability as @xmath363where @xmath364is the outage region in terms of @xmath365 in the high snr regime . since @xmath366(_resp . _ @xmath367 ) is dominated by @xmath368(_resp . _ @xmath369 ) for @xmath370 , from ( [ eq : detexp ] ) and ( [ eq : pdfjoint ] ) , we have d _ , @xmath371 with @xmath372 let @xmath373 . then , we have @xmath374the optimization problem ( [ eq : optpbl ] ) can be solved in two steps : 1 ) find optimal @xmath375 by fixing @xmath376 , and then 2 ) optimize @xmath376 . let us start from the feasible region @xmath377 in which we have @xmath378 . note that the feasibility conditions require that @xmath379 s can only move to their left in terms of their positions relative to the @xmath380 s and that @xmath379 can never be on the left of @xmath381 for @xmath382 . let @xmath383 denote the coefficients of @xmath379 in @xmath384 . the initial values of @xmath383 s are @xmath385 where @xmath386 . as long as @xmath383 is positive , @xmath379 should decrease ( pass the @xmath387 just left to it ) to make the objective function @xmath384 smaller , with the feasibility conditions being respected . each time a @xmath379 passes a @xmath387 from right to left with @xmath388 , @xmath383 decreases by @xmath175 . when @xmath389 , @xmath379 should stop decreasing . therefore , the optimal region is such that @xmath390^+,\quad i=1,\ldots , l.\ ] ] for @xmath74 such that @xmath391 , @xmath392 . for @xmath393 , @xmath394 is @xmath395 if @xmath396 and @xmath283 otherwise . note that @xmath397 s are independent of @xmath380 s and only depend on @xmath398 . this is why we can separate the optimization problem into two steps . after replacing the optimal @xmath375 in @xmath384 and some basic manipulations , we obtain @xmath399where @xmath400 is non - negative and is non - increasing with @xmath234 . hence , the optimal solution is @xmath401 and @xmath402 , from which we have @xmath403 . for @xmath404 , @xmath405for @xmath406 , @xmath407by combining ( [ eq : x ] ) and ( [ eq : xx ] ) , we get ( [ eq : dmt_pr ] ) . the main idea of the proof is to get lower bounds on the dmt by lower - bounding the mutual information of the channel defined by ( [ eq : channel_mimo ] ) and ( [ eq : channelh_mimo ] ) . since the multiplicative constants have no effects on the dmt , for simplicity of demonstration , we will neglect them and rewrite @xmath408 and @xmath409the mutual information of the channel @xmath94 is @xmath410 [ lemma : sigma ] let @xmath411 be the generalized inequality for matrices means that @xmath412 is positive semidefinite . ] , then @xmath413 and there exists a matrix @xmath85 satisfying the power constraint ( [ eq : constraintb ] ) such that @xmath414 ( [ eq : sigma ] ) comes from the definition of @xmath415 . ( [ eq : lmaxpp ] ) can be shown by construction . let us take @xmath416 then , the power constraint ( [ eq : constraintb ] ) is always satisfied with @xmath417 . since @xmath418 , we have @xmath419where @xmath420 is the exponential order of @xmath421 and is positive with probability @xmath175 in the high snr regime@xcite . by lemma [ lemma : sigma ] and the concavity of @xmath422 on positive matrices , we have @xmath423with @xmath424 therefore , with @xmath85 in ( [ eq : b ] ) , we have @xmath425 . assume that in the rest of the proof , we always consider @xmath85 being in the form ( [ eq : b ] ) . then we have @xmath426 where @xmath427 is the set of matrices @xmath85 that satisfy the power constraint ( [ eq : constraintb ] ) . define @xmath428 , we have @xmath429using the identity @xmath430and some basic manipulations , we have @xmath431where @xmath432 is positive definite . by the matrix inversion lemma @xmath433 , we have @xmath434from ( [ eq : ixy ] ) and ( [ eq : detm ] ) , we can obtain two lower bounds on @xmath435 . the first one is @xmath436whereas the second one is @xmath437since in ( [ eq : b ] ) , @xmath438 , from ( [ eq : def_p ] ) and ( [ eq : tmp ] ) , we have @xmath439 the outage probability is rcl \{_<2r } & & \{2(+ ^)2r , ( + ^^ ) 2r } + & = & \{2(+ ^)2r } + & & \{(+ ^^ ) 2r } + & & ^-(d_(r)+d_(2r ) ) where the second line follows from the independency between @xmath144 and @xmath145 . as in the case of the single - relay channel , we need two lower bounds on the mutual information . since the mutual information of the @xmath0-relay channel is the sum of that of the @xmath0 single - relay channels , these two lower bound can be obtained directly from ( [ eq : lb11 ] ) and ( [ eq : lb2 ] ) @xmath440[c]{l'l'l } \ixy & { \ \dot{\geq}\ , } & 2n\log\det({\mathbf{i}}+{{\mathsf{snr } } } { { { \uppercase{{\pmb{f}}}}}}{{{{{\uppercase{{\pmb{f}}}}}}}^{{\scriptscriptstyle \dag \ ! } } } ) \\ \ixy & { \ \dot{\geq}\ , } & \sum_{i=1}^n \log\det({\mathbf{i}}+ { { \mathsf{snr } } } { { { { { \uppercase{{\pmb{h}}}}}}}^{{\scriptscriptstyle \dag \!}}}_i{{{{{\uppercase{{\pmb{g}}}}}}}^{{\scriptscriptstyle \dag \!}}}_i{{{\uppercase{{\pmb{g}}}}}}_i{{{\uppercase{{\pmb{h}}}}}}_i ) \end{ieeeeqnarraybox}\]]the outage probability is upper bounded by @xmath441 let us denote @xmath442 and @xmath443 the set of exponential orders of the ordered eigenvalues of @xmath444 . the pdf of @xmath443 is @xmath445 where from ( [ eq : da ] ) , @xmath446 is nondecreasing with respect to the component - wise inequality , _ i.e. _ , @xmath447let us define @xmath448then , the outage probability is @xmath449where @xmath450 is @xmath451 with the second equality from the fact that the minimal elements lie always in the boundary when ( [ eq : nondec ] ) is true . for @xmath452 identically distributed for all @xmath74 , ( [ eq : lb3 ] ) is obtained by the convexity of @xmath453 . for simplicity , we prove the particular case @xmath266 here . for @xmath236 , same method applies . the lower bound is a direct consequence of theorem [ thm : dmt_mimonaf_n ] . the upper bound can be found by relaxing the half duplex constraint , _ i.e. _ , @xmath454 $ ] with all matrices being similarly defined as before . define @xmath455 $ ] . first , since @xmath456 , we have @xmath457 which means that the channel @xmath94 is asymptotically worse than the channel @xmath458 $ ] in the high snr regime . thus , we have @xmath459 . then , since @xmath460 , another bound is @xmath461 from which we have @xmath462 . when the channel is rayleigh , proposition [ prop : prod ] applies . as indicated in remark 1 , we have @xmath463 for @xmath464 and the lower bound and the upper bound in ( [ eq : max_d ] ) match . ^ _ i to prove theorem [ thm : dmt ] , it is enough to show that in the high snr regime , an error occurs with the rate-@xmath123 nvd code @xmath27 only when the channel is in outage for a rate @xmath465 . to this end , we will show that the error event set of @xmath27 is actually included in the outage event set , in the high snr regime . for a channel @xmath100 , the outage event at high snr is @xcite @xmath466let us develop the determinant as where @xmath467 s is the eigenvalues of @xmath93 . ] @xmath468where @xmath469 is the sum of @xmath470 products of @xmath74 different eigenvalues of @xmath93 . in particular , we have @xmath471 and @xmath472 . let @xmath467 denote the @xmath41 smallest eigenvalue of @xmath473 and @xmath380 denote the exponential order of @xmath467 , _ i.e. _ , @xmath474 with @xmath475 . then , we have @xmath476since @xmath477 is the smallest among all the combinations of @xmath74 different @xmath478 s . now , we are ready to write @xmath479 let us now consider the error event of a rate-@xmath123 nvd code @xmath27 . we will follow the footsteps of @xcite . using the sphere bound , the error event of ml decoding conditioned on a channel realization @xmath100 is @xmath480where @xmath83 is the awgn matrix with i.i.d . entries ; @xmath481 is the minimum euclidean distance between two received codewords , _ i.e. _ , @xmath482 ; @xmath483 is the exponential order of @xmath484 ( @xmath485 ) and @xmath486 is that of @xmath487 . therefore , the error probability conditioned on @xmath100 is @xmath488where by lemma [ lemma : pdfexp ] , we have then the average error probability becomes @xmath489therefore , we get the error event in the high snr regime @xmath490with @xmath323 being any lower bound on @xmath486 . using the same arguments as in @xcite , with a rate-@xmath123 nvd code , we can get @xmath18 lower bounds on @xmath491 @xmath492with @xmath493finally , from ( [ eq : allbound ] ) and ( [ eq : lb_dmin ] ) , we get @xmath494 which implies that @xmath495 p_^,n p_^ consider the channel @xmath496 with @xmath497 being similarly defined as @xmath94 in ( [ eq : channelh_mimo ] ) except that @xmath498 in ( [ eq : channelh ] ) are replaced by @xmath499 , respectively . since one channel use of @xmath500 is equivalent to @xmath501 channel uses of an @xmath0-relay naf channel , _ i.e. _ , @xmath502 where @xmath503 and @xmath504 are the capacities of the channel @xmath500 and the equivalent @xmath0-relay naf channel , measured by bits per channel use . therefore , we have @xmath505 where the first equality comes from the fact that the outage upper bound of the tradeoff can be achieved @xcite and the second comes from ( [ eq : capa ] ) and the definition of outage since @xmath506 on the other hand , by using a code @xmath181 defined above , an equivalent channel model of the @xmath0-relay channel is @xmath507 with @xmath508 . by theorem [ thm : dmt ] , we have @xmath509from ( [ eq : dn ] ) and ( [ eq : dccal ] ) , we obtain @xmath510 we prove , in this appendix , that @xmath256 is not a norm of an element of @xmath512 . assume that @xmath256 is a norm in @xmath200 , _ i.e. _ , @xmath513 consider now the extensions described in fig . [ cap : fields - extension ] . from ( [ eq : x_existence ] ) , by considering the left extension of fig . [ cap : fields - extension ] , we deduce that @xmath514 now , we deduce , from the right extension of figure [ cap : fields - extension ] that @xmath515 denote @xmath516 . then , the number @xmath517 has an algebraic norm equal to @xmath74 , and belongs to @xmath518 . in @xcite , it has been proved that @xmath74 was not a norm in @xmath518 . so , @xmath256 is not a norm in @xmath200 . the proof is similar to the one of appendix [ zeta8-not - norm ] . first , we assume that @xmath519 is a norm in @xmath521 , _ i.e. _ , @xmath522 we deduce that @xmath523 but we also have , @xmath524 denote @xmath516 . then the number @xmath517 has an algebraic norm equal to @xmath74 and belongs to @xmath518 , which is a contradiction . the authors would like to thank the anonymous reviewers for their valuable comments . j. n. laneman and g. w. wornell , `` distributed space - time - coded protocols for exploiting cooperative diversity in wireless networks , '' _ ieee trans . inform . theory _ 49 , no . 10 , pp . 24152425 , oct . j. n. laneman , d. n. c. tse , and g. w. wornell , `` cooperative diversity in wireless networks : efficient protocols and outage behavior , '' _ ieee trans . inform . theory _ 50 , no . 30623080 , dec . 2004 . k. azarian , h. el gamal , and p. schniter , `` on the achievable diversity - multiplexing tradeoff in half - duplex cooperative channels , '' _ ieee trans . inform . theory _ 51 , no . 12 , pp . 41524172 , dec . 2005 . p. elia , k. r. kumar , s. a. pawar , p. v. kumar , and h. lu , `` explicit , minimum - delay space - time codes achieving the diversity - multiplexing gain tradeoff , '' _ ieee trans . inform . theory _ 2004 , submitted for publication . [ online ] . available : http://ece.iisc.ernet.in/~vijay/csdpapers/explicit.pdf belfiore , g. rekaya , and e. viterbo , `` the golden code : a @xmath55 full - rate space - time code with non - vanishing determinants , '' _ ieee trans . inform . theory _ 51 , no . 4 , pp . 14321436 , apr . 2005 . s. yang and j .- c . belfiore , `` diversity - multiplexing tradeoff of double scattering mimo channels , '' _ ieee trans . inform . theory _ 2006 , submitted for publication . [ online ] . available : http://arxiv.org/pdf/cs.it/0603124

flows of dense emulsions show many complex features among which long range nonlocal effects pose problem for macroscopic characterization . in order to get round this problem , we study the flows of several dense emulsions , with droplet size ranging from 0.3 to 40 m , in a wide gap couette geometry . we couple macroscopic rheometric experiments and local velocity measurements through mri techniques . as concentration heterogeneities are expected in the wide gap couette flows of multiphase materials , we also designed a new method to measure the local droplet concentration in emulsions with a mri device . in contrast with dense suspensions of rigid particles where very fast migration occurs under shear in wide gap couette flows , we show for the first time that no migration takes place in dense emulsions even for strain as large as 100000 in our systems . as a result of the absence of migration and of finite size effect , we are able to determine very precisely the local rheological behavior of several dense emulsions . as the materials are homogeneous , this behavior can also be inferred from purely macroscopic measurements . we thus suggest that properly analyzed purely macroscopic measurements in a wide gap couette geometry can be used as a tool to study the local constitutive laws of dense emulsions . all behaviors are basically consistent with herschel - bulkley laws of index 0.5 . the existence of a constitutive law accounting for all flows contrasts with previous results obtained within a microchannel by goyon _ et al . _ [ nature * 454 * , 84 ( 2008 ) ] : the use of a wide gap couette geometry is likely to prevent here from nonlocal finite size effects ; it also contrasts with the observations of bcu _ et al . _ [ phys . rev . lett . * 96 * , 138302 ( 2006 ) ] . we also evidence the existence of discrepancies between a perfect herschel - bulkley behavior and the observed local behavior at the approach of the yield stress due to slow shear flows below the apparent yield stress in the case of a strongly adhesive emulsion .

0810.3497

emulsions are mixtures of two immiscible fluids consisting of droplets of one phase dispersed into the other , stabilized against coalescence by surfactant . at low droplets volume fraction , the emulsions have basically a newtonian behavior @xcite . when the volume fraction of the dispersed phase is increased , droplets come into contact . if a small stress is applied to a dense emulsion , the interfaces between the droplets can be strained to store surface energy ; it results in an elastic response @xcite . above a yield stress , they flow as a result of droplets rearrangements @xcite . the flow behavior of dense emulsions can be measured in classical rheological experiments , and seems to be well represented by a herschel - bulkley model @xmath1 ; in monodisperse emulsions , the exponent @xmath2 has been found to vary between @xmath3 and @xmath4 for volume fractions varying between 0.58 and 0.65 @xcite . at first glance , dense emulsions behavior thus seems to be well understood and modelled . however , it has been shown recently that the behavior of dense emulsions @xcite , and more generally of pasty materials @xcite , may be much more complex than what can be inferred from simple rheometric experiments , particularly at the approach of the yield stress . the reason is that the flow behavior of materials is usually studied in viscosimetric flows @xcite , in which one measures macroscopic quantities ( a torque @xmath5 vs. a rotational velocity @xmath6 ) . a constitutive law relating the shear stress @xmath7 to the shear rate @xmath8 in any flow can then be derived easily from these macroscopic measurements provided that the flow is homogeneous . however , in complex fluids , this last requirement may not be fulfilled as the flow may be localized , as observed in many pasty materials @xcite , even in the cone - and - plate geometry @xcite . another problem of importance , that can complicate the analysis of the rheological data , is the existence of slippage of dense emulsions at the walls @xcite even with slightly roughened walls @xcite . these problems can be got round by coupling macroscopic torque measurements and local measurements of the shear rate , e.g. through mri techniques @xcite , optical methods @xcite or sound methods @xcite , in order to account for these heterogeneities and to get a proper constitutive law . local measurements in dense emulsions have recently yielded surprising results @xcite : in some emulsions , it is impossible to find a single constitutive law compatible with all flows . the constitutive law of the material then seems to depend on the velocity at the inner cylinder in a couette cell @xcite or the pressure gradient in a poiseuille cell @xcite . three main reasons have been evoked to explain the apparent absence of a constitutive law : thixotropy , non local effects , and shear - induced migration of the droplets . thixotropy may be a cause from the apparent absence of a constitutive law . long times may indeed be needed to reach a steady state . when the measurements are not performed in this steady state , it may then leave the idea that different laws are required to describe the flow of the sample . such long times to reach steady state have been evidenced in some experiments dealing with highly adhesive emulsion @xcite . importantly , @xcite , although they claim to have reached a steady state , were unable to account for the flows of their adhesive emulsion with a single constitutive law . note that we will study again the emulsion of @xcite in this paper . the apparent absence of a single local constitutive law may also be the signature of non - local phenomena . @xcite have actually proposed a non local model to describe the flows of jammed materials ( and thus dense emulsions ) , based upon the cooperativity of the flow . in these systems , flow occurs via a succession of elastic deformation and irreversible plastic events . these localized plastic events induce a relaxation of the stress on the whole system . their zone of influence is quantified by a length @xmath9 . in presence of high stress gradient or in the vicinity of a surface , this process creates non local effects : indeed , in these situations , the rate of rearrangements of the neighboring fluid differs from what would be the bulk rearrangement rate . as a consequence , these effects affect drastically the rheological behavior when @xmath9 is comparable to the size of the confinement . such behavior has been recently pointed out by @xcite . from experimental data , the authors show that the cooperativity length @xmath9 is zero below the jamming concentration @xmath10 and is typically a few oil droplet diameters above @xmath10 . using a wide gap should then prevent us from being sensitive to these non local effects . we will check this feature by studying the flows of the @xcite emulsion in this paper . particle migration may also be a cause for the apparent lack of a local constitutive law @xcite . in multiphase materials , wide gap couette flows are indeed known to lead to concentration heterogeneities due to shear - induced migration of the dispersed elements toward the low shear zones . the migration of deformable particles , and particularly drops , has been much less studied than the one of rigid particles . single droplets have been found to migrate away from the rigid walls in any shear flow , due to asymmetric flows around deformed droplets @xcite . this leads to an equilibrium position somewhere between the walls , exactly at the center of the gap in the case of a narrow gap couette geometry , nearer from the inner cylinder in the case of a wide gap couette geometry @xcite . when dealing with a dilute emulsion , it has been observed , in contrast with dilute suspensions of rigid particles , that the equilibrium distribution of droplets is parabolic around a position that is near the equilibrium position of single droplets @xcite . the theoretical explanation is that all particles would tend to accumulate at the same equilibrium position between the walls , but their distribution is broadened by their binary collisions ( this second process may be modelled as it is for rigid particles ) . only rather dilute emulsions ( up to 10% of droplets ) have been studied so far , and nothing is known about what happens for higher droplet concentrations . in this case , one would expect the wall effect on the droplet migration to be negligible compared to the effect of the interactions between droplets ; then , one would expect migration to product the same effects as in suspensions , with the same kinetics as it has basically the same physical origin . we can thus expect that relevant information for migration in dense emulsions can be found in the literature dealing with suspensions of rigid particles . in suspensions of noncolloidal rigid particles , migration has been observed in many situations : wide - gap couette flows @xcite , parallel - plate flows @xcite , pipe flows @xcite . in wide - gap couette flows , the consequence of migration is an excess of particles near the outer cylinder . migration is related to the shear - induced diffusion of particles @xcite : the gradients in shear rate that exist in all but the cone and plate geometry generate a particle flux towards the low shear zones ( the outer cylinder in the case of the couette geometry ) , which is counterbalanced by a particle flux due to viscosity gradients . in an alternative model @xcite , particle fluxes counterbalance the gradients in normal stresses . most experiments @xcite observe that migration in suspensions of volume fraction up to 55% is rather slow , in accordance with its diffusive origin . however , it was recently found in a couette geometry that in the case of very dense suspensions ( up to 60% ) , which exhibit an apparent yield stress , migration is almost instantaneous @xcite ( it lasts for a few revolutions ) so that it may be unavoidable . a reason for this very fast kinetics may be that particles are closely packed together : any shear may then push the particles towards the outer cylinder with an instantaneous long range effect on the particle concentration . migration may then be expected to be very fast for dense emulsions exhibiting a yield stress , i.e. in which droplets are in contact as the particles in the very dense suspensions of @xcite . in this paper , we address the questions of the existence and the determination of a local constitutive law for the flows of dense emulsions . in this aim , we use a wide gap couette cell to avoid non local effects and we study in detail the question of shear - induced droplet migration in these systems . we also take care of performing our experiments in a steady state . we propose to couple macroscopic measurements and local measurements of concentration and velocity through mri techniques during the wide gap couette flows of dense emulsions . a new method designed to measure the local droplet concentration in emulsions is developed and we seek for the occurrence of migration on many formulations . local measurements of the constitutive law are then compared to purely macroscopic measurements . in sec . [ section_display ] , we present the materials and the experimental setup . we present the concentration profiles obtained on all materials after long time experiments in sec . [ section_concentration ] . the velocity profiles are shown in sec . [ section_velocity ] . the macroscopic rheometrical measurements are displayed in sec . [ section_torque ] , and a purely macroscopic determination of the constitutive law is presented . the results are analyzed in sec . [ section_local ] : we determine locally the constitutive law of the material from the velocity profiles and we compare macroscopic and local measurements of the constitutive law . as the timescale for the occurrence of migration is _ a priori _ very sensitive to the droplet size @xmath11 ( if a diffusive process is involved , it should scale as @xmath12 ) , we study dense emulsions of 4 different sizes : 0.3@xmath0 m , 1@xmath0 m , 6.5@xmath0 m , and 40@xmath0 m . moreover , as problems in defining a local constitutive law seem to be linked to the adhesion properties of the droplets @xcite , we took care of formulating adhesive and non adhesive emulsions . note in particular that we study the emulsions of @xcite and @xcite which posed problems in defining a single local constitutive law accounting for all their flows . the 0.3@xmath0 m emulsion is the adhesive emulsion of @xcite ; nearly monodisperse oil in water emulsions are prepared by shearing a crude polydisperse emulsion , composed of castor oil droplets in water stabilized by sodium dodecyl sulfate ( sds ) , within a narrow gap of 100 @xmath0 m @xcite . the surfactant concentration within the aqueous phase is set to 8% wt . which leads to an adhesive emulsion @xcite . the oil volume fraction is 73% and the droplet mean diameter is 0.3 @xmath0 m with a polydispersity of about 20% which is enough to prevent crystallization . the 1@xmath0 m emulsion is a non adhesive emulsion . the preparation of the sample is the same as the one used to get the adhesive 0.3@xmath0 m emulsion . it is composed of silicone oil in water stabilized by sodium dodecyl sulfate ( sds ) . the surfactant concentration within the aqueous phase is set to 8.5% . the viscosity of the oil is 1 pa.s and the oil volume fraction is 75% . we prepared adhesive and non - adhesive oil in water emulsions , of 6.5 @xmath0 m mean diameter with a polydispersity of about 20 @xmath13 . these emulsions are composed of silicone droplets in a mixture of glycerine and water ( 50% w glycerine , 50% w water ) stabilized by 1% wt of brij and trimethyl tetradecyl ammonium bromide , sheared in a narrow gap couette cell . the trimethyl tetradecyl concentration within the aqueous phase is set to 1 or 6.5 wt @xmath13 , which leads to respectively adhesive and non - adhesive emulsions . the oil viscosity is 1pa.s . the oil volume fraction is 75% in all cases . note that the non - adhesive 6.5@xmath0 m emulsion is the same as in @xcite . note that the use of glycerol in the continuous phase prevents from obtaining accurate concentration measurements with the method developed below . only qualitative results can be inferred from the measurements : in particular , if the intensity profiles are unchanged after a flow , it means that there is no migration . we finally studied a large droplet brine in oil non - adhesive emulsion , of 40 @xmath0 m mean diameter with a polydispersity of about 50 @xmath13 , at a 88% concentration . the emulsion is prepared by progressively adding the brine ( solution of cacl@xmath14 at 300g / l in water ) in an oil - surfactant solution ( hdf 2000 ( total solvent ) + sorbitan monooleate at 2% ) under a controlled high shear with a silverson l4rt mixer . we carried out experiments with a velocity controlled magnetic resonance imaging ( mri ) rheometer " . experiments are performed within a wide - gap couette geometry ( inner cylinder radius @xmath15 cm , outer cylinder radius @xmath16 cm , height @xmath1711 cm ) . in order to avoid slip at the walls , sandpaper is glued on the walls . rheometric measurements are performed with a commercial rheometer ( bohlin c - vor 200 ) that imposes either the torque or the rotational velocity ( with a torque feedback ) . the whole geometry is inserted in a 0.5 t vertical mri spectrometer ( 24/80 dbx by bruker ) . the mri measurements allow us to get the local orthoradial velocity and the local droplet concentration of the material anywhere in the gap , with a radial resolution of 0.5 mm . for details on the mri sequence devoted to measure the local velocity and its application to rheology , see @xcite . a method devoted to obtain the local droplet concentration was developed specifically for this study and is presented hereafter . proton nmr relies on the properties of hydrogen nuclei , which bear a magnetic dipole . in a nmr experiment , the sample is put inside a strong , static , and ( as far as possible ) homogeneous field @xmath18 . the magnetization of hydrogen nuclei then tend to align along the static field . the sample develops a non zero magnetization density @xmath19 which follows the curie law : m_o(r)=_h(r)b_o[eq_magnetization]where @xmath20 is the hydrogen density at position @xmath21 , @xmath22 is the gyromagnetic ratio of hydrogen nuclei , @xmath23 is the planck constant , @xmath24 is the boltzmann constant , and @xmath25 is the sample temperature . @xmath20 can be easily computed for any sample from its chemical formula and its density . when put out of equilibrium , magnetization of liquids often develops a threefold dynamic behavior combining a precession motion around the permanent field of the magnet , and two relaxation processes : a mono - exponential decay of transverse components of @xmath26 with a characteristic time @xmath27 , and a mono - exponential relaxation of the longitudinal component of @xmath26 towards its equilibrium value , with a characteristic time @xmath28 . nuclear magnetic resonance technique consists in series of magnetization manipulation and measurements by means of externally applied additional magnetic field , aiming at getting information about physico - chemical properties of the sample . the magnetic resonance imaging extension of nmr permits to get , as far a possible , a space resolved view of the local magnetization density , and can make such analysis local and non - perturbative . because of signal to noise ratio limitations , the space resolution of mri facilities for imaging purposes can never be much better than about one hundredth of the size of the experimental setup , that is 1 mm in our working conditions , which is much too big to resolve independent droplets . measuring the water / oil ratio in a sample then requires to have a way to distinguish oil and water magnetization inside a single pixel of a mri image . up to now , three main discrimination routes were already reported in emulsion literature . the first one relies on the so - called chemical shift difference between water and oil . actually , the exact precession frequency around the permanent field is closely related to the direct chemical neighborhood of hydrogen atoms . because all hydrogen nuclei of water are linked to oxygen atoms , and most of hydrogens in oily material are linked either to carbon or silicium atoms , water and oil magnetization always exhibit two different precession frequency . this chemical shift is quite small ( about 4 - 4.5 ppm relative difference , whenever carbon or silicon oils are used ) . nevertheless this shift can be accurately observed provided that @xmath29 inhomogeneities across the sample have a much smaller amplitude than 4.5ppm of the mean @xmath30 value ( e.g. , see spectrum by @xcite).indeed , the precession frequency is also proportional to local @xmath31 intensity , and one should avoid confusion between the original chemical shift and any spurious one originating from field heterogeneities . homogeneity constraints can be met when the fluid is embedded in a container with a simple - shape . reported studies making use of the chemical shift contrast then often deal with emulsions that are contained in a simple beaker , or even a cylindrical pipe displayed parallel to @xmath29 . quite early results deal with the use of chemical shift imaging ( csi ) for a quantitative assessment of oil / fat content in food emulsions @xcite , or the separation of oil and water components of an emulsion during a filtration process @xcite . more recently , @xcite and @xcite managed to carry out quite well resolved local measurements of water / oil concentrations and/or velocities in model suspensions flowing through a pipe in a laminar regime . unfortunately , the chemical shift contrast can hardly be transposed in the case of a couette cell . the geometry is indeed more complex and involves edges between materials with different magnetic susceptibilities , prone to generate strong local field distortions . csi applications to couette flow are thus more difficult and actually quite rare . for instance , @xcite , who studied the mixing process of oil and water in an horizontal couette like device reported quite long measuring times and only poorly resolved pictures . a second way to discriminate between water and oil in an emulsion relies on @xmath28 relaxation time . @xmath28 ( as well as @xmath27 ) indeed strongly depend on a wide set of parameters , such as fluid composition , diffusion coefficient , temperature , or the presence of dissolved species . in water and oil , they usually turn out to be quite different . for instance , in our working conditions ( 0.5 t magnetic field ) @xmath28 and @xmath27 usually range between 2 and 3s in bulk water ( depending on water origin and dissolved minerals ) , while many carbonated and silicon oils exhibit @xmath28 about few 100ms , and @xmath27 even smaller . although the use of @xmath27 contrast was already reported as a promising nmr tool of characterizing droplet size in an emulsion @xcite , previous works involving imaging possibilities of nmr prefer @xmath28 contrast . as compared with chemical shift or @xmath27 , @xmath28 physical value and @xmath28 measurement techniques are indeed fairly insensitive to field inhomogeneities . @xcite demonstrated this contrast to give quantitative information on the creaming process of an oil - water emulsion . more recently , few works showed its ability to work in more severe conditions . @xcite used @xmath28 to measure the fat fraction in an emulsion contained inside the stomach of a human volunteer . @xcite also reported the use of a @xmath28 based method to follow the droplet migration in a low concentrated emulsion under shearing in a couette cell . a third way also used to study the creaming of an emulsion @xcite , consists in preparing a water - oil emulsion based on deuterium oxide instead of normal water . because proton nmr is only sensitive to 1h nuclei , nmr or mri measurements then bring exclusive information on the oil phase . in our work , we wanted to get independent information on oil and water phase . field inhomogeneities inside our couette cell were also in many places of the same order of magnitude as the chemical shift . this was mainly due to the covering of the surfaces of the cell with sand - paper , that was used so as to avoid slip problems . @xmath28 contrast was then regarded as the best contrast source . although details of our measurement method was quite different than that of @xcite it was also based on the so - called inversion - recovery technique . starting from a magnetization in equilibrium state , it consists in putting magnetization upside down by means of an appropriate external field , and then to let it relax towards its equilibrium value . in a homogeneous sample with relaxation time @xmath28 and equilibrium magnetization density @xmath32 , the magnetization about the main magnetic field should then read m(t)=m_o(1 - 2e^-t / t_1)where @xmath33 at the time of inversion . to perform an accurate analysis , we found necessary to consider additionally that the initial inversion can never be perfect : the magnetization about the main magnetic field then actually reads m(t)=m_o(1-ie^-t / t_1)where @xmath34 reflects the actual inversion conditions : @xmath34 would be equal to 2 in perfect experimental conditions , but was actually always found to be slightly less than 2 . its exact value and space homogeneity depends on both the hardware and the sample . in a water - oil emulsion , it is admitted that water and oil behave as two independent phases : the resultant magnetization is then the sum of the individual oil and water magnetization that relax independently . the mean magnetization inside a water oil - emulsion then reads m(t)=^w m_o^w(1-i e^-t / t_1^w)+^o m_o^o(1-i e^-t / t_1^o)[eq_relaxation ] where @xmath35 and @xmath36 are the equilibrium magnetization density of bulk water and oil respectively , @xmath37 and @xmath38 are the water and oil volume fractions ( @xmath39 ) , and @xmath40 and @xmath41 are the longitudinal relaxation time of water and oil respectively . @xmath34 is supposed to be the same in oil and water . from eq . [ eq_relaxation ] , in order to observe only water droplets in a water in oil emulsion , @xcite used the fact that there is always a time during this relaxation process when the oil magnetization is zero . they made a mri picture of @xmath42 at this time , based on the assumption that @xmath43 everywhere in the sample . however , we think that the accuracy of such technique may be limited , because a wrong estimation of the actual @xmath34 factor , if not exactly 2 , or a wrong estimation of relaxation times will always lead to a wrong estimation of the time for which the oil magnetization is zero , and then , to an incomplete oil cancelling . using in situ measurements of water and oil relaxation times , we developed another approach which consists in performing 3 mri measurements of the magnetization relaxation , for 3 carefully chosen waiting times , and then in solving an equation system , letting @xmath34 as a free parameter . we then obtain space resolved measurements of @xmath44 , @xmath45 and @xmath46 as averaged at the pixel scale . the three waiting times are optimized for each sample in order to minimize the uncertainty on the magnetization values . as compared with the phase cancelling technique , we calculated that this method is much more accurate , and especially less sensitive to @xmath28 uncertainties when @xmath28 values in water and oil are clearly separated . in order to get the material volume fraction from eq . [ eq_relaxation ] , in situ measurement of relaxation times were found to be a necessary step of the experimental procedure . indeed , as far as relaxation phenomena are considered , although water and oil mainly behave as if they would be independent phases , they interact anyway together with the surfactant at the surface of each droplet . relaxation times in water may then be influenced by the presence of the oil phase . relaxation times in oil may also differ from that of pure oil because aliphatic chains of surfactant molecules mix with oil in the vicinity of the droplet surface . the sensitivity to the droplet size was reported to be quite dramatic for @xmath27 value in water / oil emulsion containing a large amount of additional impurities @xcite . we believe in such system that @xmath28 and @xmath27 should qualitatively behave the same . in our samples , which are made of quite pure water and oil phases , the @xmath28 difference between the pure phases and the droplets was found to be of order 10% . this difference is probably too small to provide a quantitative sensitivity of our measurements to droplet size changes in our systems . anyway , @xmath28 changes , as small as they may be , have to be taken into account for the accuracy of local fraction measurements . eventually , @xmath28 may also evolve with any small change in the temperature of the sample . that is why before each concentration measurement , the actual relaxation times were measured . as @xmath28 is sensitive to the details of the local structure of the material , @xmath28 measurements should be ideally space resolved . however , we worked here another way : we used an average @xmath28 measurement in the whole gap of the cell , discriminating water and oil on the basis of their chemical shift ( field inhomogeneities were too strong for a csi - based imaging , but were still ok for @xmath28 measurements ) . the average @xmath28 values for oil and water were then directly used to process the mri data . granted that processed oil and water volume fractions were then always found to be homogeneous within the gap of the cell , we had then an a posteriori confirmation that our global @xmath28 measurement were meaningful . one should be aware anyway that if any heterogeneity would have been detected in measured volume fractions , then it would have been necessary to shift to local @xmath28 measurements to ensure very accurate quantitative results ; measurements based on the global @xmath28 measurement would nevertheless provide a relevant estimation of the volume fractions . finally , in order to obtain the local volume fraction @xmath47 and @xmath48 of each phases , it is sufficient to remember that from eq . [ eq_magnetization ] : @xmath49 and @xmath50 . the absolute uncertainty on the concentration measurement performed with this method was found to be of 0.3% from measurements performed on a homogeneous emulsion . note that when the proton density of the oil phase is not known , the comparison of the @xmath45 and @xmath46 measurements performed on the homogeneous emulsion ( of known volume fraction ) allows to recover its value . note also that a simple way to check for changes in the local concentration , and to evaluate the amplitude of these changes , consists in comparing the water and oil intensity profiles within the gap after shear to the profiles measured on the emulsion just after loading . these last profiles are performed on a homogeneous emulsion , any change in these profiles induced by shear would then mean that there is migration . in the experiments presented hereafter , we control the rotational velocity of the inner cylinder . for all materials , we first apply a constant rotational velocity of 20rpm to 100rpm until 10000 to 90000 revolutions have been completed . we record the evolution of the velocity profiles and the concentration profiles in time in order to check for the occurrence of migration ( see sec . [ section_concentration ] ) . afterwards , we measure the stationary velocity profiles and the stationary torque exerted by the material on the inner cylinder for various velocities ranging from 0.1rpm to 100rpm ( see sec . [ section_velocity ] and sec . [ section_torque ] ) . in order to search for the occurrence of shear - induced migration in dense emulsions , we sheared all materials for long times . we chose to apply velocities ranging between 20 and 100 rpm ; this ensures that in all cases the whole gap is sheared ( see sec . [ section_velocity ] ) . we also checked on the 6.5@xmath0 m emulsions that the same results are obtained when the flow is localized during the experiment ( i.e. when the material is sheared at 1 rpm ) . the 0.3@xmath0 m emulsion was sheared for 10000 revs , the 1@xmath0 m emulsion was sheared for 90000 revs , both 6.5@xmath0 m emulsions were sheared for 35000 revs , and the 40@xmath0 m emulsion was sheared for 20000 revs . the corresponding local strains , extracted from the velocity profiles ( see sec . [ section_velocity ] ) range between 5 times the number of revolutions near the inner cylinder , and 0.25 times near the outer cylinder . m adhesive emulsion for 10000 revs ( squares ) , and a 40@xmath0 m non - adhesive emulsion for 20000 revs ( open circles).,width=340 ] in all the experiments , within the experimental uncertainty , we observe that the materials remain homogeneous after shear : there is no observable shear - induced migration . results for the 0.3@xmath0 m and the 40@xmath0 m emulsion are depicted in fig . [ fig_emul_concentration ] . note also that the relaxation times @xmath28 of the oil and water phases were not observed to change during all the experiments : as pointed out in sec . [ section_display ] , this is consistent with the fact that the droplet size is unchanged by shear and that all the measurements deal with the same homogeneous suspension . consistently , optical observations of the droplets after shear did not reveal size changes . as noted in sec . [ section_introduction ] , up to now only rather dilute emulsions have been studied : we can not compare our results with any result from the literature dealing with emulsions . however , the relevant mechanism involved in migration of droplets in dense emulsions should be basically the same as in suspensions of rigid particles , namely a shear - induced diffusion of particles . from this model and results from the literature of migration in dense suspension , we can thus infer whether or not migration would have been expected in our systems . in the diffusive model of migration @xcite , the particles undergo a shear - induced diffusion characterized by a self - diffusion coefficient @xmath51 @xcite , where @xmath52 is the particle volume fraction , @xmath53 the shear rate , @xmath54 the particle size , and @xmath55 is a dimensionless coefficient whose dependence on @xmath52 may theoretically be @xmath56 for small @xmath52 . the gradients in shear rate that exist in a couette geometry then generate a particle flux towards the outer cylinder , which is counterbalanced by a particle flux due to viscosity gradients . as qualitatively confirmed experimentally @xcite , one would then expect the migration phenomenon to last for a number of revolutions @xmath57 until a stationary heterogeneous profile is established , where @xmath58 and @xmath59 are respectively the outer and inner radius , and @xmath60 . within the frame of this model , we can now evaluate the expected @xmath61 for complete migration in our experiment from experimental results from literature . actually , as shown by @xcite two inconsistent set of data exist , depending on the concentration : migration has been shown to occur much more rapidly for concentrations above 55% . e.g , for moderately dense suspensions , @xcite find @xmath62 revs for 140@xmath0 m particles , at @xmath63 , with @xmath64 cm and @xmath65 cm . in the case of our materials , this would imply a value of @xmath66 revs for the 0.3@xmath0 m emulsion , @xmath67 revs for the 1@xmath0 m emulsion , @xmath68 revs for the 6.5@xmath0 m emulsions , and @xmath69 revs for the 40@xmath0 m emulsion . on the other hand for very dense suspensions , @xcite found that migration occurs during the first 50 revs within the same couette geometry as our . this would imply an expected value of @xmath70 revs for the 0.3@xmath0 m emulsion , @xmath71 revs for the 1@xmath0 m emulsion , @xmath72 revs for the 6.5@xmath0 m emulsions , and @xmath73 revs for the 40@xmath0 m emulsion in our experiments . let us recall that @xmath61 is the number of revolutions expected for _ complete _ migration , and that heterogeneities should be observable for a value of @xmath74 revolutions ( see @xcite and @xcite for measurements of transient concentration profiles in suspensions ) . if a shear - induced migration mechanism similar to the one observed in suspensions of rigid particles was acting in dense emulsions , we should therefore have observed migration in the 40@xmath0 m emulsion and in the 6.5@xmath0 m emulsions . on the other hand , our experiments can not be used to conclude directly for the failure of a diffusive model of migration in the case of the smaller droplets ( 0.3 and 1@xmath0 m ) , but we can expect to predict what happens in this case from the @xmath75 scaling of the diffusion process if this process is relevant . even if it impossible to conclude definitely for the absence of migration for any strain , we can conclude that if a diffusive process acts in dense emulsions , it is much slower than the process involved in dense suspensions of rigid particles . importantly , our results provide reliable lower bounds for the occurrence of migration in dense emulsions . for practical purposes , in most flows and particularly in most rheometrical studies of dense emulsions , it can therefore be claimed that the emulsions remain homogeneous . surprisingly , our results contrast with what was observed by @xcite in dense suspensions of hard spheres where very fast migration was found to occur . a first reason could be that jamming prevents from migration in dense emulsions . however , it was found that migration in dense suspensions can lead to regions having a concentration higher than the maximum packing fraction @xmath10 above which there is no more flow @xcite : this shows that jamming does not necessarily prevents from migration . another reason could be that deformability of the particles play a central role . the shear stresses involved in the flow are of the order of 100 to 200 pa in all materials , whereas the typical stress due to surface tension range between 1000pa for the 40@xmath0 m droplets and 120000pa for the 0.3@xmath0 m particles . the droplets may then be poorly deformable ( except for the 40@xmath0 m emulsion ) under shear , but they must be deformed due to confinement and to their high concentration . in particular they should have flat contacts which may allow the droplets to slip easily past each other ; this is a major difference with the suspensions of hard spheres in which any shear may push the particles towards the outer cylinder with an instantaneous long range effect on the particle concentration thanks to the force chains originating from direct frictional contact forces . note that consistency of our observations with the modelling of shear - induced migration based on normal stresses is difficult to check as normal stresses in dense emulsions are poorly known . it should be noted however that the normal stresses due to surface tension should be of the order of the yield stress @xcite in dense emulsions , while normal stresses tend to diverge at the approach of the maximum packing fraction in dense suspensions of rigid particles ( this difference is linked to the deformability of the droplets leading to flat contacts as pointed out above ) . it is thus not unexpected from this point of view that the kinetics of shear induced migration in a couette flow , which depends on the spatial variation of the first normal stress difference and of the radial normal stress @xcite , is very fast near the maximum packing fraction in suspensions of hard spheres while it would have to remain slow in dense emulsions . anyway , it is not possible to be more quantitative at this stage and to state if migration would have been expected to occur within this picture in our experimental conditions . in these experiments , after a 5 min preshear at 100 rpm , we measured the velocity profiles at various rotational velocities . in all systems , the stationary velocity profiles were found to develop within a few seconds and to remain stable for hours . this means that these systems are not thixotropic @xcite . in the case of the 0.3 @xmath0 m adhesive emulsion , this contrasts with the long time evolution of the velocity profiles observed by @xcite ; we will comment on this point below . the stability of the velocity profiles is also consistent with the absence of shear - induced droplet migration in all systems : as the behavior of dense emulsions depends on their concentration , changes in the concentration profiles in time would have led to changes in the velocity profiles in time . this feature is important : in the absence of a method to measure the concentration profiles , the stability of the velocity profiles can be used as a proof of the absence of migration as long as the migration process is not nearly instantaneous . in fig . [ fig_emul_velocity1 ] , we plot the dimensionless velocity profiles for the steady flows of all the emulsions for various rotational velocities ranging from 0.3 to 100rpm . m adhesive emulsion , at various rotational velocities ranging from 5 to 100rpm ; the line is the theoretical dimensionless velocity profile for a newtonian fluid . b ) same plot for a 1@xmath0 m non - adhesive emulsion . c ) same plot for a 6.5@xmath0 m adhesive emulsion . d ) same plot for a 6.5@xmath0 m non - adhesive emulsion . e ) same plot for a 40@xmath0 m non - adhesive emulsion.,title="fig:",width=302 ] m adhesive emulsion , at various rotational velocities ranging from 5 to 100rpm ; the line is the theoretical dimensionless velocity profile for a newtonian fluid . b ) same plot for a 1@xmath0 m non - adhesive emulsion . c ) same plot for a 6.5@xmath0 m adhesive emulsion . d ) same plot for a 6.5@xmath0 m non - adhesive emulsion . e ) same plot for a 40@xmath0 m non - adhesive emulsion.,title="fig:",width=302 ] m adhesive emulsion , at various rotational velocities ranging from 5 to 100rpm ; the line is the theoretical dimensionless velocity profile for a newtonian fluid . b ) same plot for a 1@xmath0 m non - adhesive emulsion . c ) same plot for a 6.5@xmath0 m adhesive emulsion . d ) same plot for a 6.5@xmath0 m non - adhesive emulsion . e ) same plot for a 40@xmath0 m non - adhesive emulsion.,title="fig:",width=302 ] m adhesive emulsion , at various rotational velocities ranging from 5 to 100rpm ; the line is the theoretical dimensionless velocity profile for a newtonian fluid . b ) same plot for a 1@xmath0 m non - adhesive emulsion . c ) same plot for a 6.5@xmath0 m adhesive emulsion . d ) same plot for a 6.5@xmath0 m non - adhesive emulsion . e ) same plot for a 40@xmath0 m non - adhesive emulsion.,title="fig:",width=302 ] m adhesive emulsion , at various rotational velocities ranging from 5 to 100rpm ; the line is the theoretical dimensionless velocity profile for a newtonian fluid . b ) same plot for a 1@xmath0 m non - adhesive emulsion . c ) same plot for a 6.5@xmath0 m adhesive emulsion . d ) same plot for a 6.5@xmath0 m non - adhesive emulsion . e ) same plot for a 40@xmath0 m non - adhesive emulsion.,title="fig:",width=302 ] the 1@xmath0 m non adhesive emulsion , the 6.5@xmath0 m adhesive and non adhesive emulsions , and the 40 @xmath0 m non adhesive emulsion share a common behavior . mri measurements show that the velocity profiles are curved , and that they occupy only a small fraction of the gap at low rotational velocities ( fig . [ fig_emul_velocity1 ] ) : in this case , the velocity tends to zero within the measurement uncertainty at some radius before the outer cylinder . the fraction of the material that is sheared increases with the rotational velocity . beyond a critical velocity @xmath76 , that depends on the emulsion and is of order 3rpm in the case of the 6.5@xmath0 m adhesive emulsion ( fig . [ fig_emul_velocity1]c ) , or of the order 50 rpm in the case of the 1@xmath0 m non - adhesive emulsion ( fig . [ fig_emul_velocity1]b ) , the whole sample is sheared . this shear localization is a classical feature of yield stress fluids flows in couette geometry where the shear stress is a decreasing function of the radius : the flow must stop at a radius @xmath11 such that the shear stress equals the material yield stress at this place . note that this behavior is not evidenced for the non adhesive 6.5@xmath0 m emulsion and for the non adhesive 40@xmath0 m emulsion : this is simply due to the fact that the applied rotational velocities are too high to probe this behavior . when all the gap is sheared , the velocity profiles are clearly different from those of a newtonian fluid ( fig . [ fig_emul_velocity1 ] ) . the curvature we observe is due to the use of a wide - gap couette geometry and is actually typical of shear - thinning fluid : the shear rate decreases more rapidly within the gap ( i.e when the shear stress decreases ) than for a newtonian fluid . the shear rate may vary here of a factor 20 within the gap while it would vary of a factor 2.1 for a newtonian fluid in this geometry ( this factor 2.1 is simply the ratio between the shear stress at the inner cylinder and the shear stress at the outer cylinder , see eq . [ eq_localstress ] in sec . [ section_torque ] ) . this implies that in the analysis of macroscopic experiments , one can not simply compute a mean shear rate within the gap to characterize the flow . finally , note that no data can be recorded close to the inner cylinder . as a consequence , the expected absence of wall slip ( due to the use of rough surfaces of roughness larger than the droplet size ) can not be proved at this stage . it can only be observed that the extrapolation of the dimensionless velocity profiles to a value of 1 at the inner cylinder is reasonable , and that wall slip if any should be of a limited amount of order 5% or less . we will show in sec . [ section_local ] that it is possible to infer the emulsion velocity at the walls from its local constitutive law , and that wall slip if any is actually less than 1% of the inner cylinder velocity . in log scale ( for readability , not all profiles are plotted ) . the solid lines of fig . [ fig_emul_velocity1log]a are the velocity profiles predicted by a herschel - bulkley law.,title="fig:",width=302 ] in log scale ( for readability , not all profiles are plotted ) . the solid lines of fig . [ fig_emul_velocity1log]a are the velocity profiles predicted by a herschel - bulkley law.,title="fig:",width=302 ] in log scale ( for readability , not all profiles are plotted ) . the solid lines of fig . [ fig_emul_velocity1log]a are the velocity profiles predicted by a herschel - bulkley law.,title="fig:",width=302 ] in log scale ( for readability , not all profiles are plotted ) . the solid lines of fig . [ fig_emul_velocity1log]a are the velocity profiles predicted by a herschel - bulkley law.,title="fig:",width=302 ] in log scale ( for readability , not all profiles are plotted ) . the solid lines of fig . [ fig_emul_velocity1log]a are the velocity profiles predicted by a herschel - bulkley law.,title="fig:",width=302 ] at first sight , the 0.3@xmath0 m adhesive emulsion presents the same behavior as the other samples . the velocity profiles are curved , and they occupy only a small fraction of the gap at low rotational velocities . however , focusing on the low velocities part and thus very low shear rates part of the profiles , a striking behavior is evidenced in the semi - logarithmic scale figure ( fig . [ fig_emul_velocity1log]a ) . a slope break in the profile is evidenced at a given radius inside the gap . the velocity then starts to decrease very slowly with the radius . there is then a large zone of the material in which there remains a very slow flow . this behavior is reminiscent of the one observed previously by @xcite and by @xcite in adhesive emulsions ( this behavior will be shortly commented in sec . [ section_local ] ) . these slowly varying slow flows occur here for velocities lower than a few [email protected]@xmath77 . this behavior seems to be a specific feature of the strongly adhesive 0.3 @xmath0 m emulsion . we measured velocity as low as [email protected]@xmath77 at the approach of flow stoppage in the 1@xmath0 m non adhesive and 6.5@xmath0 m adhesive emulsion ( fig . [ fig_emul_velocity1log]b , c ) but did not observe any slope break in these cases : there is no evidence of any slow flow in the apparently jammed zones in these emulsions . in the following , we build the constitutive laws of the emulsions accounting for their flow properties . to get this information , we use two different methods . first , focusing on the macroscopic measurements , we present an experimental procedure allowing us to get a relationship between local values of the shear stress and the shear rate by measuring only macroscopic quantities ( sec . [ section_torque ] ) . second , we analyze in detail the velocity profiles and extract a local constitutive law by differentiating the velocity field to get the local shear rate ( sec . [ section_local ] ) . in this section , we present the macroscopic rheometric measurements ( torque vs. rotational velocity ) . we then show that it is possible to extract the constitutive law of the material from a proper analysis of this set of purely macroscopic data only provided that the material is homogeneous ( absence of migration ) and that there is no wall slip , which is what we have ensured . for each rotational velocity @xmath6 , we have measured the torque exerted on the inner cylinder vs. time until a stationary state is reached ( see inset of fig . [ fig_emul_rheomacro]a for the 0.3 @xmath0 m adhesive emulsion ) . we observe that for each rotational velocity , a stationary state is reached for a strain of a few unities , consistently with the absence of evolution of the velocity profiles in time . in fig . [ fig_emul_rheomacro]a we plot the stationary torque vs. the rotational velocity for the steady flows of the 0.3 @xmath0 m adhesive emulsion . m adhesive emulsion ; inset : torque vs. angular displacement for various rotational velocities ranging between 0.01 and 50 rpm ( bottom to top ) . b ) constitutive law inferred from the purely macroscopic measurements thanks to eq.[eq_localstress ] and eq . [ eq_localrate ] ; the solid line is a herschel - bulkley fit @xmath78 of the data with @xmath79pa , @xmath80pa.s and @xmath81.,title="fig:",width=302 ] m adhesive emulsion ; inset : torque vs. angular displacement for various rotational velocities ranging between 0.01 and 50 rpm ( bottom to top ) . b ) constitutive law inferred from the purely macroscopic measurements thanks to eq.[eq_localstress ] and eq . [ eq_localrate ] ; the solid line is a herschel - bulkley fit @xmath78 of the data with @xmath79pa , @xmath80pa.s and @xmath81.,title="fig:",width=302 ] we observe that the apparent flow curve of the material is that of a yield stress shear - thinning fluid . as we have shown in sec . [ section_concentration ] that the material is homogeneous , we are allowed to use these macroscopic data to infer the constitutive law of the material . because of the strong stress and shear rate heterogeneities , in particular shear localization , standard formula relating simply the torque and rotational velocity values to a mean shear stress and a mean shear rate can not be used . however , it is possible to analyze the macroscopic data to account for these heterogeneities and to obtain the unique constitutive law consistent with this set of purely macroscopic data . the method used to analyze the wide gap couette rheometric data in order to obtain the constitutive law of the material is detailed e.g. in @xcite , we give here the main steps . first , it has to be noted that the shear stress distribution @xmath82 within the gap at a radius @xmath11 is known whatever the constitutive law of the material ; it reads ( r)=t/(2h r^2)[eq_localstress]where @xmath25 is the torque and h is the height of the couette geometry . then , the local shear rate @xmath83 is related to the rotational velocity @xmath6 through = ^r_o_r_i ( r)/r dr[eq_omega0]where @xmath59 is the inner radius and @xmath58 is the outer radius . from eqs . [ eq_localstress ] and [ eq_omega0 ] , we see that from the knowledge of the torque @xmath25 and the rotational velocity @xmath6 , we obtain a non straightforward relationship between the local shear stress and the local shear rate . in order to go one step further , eqs . [ eq_localstress ] and [ eq_omega0 ] may be combined into = -^(r_o)_(r_i ) ( ) / d[eq_omega]note that the possibility of shear localization is naturally taken into account in this equation , where @xmath84 when @xmath85 with @xmath86 the material yield stress . the constitutive law may then be derived from the whole macroscopic rheometric curve @xmath5 thanks to the differentiation of eq . [ eq_omega ] relative to @xmath25 : 2t / t=((r_i))-((r_o))[eq_localrate]finally , in order to get the shear rate @xmath87 at the inner cylinder as a function of the shear stress at the inner cylinder , one needs to eliminate the shear rate at the outer cylinder @xmath88 from this equation . this can be done by summing this relationship with a series of successive decreasing torques ( that differ by a factor @xmath89 at each step ) chosen to ensure that the shear rate at the outer cylinder is eliminated in the summation at each step , until a zero shear rate is reached at the outer cylinder : the summation stops when the shear stress computed at the outer cylinder at a given step falls below the yield stress @xcite . while this summation is in principle infinite for simple fluids , only a few steps are needed in the case of yield stress fluids , and the number of steps is more reduced for wider gaps . in the case of our couette geometry , as the ratio between the shear stress at the inner cylinder and the shear stress at the outer cylinder ( @xmath89 ) is equal to 2.1 , we see that the data of fig . [ fig_emul_rheomacro]a can be easily processed thanks to a maximum number of two summations of eq . [ eq_localrate ] . from eq . [ eq_omega ] , note that such analysis is based on the assumption that there is a local constitutive law characterizing the material and that there is negligible wall slip . the constitutive law obtained with this method is depicted in fig . [ fig_emul_rheomacro]b . we observe that the behavior of the material obtained with this procedure is basically well fitted to a herschel - bulkley law . in sec . [ section_local ] we compare this law to the constitutive law built from the local shear rates extracted from the velocity profiles . the constitutive laws of the materials accounting for their velocity profiles can be built from our experimental data , using both the velocity profiles and the torque measurements . the stress distribution @xmath82 within the gap is obtained from the macroscopic torque measurements thanks to eq . [ eq_localstress ] . the local shear rate @xmath83 in the gap is inferred from the velocity profiles @xmath90 through ( r)=v / r - v / rboth measurements performed at a given radius @xmath11 for a given rotational velocity @xmath6 thus provide a local data point of the constitutive law @xmath91 . this analysis provides a fair local measurement of the constitutive law since we measure the true local shear rate within the bulk of the material and the shear stress distribution is known from the momentum balance independently of any hypothesis ; in particular , it is independent of what happens at the interface so that a possible wall slip does no affect this analysis . we are finally allowed to combine the data measured at various radiuses because the materials are homogeneous as shown in sec . [ section_concentration ] . m emulsion , a non - adhesive 6.5@xmath0 m emulsion and an adhesive 6.5@xmath0 m emulsion ; the solid line is a herschel - bulkley fit @xmath78 of the data with @xmath81 for all materials , @xmath92pa and @xmath93pa.s for the non - adhesive 1@xmath0 m emulsion , @xmath94pa and @xmath93pa.s for the non - adhesive 6.5@xmath0 m emulsion , and @xmath95pa and @xmath96pa.s for the adhesive 6.5@xmath0 m emulsion . b)constitutive law measured locally in the gap of a couette cell for the flows of a 0.3@xmath0 m adhesive emulsion ( empty symbols ) ; the solid line is a herschel - bulkley fit @xmath78 of the data with @xmath79pa , @xmath80pa.s and @xmath81 . the crosses are the data extracted from the analysis of the purely macroscopic measurements ( see fig . [ fig_emul_rheomacro]b ) , title="fig:",width=302 ] m emulsion , a non - adhesive 6.5@xmath0 m emulsion and an adhesive 6.5@xmath0 m emulsion ; the solid line is a herschel - bulkley fit @xmath78 of the data with @xmath81 for all materials , @xmath92pa and @xmath93pa.s for the non - adhesive 1@xmath0 m emulsion , @xmath94pa and @xmath93pa.s for the non - adhesive 6.5@xmath0 m emulsion , and @xmath95pa and @xmath96pa.s for the adhesive 6.5@xmath0 m emulsion . b)constitutive law measured locally in the gap of a couette cell for the flows of a 0.3@xmath0 m adhesive emulsion ( empty symbols ) ; the solid line is a herschel - bulkley fit @xmath78 of the data with @xmath79pa , @xmath80pa.s and @xmath81 . the crosses are the data extracted from the analysis of the purely macroscopic measurements ( see fig . [ fig_emul_rheomacro]b ) , title="fig:",width=302 ] the local constitutive law extracted from the experiments performed at various rotational velocities @xmath6 on all the emulsions are plotted in fig . [ fig_emul_local1 ] ( we could not analyze the data for the 40@xmath0 m emulsion as torque data could not be measured in this case ) . it should first be noted that for each emulsion , all the shear stress vs. shear rate data fall along a single curve . this means that for a given material , our different data obtained under different boundary conditions effectively reflect the behavior of a single material with a given intrinsic constitutive law in simple shear . note also that this is consistent with the absence of migration : any heterogeneity would have apparently yielded different constitutive laws for different rotational velocities @xcite . our observation of a single local constitutive law accounting for all flows of the non - adhesive 6.5@xmath0 m emulsion ( fig . [ fig_emul_local1]a ) contrasts with the observations of @xcite on the same system in microchannel flows of up to 250@xmath0 m width : the velocity profiles observed in the channel can not be accounted for by a single constitutive law . this apparent paradox has been solved by @xcite : they showed that a rather simple non - local flow rule accounts for all the velocity profiles . they concluded that non - local effects should be observable in a zone involving up to 100 particles ; beyond this zone , this non - local law reduces to a local law . this explains why our measurements are unsensitive to such effects and why we measure a single local constitutive law : using a wide gap geometry prevents from being sensitive to these non - local effects . our observation of a single local constitutive law accounting for all flows of the adhesive 0.3@xmath0 m emulsion ( fig . [ fig_emul_local1]b ) also contrasts with the observations of @xcite . on exactly the same 0.3@xmath0 m adhesive emulsion sheared in a thin gap couette geometry , @xcite were actually unable to fit all their velocity profiles with a single constitutive law . in the @xcite study , the gap being 1 mm for 0.3@xmath0 m droplets , the non - local effects evidenced by @xcite can not be the reason for the apparent absence of a single constitutive law accounting for all flows . an important point should be noted that may explain the differences between both studies of the same material : the experimental procedures are slightly different , and the time evolutions of the emulsion properties are very different . in our experiments , the emulsion was first presheared with a mixer and then at 100 rpm for 5 minutes in a couette cell with rough surfaces . the stationary velocity profiles then measured at various rotational velocities were found to develop within a few seconds and to remain stable for hours . in the experimental procedure of @xcite , each experiment is conducted on a fresh sample which is directly sheared at the studied shear rate during at least 3 hours . a key point may then be that @xcite found their system to evolve for at least two hours , in contrast with our observations : whatever the rotational velocity , they found the apparent viscosity of the system to decrease slowly in time . this would mean that their material was in an initially structured state , and was destructured by shear , whatever the magnitude of the shear rate . then , while the authors claim that the velocity profiles no longer change significantly after 2 h , one may think that some slow relaxation still occurs and that the steady state is not really reached . this would mean that the measurements performed at different rotational velocities have been performed on different structural states of the emulsion ( in contrast with our measurements ) , naturally leading to different constitutive laws for the different velocities studied . this explanation based on the influence of the initial state of the material would be consistent with the observations of @xcite who showed that simple emulsions whose flow properties show no apparent thixotropic effects when first presheared at high shear rate are actually subject to significant aging at rest : this imply in particular that such systems have to be strongly presheared before any study . this preshear was performed in our study , it was not in the @xcite study . two other points may be noted : first , although the composition of the two emulsions is the same , the samples are different and chemical impurities may play an important role in the adhesion process ; second , the couette cell used by @xcite is smooth and huge wall slip is evidenced . focusing on the non adhesive 1@xmath0 m , the adhesive 6.5 @xmath0 m , and the non adhesive 6.5 @xmath0 m emulsions , we observe that the behavior of all materials is well fitted to a herschel - bulkley behavior @xmath78 of index @xmath81 in all the range of measured shear rates ; this is consistent with previous observations of the behavior of dense emulsions @xcite . coming back to the 0.3@xmath0 m adhesive emulsion , at moderate and high shear rates ( above 0.1s@xmath77 ) , the behavior is the same as the one encountered for the large droplets emulsions : the constitutive law is well fitted to an herschel - bulkley law on index n=0.5 . at low shear rate ( in a 0.01 - 0.1s@xmath77 range ) , the behavior of the 0.3@xmath0 m adhesive emulsion differs from the previous ones . all the points still fall on the same curve but a slight slope break is noticed in the flow curve . while a stress plateau should be reached for a simple yield stress fluid , the shear stress continues decreasing when the shear rate decreases . this is also observed on the law inferred from the macroscopic measurement . as far as we can say , this behavior is specific to the 0.3@xmath0 m adhesive emulsion : we took care of measuring the local behavior at low shear rates in the 6.5@xmath0 m adhesive emulsion but observed no slope break for shear rates as low as 10@xmath97s@xmath77 . a probable reason for the differences observed between these systems is that the 0.3 micron emulsion is more adhesive than the others , in particular because of their larger surface / volume ratio . this low shear rate behavior corresponds to the slow flows observed below a first apparent yield stress in fig . [ fig_emul_velocity1log]a . at low rotational velocity , when there is apparently shear localization , the velocity profiles are well fitted to the herschel - bulkley law predictions only for velocities larger than a few [email protected]@xmath77 . for lower velocities , at the approach of the radius where the yield stress should be reached and the velocity should tend to zero , the velocity starts to decrease very slowly with the radius ; on the other hand , at high rotational velocity , when all the gap is sheared , the velocity profile is very well fitted to the herschel - bulkley law prediction . the analysis of this behavior is difficult . one may think that the emulsion flows homogeneously and really follows a single continuous local constitutive law that corresponds to a herschel - bulkley equation at high shear rate and differs from this equation at low shear rate . one may also suggest that the law at low shear rates does not correspond to steady flows . indeed , it has to be noted that the measurements of the very low velocities in fig . [ fig_emul_velocity1log]a are averages over 1min : we can not know if they correspond to steady - flows ( the measurements may e.g. reflect a stick - slip behavior ) . these flows may thus reflect the existence of a jammed phase perturbated by some unsteady local rearrangements . in this case , this behavior would be reminiscent of shear banding , which has already been observed for strongly adhesive emulsions @xcite and is explained by a competition between structuration due to the adhesion process and destructuration by shear @xcite . the local measurements obtained for the 0.3@xmath0 m adhesive emulsion are compared in fig.[fig_emul_local1]a with the law inferred from the purely macroscopic measurements ( see sec.[section_torque ] ) . we observe that both laws are very well matched . in particular , the macroscopic measurements also allow us to evidence the slope break in the constitutive law at low shear rate . this validate the use of a wide gap couette geometry as a tool to obtain the constitutive law of dense emulsions from purely macroscopic rheometrical measurements , provided that a proper analysis of the macroscopic data is performed . we recall that this agreement is due to four important features ( i ) we have shown that all dense emulsions seem to be free from migration and thus remain homogeneous during the experiments , ( ii ) the use of a wide gap prevents from the nonlocal effects observed by @xcite , ( iii ) the material is initially destructured and a steady state is studied ( this may have not be the case in the @xcite study ) , ( iv ) wall slip has to be negligible to allow for a proper analysis of the macroscopic data : this is ensured here by the use of rough walls of roughness larger than the droplet size , as shown below . as pointed out in sec . [ section_velocity ] , our velocity measurement method does not provide the velocity at the walls : the first reliable data is obtained at around 0.5 to 1 mm from the inner cylinder . however , the local measurements of the constitutive law we have performed make it possible to study in more detail a possible wall slip effect . for a given rotational velocity of the inner cylinder , the velocity of the emulsion expected at the walls can actually be computed thanks to the knowledge of the shear rate in the emulsion corresponding to the shear stress value at the wall . the same shear stress level was actually reached somewhere in the bulk of the material for a higher rotational velocity of the inner cylinder , and the shear rate corresponding to this shear stress was then measured locally thanks to the velocity profile measurement . the velocity profile expected in the zone near the inner cylinder can then be reconstructed from the velocity profile measured at some distance from the walls and from the expected shear rate near the wall . of course , this analysis is based on the hypothesis that the material is the same as in the bulk in this 0.5 to 1 mm zone near the walls , which may be no more true in a zone where there are nonlocal effects @xcite . note also that this method can be used only when the reproducibility of the experiments is very good ( allowing data measured at a given rotational velocity to be used to predict what happens at another velocity ) . an example of this method applied to the flows of the 6.5@xmath0 m adhesive emulsion is depicted in fig . [ fig_emul_velocity2_slippage ] . m adhesive emulsion , for a 10 and a 20 rpm rotational velocity . b ) dimensionless velocity profiles expected near the walls at 10 and 20 rpm ; the velocities are computed thanks to the local shear rate measurements performed at 50 rpm.,title="fig:",width=302 ] m adhesive emulsion , for a 10 and a 20 rpm rotational velocity . b ) dimensionless velocity profiles expected near the walls at 10 and 20 rpm ; the velocities are computed thanks to the local shear rate measurements performed at 50 rpm.,title="fig:",width=302 ] we observe in fig . [ fig_emul_velocity2_slippage]b that the dimensionless velocity profiles expected near the walls have a value that is very close to 1 . this analysis was performed on several velocity profiles , and in all cases the dimensionless velocity values predicted at the inner cylinder were found to be equal to 1 within 1% . we finally conclude that , thanks to the use of rough surfaces of roughness larger than the droplet size , there is probably no wall slip in our experiments ; if any , it is of order 1% or less . this feature shows that the macroscopic measurements performed with a wide gap couette rheometer with rough walls can be trusted : this explains the very good agreement observed in fig . [ fig_emul_local1]a between the constitutive law measured locally and the constitutive law inferred from the purely macroscopic measurements . we have studied the flows of several dense emulsions in a wide gap couette geometry . we have coupled macroscopic rheometric experiments and local velocity and concentration measurements through mri techniques . the method devoted to measure the local droplet concentration was developed specifically for this study . in contrast with dense suspensions of rigid particles where very fast migration occurs under shear , we showed for the first time that no migration takes place in dense emulsions even for strain as large as 100000 in our systems . this may imply that another mechanism is involved in dense emulsions than in dense suspensions . the homogeneity of our materials under shear allows to infer their constitutive law from purely macroscopic measurements . this constitutive law is consistent with the one inferred from the velocity profiles . this contrasts with previous results obtained by @xcite within a microchannel where nonlocal finite size effects are likely to have prevented from obtaining a constitutive law . it also differs from the results of @xcite , where the loading procedure probably prevented the authors to reach a steady state . we thus suggest that properly analyzed purely macroscopic measurements in a wide gap couette geometry can be used as a tool to study dense emulsions . all behaviors we observed are basically consistent with herschel - bulkley laws of index 0.5 . however , we also evidence the existence of discrepancies with this law at the approach of the yield stress due to slow shear flows in the case of strongly adhesive emulsions , whose physical origin is unclear . finally , we have shown that there is probably no wall slip .

we construct a ramsey class whose objects are steiner systems . in contrast to the situation with general-uniform hypergraphs , it turns out that simply putting linear orders on their sets of vertices is not enough for this purpose : one also has to strengthen the notion of subobjects used from `` induced subsystems '' to something we call `` strongly induced subsystems '' . moreover we study the ramsey properties of other classes of steiner systems obtained from this class by either forgetting the order or by working with the usual notion of subsystems . this leads to a perhaps surprising induced ramsey theorem in which _ designs _ get coloured .

1607.02792

the entscheidungsproblem from mathematical logic led ramsey @xcite to the following combinatorial principle that became a cornerstone of an area now called ramsey theory . [ thm : ramsey ] for any positive integers @xmath1 , @xmath0 , and @xmath2 there exists a positive integer @xmath3 with @xmath4 i.e. , such that no matter how the @xmath0-element subsets of an @xmath3-element set @xmath5 are coloured with @xmath2 colours , there will always be an @xmath1-element subset @xmath6 of @xmath5 such that all @xmath0-element subsets of @xmath6 are the same colour . this result admits a standard reformulation in the language of hypergraphs : for a fixed integer @xmath7 , an _ @xmath0-uniform hypergraph _ , or _ @xmath0-graph _ for short , is a pair @xmath8 consisting of a vertex set @xmath9 and an edge set @xmath10 . if @xmath11 is an @xmath0-graph , we use the standard notation of writing @xmath12 , @xmath13 , @xmath14 , and @xmath15 . in case that @xmath16 has all possible edges , that is if @xmath17 , we say that @xmath16 is _ a clique _ and if additionally @xmath18 we say that @xmath16 is a @xmath19 . now theorem [ thm : ramsey ] informs us that if @xmath3 is large enough depending on @xmath1 , @xmath0 , and @xmath2 , then @xmath20 holds , meaning that in every edge - colouring of @xmath21 using @xmath2 colours there occurs a monochromatic copy of @xmath19 . this raises the question whether there is a similar result when the target hypergraph @xmath19 is replaced by an arbitrary @xmath0-graph @xmath16 . of course , when we just ask for a monochromatic appearance of @xmath16 as a subhypergraph we may apply ramsey s theorem with @xmath22 . but the problem becomes significantly more challenging when we ask for a monochromatic induced copy of @xmath16 . here , for two given @xmath0-graphs @xmath16 and @xmath23 , we say that @xmath16 is an _ induced subhypergraph _ of @xmath23 and write @xmath24 if @xmath25 and @xmath26 . by an _ induced copy _ of @xmath16 in @xmath23 we mean an induced subhypergraph @xmath27 of @xmath23 that is isomorphic to @xmath16 . [ thm : rind ] given any @xmath0-uniform hypergraph @xmath16 and any number @xmath28 of colours , there exists an @xmath0-uniform hypergraph @xmath23 with @xmath29 in the sense that for every colouring of the edges of @xmath23 with @xmath2 colours there exists a monochromatic induced copy of @xmath16 in @xmath23 . for @xmath30 this was proved independently by deuber @xcite , by erds , hajnal , and psa @xcite , and by rdl in his master thesis @xcite . for an alternative proof we refer to @xcite . that article introduces the so - called _ partite method _ on which much of the subsequent progress in this area is based , including the present work . there are even `` arithmetic '' applications of this method , see e.g. , the work of leader and russell @xcite . for general hypergraphs the proof of theorem [ thm : rind ] was obtained independently by abramson and harrington @xcite and by neetil and rdl @xcite . actually both of these articles prove considerably stronger results . shorter proofs of theorem [ thm : rind ] utilising partite structures may be found in @xcite and @xcite . the next level of generality is obtained by replacing the edge - symbol @xmath31 in formula by other @xmath0-graphs . such considerations may actually take place in a more abstract context that we are going to introduce next . [ dfn : rcl ] let @xmath32 be a class of objects endowed with an equivalence relation called _ isomorphism _ and with a transitive _ subobject _ relation . given two objects @xmath33 and @xmath16 from @xmath32 we write @xmath34 for the class of all subobjects of @xmath16 that are isomorphic to @xmath33 . for three objects @xmath35 and a positive integer @xmath2 the partition symbol @xmath36 means that no matter how @xmath37 gets coloured with @xmath2 colours , there is some @xmath38 for which @xmath39 is monochromatic . the class @xmath40 is said to have the _ @xmath33-ramsey property _ if for every @xmath41 and every @xmath2 there exists an @xmath42 with @xmath43 . finally , @xmath40 is a _ ramsey class _ if it has the @xmath33-ramsey property for every @xmath44 . for example , ramsey s theorem asserts that the class of finite sets with isomorphisms being bijections and subobjects being subsets is a ramsey class . moreover , the induced ramsey theorem tells us that the class @xmath45 of all @xmath0-uniform hypergraphs with the usual notion of isomorphism and whose subobjects are induced subhypergraphs has the @xmath31-ramsey property , where @xmath46 . it was proved in @xcite that @xmath45 has the @xmath33-ramsey property if @xmath33 is either a clique or discrete , i.e. , edgeless ( see also @xcite ) . that this condition on @xmath33 is also necessary was proved for @xmath30 in @xcite and it seems to belong to the folklore of the subject that the probabilistic approach of @xcite yields the same result for all @xmath7 ( see also theorem [ thm : main - b ] below ) . so @xmath45 is not a ramsey class , but it turns out that a slight variant of this class is ramsey : let @xmath47 be the class of all _ ordered @xmath0-uniform hypergraphs _ , that is , @xmath0-graphs endowed with a fixed linear ordering of their vertices . the isomorphisms of @xmath47 are required to respect these orderings . thus for @xmath48 with underlying ( unordered ) @xmath0-graphs @xmath33 and @xmath16 the set @xmath49 may in general correspond to a proper subset of @xmath34 . [ thm : hrc ] the class @xmath47 of all ordered @xmath0-uniform hypergraphs is a ramsey class . again this was proved in @xcite and @xcite . other known examples of ramsey classes include finite vector spaces over a fixed field @xmath33 ( see @xcite ) , and finite partially ordered sets with fixed linear extensions @xcite . the study of ramsey classes found its revival after several decades due to its connection with topological dynamics via ultrahomogeneous structures described by kechris , pestov , and todorcevic in their seminal paper ( see @xcite ) . further results in this direction were obtained in the recent work of van th @xcite and by hubika and neetril @xcite . a very readable account of the kechris - pestov - todorcevic correspondence has recently been given by solecki in his survey @xcite . for more information on structural ramsey theory in general we refer to bodirsky s survey chapter @xcite . throughout the rest of this article , we consider classes of @xmath0-graphs called steiner systems . for fixed integers @xmath50 , by a _ steiner @xmath51-system _ we mean an @xmath0-uniform hypergraph @xmath11 with the property that every @xmath52-element subset of @xmath9 is contained in at most one edge of @xmath16 . such objects are also called `` partial steiner systems '' in the design - theoretic literature , while the term `` steiner system '' is reserved there to what we will call `` complete steiner systems '' ( see definition [ dfn : cmpl ] below ) . we denote the class of all steiner @xmath51-systems with subobjects again being induced subhypergraphs by @xmath53 . for example , the members of @xmath54 are sometimes referred to as _ linear hypergraphs _ while @xmath55 . the following generalisation of the induced ramsey theorem was obtained by neetil and rdl in @xcite . [ thm : edgesteiner ] for any integers @xmath56 the class @xmath53 has the edge - ramsey property . as remarked in @xcite the proof described there does also show that the corresponding ordered class @xmath57 has the edge - ramsey property . the members of this class are , of course , steiner @xmath51-systems with fixed linear orderings of their vertex sets and the isomorphisms of @xmath57 are required to preserve these orderings . in the light of theorem [ thm : hrc ] it is natural to wonder whether these classes @xmath57 are ramsey classes as well , but it turns out that for @xmath58 they are not ( see also corollary [ cor : ind - st - ord ] below ) . the main result of this article , however , asserts that those classes can be made ramsey by changing the subobject relation as follows . [ dfn : strong ] given two steiner @xmath51-systems @xmath16 and @xmath23 , we say that @xmath16 is a _ strongly induced subsystem _ of @xmath23 and write @xmath59{{$\blacktriangleleft$}}}\hskip.4em}h$ ] if 1 . [ it : str - a ] @xmath24 , i.e. , @xmath16 is an induced subsystem of @xmath23 2 . [ it : str - b ] and moreover @xmath60 holds for all @xmath61 . the set of all strongly induced copies of @xmath16 in @xmath23 is denoted by @xmath62{{$\blacktriangleleft$}}}}$ ] . this concept leads to two further classes of steiner systems . [ dfn : str - cl ] let @xmath63{{$\blacktriangleleft$}}}}(r , t)$ ] be the class of all steiner @xmath51-systems with isomorphisms as usual and whose subobjects are strongly induced subsystems . similarly , @xmath63{{$\blacktriangleleft$}}}}_<(r , t)$ ] refers to the corresponding class of ordered steiner @xmath51-systems . we may now announce the first main result of this article . [ thm : main - a ] the class @xmath63{{$\blacktriangleleft$}}}}_<(r , t)$ ] is ramsey . in order to gain a better understanding as to why this is true , whilst in general the class @xmath53 fails to be ramsey , we will also determine for each of the four classes of steiner systems introduced so far for which objects @xmath33 they have the @xmath33-ramsey property . in this manner we can separately study the effects of ordering the vertices and of changing the definition of subobjects . the adjectives _ weak _ , _ strong _ , _ unordered _ , and _ ordered _ will be applied to these classes as indicated by the following table . [ cols="^,^,^",options="header " , ] it will turn out that the property of steiner systems relevant to unordered classes is that of homogeneity . [ dfn : rigid ] a steiner @xmath64-system @xmath33 is said to be _ homogeneous _ if every permutation of @xmath65 induces an automorphism of @xmath33 . thus a member of @xmath66 is homogeneous if either it is of the form @xmath67 for some @xmath68 or if it has no edges . it should be observed , however , that except for the edge @xmath31 the homogeneous @xmath0-graphs of the former type do not belong to @xmath69 as long as @xmath58 . similarly for @xmath58 the weak classes demand completeness . [ dfn : cmpl ] a steiner @xmath64-system @xmath33 is called _ complete _ if for every @xmath70 there is an edge @xmath71 with @xmath72 . trivial cases for this to happen are that @xmath33 consists of less than @xmath52 isolated vertices or that @xmath73 . non - trivial examples of complete steiner systems , also known as _ designs _ in the literature , are very hard to come up with ( at least for @xmath74 , say ) . but recently keevash @xcite established an important 160 year old conjecture of steiner regarding their existence . with this terminology the second main result of this article reads as follows . [ thm : main - b ] let @xmath56 , @xmath75{{$\blacktriangleleft$ } } } } , \ccs^{\,{\scalebox{1.1}[1.1]{{$\blacktriangleleft$}}}}_<\bigr\}$ ] , and @xmath76 . then @xmath77 has the @xmath33-ramsey property if both of the following conditions hold : 1 . [ it : main - uno ] if @xmath77 is unordered , then @xmath33 is homogeneous , 2 . [ it : main - weak ] if @xmath77 is weak and @xmath78 , then @xmath33 is a complete steiner system . if on the other hand [ it : main - uno ] or [ it : main - weak ] fails , then @xmath77 fails to have the @xmath33-ramsey property . notice that this generalises all other structural ramsey - theoretic theorems stated above . theorem [ thm : main - a ] follows , as the above clauses [ it : main - uno ] and [ it : main - weak ] hold vacuously for @xmath79{{$\blacktriangleleft$}}}}_<$ ] . the case @xmath80 and @xmath73 yields theorem [ thm : edgesteiner ] , and the induced ramsey theorem corresponds to the case @xmath81 , @xmath80 , and @xmath73 . for the readers convenience we would now like to state more explicitly what theorem [ thm : main - b ] says for @xmath75{{$\blacktriangleleft$}}}}\bigr\}$ ] . [ cor : ind - st ] the class @xmath53 has the @xmath33-ramsey property if and only if one of the following two cases holds : 1 . @xmath82 and either @xmath33 is an edge or @xmath83 , 2 . @xmath81 and @xmath33 is homogeneous . [ cor : ind - st - ord ] the class @xmath84 has the @xmath33-ramsey property if and only if @xmath81 or @xmath33 is a complete steiner @xmath64-system ( the cases @xmath85 and @xmath73 are included ) . [ cor : ind - st - str ] the class @xmath63{{$\blacktriangleleft$}}}}(r , t)$ ] has the @xmath33-ramsey property if and only if 1 . @xmath33 is an edge 2 . or @xmath33 is discrete , 3 . or if @xmath81 and @xmath33 is a clique as our proof of theorem [ thm : main - a ] is not so short and proceeds in several steps , we would like to begin with an informal discussion of some of its central ideas and how they relate to two earlier arguments yielding special cases , namely to the first of the two proofs of the induced ramsey theorem presented in @xcite and to the proof of theorem [ thm : edgesteiner ] from @xcite . a common theme occurring in all these proofs is that they rely on the _ partite construction _ introduced in @xcite . generally speaking , this construction is a versatile iterative amalgamation technique that allows in many situations to strengthen `` weak '' ramsey theoretic facts . for instance , ramsey s theorem trivially yields a version of theorem [ thm : rind ] where we just demand the monochromatic copy of @xmath16 to be a subhypergraph that is not necessarily induced . still , this can be used in a partite construction generating a bigger ramsey object for @xmath16 of the kind that is actually desired by theorem [ thm : rind ] . the two main ingredients of such a partite construction are 1 . a so - called _ partite lemma _ 2 . and _ partite amalgamations_. in both @xcite and @xcite the partite lemma essentially says that the result to be proved holds whenever @xmath16 is @xmath0-partite and it is proved by an application of the hales - jewett theorem discussed in subsection [ subsec : hj ] below . the partite amalgamations were straightforward in @xcite but in the case of steiner systems @xcite some arguments were needed to show that no pairs of distinct edges intersecting in @xmath52 or more vertices were created . this verification was in turn based on a study of the intersection properties of so - called `` canonical copies '' that correspond to the combinatorial lines of the hales - jewett cube ( see @xcite*lemma 2.8 ) . when one attempts to base a proof of theorem [ thm : main - a ] on the same ideas , there arises the following problem : while it is perfectly possible to use the hales - jewett theorem again for proving a partite lemma ( see the `` preliminary partite lemma '' in subsection [ subsec : ppl ] below ) it turns out that its intersection properties are not strong enough for enabling us to perform partite amalgamations in sufficiently general situations . our way of coping with that difficulty is that we run the partite construction twice . the first time we just aim at getting a better partite lemma , called the `` clean partite lemma '' in subsection [ subsec : pl ] below . owing to its more useful intersection properties this lemma can then be applied in a second partite construction that proves theorem [ thm : main - a ] ( see subsection [ subsec : apc ] below ) . for experts on the partite method it might also be interesting to observe that while our proof of the clean partite lemma uses a special feature of the hales - jewett cube ( see clause [ it : ppl2 ] of lemma [ subsec : ppl ] below ) , it only does so `` in the projection '' . throughout the rest of this section we fix integers @xmath56 and a number of colours @xmath28 . in this short subsection we will fix some terminology regarding the hales - jewett theorem . for a finite nonempty set @xmath86 and a positive integer @xmath87 , the hales - jewett cube @xmath88 is defined to be the cartesian power @xmath89 . for any @xmath90 $ ] we let @xmath91 denote the natural projection onto the @xmath92-th coordinate defined by @xmath93 for every @xmath94 . now consider all partitions @xmath95=c { { \mathbin{\ifx\mathbin\mathop\vphantom{\cup}\fi \mathpalette\mov@rlay{\cup\cr\cdot } } \ifx\mathbin\mathop\expandafter\displaylimits\fi}}m$ ] with @xmath96 and for each of them all functions @xmath97 . we will denote the set of all pairs @xmath98 arising in this way by @xmath99 . its members encode in the following way so - called combinatorial lines that are subsets of @xmath88 . for each @xmath100 we define the embedding @xmath101 such that for every @xmath90 $ ] and @xmath102 we have @xmath103 the range of this map , i.e. , the set @xmath104 $ ] , is called the _ combinatorial line _ associated with the pair @xmath98 . it is convenient to think of the coordinates from @xmath105 as being _ constant _ and of those from @xmath3 as being _ moving_. we may now state the hales - jewett theorem @xcite ( see also @xcite for a beautiful alternative proof ) . for any positive integers @xmath106 and @xmath2 there exists a positive integer @xmath107 such that whenever @xmath86 is a @xmath106-set and @xmath108 is an integer , every colouring of @xmath88 with @xmath2 colours contains a monochromatic combinatorial line . let @xmath109 be any positive integer . a steiner @xmath64-system @xmath5 is said to be _ @xmath109-partite _ if its vertex set may be partitioned into @xmath109 classes @xmath110 such that all edges @xmath111 are _ crossing _ in the sense that @xmath112 holds for all @xmath113 $ ] . we shall write @xmath114 for the class of all @xmath109-partite steiner @xmath51-systems @xmath115 having one such @xmath109-partition of its vertex set distinguished . of course , if @xmath116 then such a @xmath109-partite steiner system can not have any edges . but the additional structure on such partite steiner systems we need to deal with below may still be non - trivial . associated with each @xmath117 we have a _ projection _ @xmath118\ ] ] sending the vertices from @xmath119 to @xmath120 for every @xmath113 $ ] . in terms of this map , the crossing property of the edges means that @xmath121 is injective on every edge of @xmath5 . let us fix some @xmath122 with @xmath123 $ ] for the rest of this subsection . given any @xmath117 we shall write @xmath124{{$\blacktriangleleft$}}}}$ ] for the set of all those strongly induced copies @xmath125 of @xmath33 in @xmath5 which have the property that @xmath121 is an isomorphism from @xmath125 to @xmath33 . in particular , these copies of @xmath33 are `` crossing '' in @xmath5 and contain precisely one vertex from each set of the form @xmath119 . the little cross `` @xmath126 '' in the notation @xmath124{{$\blacktriangleleft$}}}}$ ] is intended to remind us of this fact . we are now ready to define the class of objects to which our partite lemmata will apply . [ dfn : fhyp ] for @xmath127 an _ @xmath33-hypergraph _ is a pair @xmath128 with @xmath117 and @xmath129{{$\blacktriangleleft$}}}}$ ] , such that @xmath130\in e(f ) \text { holds for every } e\in e(x)\,.\ ] ] the class of all @xmath33-hypergraphs is denoted by @xmath131 . the dotted `` edge '' must not exist . notice that not all copies of @xmath33 are in @xmath86 ( green ) . ] about two @xmath33-hypergraphs @xmath128 and @xmath132 we say that @xmath128 is _ strongly induced _ in @xmath132 and write @xmath133{{$\blacktriangleleft$}}}\hskip.4em}(y , \cr)$ ] if @xmath134{{$\blacktriangleleft$}}}\hskip.4em}y$ ] and @xmath135{{$\blacktriangleleft$}}}}\cap\cr$ ] hold . by @xmath136 we denote the set of all strongly induced copies @xmath137 of @xmath128 in @xmath132 . if @xmath138 is a system of such copies and @xmath2 is a positive integer , then the partition symbol @xmath139 means that for every colouring of @xmath140 with @xmath2 colours there exists a copy @xmath141 such that @xmath142 is monochromatic . evidently if @xmath132 supports any such system @xmath138 , then @xmath143 holds as well however , we may still gain something by considering the more general partition property because a system of copies @xmath144 for which it is valid might also have some additional properties which the full system @xmath136 possibly lacks . these additional properties can then be exploited in future arguments . for example , clause [ it : ppl2 ] of the preliminary partite lemma stated below is not hard to verify for the system of copies @xmath144 we get from the hales - jewett theorem , but it seems to be less clear whether @xmath136 would satisfy it as well . roughly speaking , this clause asserts that the property of a @xmath52-set to be contained in a relevant copy of @xmath33 reflects from @xmath132 to members of @xmath144 . [ lem : ppl ] for every @xmath33-hypergraph @xmath128 and every positive integer @xmath2 there exists an @xmath33-hypergraph @xmath132 together with a system of copies @xmath138 such that the following statements are true : 1 . [ it : ppl1 ] @xmath145 2 . [ it : ppl2 ] whenever @xmath141 , @xmath146 , and @xmath147 is a @xmath52-subset of @xmath148 , there is some @xmath149 with @xmath150 . from the preliminary partite lemma ] if @xmath151 we may just take @xmath152 and @xmath153 , and the assertion holds vacuously . so let us suppose @xmath154 from now on and set @xmath155 . let us write @xmath156 . we begin by defining a @xmath109-partite steiner @xmath51-system @xmath157 in the following way : 1 . for @xmath113 $ ] we let @xmath158 be the @xmath87-th cartesian power of @xmath159 . + a typical vertex from this class will be written as @xmath160 . for @xmath90 $ ] and @xmath113 $ ] , we let @xmath161 be the natural projection onto the @xmath92-th coordinate defined by @xmath162 for any @xmath163 as in the preceding sentence . a crossing @xmath0-subset @xmath164 of @xmath165 is declared to be an edge of @xmath6 if and only if @xmath166\in e(x)$ ] holds for all @xmath90 $ ] , i.e. , @xmath167 , \text { then } \pi_h[f ] \in e(x)\bigr\}\,.\ ] ] let us check that the @xmath109-partite hypergraph @xmath6 thus defined is indeed a steiner @xmath64-system . if @xmath168 are distinct , then @xmath169 . assume for the sake of contradiction that @xmath170 . by projection we get @xmath171 \cap \pi_h[f_2]|\ge t$ ] for every @xmath90 $ ] . since @xmath5 is a steiner @xmath64-system , it follows that @xmath172 = \pi_h[f_2]$ ] holds for every @xmath90 $ ] . this in turn yields @xmath173 . the above definition of the edges of @xmath6 shows that @xmath174\in e(f)\,,\ ] ] as required by definition [ dfn : fhyp ] . in what follows it will be convenient to denote the unique vertex in @xmath159 of a copy @xmath175{{$\blacktriangleleft$}}}}$ ] by @xmath176 , so that @xmath177\bigr\}$ ] holds for such copies @xmath125 . we are now ready to define a collection @xmath178{{$\blacktriangleleft$}}}}$ ] . to this end we consider any sequence @xmath179 of not necessarily distinct members of @xmath86 . for each @xmath113 $ ] we may look at the vertex @xmath180 from @xmath181 . evidently for each @xmath71 we have @xmath182 . together with this implies that the vertices @xmath183 span a crossing strongly induced copy of @xmath33 in @xmath6 , which we will denote by @xmath184 in the sequel . finally we set @xmath185 observe that @xmath186 yields a natural bijective correspondence between the hales - jewett cube @xmath187 and @xmath140 . next we address the subhypergraphs of @xmath6 corresponding to combinatorial lines . consider a partition @xmath95=c { { \mathbin{\ifx\mathbin\mathop\vphantom{\cup}\fi \mathpalette\mov@rlay{\cup\cr\cdot } } \ifx\mathbin\mathop\expandafter\displaylimits\fi}}m$ ] with @xmath96 as well as a function @xmath188 . observe that if @xmath189 , then @xmath190 is a copy of @xmath33 in @xmath5 . to describe the line encoded by the pair @xmath98 we will first introduce a partite map @xmath191 such that for @xmath192 and @xmath90 $ ] we have @xmath193 let @xmath194 be the induced @xmath109-partite subsystem of @xmath6 spanned by the range of @xmath195 . [ clm : line - ind ] if @xmath100 , then @xmath196{{$\blacktriangleleft$}}}\hskip.4em}y$ ] , where the isomorphism is given by @xmath197 . the injectivity of @xmath197 follows from @xmath96 . next we show that if @xmath31 is an edge of @xmath5 , then @xmath198 $ ] is an edge of @xmath194 . notice that @xmath198 $ ] is a crossing @xmath0-subset of @xmath6 . we need to verify that @xmath199\in e(x)$ ] holds for all @xmath90 $ ] . for @xmath200 we have @xmath199=e$ ] . on the other hand , if @xmath189 , then @xmath190 is some copy @xmath175{{$\blacktriangleleft$}}}}$ ] . since we have @xmath201\in e(f)$ ] in view of , it follows that @xmath199\in e\bl \til{f}\br$ ] . there are two things that remain to be shown , namely that all edges of @xmath194 have preimages in @xmath5 and that the inducedness of @xmath194 in @xmath6 is strong . observe that both of them are implied by the following statement : @xmath202=f$}\,.\ ] ] to prove this we put @xmath203 and let @xmath204 be the preimage of @xmath205 with respect to @xmath195 . for every @xmath200 we have @xmath206\in e(x)$ ] . since @xmath5 is a steiner @xmath64-system and @xmath207 , it follows that all edges of the form @xmath166 $ ] with @xmath200 must be the same . in other words there is an edge @xmath208 with @xmath209 $ ] for every @xmath200 . clearly @xmath31 is as desired . we keep considering @xmath105 , @xmath3 , and @xmath210 as above . let @xmath211 be the map given by . now @xmath212 $ ] is a subset of @xmath140 and one confirms easily that all vertices of the copies of @xmath33 belonging to this set lie in @xmath213 . the next claim asserts that combinatorial lines correspond to strongly induced copies of @xmath128 in @xmath132 . [ clm : combi - line ] if @xmath100 , then @xmath214{{$\blacktriangleleft$}}}\hskip.4em}(y , \cr)$ ] . owing to claim [ clm : line - ind ] the only thing that needs to be checked is @xmath215{{$\blacktriangleleft$}}}}\cap\cr$ ] . we leave the details to the reader . now we contend that @xmath132 and the system of copies @xmath216 have the properties [ it : ppl1 ] and [ it : ppl2 ] demanded by the preliminary partite lemma . the first of them is a direct consequence of the hales - jewett theorem and claim [ clm : combi - line ] . let us now prove [ it : ppl2 ] for some @xmath141 , @xmath217 , and @xmath52-set @xmath218 . choose a partition @xmath95=c { { \mathbin{\ifx\mathbin\mathop\vphantom{\cup}\fi \mathpalette\mov@rlay{\cup\cr\cdot } } \ifx\mathbin\mathop\expandafter\displaylimits\fi}}m$ ] with @xmath96 and a function @xmath188 such that @xmath219 and let @xmath220 . the assumption @xmath221 yields @xmath222\subseteq v(f_h)$ ] for every @xmath90 $ ] . in order to define @xmath223 we select an arbitrary @xmath224 , set @xmath225 for every @xmath90 $ ] , and finally we let @xmath226 . one sees immediately that @xmath227 and @xmath150 . we will now provide an abstract description of the partite construction that is general enough for our intended applications . since steiner @xmath64-systems are in general not closed under the kind of amalgamation we need to perform , we will explain everything with general @xmath0-uniform hypergraphs instead . actually it will be among the main difficulties encountered later on to formulate appropriate side conditions that allow us to maintain steinerness throughout the partite construction . the material that follows splits naturally into four parts . we begin by setting up some terminology regarding the `` pictures '' that the partite construction generates . next we introduce the so - called `` picture zero '' it is initialised with . in the subsubsection that follows we discuss the `` amalgamations '' that bring us from one picture to the next . finally we will be in a position to say precisely how the partite construction proceeds and what its main partition property ( see lemma [ lem : pcrams ] below ) asserts . for the purposes of this subsection , we fix an @xmath0-uniform hypergraph @xmath33 with @xmath109 vertices , say . the pictures we need to deal with for proving theorem [ thm : main - a ] will be three - layered structures consisting of a partite hypergraph , a system of distinguished copies of @xmath33 , and a system of distinguished copies of the object for which we intend to find a ramsey object . as a first step towards the definition of these pictures , we talk about hypergraphs with a distinguished system of copies of @xmath33 . by an _ @xmath33-system _ we mean a pair @xmath128 consisting of an @xmath0-uniform hypergraph @xmath5 and a collection @xmath228 of induced copies of @xmath33 in @xmath5 . this concept should not be confused with the @xmath33-hypergraphs from definition [ dfn : fhyp ] . the differences are that @xmath33-systems do not come with a @xmath109-partite structure and that the hypergraph @xmath5 is not required to be a steiner @xmath64-system for any @xmath58 . accordingly we can not demand the copies of @xmath33 belonging to @xmath86 to be strongly induced as we did it in the case of @xmath33-hypergraphs . in the sequel we will need to work with two different kinds of @xmath33-subsystems . for two @xmath33-systems @xmath128 and @xmath132 we say that the former is a _ semiinduced subsystem _ of the latter if @xmath229 and @xmath230 hold . if moreover @xmath231 we call @xmath128 an _ induced subsystem _ of @xmath132 and write @xmath232 . for a collection @xmath144 of semi - induced copies of @xmath128 in @xmath132 the partition symbol @xmath233 means that for every colouring of @xmath140 with @xmath2 colours there is some @xmath234 for which @xmath142 is monochromatic . for the remainder of this subsection we fix two @xmath33-systems @xmath128 and @xmath132 as well as a system of semi - induced copies @xmath144 such that holds . there arises no loss of generality by assuming @xmath235\,,\ ] ] where @xmath236 . let us enumerate @xmath140 and @xmath144 as @xmath237 and @xmath238 respectively . we may now describe the first two layers of our pictures . in the definition that follows , the projection @xmath239 $ ] is defined as in . -hypergraph ] [ dfn : yr - hg ] a _ @xmath132-hypergraph _ is a pair @xmath240 , 1 . [ it : yr1 ] where @xmath241 is an @xmath1-partite @xmath0-uniform hypergraph , 2 . [ it : yr2 ] such that if @xmath242 , then @xmath243\in e(y)$ ] 3 . [ it : yr3 ] and @xmath244 is a system of crossing induced copies of @xmath33 in @xmath241 , 4 . [ it : yr4 ] such that if @xmath245 , then @xmath246 . together with such @xmath132-hypergraphs @xmath240 , the partite construction will also generate systems of so - called good copies of @xmath128 within them . here a _ good copy _ of @xmath128 in @xmath240 is a crossing induced copy of @xmath128 that is isomorphic to a member of @xmath144 via @xmath247 . [ dfn : pic ] a _ picture _ is a triple @xmath248 such that 1 . [ it : pic1 ] @xmath240 is a @xmath132-hypergraph 2 . [ it : pic2 ] and @xmath249 is a system of good copies of @xmath128 in @xmath240 . if @xmath250 is a picture , we will write @xmath251 , @xmath252 , @xmath253 , @xmath254 , and @xmath255 . the next definition clarifies that for two pictures @xmath256 and @xmath257 the notation @xmath258 has its expected meaning . for two pictures @xmath250 and @xmath259 we write @xmath258 if the following hold : 1 . for every @xmath260 $ ] we have @xmath261 . 2 . @xmath240 in an induced @xmath33-subsystem of @xmath262 . 3 . a copy @xmath263 of @xmath128 with @xmath264 belongs to @xmath249 if and only if it belongs to @xmath265 . the starting point of the partite construction is a so - called `` picture zero '' @xmath266 , which has associated with each member @xmath267 of @xmath144 its own good copy @xmath268 of @xmath128 . these good copies are to be mutually vertex disjoint . moreover , for each @xmath269 $ ] the good copy @xmath270 is to be placed on the vertex classes @xmath271 of @xmath272 in such a way that @xmath273 holds if and only if @xmath274 . more exactly , we demand @xmath268 and @xmath267 to be isomorphic via the projection @xmath275 . the formal definition that follows summarises this description . [ dfn : picz ] _ picture zero _ is a picture @xmath266 with the property that we can write @xmath276 in such a way that the following holds : 1 . [ it : p01 ] @xmath277 2 . [ it : p02 ] @xmath278 3 . [ it : p03 ] @xmath279 4 . [ it : p04 ] if @xmath269 $ ] , then @xmath268 is isomorphic to @xmath267 via @xmath275 . the partite construction itself proceeds in @xmath280 successive amalgamation steps . to explain what happens in one such step , suppose that we have a picture @xmath250 as well as an integer @xmath281 $ ] . let @xmath282 list the vertices of @xmath283 in increasing order . define @xmath284 to be the @xmath109-partite @xmath33-system with @xmath285\ , , \\ e(z_\rho ) & = \bigl\{e\in e(z)\,\big|\,\psi_z[e]\in e(f_\rho)\bigr\}\ , , \\ \text { and } \quad \quad \quad \quad \quad \cs_\rho & = \bigl\{\til{f}\in \cs\,\big|\,\psi_{z}\bl \til{f}\br = f_\rho\bigr\}\,.\end{aligned}\ ] ] observe that @xmath284 is an induced @xmath33-subsystem of @xmath240 due to @xmath286 and the conditions [ it : yr2 ] , [ it : yr4 ] of definition [ dfn : yr - hg ] . now suppose @xmath287 to be a further @xmath109-partite @xmath33-system admitting a system @xmath288 of @xmath109-partite induced copies of @xmath284 with @xmath289 later on we will use either lemma [ lem : ppl ] or lemma [ lem : cpl ] to obtain such a system @xmath288 . the _ amalgamation _ we have in mind leads to a new picture @xmath290 the formal definition of which will cover the remainder of this subsubsection . the underlying idea is that starting from @xmath287 we extent every copy @xmath291 of @xmath284 to its own copy @xmath292 of the picture @xmath256 , keeping these copies of @xmath256 as disjoint as possible , i.e. , such that two distinct such copies do only share vertices of @xmath293 with each other . to begin with , the vertex classes of the desired hypergraph @xmath294 are going to be @xmath295 \cr v^j(z)\times \gw & \text { if } j\not\in\{j(1 ) , \ldots , j(k)\}\ , . \end{cases}\ ] ] now for every copy @xmath296 we fix a bijection @xmath297 establishing an isomorphism between @xmath284 and @xmath298 respecting the @xmath109-partite structure and then we extend @xmath299 to an injective map @xmath300 given by @xmath301 further , we let the picture @xmath292 be the image of @xmath256 under @xmath302 . finally @xmath303 is defined to be the union of all these pictures @xmath292 as @xmath304 varies over @xmath288 , so explicitly we stipulate @xmath305 it is easy to check that @xmath306 is again a picture and that we have @xmath307 the pictures of the form @xmath292 will be referred to as the _ canonical copies _ of @xmath256 in @xmath306 . having thus defined amalgamations , we may proceed by describing the partite construction itself . suppose that we construct a sequence @xmath308 of pictures starting with picture zero and such that for each @xmath281 $ ] we have @xmath309 then we say that the picture @xmath310 has arisen by means of a _ partite construction _ and about @xmath132 itself we say that it was put _ senkrecht_. we conclude this subsection by stating an important property of this construction . [ lem : pcrams ] if the picture @xmath250 arises by a partite construction , then @xmath311 keeping the above notation in force we have @xmath312 , @xmath313 , and @xmath314 . consider any colouring @xmath315 $ ] . when we go backwards through the partite construction , the first partition property we may invoke is @xmath316 . it leads to a canonical copy @xmath317 of @xmath318 and to a colour @xmath319 $ ] such that @xmath320 holds for all @xmath321 with @xmath322 . iterating this argument @xmath323 further times we may use the partition properties of @xmath324 in turn and ultimately obtain a function @xmath325\lra [ c]$ ] as well as a copy @xmath326 of picture zero such that @xmath327 next we apply to the colouring @xmath328 of @xmath140 and get some @xmath269 $ ] such that @xmath329 is monochromatic under this colouring . due to this means that the good copy @xmath330 of @xmath128 corresponding to @xmath331 has @xmath332 monochromatic under @xmath333 . now we analyse what happens when we clean the preliminary partite lemma by means of the partite method . resuming the discussion of subsection [ subsec : ppl ] , we suppose again that @xmath33 is a steiner @xmath64-system with @xmath123 $ ] . [ lem : cpl ] given an @xmath33-hypergraph @xmath128 and an integer @xmath2 there is an @xmath33-hypergraph @xmath240 and a system of copies @xmath334 such that we have : 1 . [ it : cpl1 ] @xmath335 2 . [ it : cpl2 ] if @xmath336 and @xmath337 are distinct members of @xmath249 and @xmath338 has size @xmath52 , then there are @xmath339 and @xmath340 with @xmath341 . owing to the preliminary partite lemma , there exists an @xmath33-hypergraph @xmath132 and a system of copies @xmath138 with @xmath145 such that @xmath342\,.\ ] ] as this situation does not change by adding isolated vertices to @xmath6 we may suppose for notational simplicity that this @xmath109-partite steiner system is balanced , i.e. , that @xmath343 holds for some positive multiple @xmath1 of @xmath109 . moreover , we may relabel the vertices of @xmath6 so as to obtain @xmath344 \quad \text{for all } i\in [ k]\ ] ] and , hence , @xmath345 $ ] . now we put @xmath132 senkrecht and run the partite construction . reusing the notation of subsection [ subsec : pc ] we enumerate @xmath140 and @xmath144 as in and , and let @xmath266 be picture zero as described in definition [ dfn : pic ] . the goal is to construct recursively a sequence @xmath346 of pictures with @xmath347 for each @xmath281 $ ] , where @xmath348 should be some ramsey system with @xmath349 . we intend to maintain throughout the construction 1 . [ it : alpha ] that each @xmath350 is a steiner @xmath64-system 2 . [ it : beta ] and that the copies of @xmath33 belonging to @xmath351 are strongly induced in @xmath350 . notice that picture zero has these properties because @xmath5 is a steiner @xmath64-system and the copies @xmath352 are strongly induced in @xmath5 . now suppose that for some @xmath353 $ ] we have already managed to construct a picture @xmath354 satisfying [ it : alpha ] and [ it : beta ] . then @xmath355 is an @xmath33-hypergraph and the preliminary partite lemma allows us to choose @xmath348 in such a way that @xmath356 is an @xmath33-hypergraph and that the copies @xmath357 are strongly induced in @xmath356 . it remains to be checked that the picture @xmath347 has the properties [ it : alpha ] and [ it : beta ] as well . starting with [ it : alpha ] we consider any two edges @xmath31 and @xmath358 of @xmath350 with @xmath359 . we are to prove that @xmath360 . let @xmath361 and @xmath362 be the canonical copies of picture @xmath363 with @xmath364 and @xmath365 . if @xmath366 we get the desired conclusion from and the induction hypothesis , so we may suppose @xmath367 from now on . this yields @xmath368 , i.e. , @xmath369\subseteq v(f_\rho)$ ] . using @xmath359 and @xmath370{{$\blacktriangleleft$}}}\hskip.4em}y$ ] we deduce @xmath371 and , as @xmath372 is a steiner @xmath64-system , this entails indeed @xmath360 . in view of @xmath373{{$\blacktriangleleft$}}}}$ ] and definition [ dfn : yr - hg][it : yr4 ] picture @xmath374 satisfies [ it : beta ] as well . this completes the proof that the partite construction we were aiming at can indeed be carried out . since the members of @xmath144 are strongly induced in @xmath132 , a similar argument yields @xmath375{{$\blacktriangleleft$}}}\hskip.4em}z^{|\cr| } \text { for all } \bl \til{x } , \til{\cq}\br \in \gz^{|\cr|}\,.\ ] ] the @xmath33-hypergraph @xmath240 and the system @xmath249 promised by the clean partite lemma will essentially be a @xmath109-partite reorganisation of the last picture @xmath310 and the partition property [ it : cpl1 ] will be an easy consequence of lemma [ lem : pcrams ] . to get [ it : cpl2 ] as well we prove that all the pictures we have generated satisfy this intersection property . in other words we contend that for every nonnegative @xmath376 we have : _ .3 cm @xmath377 .2 cm if @xmath378 are distinct and @xmath338 is a @xmath52-set , then there are @xmath339 and @xmath340 with @xmath341 . _ let us prove this by induction on @xmath379 . the base case @xmath380 is clear because by definition [ dfn : picz][it : p01 ] there are no distinct members @xmath381 with @xmath382 . for the inductive step we suppose that @xmath383 holds for some @xmath384 $ ] and that @xmath336 , @xmath337 as well as @xmath147 are as above . let @xmath361 and @xmath362 be the canonical copies of picture @xmath363 with @xmath385 and @xmath386 . if @xmath366 the desired conclusion can be drawn from the induction hypothesis . otherwise @xmath387 projects @xmath147 to a @xmath52-subset of @xmath388 and @xmath336 onto some member of @xmath144 , say @xmath267 . applying to @xmath267 , @xmath283 , and @xmath389 $ ] we get some @xmath390 with @xmath389\subseteq v(\til{f}')$ ] . the member @xmath391 of @xmath392 that @xmath393 projects to @xmath223 satisfies @xmath394 and for similar reasons there is some @xmath340 with @xmath395 . this completes the inductive step . finally we put everything together : since @xmath396 is a @xmath132-hypergraph , tells us that there is a @xmath109-partite steiner @xmath64-system @xmath241 with @xmath397{$\ , { { \mathbin{\ifx\mathbin\mathop\vphantom{\cup}\fi \mathpalette\mov@rlay{\cup\cr\cdot } } \ifx\mathbin\mathop\expandafter\displaylimits\fi}}\,$}}\ldots { \scalebox{1.1}[1.1]{$\ , { { \mathbin{\ifx\mathbin\mathop\vphantom{\cup}\fi \mathpalette\mov@rlay{\cup\cr\cdot } } \ifx\mathbin\mathop\expandafter\displaylimits\fi}}\,$}}v^{im / k}\bl z^{|\cr|}\br\ ] ] for all @xmath113 $ ] and @xmath398 . so informally @xmath241 is the same as @xmath399 except for having a different partite structure . setting @xmath313 and @xmath400 , the same argument shows that @xmath240 is actually an @xmath33-hypergraph and that we have @xmath334 by . now @xmath240 and @xmath249 have the desired properties , because [ it : cpl1 ] follows from lemma [ lem : pcrams ] and [ it : cpl2 ] holds in view of @xmath401 . now we are ready to prove that @xmath402{{$\blacktriangleleft$}}}}_<(r , t)$ ] is a ramsey class . to this end let any @xmath403{{$\blacktriangleleft$}}}}_<(r , t)$ ] be given . we are to find some @xmath404{{$\blacktriangleleft$}}}}_<(r , t)}$ ] with @xmath405 . recall that by theorem [ thm : hrc ] there is an ordered @xmath0-uniform hypergraph @xmath406 with @xmath407 but this is not enough because it is neither clear whether @xmath406 is a steiner @xmath64-system nor , if it actually is , whether the monochromatic copy of @xmath408 it leads to could always be chosen to be strongly induced . to overcome these problems we define @xmath140 to be the subset of @xmath409 corresponding to @xmath410 , put @xmath132 senkrecht , and try to run the partite construction using the clean partite lemma in every amalgamation step . for this purpose we suppose that @xmath411 $ ] and @xmath412 $ ] hold for @xmath413 and @xmath236 , and that the orderings of @xmath414 and @xmath406 agree with the natural orderings of @xmath415 $ ] and @xmath416 $ ] , respectively . moreover we let @xmath86 be the subset of @xmath417{{$\blacktriangleleft$}}}}$ ] corresponding to @xmath418{{$\blacktriangleleft$}}}}$ ] . by there exists a system @xmath144 of semi - induced copies of @xmath128 in @xmath132 with @xmath419 enumerate @xmath140 and @xmath144 as in and , and let the picture zero corresponding to this situation be given as in definition [ dfn : picz ] by @xmath266 . we intend to run the partite construction , thus generating a sequence @xmath420 of pictures . suppose that for some @xmath281 $ ] we have already managed to obtain , after @xmath421 steps , the picture @xmath354 such that the following conditions hold : 1 . the hypergraph @xmath422 is a steiner @xmath64-system . every member of @xmath423 is strongly induced in @xmath422 . 3 . if @xmath424 , then @xmath425 is strongly induced in @xmath422 . then in particular the @xmath109-partite @xmath33-system @xmath426 is an @xmath33-hypergraph by @xmath427 and @xmath428 . owing to the clean partite lemma there exists an @xmath33-hypergraph @xmath429 together with a system @xmath348 of partite , strongly induced copies of @xmath426 with @xmath430 such that 1 . if @xmath431 are distinct and @xmath432 is a @xmath52-set , then there are @xmath433 and @xmath434 with @xmath435 . we define @xmath436 and contend that this picture satisfies @xmath437 , @xmath438 , and @xmath439 . it should be clear that these claims easily follow from the induction hypothesis and the following statement : @xmath440{{$\blacktriangleleft$}}}\hskip.4em}z(\pi^\rho)\,.\ ] ] before proving this , we record two other properties of our construction . first , the copies from @xmath348 being strongly induced in @xmath429 entails : 1 . if @xmath441 and if @xmath361 is a canonical copy of @xmath363 with @xmath442 , then @xmath443 . second we reformulate @xmath377 in a more picturesque way : 1 . if @xmath444 are distinct canonical copies of @xmath363 in @xmath374 and @xmath445 is a @xmath52-set , then there are @xmath446 and @xmath447 with @xmath435 and @xmath448{{$\blacktriangleleft$}}}\hskip.4em}z^\rho_\rho$ ] . now we are ready to confirm . to this end , let a canonical copy @xmath361 , an edge @xmath449 and a @xmath52-set @xmath450 be given . we are to prove that @xmath443 . let @xmath362 be the canonical copy of @xmath363 with @xmath451 . since we are otherwise done , we may suppose that @xmath361 and @xmath362 are distinct . owing to @xmath452 there exists some @xmath447 with @xmath453 and @xmath454{{$\blacktriangleleft$}}}\hskip.4em}z^\rho_\rho$ ] . by @xmath428 we also have @xmath455{{$\blacktriangleleft$}}}\hskip.4em}z\bl \wh{\pi}^{\rho-1}\br$ ] . together with @xmath456 this implies @xmath457 . thus @xmath458 yields @xmath443 and consequently @xmath459 is indeed strongly induced @xmath460 . we have thereby completed the proof of and , hence , the proof of @xmath437 , @xmath438 , and @xmath439 . since picture zero evidently satisfies @xmath461 , @xmath462 , and @xmath463 , we have thus shown that the envisaged partite construction can indeed be carried out . recall that lemma [ lem : pcrams ] gives @xmath464 now let @xmath465 be the ordered steiner @xmath64-system obtained from @xmath399 by ordering the vertices in any way satisfying @xmath466 and forgetting the partite structure . then we have @xmath467{{$\blacktriangleleft$}}}}_<(r , t)$ ] by @xmath468 . moreover it is easy to deduce from @xmath469 , @xmath470 , and that @xmath471 holds in the sense of @xmath472{{$\blacktriangleleft$}}}}_<(r , t)$ ] . this means that @xmath465 has the desired ramsey property and the proof of theorem [ thm : main - a ] is complete . in this section we shall prove theorem [ thm : main - b ] . to this end we need to show , on the one hand , that under certain conditions a class @xmath473 has the @xmath33-ramsey property and , on the other hand , that there is a counterexample if these conditions fail . we call the results of the former type _ positive _ and those of the latter type _ negative_. the positive part of theorem [ thm : main - b ] is actually a direct corollary of theorem [ thm : main - a ] . the extra assumptions [ it : main - uno ] and [ it : main - weak ] from theorem [ thm : main - b ] are in the following way helpful for seeing this : 1 . if @xmath127 is homogeneous , then there is a unique ordered version @xmath414 of @xmath33 and for every @xmath474 there is natural bijective correspondence between @xmath49 and @xmath34 . 2 . if @xmath81 or if @xmath33 is a complete steiner @xmath64-system , then every induced copy of @xmath33 is in fact strongly induced . one way to show that these observations and theorem [ thm : main - a ] imply the positive part of theorem [ thm : main - b ] is to look separately at the three cases @xmath80 , @xmath475 , and @xmath79{{$\blacktriangleleft$}}}}$ ] . ( recall that the case @xmath79{{$\blacktriangleleft$}}}}_<$ ] was already covered by theorem [ thm : main - a ] ) . let us illustrate this by treating the class @xmath69 . suppose to this end that @xmath476 are given , where @xmath33 obeys conditions [ it : main - uno ] and [ it : main - weak ] . that is , @xmath33 is homogeneous and , in case @xmath58 , it is also a complete steiner @xmath64-system . take any ordered versions @xmath414 and @xmath477 of @xmath33 and @xmath16 , and let @xmath478{{$\blacktriangleleft$}}}}_<(r , t)$ ] with @xmath479 be given by theorem [ thm : main - a ] . it suffices to confirm that @xmath23 is as desired . so consider any @xmath2-colouring of @xmath37 . of course , this induces a @xmath2-colouring of @xmath480{{$\blacktriangleleft$}}}}$ ] and by our choice of @xmath23 there is a strongly induced copy @xmath481 of @xmath477 in @xmath482 for which @xmath483{{$\blacktriangleleft$}}}}$ ] is monochromatic . now @xmath484 is in particular an induced copy of @xmath16 in @xmath23 and by the observations above @xmath34 is monochromatic with respect to the colouring we started with . [ lem : order ] for every @xmath486 there is some @xmath487 such that no matter how we assign orderings to the vertex sets of @xmath16 and @xmath488 , thus obtaining @xmath489 , we will always have @xmath490{{$\blacktriangleleft$}}}}\not = \varnothing$ ] . due to the similarity with the argument from @xcite we only give a sketch . if @xmath488 is homogeneous we may take @xmath491 , so suppose from now on that the number @xmath1 of linear orderings on @xmath492 that lead to different ordered steiner systems is greater than @xmath493 . if the integer @xmath494 is large enough depending on @xmath495 , then an easy probabilistic argument shows that there is some @xmath496 with @xmath497 and @xmath498 . let @xmath482 be any ordered version of @xmath23 and let @xmath499 be the random ordered steiner system obtained by inserting independently and uniformly at random one of the @xmath1 ordered versions of @xmath488 into each edge of @xmath23 . if @xmath494 and @xmath23 were chosen so large that @xmath500 then with positive probability @xmath16 will be as desired . lemma [ lem : order ] says that the class @xmath402{{$\blacktriangleleft$}}}}(r , t)$ ] has the so - called _ ordering property_. there is an alternative proof of this fact using theorem [ thm : main - a ] . for a similar argument we refer to @xcite . for every @xmath501 that is not homogeneous there is some @xmath487 such that for every @xmath502 there is a red - blue colouring of @xmath37 such that for no @xmath503 the set @xmath504{{$\blacktriangleleft$}}}}$ ] is monochromatic . to confirm this we consider any two distinct ordered versions of @xmath33 , say @xmath505 and @xmath506 let @xmath502 be arbitrary and let @xmath482 be any ordering of @xmath23 . now colour all members of @xmath37 that are isomorphic to @xmath505 under this ordering _ red _ , those isomorphic to @xmath506 _ blue _ , and the remaining ones arbitrarily either red or blue . now look at any @xmath503 . notice that @xmath27 inherits an ordering from @xmath482 , thus becoming some @xmath507 . let @xmath508 be obtained from @xmath488 by ordering its vertices in such a way that its first @xmath509 vertices form a copy of @xmath505 while its remaining vertices from a copy of @xmath506 . by the choice of @xmath16 there is a strongly induced copy @xmath510 of @xmath508 in @xmath511 . by the construction of @xmath508 the set @xmath512{{$\blacktriangleleft$}}}}$ ] contains copies of both colours and , hence , so does @xmath504{{$\blacktriangleleft$}}}}$ ] . finally we need to show that condition [ it : main - weak ] from theorem [ thm : main - b ] is necessary . it seems convenient to deal with the cases @xmath475 and @xmath80 separately . we begin with the easier , ordered case . since @xmath414 is not complete , there is a @xmath52-set @xmath515 that does not appear in any edge of @xmath414 . we want to form new ordered steiner systems by adding an edge to @xmath414 through @xmath147 and @xmath516 new vertices . in view of @xmath58 , this can be done in at least two nonisomorphic ways . let @xmath505 and @xmath506 be two distinct ordered steiner @xmath64-systems that can arise in this way and let @xmath477 be their disjoint union . now if any @xmath517 is given , we may colour @xmath518 in such a way that exactly those copies of @xmath414 that sit in copies of @xmath505 are red whilst all others are blue . as the red copies of @xmath414 can not sit in copies of @xmath506 as well , there is no monochromatic copy of @xmath477 under this colouring . again let @xmath520 be a @xmath52-set not contained in any edge of @xmath16 and let @xmath391 be obtained from @xmath33 by adding a new edge containing @xmath147 and @xmath516 new vertices . define @xmath521 to be the disjoint union of two copies of @xmath391 and let @xmath522 be obtained by applying lemma [ lem : order ] to @xmath488 . we contend that there is no @xmath523 with @xmath524 . to see this , let any @xmath523 be given , order it arbitrarily to get some @xmath525 and consider the red - blue colouring of @xmath482 from the proof of lemma [ lem:34 ] . by our choice of @xmath16 every member of @xmath526 contains a copy of @xmath488 that in turn contains two copies of @xmath33 with different colours . so in particular no @xmath527 is monochromatic .

we construct series expansions for the scaling variables ( which transform multiplicatively under a renormalization group ( rg ) transformation ) in examples where the rg flows , going from an unstable ( wilson s ) fixed point to a stable ( high - temperature ) fixed point , can be calculated numerically . the examples are dyson s hierarchical model and a simplified version of it . we provide numerical evidence that the scaling variables about the two fixed points have overlapping domain of convergence . we show how quantities such as the magnetic susceptibility can be expressed in terms of these variables . this procedure provide accurate analytical expressions both in the critical and high - temperature region .

hep-lat0007037

in many statistical mechanics or field theory problems one needs to calculate the macroscopic features of a system in terms of some microscopic parameters . in field theory language , the microscopic parameters are called the bare parameters and the macroscopic behavior is encoded in the low momentum @xmath0-point functions which can be used to define the renormalized quantities . expressing the renormalized quantities in terms of the bare ones is a notoriously difficult problem . the renormalization group ( rg ) method @xcite was designed in part to tackle this problem . however , its practical implementation for realistic lattice models such as spin models or lattice gauge theories is still a formidable technical enterprise . near rg fixed points , expansions are often available . the main problem consists in interpolating between fixed points . an interesting attempt to model the interpolation between the weak and strong coupling regime of an asymptotically free gauge theory can be found in a recent publication of the qcd - taro collaboration @xcite . the interpolation was performed using the monte carlo simulation following a method developed in ref . . one delicate step of this method is the choice of a finite number of operators which can be used to express the renormalized action after successive rg transformations . in this article , we address the question of the interpolation between fixed points in models where the block spin method closes exactly and where the rg transformation can be calculated easily by numerical methods . one model satisfying these requirements is dyson s hierarchical model @xcite , for which only the local measure is renormalized according to a rather simple integral equation . this integral equation , briefly reviewed in section [ sec : hm ] can be treated with numerical integration methods @xcite , or by algebraic methods @xcite based on the fourier transform of the original equation . the second method leads to very accurate determinations of the critical exponents @xcite . one could say that the hierarchical model ( hm ) is `` numerically solvable '' , in the sense that for a given set of bare parameters , one can calculate the zero - momentum @xmath0-point functions . however , one has to repeat the tedious numerical procedure for every particular calculation and analytical expressions are more desirable . in this article , we show that this goal can be achieved by constructing the scaling variables corresponding to the fixed points governing the rg flows in the high - temperature ( ht ) phase . we use the terminology `` scaling variables '' for quantities which transform multiplicatively under a rg transformation . in ref . @xcite , where these quantitites were introduced , wegner call them `` scaling fields '' . indeed , in order to emphasize that they are functions of the bare parameters , we would prefer to call them `` scaling functions '' . however since this term is already reserved for other quantities , we will use the terminology `` scaling variables '' which seems to have prevailed @xcite over the years . the scaling variables can in principle be constructed in the vicinity of any fixed point . near a given fixed point , they are simply the eigen - directions of the linearized rg transformation . when moving away from the fixed point , non - linear terms need to be added order by order to maintain a multiplicative renormalization . what is the domain of convergence of this expansion ? can we combine two expansions in order to follow the flows in a crossover region ? these questions are in general very difficult to answer , but if the domains of convergence of the expansions fail to overlap , the whole approach is useless . the empirical results that we will present indicate that the domains of convergence of various expansions do overlap . these results should be considered as an encouragement to pursue this approach for other models . to limit the technical difficulties of constructing these functions to orders large enough to get an idea about the asymptotic behavior of their series expansions , we have imposed some limitations on the calculations presented below . first , we only consider the ht phase . second , we mostly consider the flows from the unstable fixed point to the ht fixed _ starting along the unstable direction_. the perturbative relaxation of this second condition will be briefly discussed . third , we only discuss the magnetic susceptibility and not the higher moments . the article is organized as follows . in section [ sec : hm ] , we review the basic facts about the hm and its rg transformation . we present a variant of the truncation method proposed in ref . @xcite which will be used in the rest of the article . we also clarify the relationship between the truncation and the ht expansion . in the next four sections , we present , explain and illustrate the main ideas with a simplified one - variable model @xcite where all the calculations are not too difficult . this model is simply a quadratic map with two fixed points , one stable and one unstable . the one - variable model is presented and motivated in section [ sec : toy ] , the scaling variables are constructed in section [ sec : scalingtm ] and their convergence studied in section [ sec : largetm ] . we want to make clear that all our assertions concerning the convergence of series are based on the analysis of numerical values of the coefficients rather than on analytical results concerning these coefficients . the susceptibility is calculated in section [ sec : sustm ] . the most important result for these four sections is illustrated fig . [ fig : tmoverlap ] which shows that scaling variables can be constructed accurately in overlapping regions . the rest of the article is devoted to generalizing the construction for the hm , for which the rg transformation can be approximated by a multivariable quadratic map . in section [ sec : multi ] , we show how to choose the coordinates in order to solve the linear problem . in section [ sec : impl ] , we present approximations which allow one to calculate nonlinear expansions for the scaling variables . finally , the questions of convergence are discussed in section [ sec : hmseries ] . the most important result is illustrated in fig . [ fig : hmoverlap ] which indicates overlapping domains of convergence . in more general terms , our article addresses the question of the non - linear behavior of the rg flows in models where non - linear expansions are calculable and can be compared with numerical solutions . in the first approximation , we have a linear theory . trying to go beyond this first approximation has some similarities with trying to make perturbation near integrable systems in classical mechanics . in this section , we review the basic facts about the rg transformation of the hm to be used in the rest of the paper . in order to avoid useless repetitions , we will refer the reader to ref . @xcite for a more complete discussion . in the following , we will emphasize new material such as the various possibilities available for the truncation procedure and the relationship between the truncation and the ht expansion . the energy density ( or action in the field theory language ) of the hm has two parts . one part is non - local ( the `` kinetic term '' ) and invariant under a rg transformation . its explicit form can be found , for instance , in section ii of ref . the other part is a sum of local potentials given in terms of a unique function @xmath1 . the exponential @xmath2 will be called the local measure and denoted @xmath3 . for instance , for landau - ginsburg models , the measures are of the form @xmath4 , but we can also consider limiting cases such as a ising measure @xmath5 . under a block spin transformation which integrates the spin variables in `` boxes '' with two sites , keeping their sum constant , the local measure transforms according to the intergral formula @xmath6 where @xmath7 is the inverse temperature ( or the coefficient in front of the kinetic term ) and @xmath8 is a normalization factor to be fixed at our convenience . we use the fourier transform @xmath9 we introduce a rescaling of @xmath10 by a factor @xmath11 , where @xmath12 and @xmath13 are constants to be fixed at our convenience , by defining @xmath14 in the following , we will use @xmath15 . for @xmath16 , this corresponds to the scaling of a massless gaussian field in @xmath17 dimensions . contrarily to what we have done in the past , we will here absorb the temperature in the measure by setting @xmath18 . with these choices , the rg transformation reads @xmath19\left[r_{n } \left({\sqrt{2}k\over 2 } \right ) \right]^2 \ . \label{eq : rec}\ ] ] we fix the normalization constant @xmath20 so that @xmath21 . for an ising measure , @xmath22 , while in general , we have to numerically integrate to determine the coefficients of @xmath23 expanded in terms of @xmath10 . if we taylor expand about the origin , @xmath24 ( where @xmath25 ) then the finite - volume susceptibility is @xmath26 the susceptibility @xmath27 is defined as @xmath28 the susceptibility tends to a finite limit for @xmath29 , where @xmath30 is a constant depending on @xmath31 . for @xmath7 equal to or larger than @xmath32 , the definition of @xmath27 requires a subtraction ( see e. g. , ref . @xcite for a practical implementation ) . in the following , we will only consider the ht phase ( @xmath33 ) . we have eliminated @xmath7 from the mapping , moving it to the initial local measure . the mapping has then fixed points independent of the temperature . one of them is the `` universal function '' @xmath34 found empirically in ref . @xcite and which can be written with great accuracy using numerical coefficients provided by koch and wittwer in ref . the initial measure as a function of @xmath7 determines a line of initial conditions in the parameter space , running from the ht fixed point , where all the initial parameters are zero , to the critical surface ( @xmath35 ) . we can derive an explicit form for @xmath36 in terms of @xmath37 . @xmath38 where @xmath39 to study the susceptibility not too far from the ht fixed point , we can expand @xmath27 in terms of @xmath7 . we need to expand each of the parameters to some power in @xmath7 at each level to find @xmath27 . since we choose the scaling factor @xmath12 so that @xmath7 is eliminated from the recursion , we find that @xmath40 . from the form of the recursion , eq . ( [ eq : recursion ] ) , we can see that @xmath37 will always have @xmath41 as the leading power in its ht expansion ( since @xmath42 ) . if we want @xmath43 expanded to order @xmath44 , we will use the truncated recursion formula @xmath45_{m_{max}}=\sum_{i=0}^{m_{max}-l } \frac{(-\frac{1}{2})^{i } ( 2(l+i ) ) ! } { ( 4/c)^{(l+i ) } i ! ( 2 l ) ! } \sum_{p+q = l+i}\left [ a_{n , p } a_{n , q}\right ] _ { m_{max } } \ , \label{eq : httruncation}\ ] ] where the notation @xmath46_{m_{max}}$ ] means that the expression in brackets should be expanded up to order @xmath47 in @xmath7 . we define the coefficients of the expansion of the infinite - volume susceptibility by @xmath48 we define @xmath49 , the ratio of two successive coefficients , and introduce quantities @xcite , called the extrapolated ratio ( @xmath50 ) and the extrapolated slope ( @xmath51 ) , defined by @xmath52 and @xmath53 where @xmath54 is called the normalized slope . if we use the expansion @xmath55 and assume that @xmath56 and @xmath57 are constants , we find @xmath58 where @xmath59 is a constant . however , if we calculate this quantity in @xmath60 with ising and landau- ginsburg measures , we find oscillations ( ref . @xcite ) . for comparison with results described later , fig . [ fig : hmexslope ] shows these oscillations in the extrapolated slope . the ht expansion can be calculated to very high order , however , due to a large number of subleading corrections , this is a very inefficient way to obtain information about the critical behavior . in ref . @xcite , it was found that one can obtain much better results by altering eq . ( [ eq : httruncation ] ) . first , we can retain a much smaller number of terms in the sum ( originally from 1 to @xmath61 ) than in the expansions @xmath46_{m_{max}}$ ] . as an example , one can calculate the 1000-th ht coefficient of @xmath27 with 16 digits of accuracy using only 35 terms in the sum . second , we can simply replace the expansions @xmath46_{m_{max}}$ ] by numerical values . this is equivalent to consider the polynomial approximation @xmath62 for some integer @xmath63 at each @xmath0-step . the mapping is then confined to an @xmath63-dimensional space . there remains to decide if one should or not truncate to order @xmath64 after squaring @xmath65 . this makes a difference since the exponential of the second derivative has terms with arbitrarily high order derivatives . numerically , one gets better results at intermediate values of @xmath63 by keeping all the terms in @xmath66 . in addition , for the calculations performed later , the intermediate truncation pads the `` structure constants '' of the maps ( see sec . [ sec : multi ] ) with about fifty percent of zeroes . a closer look at section [ sec : multi ] , may convince the reader that not truncating after squaring is more natural because we obtain correct ( in the sense that they keep their value when @xmath63 is increased ) structure constants in place of these zeroes . we have thus followed the second possibility where we truncate only once at the end of the calculation . with this choice @xmath67 compared to the ht expansion , the inital truncation to order @xmath63 is accurate up to order @xmath68 . after one iteration , we will miss terms of order @xmath69 but we will also generate some contributions of order @xmath70 . after @xmath0 iterations we generate some of the terms of order @xmath71 as in superconvergent expansions ( such as newton s method to calculate the roots of a polynomial ) . before attacking the multivariable expansions of the scaling variables , we would like to illustrate the main ideas and study the convergence of series with a simple one variable example which retains the important features : a critical temperature , rg flows going from an unstable fixed point to a stable one , and log - periodic oscillations in the susceptibility . in order to obtain a simple one - variable model , we first consider the @xmath72 truncation using eq . ( [ eq : truncation2 ] ) . the mapping is then reduced to only one variable . the mapping takes the form : @xmath73 expanding the denominator and keeping terms in the mapping only up to order @xmath74 in @xmath75 , we obtain @xmath76 from eq . ( [ eq : sus ] ) , we can put this recursion directly in terms of the ( truncation approximated ) susceptibility : @xmath77 this approximate equation was successfully used in ref . @xcite to model the finite - size effects . if we expand @xmath27 in @xmath7 , and define the exptrapolated slope , @xmath78 , as in eq . ( [ eq : shat ] ) , we see oscillations in fig . [ fig : tmexslope ] quite similar to those in the hm . [ fig : hmexslope ] ) . our goal is to obtain the susceptibility ( in the one - variable model ) as a function of @xmath7 and @xmath79 . despite the simple appearance of the susceptibility recursion , the behavior of the large-@xmath0 limit is not immediately apparent . simply running the recursion and examining the result yields little , as the expression becomes very complicated in short order . instead , we approximate the susceptibility near to either @xmath80 or @xmath35 , then successively build up more accurate approximations as we move the temperature away from this point . in the following , we use the notation @xmath81 and we only consider the case @xmath82 as in the ising model . our analysis will be simplified if we remove the @xmath0-dependence of the recursion . for this purpose , we define a new quantity , @xmath83 , such that @xmath84 where @xmath85 is an arbitrary constant . this gives the recursion @xmath86 the fixed points of this map are @xmath87 we can choose @xmath85 so that the non - zero fixed point is equal to one : @xmath88 this has the nice effect of making the fixed points independent of @xmath89 and @xmath7 . also , @xmath7 is removed entirely from the map , making it only dependent on @xmath89 : @xmath90 we call this map the `` @xmath91-map '' . we recover the susceptibility from : @xmath92 recalling that @xmath82 , we have @xmath93 allowing us to write @xmath94 for non - zero @xmath95 ( @xmath96 means that @xmath80 and @xmath97 for all @xmath0 ) . if we iterate the recursion eq . ( [ eq : hmap ] ) , we find that for @xmath98 , @xmath99 as @xmath100 . correspondingly , for @xmath100 , @xmath101 approaches a constant , @xmath27 , as we saw by using the original recursion . for @xmath102 , @xmath103 for all @xmath0 . for @xmath104 , @xmath105 for @xmath100 . this is the same behavior of the susceptibility expected as the temperature ( @xmath7 ) crosses the critical value . the adjustable parameter in @xmath95 is @xmath7 , so we see that @xmath102 corresponds to @xmath35 , consequently @xmath106 in fig . [ fig : h5 ] , we show the values of @xmath107 for different values of @xmath108 . this figure indicates that the shape of the function is independent of @xmath108 . to understand the behavior of the @xmath91-map , we need to investigate the stability of the fixed points . we find : @xmath109 and @xmath110 in the following , we only consider @xmath111 , so the zero fixed point is stable , while the other is unstable . all the @xmath95 between zero and the unstable fixed point are attracted toward the stable fixed point . also note that for all @xmath95 greater than the unstable fixed point , @xmath112 . for @xmath95 at the unstable fixed point , @xmath103 for all @xmath0 . so then @xmath113 . this diverges , since @xmath114 . note that the @xmath91-map can be put in the form used in collet and eckmann monography @xcite , @xmath115 with @xmath116 by using a linear transformation . with their parametrization , the first bifurcation where the fixed point becomes unstable and a cycle 2 develops occurs at @xmath117 . this corresponds to @xmath89=3 or -1 which is outside of the region where we will study the map in the following ( clearly , a cyclic behavior means no thermodynamics limit ) . we can expand the map about the unstable fixed point , @xmath103 . @xmath118 the usual notation for the eigenvalue near the critical point is @xmath119 , so we have @xmath120 . if we define @xmath121 then : @xmath122 with the starting value @xmath123 . we call this map the `` @xmath124-map '' . note the similarity of the @xmath124-map to the original @xmath91-map . we can introduce a duality transformation @xcite between the two maps which interchanges @xmath125 and @xmath126 . if the duality transformation is applied twice , we return to the original quantities . for @xmath127 , we also have @xmath128 with small values ( approaching 0 from above ) in one variable corresponding to large values ( approaching 1 from below ) values in the dual variable . we would like to construct the susceptibility , @xmath27 , as a function of @xmath129 . as a first approximation , we use eq . ( [ eq : dmap ] ) to linearize near the unstable fixed point ( @xmath130 ) , finding the critical exponent @xmath131 in the process . beginning with a value of @xmath132 very small ( @xmath7 close to @xmath32 ) , then for a certain number of iterations @xmath133 , so that @xmath134 . so as long as @xmath135 , then @xmath136 and @xmath137 . if we assume that @xmath83 stays near 1 for some number of iterations , and then drops quickly , near some @xmath138 , to the region where @xmath139 ( thus stabilizing @xmath27 ) , then @xmath140 . defining @xmath141 by @xmath142 , we have @xmath143 which gives @xmath144 using the usual notation for the critical exponent , the leading singularity is given by @xmath145 . so we have @xmath146 or , equivalently : @xmath147 we can divide the leading singularity out of @xmath27 and plot the remainder near to the critical point ( fig . [ lbet ] ) . we see periodic oscillations with respect to the variable @xmath148 which have period of @xmath149 . the linear approximation gives no clue as to the origins of the oscillations . to understand this , as well as the higher - order corrections to the leading singularity , we need to expand @xmath150 to higher order in @xmath132 . in this section , we show that the idea of scaling variable comes naturally as a way to express the susceptibility of the one - variable model in terms of the input parameters . we can find @xmath150 in terms of @xmath132 by expanding @xmath152 and plugging into eq . ( [ eq : rec ] ) . we find @xmath153 given the initial coefficients , i.e. @xmath154 , @xmath155 , and @xmath156 for @xmath157 , we can find all the coefficients at any @xmath0 . for example , @xmath158 finding forms for the coefficients explicitly in terms of @xmath119 and @xmath0 by this direct method quickly becomes difficult . we will show that this task can be simplified by introducing the scaling variables . as noticed before , the transition region from @xmath159 to @xmath160 has a shape which looks independent of @xmath132 ( see fig . [ fig : h5 ] and use eq . ( [ eq : doh ] ) ) . the shape seem to depend only on the value of @xmath119 . we can prove this in a restricted way by imagining a new initial value of `` @xmath132 '' which is identical to the current @xmath161 . the shape of the curve to the right would then be identical to the current case . likewise , we can reverse the map , getting new values for `` @xmath132 '' which nonetheless generate all the values in the current series @xmath162 . this inverse map is unique , assuming we confine it to only positive values for @xmath150 , and has the form @xmath163 in this case , the renormalization group is a group in the strict sense , since we have defined a unique inverse . what we would like then is a way of parameterizing @xmath150 in terms of a function independent of @xmath0 . regardless of our actual @xmath132 , we can extrapolate backwards , as suggested above , to another `` @xmath132 '' , which is small enough so that the linearized method works , where each new `` @xmath150 '' scales approximately as @xmath164 times the `` @xmath132 '' . this suggests using a new parameter , which we call @xmath165 , that scales exactly like this , so that if @xmath150 corresponds to @xmath165 , then @xmath166 corresponds to @xmath167 . this is our first covariant quantity , so - called since it scales exactly the same , regardless of the value of @xmath0 . the idea of using variables with simple transformation properties has a long history , for instance the angle - action variables in classical mechanics and the normal form of differential equations appearing in poincar s dissertation @xcite . for continuous rg transformations , wegner@xcite introduced the notion of `` scaling field '' . we discuss here its analog for discrete rg transformations . let us define a function @xmath124 such that @xmath168 . from the explicit form of the recursion formula eq . ( [ eq : rec ] ) this requirement implies @xmath169 if we let @xmath170 we find that @xmath171 the first coefficient is undetermined . we let @xmath172 , so that for small @xmath165 , we have @xmath173 . the first few coefficients give @xmath174 in fig . [ dy ] , we plot @xmath175 for several values of @xmath119 . in the limit where @xmath176 , we can calculate explicitly @xmath175 . this corresponds to @xmath177 , so that @xmath178 this is satisfied if @xmath179 , where @xmath180 is arbitrary . so we have @xmath181 . to satisfy @xmath182 for small @xmath150 , we need @xmath183 . finally , we have @xmath184 the expression of @xmath151 as @xmath185 allows us to interpolate between integral values of @xmath0 . in fig . [ dn ] , we show the curve for @xmath175 superimposed on points generated from the @xmath124-map for @xmath186 . 0.2pt [ dn ] we can invert the series for @xmath175 in terms of @xmath165 , defining a new function @xmath187 , so that @xmath188 . since we know the coefficients of the @xmath175 series , we can invert to get @xmath187 in terms of @xmath150 . more directly , we can use @xmath167 , or @xmath189 along with the constraint @xmath190 for small values of @xmath150 , to construct a series solution . if we let @xmath191 we find that @xmath192 with @xmath193 . the notation @xmath194 means to use the integer part of @xmath195 . the first few terms give : @xmath196 taking our @xmath176 example , we can immediately invert our result for @xmath175 , to find @xmath197 with the @xmath187-function , we can now reconstruct @xmath150 in terms of @xmath132 : @xmath198 because of the duality between our two maps , we can easily reproduce all of the above results of the @xmath124-map for the @xmath91-map . just as with @xmath150 , we can parameterize @xmath83 in terms of a covariant variable , @xmath199 , dual to @xmath165 . similarly to before , we define @xmath200 , with @xmath201 . we can also define the inverse function for @xmath91 , @xmath202 , where @xmath203 we can immediately carry over all of the results we found for @xmath175 and @xmath204 to @xmath205 and @xmath206 ) , replacing @xmath119 with @xmath89 . for example , @xmath207 while @xmath208 as with @xmath150 , we can find @xmath83 to any order in @xmath95 . the first few terms in the expansion give : @xmath209 the new covariant variable behaves similarly to @xmath165 . however , since @xmath210 , we see that as @xmath100 , @xmath211 , while as @xmath212 , @xmath213 . as with the @xmath124-map , @xmath199 goes to zero with @xmath83 , while blowing up as @xmath83 approaches @xmath214 . thus , for various @xmath89 , @xmath91 versus @xmath199 and @xmath202 versus @xmath83 will have at least qualitatively similar shapes to the corresponding @xmath124-map plots . in this section we give empirical results concerning the large order behavior of @xmath215 , @xmath216 and their inverses . from this , we infer the domain of convergence of the expansions . we would like to know if the series converge for the ranges of interest . the allowed ranges are @xmath217 to infinity for @xmath165 and @xmath199 , and @xmath217 to @xmath214 for @xmath150 and @xmath83 . for the @xmath124-map , we first examine the extreme allowed values for @xmath119 , @xmath214 and @xmath74 . for @xmath219 ( @xmath220 ) , we already found that @xmath221 . ( [ eq : coef ] ) converges in the whole complex plane . at the other end of our range , for @xmath222 ( @xmath223 ) , ( [ eq : coef ] ) becomes @xmath224 so that @xmath225 this converges only for @xmath226 , to @xmath227 . we use the ratio test to determine convergence for our series . the series will converge in the entire complex plane if @xmath228 increases without bound as @xmath229 . for all tested values of @xmath230 , we found that we could fit a straight line to plots of @xmath231 versus @xmath232 for large enough @xmath233 ( fig . [ line ] ) . 0.2pt thus for large @xmath233 , the coefficients follow the rule @xmath234 where we find that @xmath235 is always of order @xmath214 ( in fact , @xmath236 ) and @xmath237 . the value of @xmath10 can be determined by a linear fit . it seems clear that with @xmath230 , the series for @xmath218 converges for all @xmath238 . in agreement with the above exact series , for @xmath222 , @xmath239 and @xmath240 , while for @xmath176 , @xmath239 and @xmath241 . the intermediate values are given in fig . [ fig : slope ] . the form that the ratio of successive coefficients takes also gives us an idea of the number of coefficients we will need to approximate @xmath218 for a given value of @xmath187 . assuming that a particular term in the series becomes significant when it is the same size as the previous term , we find that the maximum order that we need in our approximation goes like @xmath242 thus for @xmath119 near @xmath74 , the number of terms needed will grow approximately linearly with @xmath187 . on the other hand , for @xmath119 near 1 , where @xmath243 , the number of terms will increase much more quickly with @xmath187 . we now turn to the dual function , @xmath91 . above , we found the series in @xmath218 for @xmath222 . from this , we can immediately write down @xmath244 when @xmath245 . the series is identical in form , with @xmath119 replaced by @xmath89 , so that @xmath246 which converges for @xmath247 to @xmath248 . for the general case , we again examine the ratios of successive coefficients , @xmath249 . we find the ratios flatten to constant values , for large enough @xmath233 , indicating finite radii of convergence . the radii get smaller and vanish as @xmath89 approaches zero as shown in fig . [ fig : radius ] . 0.2 in note that for any value of @xmath89 tried , we found good evidence that @xmath250 , indicating a @xmath251 behavior . this is consistent with the existence of a quadratic minimum for the inverse function discussed below . we already found above that for @xmath252 , @xmath253 . this series converges for @xmath254 , which is the required range . ( since @xmath187 must diverge as @xmath255 , it is not surprising that our series breaks down at this point . ) for @xmath222 , we have , inverting the @xmath124-function found above , that @xmath256 which also converges for @xmath254 . we find empirically from the analysis of ratios that for all @xmath230 , @xmath215 converges in the region @xmath254 . the analysis of the extrapolated slope for various @xmath119 gives convincing evidence that the main singularity has the form @xmath257 this can be seen with short series when @xmath119 is close to one and requires larger and larger series as @xmath119 gets close to 2 . the @xmath216 series behaves similarly to the @xmath215 , having a radius of convergence of @xmath214 for all @xmath89 . in this case , the ratio of coefficients @xmath258 approaches @xmath214 smoothly from below as @xmath233 increases , for larger @xmath89 around @xmath259 , small oscillations begin to appear in the curve . as @xmath89 becomes smaller , these oscillations grow and become quite large for @xmath89 smaller than @xmath260 ( fig . [ ytco2 ] ) . however , in all cases studied , the ratio eventually approaches @xmath214 , for large enough @xmath233 . a detailed study shows that if we continuate @xmath216 for negative values of @xmath91 using the series expansion , the function develops a quadratic minimum when @xmath261 . the absolute value of this minimum found in all examples studied is compatible with the radius of convergence of the inverse function and the square root singularity of the @xmath244 discussed above . as one can guess by looking at fig . [ fig : tmexslope ] , the analysis of the extrapolated slope is intricated . however , if we calculate enough terms and if @xmath119 is not too close to 2 , one can get approximate results which are consistent with a main singularity of the form @xmath262 for instance , just by looking at the asymptotic behavior of fig . [ fig : tmexslope ] , one can see that @xmath263 , as expected , with errors of the order @xmath264 . now that we can calculate @xmath83 and @xmath150 , we can construct the ( infinite - volume limit ) susceptibility . if we focus on the ht fixed point and the @xmath91-map , we get the ht expansion in terms of @xmath265 . on the other hand , to understand the susceptibility s behavior near to the fixed point , we need to expand in terms of @xmath266 . finally , we consider the possibility of combining the two expansions in the crossover region . we could find the ht expansion from the finite - volume susceptibility found in the previous section , by taking the large-@xmath0 limit . since @xmath267 , all of the @xmath268 terms will drop out . however , we can also construct the large-@xmath0 limit directly . recalling that @xmath269 , and using @xmath201 , we find that @xmath270 the infinite - volume limit becomes : @xmath271 using the @xmath202-expansion for @xmath91 , we find @xmath272 as @xmath100 , @xmath273 , so the infinite volume limit of the susceptibility is : @xmath274 using the @xmath91-expansion for @xmath202 , we get the ht expansion for the susceptibility : @xmath275 using @xmath269 , we find the ht expansion for the finite - volume susceptibility ( recall that @xmath276 ) : @xmath277 we can find the error in the finite - volume susceptibility as an approximation to @xmath27 . for large @xmath0 ( small @xmath199 ) , we get @xmath278 . the relative error is then : @xmath279 this behavior is observed @xcite in the approach to infinite volume for the hm . in addition of using the ht expansion ( expressing @xmath27 in terms of @xmath95 ) , we would also like to have an expansion in terms of the dual variable @xmath132 . multiplying the numerator and the denominator of eq . ( [ eq : suscep ] ) by @xmath280 , we obtain @xmath281 with @xmath282 but since @xmath283 , we have also @xmath284 for any @xmath0 . in other words , @xmath285 is rg invariant . due to the discrete nature of our rg transformation , @xmath285 is not exactly a constant . if expressed in terms of ln(@xmath286 ) , @xmath285 is a periodic function of period ln@xmath119 . however , for @xmath119 not too close to 2 , the non - zero fourier modes are very small @xcite . as we have seen above , @xmath287 is the same for every @xmath0 . we can thus pick @xmath0 such that we are just in the crossover region and @xmath288 expansions have a reasonable chance to be accurate . in order to test the accuracy of the approximations @xmath289 ( series expansion up to a certain order ) for various @xmath0 , we have prepared an empirical sequences of @xmath151 starting with @xmath290 . we have then tested the scaling properties by calculating @xmath291 - 1|\ , \label{eq : bdn}\ ] ] where @xmath286 was calculated with 16 digits of accuracy . optimal approximations are those for which @xmath292 . for such approximation , scaling is as good as it can possibly be . indeed , due to the peculiar way numerical errors propagate @xcite , one does not reach exactly the expected level @xmath293 ( more about this question in section [ sec : hmseries ] ) . we can define a similar dual quantity by replacing @xmath124 by @xmath91 and @xmath187 by @xmath202 . in this case , @xmath294 is estimated with the same accuracy as @xmath286 by stabilizing @xmath295 , for large enough @xmath0 . we have performed this calculation for @xmath296 1.1 , 1.5 and 1.9 . the conclusions in the three cases are identical . for @xmath0 large enough , the @xmath297 of @xmath187 starts increasing from @xmath293 until it saturates around 1 . by increasing the number of terms in the expansion , we can increase the value of @xmath0 for which we start losing accuracy . similarly , for @xmath0 low enough , the @xmath297 of @xmath202 starts increasing etc ... we want to know if it is possible to calculate enough terms in each expansion to get scaling with some desired accuracy for both functions . the answer to this question is affirmative according to fig . [ fig : tmoverlap ] for @xmath298 . 0.2 in one sees , for instance , that with 10 terms in each series , we have scaling with about 1 part in 1000 near @xmath299 for both expansions . the situation can be improved . for 70 - 70 expansions , an optimal accuracy is reached from @xmath300 to 46 . for the other values of @xmath301 quoted above , similar conclusions are reached , the only difference being the optimal values of @xmath0 . another evidence for overlapping convergence is that we can calculate the rg invariant @xmath285 for a certain range of @xmath302 . to evaluate @xmath285 , we use the series expansions for @xmath202 and @xmath124 , cutting each off at some order : @xmath303 where @xmath304 and @xmath305 are the @xmath233th coefficients in the @xmath124 and @xmath202 series , respectively . the @xmath124 series is more accurate the smaller is the value of @xmath187 , while the @xmath202 series is more accurate for @xmath218 close to @xmath214 , which means large values of @xmath187 . thus for @xmath187 very large , we need many terms in the @xmath124 series , while we need less terms in the @xmath202 series . on the other hand , when @xmath187 is very small , we need a large value for @xmath306 and a relatively small @xmath307 . we have found that , given a fixed value of @xmath308 , the most accurate values for @xmath285 are obtained when @xmath309 . in fig . [ fig : all ] , we show @xmath285 calculated by keeping @xmath310 terms each in the expansions for @xmath187 and @xmath202 . the result is plotted against @xmath311 . we used @xmath312 , which makes the oscillations much larger than , for example , near to @xmath313 . near the fixed points , we need more terms in the appropriate series to get accurate results . we can study the oscillation we see in @xmath285 by looking at its fourier expansion . since @xmath285 is periodic in @xmath314 , we can express @xmath315 where @xmath316 . the coefficients are given as @xmath317 for some @xmath318 of our choice . as an example , we calculate @xmath319 for @xmath312 where the oscillations are substantial . the choice of the interval can be infered from fig . [ fig : all ] . if we had infinite series , the function would be exactly periodic . for finite series , we see that @xmath318 can not be too large or too small . for intermediate values , we obtain @xmath320 . in fig . [ allm0 ] , we show this constant term subtracted from the @xmath285 we evaluated above . what remains is oscillations about @xmath217 . we also calculated @xmath321 ( using @xmath322 so that we keep information about the phase constant as well as the amplitude ) . we construct the first oscillating mode from this , and in fig . ( [ allm0m1 ] ) we show the result of subtracting out this in addition to the constant term . we are left with even smaller oscillations , from the @xmath323 term . in fig . [ allm0m1 ] , one notices that when @xmath324 is of the order of 1.7 , the breakdown of the approximation starts with small numerical noise before the complete breakdown near @xmath325 is observed ( see fig . [ fig : all ] ) . in the one - variable model , we learned that we could express the basic mapping parameter , @xmath83 , the distance from the ht fixed point , in terms of either of two new variables , @xmath165 or @xmath199 . the first of these variables transforms multiplicatively with @xmath119 , the eigenvalue near the unstable fixed point , while the second transforms multiplicatively with @xmath89 , the eigenvalue near the ht fixed point . we would like to extend these methods to dyson s hm . we will express the parameters of the model in terms of two new sets of variables @xmath326 and @xmath327 , where @xmath63 is equal to the number of parameters we keep in the mapping . these are the scaling variables . near to the unstable fixed point we find a set of eigenvectors and eigenvalues which control the rg flows in the linearized region . we use the notation @xmath328 for the @xmath329th eigenvalue , ordered in size from greatest to least . for each of these eigenvalues , we introduce a variable @xmath330 which transforms under the rg transformation as @xmath331 . we denote @xmath332 for the eigenvalues in the linear ht region , similarly defining variables which transform as @xmath333 . then we can expand each of the parameters in terms of each set of scaling variables . our first task will be to use special coordinates where the problem will be solved at the linear level . to use a relativistic analogy , the transformations leading to such a system of coordinates ( the `` linear scaling variables '' ) , can be interpreted as a `` translation '' or shift and a `` rotation '' or more generally a linear transformation . we previously considered truncation in the @xmath37 parameters . unfortunately , this approximation becomes less accurate if we are interested in the critical region . we would like to investigate a truncation that works near to the nontrivial fixed point . we re - express the basic recursion in terms of a new set of parameters , which give the distance from the nontrivial fixed point . for notational convenience , we first rewrite the unnormalized recursion given in eq . ( [ eq : truncation2 ] ) , using the `` structure constants '' : @xmath334 with @xmath335 these zeroes can be understood as a `` selection rule '' associated with the fact that @xmath37 is of order @xmath41 as explained in section [ sec : hm ] . if we follow the truncation procedure explained in section [ sec : hm ] , the indices simply run over a finite number of values . we use `` relativistic '' notations . repeated indices mean summation . the greek indices indices @xmath336 and @xmath337 go from @xmath217 to @xmath63 , while latin indices @xmath329 , @xmath338 go from 1 to @xmath63 . obviously , @xmath339 is symmetric in @xmath336 and @xmath337 . by construction , @xmath25 and we can write the normalized recursion in the form : @xmath340 with @xmath341 let assume that we know a fixed point of the recursion formula @xmath342 . we then introduce intermediate variables @xmath343 which are zero at the fixed point : @xmath344 plugging this relation into eq . ( [ eq : general ] ) , using the fixed point properties , subtracting the fixed point and reducing to the same denominator , we obtain a recursion formula for the @xmath343 having the same form as eq . ( [ eq : general ] ) with the substitutions @xmath345 and @xmath346 where @xmath347 . these equations express the transformation properties of the structure constants under a translation ( shift ) of the coordinates . in particular if @xmath348 ( as for the ht fixed point ) , one can see that since @xmath349 , the transformation reduces to the identity . we will now study the transformation properties under changes of coordinates which will diagonalize the linear rg transformation matrix @xmath350 . the diagonalization of the linear @xmath350 near the ht fixed point is quite simple because it is of the upper triangular form . the eigenvalues are just the diagonal terms . from eq . ( [ eq : struct ] ) , one sees that @xmath233th diagonal term is given as @xmath351 this spectrum was obtained in ref . @xcite with a different method . this means that an @xmath352 truncated version of the matrix will have eigenvalues which are identical to the first @xmath233 eigenvalues of the full hm map . furthermore , the @xmath233th eigenvector will contain only @xmath233 non - zero entries . this means that if we truncate to @xmath353 , we are simultaneously truncating to the parameter space to a subspace spanned by the first @xmath63 eigenvectors . this `` stability '' is due to the special relationship existing between the ht expansion and the polynomial truncation explained in section [ sec : hm ] . on the other hand , the linearized map near the nontrivial fixed point does not have these simple properties . the eigenvalues may be determined numerically from a truncated version of the linearized matrix . with a large enough truncation , we can find a certain number of the eigenvalues to any desired precision . we find that there is one eigenvalue larger than @xmath214 , with all the rest less than @xmath214 . though we do not have a simple formula for these eigenvalues , we know that they do shrink in size quickly , similarly to the eigenvalues of the ht fixed point . for instance the numerical values for @xmath354 are @xmath355 , @xmath356 . a more complete list is given in ref . @xcite . since @xmath357 is not a symmetric matrix the left and right eigenvector are distinct . @xmath358 and @xmath359 the notation @xmath360 means that there is no sum on @xmath361 . since all the eigenvalues are distinct , one can normalize the eigenvectors in such a way that @xmath362 similarly a completeness relation ( decomposition of the identity ) can be obtained by summing over @xmath361 : @xmath363 in numerical calculations , the relations of orthogonality and completeness provide a reliable way to check the accuracy of the calculations . we can thus diagonalize the linearized rg : @xmath364 if we start with the transformation expressed in the general form eq . ( [ eq : general ] ) but with @xmath180 replaced by @xmath365 to indicate that the fixed point is at the origin , as in the previous subsection , we can define a new set of coordinates @xmath366 , such that @xmath367 or , equivalently , @xmath368 with these notations , all the structure constants transform according to the familiar rules of tensor analysis . since @xmath369 and @xmath370 are inverse of each other , we can think of lower indices as covariant and upper indices as contravariant . as an example of transformation under eq . ( [ eq : transf ] ) , we have @xmath371 the other structure constants transform according to the same rules . the eigenvectors are not determined uniquely even after normalizing . if @xmath372 and @xmath373 are components of normalized right and left eigenvectors , then @xmath374 and @xmath375 , where @xmath376 is a constant , will work equally well . in the following , we will fix this ambiguity by requiring that the `` other '' fixed point be at @xmath377 in the new coordinates . in summary , we can choose a system of coordinate @xmath378 where the unstable fixed point will be at the origin of the coordinate and the ht fixed point at @xmath379 , and such that the linear rg transformation is diagonal . in this system of coordinates , the rg transformation reads @xmath380 where the new structure constant are calculated from the original ones according to the transformation laws discussed in the previous two subsections . similarly , we can introduce new coordinates @xmath381 , so that @xmath382 and take the eigenvectors so that the unstable fixed point is at @xmath383 in the new system of coordinates . note that the form of the eigenvectors guarantees that @xmath384 is of order @xmath41 . this can be seen by inverting eq . ( [ eq : atoh ] ) using the matrix of left eigenvalues . due to the upper - diagonal form , the second left eigenvector has its first entry equal to zero , the third its first two entries etc ... . in this new system , the rg transformation reads @xmath385 we would like to express the evolution of the canonical coordinates in terms of functions of the scaling variables : @xmath386 where the sums over the @xmath329 s run from @xmath217 to infinity in each variable and @xmath387 . in practice , we expand in terms of some finite number of the scaling variables , less than or equal to the number of @xmath124 parameters we have truncated to . using the notation @xmath388 ) and the product symbol , we may rewrite the expansion as @xmath389 using the transformation law for the scaling variables , we have @xmath390 each constant term , @xmath391 , is zero , as the scaling variables vanish at the fixed point . from eq . ( [ eq : hmdrules ] ) , we see that all but one of the linear terms are zero for each value of @xmath361 . the remaining term is the one proportional to the @xmath361th scaling variable . we take these coefficients to be @xmath214 , so that the @xmath392 for small @xmath393 . for the higher - order terms , we obtain the recursion @xmath394 the calculation can be organized in such way that the r. h. s. of the equations are already known . this will be the case for instance if we proceed order by order in @xmath395 , the degree of non - linearity . as one can see , this expansion may suffer from `` small denominator problems '' . this issue will be discussed elsewhere @xcite . we can likewise expand each @xmath396 in terms of scaling variables @xmath397 . the derived recursions are identical in form to those derived above . from eq . ( [ eq : hteigenv ] ) , one sees that the denominator will vanish for some equations and unless the numerator is also zero , the expansion is ill - defined . again this question will be discussed elsewhere @xcite . in the following , we will only use the expansion for the @xmath124-functions . one can likewise find expansions of the scaling variables in terms of the canonical coordinates , by setting @xmath398 and requiring that when @xmath399 is replaced by @xmath400 , the function is multiplied by @xmath401 . since @xmath400 has a denominator , it needs to be expanded for instance in increasing order of non - linearity . the same considerations applies for @xmath202 . note that small denominators may also be present in these calculations . however , the calculation of @xmath402 and @xmath403 is free of such a problem since the largest eigenvalue can not be written as product of lower eigenvalues smaller than 1 . it is in principle simple to obtain these expansions order by order in the degree of non - linearity , however there exists some practical limitations . for instance , if we want to calculate all the non - linear terms of order 10 in any of the variables , with @xmath63=30 , we need to calculate and store @xmath404 numbers . in the next section , we show that one can organize these calculations in a way which allows accurate answers for the susceptibility . from eqs . ( [ eq : sus ] ) and ( [ eq : atoh ] ) , we obtain @xmath405 for @xmath0 large enough , the linear behavior applies and the the @xmath396 get multiplied by @xmath406 at each iteration . in the large @xmath0 limit , only the @xmath407 term survives and consequently , @xmath408 using the same method as in the one - variable model , we can in the limit replace @xmath409 by @xmath410 and obtain @xmath411 this allows us to write @xmath412 where @xmath413 and @xmath285 the rg invariant @xmath414 for reference , in the case @xmath354 , @xmath415 . one can calculate the subtracted four - point function following the same procedure , namely expressing the @xmath37 in terms of the @xmath384 . however , @xmath416 and one needs to go beyond the linear expansion to calculate these quantities . in this section , we show that it is possible to design approximations such that one can calculate the susceptibility using eq . ( [ eq : susyy ] ) . for this purpose we have calculated an empirical series of @xmath37 with @xmath354 , @xmath417 and an initial ising measure . detail relevant for this calculations can be found in refs . the calculations have been performed with @xmath418 , a value for which at the @xmath7 considered , the errors due to the truncation are of the same order as those due to the numerical errors . these errors are small enough to allow a determination of the susceptibility with seven significant digits if we use double precision . the empirical flow proceeds in two steps . first , the flow goes from the initial mesure to close to the unstable fixed point . second , the flow goes from close to the unstable fixed point to the ht fixed point . the first step depends on the choice of the initial measure and will not be discussed in full detail . our main goal will be to show that it is possible to construct nonlinear expansions which allow to describe accurately the second step . we now discuss the flow chronologically . in order to keep track of the chronological sequence , we postponed the technical discussion of the convergence of the series to section [ sec : hmseries ] . in this section , the results are simply quoted . our choice of @xmath7 means that we start near the stable manifold . after about 25 iterations , we start approaching the unstable fixed point and the linear behavior @xmath419 becomes a good approximation . during the next 20 iterations , the irrelevant variables die off at the linear rate and at the same time we move away from the fixed point along the unstable direction , also at the linear rate . at @xmath420 we are in good approximation on the unstable manifold and @xmath421 becomes proportional to @xmath422 . in other words , the non - linear terms are taking over . at this point , we can approximate the @xmath366 as functions of @xmath402 only : @xmath423 . this approximation is consistent in the sense that if @xmath424 at @xmath425 then it is also the case for positive @xmath0 . we have calculated @xmath426 up to order 80 in @xmath402 using eq . ( [ eq : sca ] ) . we have then inverted @xmath427 to obtain @xmath402 . given the empirical @xmath428 we then calculated the approximate @xmath429 and then used the other functions @xmath430 ( with @xmath431 ) to `` predict '' @xmath366 . comparison with the actual numbers were good in a restricted range . for @xmath432 , the relative errors were less than a percent . they kept decreasing to less than one part in 10,000 for @xmath433 and then increased again . it will be shown in section [ sec : hmseries ] that this corresponds to the fact that when @xmath402 becomes too large ( a value of 3.7 first exceeded at @xmath434 ) , the series expansion of @xmath427 diverges , unlike the one - variable model for which @xmath218 is analytical . it will also be shown later that the quality of the approximation between @xmath299 and @xmath435 can be improved by by treating @xmath436 , @xmath437 perturbatively . near @xmath435 , the presence of the ht fixed point starts dominating the flow but we are still far away from the linear regime . to take into account the non - linear effects , we have calculated @xmath438 up to order 11 in @xmath7 . the reason for this modest order is that this is a multi - variable expansion . recalling the discussion about the ht expansion in section [ sec : hm ] , we can count the number of terms at each order in @xmath7 . at order two , we have @xmath439 and @xmath440 , but since the linear transformation is diagonalized , @xmath440 will only appear in @xmath441 . at order three , we have @xmath442 , @xmath443 and not @xmath444 . it is easy to see that at order @xmath233 , one has @xmath445 terms , where @xmath446 is the number of partitions of @xmath233 . it has been known from the work of hardy and ramanujan that @xmath447 it seems thus difficult to get very high order in this expansion . in order to fix the ideas , there are 41 terms at order 11 , 489 at order 20 and 13,848,649 at order 80 . as explained in section [ sec : hmseries ] , the expansion up to order 11 has a sufficient accuracy to take over at @xmath435 . it also provides optimal ( given our use of double precision ) results for @xmath448 . as @xmath0 increases beyond 60 , one can see the effect of each order disappear one after the other . finally , the linear behavior becomes optimal near @xmath449 . this concludes our chronological discussion . the main point made was that there exists a region in the crossover where both expansions are valid . we now proceed to justify empirically the claims made above . the behavior of @xmath450 obtained by the procedure described above , has been studied following the same methods as for the one - variable model . in order to provide a comparison , we have calculated the same quantities for the one - variable model with @xmath451 . in the following , we call this model the `` simplified model '' ( sm ) . for this special value , the critical exponents @xmath131 of the two models coincide with five significant digits . we have good evidence that both series have a radius of convergence 1 as indicated by the extrapolated ratio defined in eq . ( [ eq : rhat ] ) reaching 1 at an expected rate ( fig . [ fig : rhdone ] ) . 0.2 in similarly , their extrapolated slopes seem to converge to the same value @xmath452 as shown in fig . [ fig : shdone ] . 0.2 in in conclusion , the function @xmath215 for the sm is a reasonably good model to guess the asymptotic behavior of @xmath450 . the situation is different for the inverse function @xmath453 . a first look at the differences is given in fig . [ fig : logco ] . 0.2 in the quantity plotted in this figure will be used to discriminate between a finite and an infinite radius of convergence . if @xmath454 as for a radius of convergence @xmath455 , then we have @xmath456 for some constant @xmath457 . on the other hand , if @xmath458 as for an infinite radius of convergence , then we have @xmath459 . in the following , we will compare fits of the form @xmath460 and @xmath461 . we first consider the case of sm , where according to the our study in section [ sec : largetm ] , we should have an infinite radius of convergence . this possibility is highly favored as shown in fig . [ fig : toyfit ] . one sees clearly that the solid line is a much better fit . the chi - square for the solid line fit is 200 times smaller . in addition @xmath462 as expected . in conclusion , this analysis comfirms the ratio analysis done previously and favors strongly the inifite radius of convergence possibility . the analysis for the hm is more delicate . one observes periodic `` dips '' in fig . [ fig : logco ] which make the ratio analysis almost impossible . we have thus only considered , the `` upper envelope '' by removing the dips from fig . [ fig : logco ] . the fit represents an upper bound rather than the actual coefficients . the fits of the upper envelope are shown in fig . [ fig : hmfit ] the possibility of a finite radius of convergence is slightly favored , the chi - square being 0.4 of the one for the other possibility . also , the second fit does not have the @xmath462 property . from @xmath463 , we estimate that the radius of convergence is about 3.7 . as explained in section [ sec : impl ] , one can calculate @xmath438 using an expansion in @xmath7 . as in section [ sec : largetm ] , we will use an empirical series @xmath37 , calculate the corresponding @xmath384 an plug them in the scaling variables . this empirical series was calculated with an initial ising measure and @xmath464 ( see ref . again we define a quantity @xmath297 as in eq . ( [ eq : bdn ] ) which is very small when we have good scaling and increases when the approximation breaks down . the results are shown in fig . [ fig : app ] for successive orders in @xmath7 . 0.2 in 0.2 in the solid line on the right is the linear approximation . it becomes optimal near @xmath449 . the next line ( dashes ) is the second order in @xmath7 expansion . it becomes optimal near @xmath465 . each next order gets closer and closer to be optimal near @xmath466 . the last curve on the left is the order 9 approximation . it is hard to resolve the next two approximations on this graph . the asymptotic value is stabilized with 16 digits and one may wonder why we get only scaling with 14 or 15 digits in fig . [ fig : app ] . the reason is that we use empirical data and that numerical errors can add coherently as explained in ref . this can be seen directly by considering the difference between two successive values of @xmath297 as shown in fig . [ fig : numerr ] . one sees that the numerical errors at each step tend to be negative more often than positive , and consequently there is a small `` drift '' which affects the last digits . we can now look at the @xmath297 defined as in eq . ( [ eq : bdn ] ) for @xmath402 and @xmath403 together in fig . [ fig : hmoverlap ] . if we use an expansion of order 5 in @xmath7 for @xmath202 and of order 10 in @xmath427 for @xmath402 , we can get scaling within a few percent for both variables at @xmath433 . we can go below 1 part in 1000 , with an expansion of order 11 in @xmath7 and order 80 in @xmath427 . at this point , the main problem is that the effects of the subleading correction makes the scaling properties _ worse _ when @xmath467 and @xmath0 decreases . one can improve the scaling properties by taking the effects of @xmath436 into account . a detailed study shows that one can estimate the subleading effects between @xmath468 and @xmath299 . one finds that @xmath469 it is thus possible to get a function scaling better by subtracting these correction . this improve the scaling properties by almost one order of magnitude near @xmath433 and by almost two order of magnitude near @xmath299 . it is likely that our approximation of having a single scaling variable can be corrected order by order in @xmath436 , @xmath470 etc ... . we have shown in two examples that the susceptibility can be expressed in terms of the scaling variables corresponding to the two fixed points governing the ht flows . we have given convincing evidence that the expansions of these variables have overlapping domains of convergence . several interesting questions remain to be discussed . we have not discussed in detail the initial approach of the unstable fixed point . we have just shown that it can in principle be incorporated by calculating the subleading corrections in @xmath436 , @xmath470 etc ... this study depends on the details of the initial measure . a particularly interesting set of initial measures are the landau - ginzburg models in the vicinity of the gaussian fixed point . we can use the usual perturbation theory in the quartic ( or higher orders ) coupling constant to construct the scaling variables associated with the gaussian fixed point following the procedure described above , and interpolate between the gaussian fixed point and the unstable fixed point . the question of small denominators and resonances have been circumvented by using only quantities which are free of this problem ( namely @xmath402 and @xmath403 ) in our calculation of the susceptibility . the effects of the small denominators of the other scaling variables is a complex topic presently under study @xcite . we emphasize that the complete knowledge of the scaling functions and their inverse provides analytical expressions for all the thermodynamic quantities at any volume ( see eq . ( [ eq : dsubn ] ) in the simplified model ) . in practical calculations , one is naturally led to combine different expansions and it is thus important to use sets of interactions which are compatible , as done in ref . @xcite for gauge theories . this research was supported in part by the department of energy under contract no . fg02 - 91er40664 . y. m. thanks the aspen center for physics for its hospitality in summer 1999 , where discussions with p. de forcrand motivated the completion of this work and in summer 2000 while this manuscript was written . p. collet and j .- p . eckmann , _ a renormalization group analysis of the hierarchical model in statistical mechanics _ , edited by j. ehlers et al . , lecture notes in physics * 74 * , ( springer - verlag , berlin , 1978 ) .

superconductivity in a novel class of layered materials , ti - based pnictide oxides , was recently discovered . these compounds have attracted interest since they combine features of copper oxide and iron pnictide superconductors . here the transition metal ( titanium ) forms two - dimensional tio layers ( anti structure to the cuo planes ) , capped by pnictogen ions ( similar to feas layers ) . the pnictide oxide compounds show a spin or charge density wave phase which coexists with superconductivity in some members of the family . unlike the cuprates , but similar to iron pnictides , the parent compounds of pnictide oxides are metals with specific nesting properties of the fermi surface which leads to the density wave instability . the nature of the superconductivity , coexisting with the density wave order , and the possible competition or mutual interaction between both states is one of the central questions of recent studies . this short review summarizes the current knowledge from an experimental as well as theoretical point of view and discusses some of the open questions and possible future developments .

1406.0442

superconductivity in transition metal oxides ( copper oxides or cuprates ) , pnictides ( iron arsenides ) , or chalcogenides ( fese , kfe@xmath0se@xmath0 ) have attracted increasing attention recently , not only because of the extraordinary high transition temperatures achieved in some systems , but also because of the proximity of those systems to long - range magnetic orders in form of antiferromagnetism ( cuprates)@xcite or spin density waves ( pnictides).@xcite as a consequence , `` unconventional '' pairing symmetries ( d - wave in cuprates , s@xmath1 in pnictides ) mediated by spin fluctuations have been proposed to explain the superconducting properties of these systems.@xcite the magnetic fluctuations play a significant role@xcite and evidence for a magnetic quantum critical point , hidden in the superconducting dome , was found in several pnictides.@xcite structurally , cuprates and pnictides have in common the layered structure with active superconducting layers ( cuo@xmath0 in cuprates and feas in pnictides ) separated by charge reservoir blocks . superconductivity commonly arises through charge doping from a nonsuperconducting parent compound which , in most cuprates and pnictides , exhibits magnetic order in the ground state . charge doping can be achieved by either varying the oxygen content ( in cuprates the oxygen content can be controlled by synthesis and annealing , in pnictides oxygen can be replaced by fluorine or hydrogen ) or by replacing cations with ions of different valency ( for example , sr for la in la@xmath0cuo@xmath2 or k for ba in bafe@xmath0as@xmath0 ) . however , there are exemptions from the rule . lifeas was found to be a stoichiometric self - doped superconductor which does not need additional doping.@xcite isovalent doping of phosphorous for arsenic@xcite as well as hydrostatic pressure@xcite have also been shown to induce superconductivity in the parent compounds of some pnictides . in view of the exciting superconducting properties of layered cuprates and pnictides as well as the wealth of interesting physical phenomena that originate from the strong correlations between magnetism and superconductivity , various attempts have been made to search for other layered transition metal oxides / pnictides where superconductivity could be established near magnetic or other instabilities . one of the compound families with interesting magnetic and electronic properties are the ti - based layered pnictide oxide systems of the general formula @xmath3ti@xmath4o , with @xmath3=na@xmath0 , ba , ( srf)@xmath0 , etc . and @xmath5=as , sb , bi . the first synthesis of na@xmath0ti@xmath0as@xmath0o and na@xmath0ti@xmath0sb@xmath0o by adam and schuster@xcite in 1990 revealed an interesting anti - k@xmath0nif@xmath2 structure , the existence of magnetic interactions in the ti@xmath0o layers resulting in a magnetic transition , but no indication of superconductivity was found.@xcite these materials are members of a broader pnictide oxide class of compounds the structure and properties of which have been discusses about two decades ago.@xcite the search for superconductivity in pnictide oxides was not successful until very recently where superconductivity was discovered in bati@xmath0sb@xmath0o and ba@xmath6na@xmath7ti@xmath0sb@xmath0o.@xcite the existence of a superconducting state in pnictide oxides at low temperatures is of particular interest since superconductivity is preceded by a density wave transition at about 50 k and the possible coexistence and mutual interaction of both orders has inspired more theoretical and experimental work . in the following sections we present a brief review of the state of the art of synthesis , structure , physical properties , and superconductivity research of ti - based pnictide oxide compounds . the synthesis of na@xmath0ti@xmath0@xmath8o ( @xmath5=as , sb ) was first described by adam and schuster.@xcite the solid state synthesis was achieved by mixing the precursor compounds na@xmath0o and tias ( tisb ) at a ratio of 1:2 and annealing in argon atmosphere for several hours at 800@xmath9c to 900@xmath9c . the resulting synthesis products , na@xmath0ti@xmath0as@xmath0o and na@xmath0ti@xmath0sb@xmath0o , crystallize in the tetragonal structure , space group i4/mmm ( no . the material is unstable when exposed to air or moisture.@xcite similar routes to synthesize na@xmath0ti@xmath0sb@xmath0o have been chosen by reacting stoichiometric amounts of na@xmath0o , ti , and sb powder in sealed ta tubes at a temperature of 1000@xmath9c for three days.@xcite liu et al . used carefully prepared na@xmath5 powder mixed with powders of tio@xmath0 and ti in the stoichiometric ratio to synthesize na@xmath0ti@xmath4o through sintering at 800@xmath9c in argon atmosphere for 50 hours.@xcite in order to avoid contamination , all preparations had to be carried out in a glove box filled with high purity argon gas . single crystals of na@xmath0ti@xmath0sb@xmath0o can be grown by a nasb flux method , as described in detail by ozawa and kauzlarich.@xcite the polycrystalline sample of na@xmath0ti@xmath0sb@xmath0o , prepared by solid state reaction , and the nasb flux were placed in a ta tube enclosed in a fused silica ampoule . crystal growth was achieved in a vertical furnace . the best crystals were obtained at a reaction temperature of 1100@xmath9c , slow cooling at a rate of 5 to 10@xmath9c / h , and quenching from 550@xmath9c . this procedure yielded plate - like single crystals of several millimeters in size . very recently , the flux method was also employed by shi et al . to grow similarly sized single crystals of na@xmath0ti@xmath0as@xmath0o and na@xmath0ti@xmath0sb@xmath0o from naas and nasb flux , respectively.@xcite the structure type of na@xmath0ti@xmath4o is that of eu@xmath2as@xmath0o ( anti - k@xmath0nif@xmath2 ) with na and ti occupying the eu ( 4e ) and eu ( 4c ) sites , respectively . the tetragonal lattice of na@xmath0ti@xmath4o ( @xmath10=4.070 , @xmath11=15.288 @xmath12 for @xmath5=as ) is shown in fig . the lattice parameters for @xmath5=sb are slightly larger , @xmath10=4.144 @xmath12 and @xmath11=16.561 .@xcite the tetragonal structure of na@xmath0ti@xmath4o consists of different layers stacked along the @xmath11-axis . titanium and oxygen ions form a square lattice with oxygen occupying the nodes and titanium ions located half way between two oxygens . the @xmath5 ions are located in the center of the squares above and below the ti@xmath0o lattice . two @xmath5 and four ti form octahedra sharing corners only , as shown in the center of fig . the na ions fill into the voids above and below the oxygen ions , forming a double layer which separates two neighboring @xmath5-ti@xmath0o-@xmath5 slabs . there are two variable parameters within the unit cell , the distance of the @xmath5 ions above and below the ti@xmath0o plane and the displacement of the na ions from the center plane between two adjacent @xmath5-ti@xmath0o-@xmath5 slabs . it is worth noting that the ti@xmath0o square lattice is similar to the cuo@xmath0 lattice of the high - temperature superconducting cuprates , but the lattice positions of the transition metal and oxygen are interchanged ( anti - k@xmath0nif@xmath2 structure ) . both square lattices are compared in fig . the similarity of the planar structures has raised the question if superconductivity could be induced into the ti@xmath0o plane through proper doping.@xcite the successful replacement of the ( na@xmath13)@xmath0 double layer in na@xmath0ti@xmath0as@xmath0o by a single layer of divalent ba@xmath14 was first reported by wang et al.@xcite the synthesis resulted in a new layered pnictide oxide compound : bati@xmath0as@xmath0o . the solid state synthesis was achieved by mixing the precursors bao , ti , and as in a stoichiometric ratio and annealing the mixture , wrapped in tantalum foil and sealed in quartz tubes , at 850@xmath9c for 40 hours . a polycrystalline sample of composition bati@xmath0as@xmath0o was obtained . since ba@xmath14 substituting for two na@xmath13 does not significantly change the charge state of the as - ti@xmath0o - as block , a compound with similar physical properties was obtained . the sb - analogue , bati@xmath0sb@xmath0o , was synthesized by conventional solid state reaction , mixing the precursors bao , ti , and sb at a ratio of 1:2:2 . two similar approaches have been reported : ( i ) yajima et al.@xcite used tantalum foil to wrap the precursor material and sealed it in a quartz tube . the reaction was started by heating to 1000@xmath9c for 40 hours , followed by cooling to room temperature at a rate of 25@xmath9c / h . ( ii ) doan et al.@xcite sealed the precursors in nb tubes under argon gas and enclosed the tubes in sealed quartz ampoules . the samples were annealed for 3 days after heating ( 2@xmath9c / min ) to 900@xmath9c . slow cooling ( 1@xmath9c / min ) to 200@xmath9c and regrinding and sintering at 900@xmath9c for another three days did ensure the homogeneity of the synthesis product . the hole doped ba@xmath6na@xmath7ti@xmath0sb@xmath0o was successfully synthesized following a similar procedure using bao , bao@xmath0 , na@xmath0o , na@xmath0o@xmath0 , ti , and sb mixed in stoichiometric amounts as precursor materials.@xcite hole doping was also achieved using k or rb to partially replace ba . pachmayr and johrendt describe the synthesis of ba@xmath6k@xmath7ti@xmath0sb@xmath0o from different precursor materials , depending on the potassium content.@xcite for [email protected] , ba , bao@xmath0 , k , sb , and ti were sealed in niobium tubes within silica ampoules ( in argon ) and slowly heated to 600@xmath9c , held at this temperature for 15 hours , and cooled to room temperature . the reaction product was repeatedly ground , pelletized , and sintered at 900@xmath9c for 3 days , followed by slow cooling to 200@xmath9c to improve the chemical and structural homogeneity . for x@xmath16 0.5 , ba , k , sb , ti , and tio@xmath0 have been used as precursors and the reaction and sintering conditions were slightly modified.@xcite von rohr et al . synthesized ba@xmath6rb@xmath7ti@xmath0sb@xmath0o from a stoichiometric mixture of bao , rb@xmath0o , ti , tio@xmath0 , and sb.@xcite the precursor pellets were sealed in niobium tubes and sintered at 1000@xmath9c for 3 days . regrinding and further sintering after the initial reaction for 24 hours at the same temperature ensured homogeneity of the final synthesis product . polycrystalline samples of bati@xmath0bi@xmath0o as well as solid solutions of bati@xmath0(as / sb)@xmath0o , bati@xmath0(sb / bi)@xmath0o , and bati@xmath0(sb / sn)@xmath0o have also been prepared by solid state reaction very recently.@xcite the bati@xmath4o compounds crystallize in the tetragonal structure with space group p4/mmm ( no . 123 ) . the structure , shown in fig . [ f2 ] , has the identical ti@xmath0o square lattice capped above and below by @xmath5 ions as na@xmath0ti@xmath4o ( fig . [ however , the stacking of the @xmath5-ti@xmath0o-@xmath5 slabs along the @xmath11-axis is different , every second slab is shifted along the [ 1,1,0 ] direction so that two neighboring slabs are perfectly aligned along the @xmath11-direction . the ba@xmath14 ions then fit exactly halfway in between two @xmath5-ti@xmath0o-@xmath5 blocks forming a square lattice on their own . thus the double layer of na@xmath0 in na@xmath0ti@xmath4o is replaced by a plane of ba ions in bati@xmath4o . this results in a reduction of the @xmath11-axis and the unit cell volume by about 50% , as shown in fig . the symmetry leaves only one adjustable parameter in the unit cell , the distance of the @xmath5 ions from the ti@xmath0o plane . the lattice parameters of bati@xmath4o increase from @xmath5=as ( @xmath10=4.0456 , @xmath11=7.2723 )@xcite via @xmath5=sb ( @xmath10=4.1196 , @xmath11=8.0951 )@xcite to @xmath5=bi ( @xmath10=4.12316 , @xmath11=8.3447 )@xcite with the ionic size of the pnictogen , as expected . the lattice parameters of the solid solutions bati@xmath0(as@xmath6sb@xmath7)@xmath0o and bati@xmath0(sb@xmath6bi@xmath7)@xmath0o change linearly with x following vegard s law , indicating a homogeneous mixing of ions on the pnictogen site.@xcite the physical properties of pnictide oxide compounds are mainly determined by the charges and spins of the titanium @xmath17-orbitals and the hybridized pnictogen @xmath18-states . thereby , the major contribution to the fermi surface arises from ti @xmath17-states with a sharp peak of the density of states near the fermi energy @xmath19.@xcite the simplified ionic picture with valencies of oxygen ( 2- ) , pnictogen ( 3- ) , ba or na@xmath0 ( 2 + ) leaves the valence state of ti ( 3 + ) as 3@xmath20 . the one 3@xmath17 electron of titanium is therefore largely responsible for the transport and magnetic properties of the compounds . the existence of 3@xmath20 electrons with spin 1/2 at the fermi energy may result in magnetic fluctuations through superexchange interactions involving the oxygen or pnictogen orbitals , as well as direct exchange between two ti ions . the first measurements of magnetic properties of na@xmath0ti@xmath0sb@xmath0o by adam and schuster have indeed revealed an anomaly of the susceptibility near a critical temperature of t@xmath21120 k.@xcite the exact nature of the transition , however , is still under discussion ( see section [ dw ] ) . the existence of a phase transition in na@xmath0ti@xmath0sb@xmath0o was confirmed by axtell et al . in revealing a spin - gap like behavior of the magnetic susceptibility.@xcite the susceptibility shows a sharp drop below 114 k possibly indicating the opening of a spin gap in the magnetic excitation spectrum . at the transition temperature , the metallic resistivity displays a sharp increase with a subsequent decrease toward lower temperatures . the original results of axtell et al . are shown in fig . other physical quantities , e.g. thermoelectric power and heat capacity , show clear anomalies at t@xmath22.@xcite a large magnetoresistance effect of up to 40 % was found at low temperatures and in fields up to 14 tesla which hints for a coupling of the 3@xmath17 electrons to the external field . the possible origin of the instability in na@xmath0ti@xmath0sb@xmath0o was addressed by pickett@xcite based on first - principles calculations of the electronic structure ( see also section [ fpc ] ) . the metallic character of na@xmath0ti@xmath0sb@xmath0o was explained by a strong mixing of the ti @xmath17-states with the @xmath18-states of sb and o and the direct @xmath17-@xmath17 overlap . the fermi surface , dominated by the delocalized ti 3@xmath17 states , exhibits a nesting property which apparently leads to the observed instability at t@xmath22 . the low - temperature state could be a spin - density wave ( sdw ) or a charge - density wave ( cdw ) state , but the calculations did not allow to decide which order may be realized in na@xmath0ti@xmath0sb@xmath0o . a similar drop of the magnetic susceptibility , although not as sharp , was found in na@xmath0ti@xmath0as@xmath0o at about 320 k,@xcite but a clear anomaly in the resistivity could not be resolved.@xcite the overall transport characteristics of na@xmath0ti@xmath0as@xmath0o shows a negative temperature coefficient , in contrast to the metallic behavior of na@xmath0ti@xmath0sb@xmath0o . the significantly higher t@xmath22 of na@xmath0ti@xmath0as@xmath0o shows that the pnictogen ion has a strong influence on the stability range of the density wave phase . bati@xmath0as@xmath0o , first synthesized by wang et al.@xcite in 2010 , exhibits the density wave transition at t@xmath22=200 k , significantly lower than na@xmath0ti@xmath0as@xmath0o . a clear drop of the magnetic susceptibility and a sharp increase of the resistivity , followed by a metallic - like decrease to lower temperature , as well as a pronounced peak of the heat capacity signal the onset of the density wave order . resistivity and magnetic susceptibility data from ref . [ 26 ] are shown in fig . attempts to induce superconductivity by charge doping into the ti@xmath0o plane through substitution of barium by potassium or sodium as well as intercalation of lithium have not been successful . however , the density wave transition was shifted to lower temperature in the intercalated li@xmath7bati@xmath0as@xmath0o . the question whether or not superconductivity can be induced in the ti - based layered pnictide oxide compounds could be related to the stability of the density wave phase . if the spin or charge ordered states compete with superconductivity , the suppression of the sdw / cdw phase should be in favor of a superconducting state . comparing t@xmath22 of na@xmath0ti@xmath4o for @xmath5=as and @xmath5=sb , it becomes obvious that the replacement of as by sb reduces the critical temperature significantly , namely from 320 k ( as ) to 114 k ( sb ) . on the other hand , the @xmath23 of bati@xmath0as@xmath0o ( 200 k ) is also much smaller than that of na@xmath0ti@xmath0as@xmath0o ( 320 k ) . the replacement of as by sb or bi in bati@xmath0as@xmath0o could further reduce or completely suppress the density wave state . the physical properties of bati@xmath0sb@xmath0o , first synthesized in 2012,@xcite indeed show the decrease of t@xmath22 to 54 k. resistivity and magnetic susceptibility data for bati@xmath0sb@xmath0o are shown in fig . the resistivity exhibits a hump - like feature with a sharp increase at t@xmath22 and a large magnetoresistance effect at lower temperatures , as shown in the upper inset of fig . [ f5]a , similar to the results for bati@xmath0as@xmath0o ( fig . the magnitude of the magnetoresistance at 50 koe exceeds the one observed in the as analogue significantly.@xcite the percentage increase , @xmath24 , at 2 k and 50 koe reaches 14 % . the density wave transition in bati@xmath0sb@xmath0o is also revealed by a sharp drop of the magnetic susceptibility ( fig . [ f5]b ) . it is obvious that the critical temperature of the density wave phase is significantly lower in bati@xmath4o and it is further reduced by replacing arsenic with antimony . it should be noted that this phase is completely absent at all temperatures in bati@xmath0bi@xmath0o.@xcite a systematic decrease of t@xmath22 in bati@xmath0(as@xmath6sb@xmath7)@xmath0o from 200 k to 54 k and in bati@xmath0(sb@xmath6bi@xmath7)@xmath0o from 54 k to zero temperature prove the suppression of the dw phase with increasing ( average ) size of the pnictogen ion.@xcite the phase diagrams for bati@xmath0(as@xmath6sb@xmath7)@xmath0o and bati@xmath0(sb@xmath6bi@xmath7)@xmath0o are shown in fig . [ f6]a and fig . [ f6]b , respectively . the extrapolation of t@xmath22 to zero temperature yields a critical value of the bi content between x=0.15 and x=0.2 . superconductivity in the pnictide oxides was discovered simultaneously in bati@xmath0sb@xmath0o with a critical temperature of t@xmath25=1.2 k@xcite and in the sodium substituted ba@xmath6na@xmath7ti@xmath0sb@xmath0o with t@xmath25 up to 5.5 k.@xcite the bulk nature of the superconducting state in bati@xmath0sb@xmath0o is proven by a significant magnetic shielding signal and a pronounced peak of the heat capacity , as shown in figs . [ f7]a and [ f7]b , respectively . the drop of the resistivity to zero confirms the existence of a superconducting phase below 1.2 k. replacing ba by na removes electrons from the active layer , i.e. na substitution results in hole doping of the ti@xmath0o plane . the synthesis of ba@xmath6na@xmath7ti@xmath0sb@xmath0o was successful for [email protected] . x - ray spectra for x=0 , 0.1 , 0.3 are shown in fig . the systematic shift of the [ 200 ] and [ 004 ] peak positions , shown in the inset , indicate that the na ions dope into the ba positions . the solubility limit for na in the p4/mmm structure was found at about 35% . magnetic susceptibility data in fig . [ f8]b prove the existence of superconductivity in ba@xmath6na@xmath7ti@xmath0sb@xmath0o with t@xmath25 systematically increasing with the na content up to x=0.2 . the large shielding signal and a significant meissner effect indicate the bulk nature of the superconducting phase . similar susceptibility data for ba@xmath6na@xmath7ti@xmath0sb@xmath0o have been reported recently and confirm the results of fig . [ f8]b with regard to the magnitude of magnetic shielding as well as the critical temperatures.@xcite the increase of t@xmath25 with the sodium content shows that the stability of the superconducting state is controlled by the charge carrier density , similar to copper and iron based high - temperature superconductors . more interestingly , the characteristic signature of the dw transition , a hump - like anomaly in the resistivity curves , is still observed for sodium contents up to 25% , as shown in fig . the sizable peak of the heat capacity and it s characteristic dependence on magnetic fields , shown in fig . [ f9]b for x=0.15 , proves the bulk nature of superconductivity in the sodium doped compound . the superconducting phase coexists with the density wave phase at low temperatures . the critical temperature t@xmath22 , as determined from the dip in @xmath27 ( right inset in fig . [ f9]a ) , seems to level off with increasing x at about 30 k. this is in contrast to the spin density wave phase of the iron pnitide superconductors where the sdw transition is sharply suppressed upon charge doping and it extrapolates to zero at a critical doping , revealing a ( hidden ) quantum critical point . for na concentrations above 25 % , the dw anomaly could not be resolved any more in resistivity measurements . the phase diagram of ba@xmath6na@xmath7ti@xmath0sb@xmath0o is shown in fig . [ f10]a for @xmath28 values up to the na solubility limit . hole doping by alkali metal substitution for barium was also achieved using potassium and rubidium . the increase of t@xmath25 and the decrease of t@xmath22 , reported in ba@xmath6k@xmath7ti@xmath0sb@xmath0o,@xcite are shown in the phase diagram of fig . the maximum of [email protected] k was obtained at x@xmath30=0.12 in ba@xmath6k@xmath7ti@xmath0sb@xmath0o with a quickly decreasing superconducting shielding signal at higher doping concentrations . the decrease of t@xmath22 with x@xmath30 with a concave curvature and an apparent saturation above x@xmath30=0.1 mimics the phase diagram of ba@xmath6na@xmath7ti@xmath0sb@xmath0o . doping rubidium into the barium site increases t@xmath25 of ba@xmath6rb@xmath7ti@xmath0sb@xmath0o up to 5.4 k at optimal doping of x=0.2 , similar to ba@xmath6na@xmath7ti@xmath0sb@xmath0o.@xcite the critical temperature of the dw phase , however , appeared to decrease faster and t@xmath22 was extrapolated to zero at a critical doping of x=0.12 . this behavior seems to be different from other hole - doped bati@xmath0sb@xmath0o compounds , however , one should take into account that the definition of t@xmath22 from resistivity data for larger x values becomes less certain due to the disappearance of the characteristic hump - like anomaly . whereas alkali metal ions like na , k , rb replace the ba ions outside the sb - ti@xmath0o - sb layer with a minimal structural impact on the active superconducting block , hole doping can also be achieved by substituting tin for antimony in bati@xmath0(sb@xmath6sn@xmath7)@xmath0o.@xcite the doping - induced increase of t@xmath25 , however , was significantly smaller with a maximum of 2.5 k at x=0.3 . the phase diagram and the decrease of t@xmath22 with the concave curvature is very similar to the cases where ba was replaced by na and k. the smaller t@xmath25 in bati@xmath0(sb@xmath6sn@xmath7)@xmath0o may be due to disorder effects introduced into the active sb - ti@xmath0o - sb block . the partially covalent bond between the ti and sb ions are perturbed by sn substitution which gives rise to chemical disorder affecting the hybridization of titanium and pnictogen states at the fermi energy . the covalent character of the ti - sb bond was also suggested to reduce the effective , tin doping - induced charge transfer to the ti@xmath0o plane.@xcite the existing gross similarities , but also the subtle differences , in the physical properties and the superconducting / density wave phase diagrams of hole - doped pnictide oxides raise the question whether or not changes of the structural parameters induced by the ionic substitutions are essential , besides the obvious doping of holes into the active layer . the effects of potassium , sodium , and rubidium doping on the lattice constants of ba@xmath6a@xmath7ti@xmath0sb@xmath0o ( a = na , k , rb ) are qualitatively similar with @xmath10 decreasing and @xmath11 increasing with x , however , the overall volume change is different . in ba@xmath6na@xmath7ti@xmath0sb@xmath0o the unit cell volume clearly decreases whereas ba@xmath6k@xmath7ti@xmath0sb@xmath0o shows a slight but continuous increase with x.@xcite the volume of ba@xmath6rb@xmath7ti@xmath0sb@xmath0o also decreases , but the rate of decrease is much smaller than in the na - doped compound.@xcite the effect of tin doping is different , both lattice parameters @xmath10 and @xmath11 decrease with x , resulting in a volume decrease comparable with that of ba@xmath6na@xmath7ti@xmath0sb@xmath0o.@xcite despite the different responses of the unit cell parameters to doping , the stability of the superconducting and dw phases and the resulting phase diagrams are very similar , as shown in fig . this indicates that the charge transfer to or from the active layer is the most relevant parameter determining the properties of hole - doped bati@xmath0sb@xmath0o . isovalent doping of bati@xmath0sb@xmath0o with bi was reported to increase the superconducting critical temperature of bati@xmath0(sb@xmath6bi@xmath7)@xmath0o up to 3.7 k at the optimal x=0.17 . at higher x values ( up to x=0.4 ) , t@xmath25 decreased again and the temperature coefficient of the resistivity changed sign , indicating a metal - nonmetal transition near x=0.3.@xcite the dw phase was further suppressed with t@xmath22 decreasing to about 40 k at x=0.13 and the characteristic anomaly in resistivity and magnetization could not be discerned any more at x=0.17 . the phase diagram derived from resistivity and magnetization measurements is shown in fig . the decrease of t@xmath22 follows the trend with increasing ionic radius of the pnictogen , starting from bati@xmath0as@xmath0o ( t@xmath22=200 k ) to bati@xmath0sb@xmath0o ( t@xmath22=54 k ) and finally to bati@xmath0(sb@xmath31bi@xmath32)@xmath0o ( t@xmath22=40 k ) . it is important to consider any possible impurity phase which might be superconducting in bi containing compounds . it is known , for example , that babi@xmath33 is superconducting at 5.7 k.@xcite pure bismuth , not superconducting in the crystalline state , can become superconducting in its amorphous form , with t@xmath25 above 6 k.@xcite in bati@xmath0(sb@xmath6bi@xmath7)@xmath0o , the onset of a resistance drop as well as a small diamagnetic signal at higher temperatures ( near 6 k ) was indeed observed and attributed to bi impurities , also resolved in x - ray spectra.@xcite yajima et al . have synthesized the complete series of solid solutions of bati@xmath0(sb@xmath6as@xmath7)@xmath0o and bati@xmath0(sb@xmath6bi@xmath7)@xmath0o for 0@xmath26x@xmath261.@xcite the smooth increase of both lattice parameters with x , following vegard s law over the whole concentration range , shows that the compounds form solid solutions . the complete phase diagram is shown in fig . t@xmath22 decreases smoothly from 200 k to 54 k in bati@xmath0(sb@xmath6as@xmath7)@xmath0o , as expected . the superconducting t@xmath25 , in bati@xmath0sb@xmath0o at 1.2 k , decreases to zero when a small amount of as is substituted for sb . the dome - shaped stability region of the superconducting phase for bi doping up to x=0.4 was also observed , similar to the report by zhai et al.@xcite however , at higher bi doping ( [email protected] ) , a reentrant superconducting phase was found.@xcite the t@xmath25 of this phase reaches a maximum of 4.6 k for x=1 ( bati@xmath0bi@xmath0o ) . based on magnetic ( zero field cooled ) measurements , showing a relatively large shielding signal , the authors suggested that bulk superconductivity in this phase is intrinsic . in view of the unconventional superconducting properties and pairing symmetries of the layered cuprates ( d - wave pairing ) and iron arsenides ( s@xmath35 pairing , two - gap superconductivity ) , the physical nature and the details of the superconducting state and the symmetry of the order parameter in pnictide oxides are of essential interest . since in bati@xmath0sb@xmath0o and it s hole doped variants superconductivity coexists with the density wave order , the possible competition or mutual interaction of both states may result in unconventional pairing mechanisms . two possible scenarios have been discussed theoretically in recent first - principles calculations of bati@xmath0sb@xmath0o and it s hole - doped analogue.@xcite interestingly , the possible symmetry of the superconducting gap depends on the nature of the density wave phase , spin ( sdw ) or charge density wave ( cdw ) . a magnetic instability , caused by fermi surface nesting , would result in the formation of a spin density wave and allow for spin - fluctuation mediated superconductivity and the possibility of a sign - changing s - wave superconducting state.@xcite contrary , when the fermi surface nesting results in a charge density wave , electron - phonon mediated superconductivity with the possibility of multiband character was suggested based on first - principle calculations of the phonon dispersions and the electron - phonon coupling.@xcite details of the first - principles calculations will be discussed in section [ fpc ] . the heat capacity of the superconducting state is a sensitive quantity , reflecting unconventional pairing symmetries or multigap superconductivity . data available so far will be discussed in the following section . because of the low superconducting transition temperatures ( 1.2 k to 5.5 k ) , meaningful heat capacity ( c@xmath36 ) measurements have to be extended to temperatures below 1 k. a systematic study of the superconducting c@xmath36 to @xmath37he temperatures has been conducted for the hole - doped ba@xmath6na@xmath7ti@xmath0sb@xmath0o.@xcite the results for x=0 and x=0.15 are shown in fig . the heat capacity of both compounds can be fit to a single - gap weak - coupling model . for 15% na doping , the data fit nearly perfectly to the bcs model , as shown by the dashed line in fig . [ f11]b . the data for the undoped bati@xmath0sb@xmath0o , however , deviate significantly from the bcs function ( fig . [ f11]a ) . the best fit is provided by the @xmath38-model@xcite which varies the zero - temperature superconducting gap value @xmath39 while maintaining the same relative temperature dependence @xmath40 according to the bcs model . the parameter @xmath38 is defined as @xmath41 and the best fit for bati@xmath0sb@xmath0o is obtained for @xmath421.4 ( note that @xmath38=1.764 in the bcs theory ) . the normal state sommerfeld constant , @xmath43 , is obtained as 10.9 mj/(mol k@xmath44 ) and 13 mj/(mol k@xmath44 ) for x=0 and x=0.15 , respectively , in good agreement with calculated values.@xcite the heat capacity data of bati@xmath0sb@xmath0o , shown in fig . [ f11]a , are consistent with data reported by yajima et al.@xcite , however , in the latter case the large width of the superconducting transition made it difficult to fit c@xmath36(t ) to a more detailed model ( see fig . [ f7]b ) . results of heat capacity measurements of other hole - doped bati@xmath0sb@xmath0o compounds are consistent with the above conclusions . in ba@xmath45rb@xmath46ti@xmath0sb@xmath0o , the heat capacity jump at t@xmath25=5.4 k , @xmath47=1.57 , is compatible with the bcs model and the normal state @xmath43=14 mj/(mol k@xmath44 ) is nearly the same as the value for ba@xmath48na@xmath49ti@xmath0sb@xmath0o discussed above.@xcite a more detailed fit to the bcs or @xmath38-models seems not feasible since the data are limited to t@xmath162 k. heat capacity data of bati@xmath0(sb@xmath50sn@xmath51)@xmath0o have also been found to be consistent with a single - gap bcs model with @xmath47=1.39 at t@xmath25=2.3 k and @xmath43=17.1 mj/(mol k@xmath44).@xcite all heat capacity data published so far are consistent with a weak - coupling single - gap bcs description , although the unconventional , spin - fluctuation mediated pairing mechanism@xcite can not be completely ruled out . furthermore , the data do not show any signature of multiple superconducting gaps , as suggested by subedi.@xcite further evidence for conventional weak - coupling superconductivity in pnictide oxides was derived from muon spin rotation measurements ( london penetration depth ) of ba@xmath6na@xmath7ti@xmath0sb@xmath0o@xcite and bati@xmath0(as@xmath6sb@xmath7)@xmath0o,@xcite with no clear indication of multiple superconducting gaps . support for an ordinary s - wave superconducting state in bati@xmath0sb@xmath0o was also derived from @xmath52sb nuclear magnetic and quadrupole resonance experiments revealing the existence of a coherence peak at t@xmath25 in the 1/t@xmath53 vs. temperature data.@xcite the upper critical field , h@xmath54 , can be determined from field - dependent resistivity measurements . an example is shown in fig . [ f12 ] for the system ba@xmath6na@xmath7ti@xmath0sb@xmath0o . the extrapolation to zero temperature ( dashed lines in fig . [ f12]c ) yields h@xmath54 values between 0.8 koe ( x=0 ) and 24 koe ( x=0.25 ) . these numbers can be used to estimate the ( average ) coherence length @xmath55 which decreases from 640 for x=0 to 120 for x=0.25 . these data are in excellent agreement with other hole - doped bati@xmath0sb@xmath0o compounds.@xcite the lower critical field h@xmath56 , which is related to the london penetration depth @xmath57 , was measured for the optimally doped ba@xmath6rb@xmath7ti@xmath0sb@xmath0o.@xcite with the value of h@xmath58 38 oe and the upper critical field h@xmath54=23 koe , the ginzburg - landau parameter can be estimated as @xmath5935 , revealing that pnictide oxides are strongly type - ii superconductors . accordingly , the magnetic penetration depth for this compound is @xmath604200 .@xcite the effect of hydrostatic pressure on the superconducting state was studied extensively for conventional as well as unconventional superconductors.@xcite the control of fundamental parameters , like the phonon energy , electron - phonon coupling strength , charge transfer and chemical bonds , etc . by external pressure result in complex responses of the superconducting state to the lattice compression . on the other hand , a systematic study of the stability of the superconducting state with respect to pressure can also illuminate the microscopic mechanisms and fundamental interactions leading to superconductivity . in superconducting pnictide oxides , the effects of pressure on the superconducting t@xmath25 and the density wave order ( t@xmath22 ) are of specific interest and have been investigated very recently.@xcite the hole - doped ba@xmath6na@xmath7ti@xmath0sb@xmath0o was studied at three typical doping states , for x=0 ( undoped ) , x=0.1 ( underdoped ) , and x=0.15 ( optimally doped ) . the density wave phase was suppressed for all doping levels , however , the rate of decrease of t@xmath22 showed a strong dependence on the na content x , with the largest rate at x=0 and only a very minute decrease of t@xmath22 at x=0.15 . the pressure effect on the density wave transition temperature is summarized in fig . [ f13]a . the relative change of the superconducting t@xmath25 with pressure is shown in fig . similar to most cuprate and iron arsenide superconductors , the pressure coefficient of t@xmath25 strongly depends on the doping state . the critical temperature of the undoped compound ( bati@xmath0sb@xmath0o ) increases by 40% with compression to 16 kbar . at 10% sodium doping , t@xmath25 initially increases at low p but then saturates with further increasing pressure . the overall t@xmath25 increase does not exceed about 10% . for the optimally doped sample , however , the opposite trend is observed , the superconducting t@xmath25 decreases linearly with increasing pressure . it should be noted that a similar doping dependence of the pressure coefficient of t@xmath25 was reported in high - pressure studies of cuprate high - temperature superconductors@xcite as well as iron arsenide superconductors.@xcite in cuprate superconductors , the pressure effects have been understood as a charge transfer to the active superconducting layer induced by pressure and the general dome - shaped t@xmath25 vs. doping phase diagram . in iron arsenide superconductors , the proximity of the superconducting state to a spin density wave phase , the possible competition between both phases , and the pressure effect on this phase was considered as an additional circumstance that had to be taken into account . in ba@xmath6na@xmath7ti@xmath0sb@xmath0o the active superconducting layer is the sb - tio@xmath0-sb slab with charge doping provided by the na ions occupying ba sites in between the active layers . a charge transfer induced by pressure to the active layer seems unlikely , but not impossible . however , since the sb @xmath61 states mix with the ti @xmath62 states at the fermi surface ( sb - ti bonds ) , an effect of pressure on the fermi surface resulting in a redistribution of charges as well as a change of the nesting property can not be ruled out . in addition , the suppression of the density wave state by pressure , which is particularly pronounced at low doping levels , can contribute to the rise of t@xmath25 of the undoped or underdoped compounds . the critical temperature of the density wave state is almost not affected by pressure for optimal doping ( fig . [ f13]a ) and other effects such as a reduction of the density of states at the fermi level due to pressure - induced band broadening etc . can dominate and result in the observed t@xmath25 decrease for x=0.15 . the apparent anticorrelation between the pressure changes of t@xmath22 and t@xmath25 lead to the conjecture that both ordered phases actually compete with one another , leading to a t@xmath25 enhancement whenever the density wave phase is suppressed by pressure ( as in the undoped bati@xmath0sb@xmath0o ) . this tendency is similar to the effect of doping on both states , with increasing hole doping t@xmath22 decreases and t@xmath25 increases . a more fundamental and complete understanding of the effects of external pressure on both states , however , needs additional work , also involving superconducting pnictide oxides with isovalent substitutions . the nature of the density wave phase is crucial for the understanding of superconductivity in pnictide oxides and how both states coexist and accommodate each other . the strong nesting of the fermi surface , first discussed by pickett based on first - principles calculations of na@xmath0ti@xmath0sb@xmath0o,@xcite can explain the observed instability ( near t@xmath22=114 k in na@xmath0ti@xmath0sb@xmath0o ) as the formation of a spin or charge density wave with corresponding anomalies of the magnetic susceptibility and the resistivity ( see fig . [ f3 ] ) . the question of whether the phase below t@xmath22 is defined by a spin or a charge density wave has been a matter of discussion since then . early studies revealing the sharp drop of the magnetic susceptibility ( see figs . [ f3]b , [ f4]b , and [ f5]b ) suggested the opening of a spin gap , possibly due to a spin - peierls transition.@xcite the high - temperature property of the susceptibility implies that the main magnetic exchange interactions are ti - sb - ti superexchange , but not ti - o - ti interactions . theoretical calculations confirm the hybridization of ti @xmath17 states with sb @xmath18 states but also emphasize on the importance of direct ti @xmath17-@xmath17 overlap.@xcite neutron scattering experiments , however , have not detected a sizable magnetic moment that could be associated with static magnetic order.@xcite the absence of static magnetic order is further suggested by muon spin relaxation studies of bati@xmath0(as@xmath6sb@xmath7)@xmath0o and ba@xmath6na@xmath7ti@xmath0sb@xmath0o.@xcite it should be noted that the absence of a sizable static magnetically ordered moment does not exclude the possibility of the formation of a spin density wave state if the ordered magnetic moment is very small , below the resolution of the neutron or muon spin relaxation experiments . however , the missing proof of a detectable magnetic moment below t@xmath22 is in favor of the formation of a charge density wave , with strong coupling of the carriers to the lattice . anomalies of both lattice constants near t@xmath22 have in fact been observed in neutron scattering studies of na@xmath0ti@xmath0sb@xmath0o although no lattice dimerization or superstructures have been reported.@xcite the structural distortion mainly increases the ti - ti distance and reduces the direct overlap with a partial opening of a band gap . the ti @xmath20 electrons become slightly more localized which explains the increase of the resistivity just below t@xmath22 . recent optical spectroscopy experiments show the opening of a density wave - like energy gap at t@xmath22 removing most of the free carrier spectral weight with a simultaneous reduction of the scattering rate.@xcite the nature of the density wave order in pnictide oxides ( with @xmath5=as , sb ) is one key to understand the superconductivity in these compounds . first theoretical work by pickett did reveal the nesting property in form of a zone corner ( m - point ) box - like fermi surface.@xcite the local spin - density approximation , applied to na@xmath0ti@xmath0sb@xmath0o , confirmed the nearly ionic picture with na@xmath63 , ti@xmath64 , sb@xmath65 , and o@xmath66 . the valence electrons of ti@xmath64 are in 3@xmath20 state , i.e there is one 3@xmath17 electron per ti ion with total spin 1/2 . note the apparent analogy to the 3@xmath67 state of cu@xmath14 in the copper oxide superconductors with one spin-1/2 hole per cu at the fermi level . the direct @xmath17-@xmath17 orbital overlap and the hybridization with @xmath18 states from antimony and oxygen result in the delocalization of the ti @xmath17 states and the metallic properties of na@xmath0ti@xmath0sb@xmath0o . accordingly , the density of states at the fermi energy is mostly composed of ti @xmath17 states with some contribution from the sb and o @xmath18 orbitals . the simulation of different kinds of ferromagnetic or antiferromagnetic long range order by pickett@xcite have not revealed a stable magnetic ground state with ordering of the magnetic ti 3@xmath20 moments . therefore , the formation of a spin- or charge - density wave which originates from the nesting property of the box - like section of the fermi surface with mostly ti @xmath17 character was suggested . the existence of nested sections of the fermi surface in nati@xmath0sb@xmath0o around the m point with a pseudo one - dimensional character was confirmed by a tight - binding band structure calculation ( extended hckel method).@xcite it was suggested that a charge density wave with an incommensurate modulation and an associated lattice distortion would lead to the anomaly of the resistivity observed at t@xmath22=120 k for this compound ( fig . the structural distortion affecting the @xmath10 and @xmath11 axis lattice parameters and the @xmath68 ratio was indeed observed in powder neutron scattering experiments,@xcite but no superlattice reflections of magnetic or lattice origin could be resolved . for a better understanding of the electronic structure , the character of the occupied ti @xmath17 bands is essential . the nested box fermi surface is comprised of almost exclusively ti @xmath69 states with a significant overlap between the orbitals of neighboring ti ions.@xcite the ti orbitals hybridize with the sb @xmath18 bands closer to the center of the brillouin zone . it should be noted that this is different from the cu @xmath62 orbitals in the copper oxide superconductors which have prominently @xmath70 character and mix with the oxygen @xmath18 bands . the problem of possible magnetic order and its consequences for superconductivity in ba@xmath6na@xmath7ti@xmath0sb@xmath0o was discussed by singh based on first - principles density functional calculations.@xcite similar to na@xmath0ti@xmath0sb@xmath0o , discussed in the previous paragraphs , the nearly ionic picture ( ba@xmath14 , ti@xmath64 , sb@xmath65 , o@xmath66 ) holds very well with ti @xmath17 states hybridized with sb @xmath18 states providing most of the density of states near the fermy energy . the ti @xmath17 orbitals at the fermi surface are a mixture of @xmath71 , @xmath70 , and @xmath69 , with a strong @xmath69 character . the fermi surface obtained is rather complex with electron- as well as hole - like sheets present . most importantly , there exists a 2d square section stretching along the zone edge line ( m - a ) giving rise to the nesting property . another nested region of the fermi surface extends around the x point . the nesting wave vector along the orientation of the ti square lattice is ( 0.24,0.24,0)(2@xmath72/a ) , in good agreement with earlier calculations.@xcite the calculated fermi surface is shown in fig . [ f14 ] for different positions of the fermi energy ( corresponding to different doping levels ) . the nesting property of the box - shaped section around m - a is remarkably stable with respect to a variation of the fermi energy @xmath19 . decreasing @xmath19 ( @xmath19 is set to zero for bati@xmath0sb@xmath0o ) below zero , corresponding to hole doping with na , k , or rb , decreases the size but the square shape is approximately preserved . this may explain the extraordinary stability of the density wave phase upon hole doping ( see fig . [ f10 ] ) , as first observed experimentally in ba@xmath6na@xmath7ti@xmath0sb@xmath0o.@xcite a magnetic instability was found ( due to the fermi surface nesting ) resulting in an ordered ground state with a double stripe spin structure which also lowers the symmetry from tetragonal to orthorhombic.@xcite the ordered magnetic moment , however , is small ( @xmath73 ) , according to the calculations . nevertheless , the coupling of the spin to the charge carriers appears to be significant with possible consequences for the superconducting state . the possibility of spin - fluctuation mediated superconductivity was consequently discussed and the pairing symmetry , consistent with the structure of the fermi surface , was suggested to be a sign changing @xmath74-wave superconducting state . for this `` unconventional '' superconductivity it is important that the spin fluctuation dominate and the transition at t@xmath22 results in the formation of a spin density wave . more recent first - principles calculations of na@xmath0ti@xmath4o ( @xmath5=as , sb ) have also concluded that the ground state is magnetically ordered.@xcite the calculated band structures and fermi surfaces are consistent with earlier calculations.@xcite the magnetic structure of the ground state depends on the pnictogen atom . for @xmath5=as , a checkerboard spin configuration was found to be stable , with four spins on a plaquette of neighboring ti ions aligned with one another , forming a block spin with an antiferromagnetic order between different blocks . for @xmath5=sb the magnetic order is the double stripe structure described by singh for bati@xmath0sb@xmath0o . the fermi surface nesting as the origin of the magnetic orders is confirmed . a slightly different picture of the magnetic ground state was derived from electronic structure calculations of bati@xmath0as@xmath0o.@xcite only two of four ti spins were found to order with antiparallel moments , driven by the fermi surface nesting and the coulomb interactions . no modulation of the charge density was found , favoring the formation of a spin density wave at the ordering temperature t@xmath22 . it appears interesting that first - principles calculations of various pnictide - oxides reveal different ground states with double stripe , block spin , or partial magnetic orders . these magnetic structures are very close in energy , but they are all commensurate with the lattice . the nesting wave vector , however , is most likely incommensurate with the atomic structure and the true magnetic order may well be an incommensurate spin density wave . with the small ordered moment per ti ion , it could be a challenge to detect the possible order in neutron scattering or other experiments . first - principles calculations including the phonon dispersion spectrum and the electron - phonon interactions of bati@xmath0sb@xmath0o have been presented by subedi.@xcite it was concluded that , despite the tendency to a magnetic instability , the phonon dispersions may give rise to a lattice instability near the corners of the brillouin zone which stabilizes a charge density wave phase . the electron - phonon coupling is strong enough to explain the superconducting phase in this compound as conventional phonon - mediated s - wave superconductivity . the calculated electron - phonon coupling constant , @xmath75=0.55 , is in good agreement with the experimentally derived value.@xcite an analysis of the electronic structure and chemical bonding of na@xmath0ti@xmath4o and bati@xmath4o ( @xmath5=as , sb ) has in part confirmed the results from earlier work.@xcite the bonding between the ti@xmath4o and na@xmath0 ( ba ) layers is mainly of ionic character , whereas the bonding within the ti@xmath4o has also covalent character . ti @xmath69 , @xmath70 , and @xmath71 states dominate the fermi surface . interestingly , the charge transfer to the ti@xmath4o layer is smaller than expected from the nominal charges na@xmath63 and ba@xmath14 . it decreases from 1.6 electrons per na@xmath0 in na@xmath0ti@xmath0as@xmath0o to 1.23 electrons per ba in bati@xmath0sb@xmath0o.@xcite this trend in the charge transfer from the cation to the ti@xmath4o layer could be one relevant parameter which controls the superconducting state in pnictide oxides . it is interesting that superconductivity occurs among the four compounds only in bati@xmath0sb@xmath0o ( t@xmath25=1.2 k ) with the smallest electron transfer and t@xmath25 was enhanced by na ( hole ) doping , i.e. by a further reduction of the electron number in the active layer . similar calculations for bati@xmath0bi@xmath0o , a pnictide oxide with no density wave phase but a relatively high superconducting t@xmath25=4.6 k , show an even smaller charge transfer of 1.18 electrons per ba ion,@xcite which is consistent with the general trend mentioned above . the role of the charge density in the active ti@xmath4o layer is not completely understood and needs further studies . within the rigid band model , the decrease of the electron number will change the density of states ( dos ) at the fermi energy , which is expected to increase since the fermi energy is just on the high - energy slope of a peak of the dos . on the other hand , the superconducting state appears to compete with the density wave state and reducing t@xmath22 or suppressing the dw state completely leads to a superconducting phase , observed in bati@xmath0sb@xmath0o , its hole - doped analogues , and bati@xmath0bi@xmath0o . further experimental and theoretical studies are needed to deepen the understanding of the coexistence and mutual interactions of the density wave and superconducting states in the new ti - based pnictide oxide superconductors . superconductivity in a novel transition metal pnictide oxide has attracted attention recently . this class of compounds with the chemical compositions na@xmath0ti@xmath4o and bati@xmath4o ( @xmath5=as , sb , bi ) form layered structures with characteristic ti@xmath4o slabs separated by either a double layer of na@xmath0 or a single layer of ba . here the transition metal is ti with one 3@xmath17 electron in the formal ionic ti@xmath64 state . structurally , the ti@xmath0o form a 2d lattice which is the anti - structure to the cuo@xmath0 planes of the copper oxide high temperature superconductors , i.e. the titanium and oxygen ions occupy the oxygen and copper sites in the cuo@xmath0 structure , respectively . the pnictogen ions are located above and below the ti@xmath0o plane . with the exception of bati@xmath0bi@xmath0o , all other compounds exhibit an instability at higher temperature which is most probably caused by a nesting feature of sections of the fermi surface . the nature of this instability is not completely resolved yet , but there is evidence for the formation of a density wave state , through a modulation of either charge or spin density . the electronic states at the fermi energy are dominated by ti @xmath17 states partially hybridized with @xmath18 states of @xmath5 and o. superconductivity was discovered in bati@xmath0sb@xmath0o at 1.2 k and t@xmath25 did increase through na doping to a maximum of 5.5 k. interestingly , the density wave state was only slightly suppressed by na ( hole ) doping and it coexists with the superconducting state in ba@xmath6na@xmath7ti@xmath0sb@xmath0o up to x=0.35 , the solubility limit of na in the structure of bati@xmath0sb@xmath0o . similar results have been obtained for other hole - doped samples using k and rb replacing ba , or sn replacing sb . the nature of the superconducting pairing has been a matter of discussion . it may be closely related to the nature of the density wave phase , spin or charge density . various experimental techniques , designed to reveal information about the symmetry of the superconducting order parameter , have provided evidence in favor of a conventional bcs - like @xmath74-wave superconductivity . furthermore , no signature of a magnetically ordered state could be detected although first - principles calculations did find stable magnetic ground states in different pnictide oxide compounds . superconductivity and the density wave order most likely compete with one another . evidence for this competition is derived from the fact that a superconducting state is only observed when the critical temperature of the density wave transition is lowered to 54 k ( in bati@xmath0sb@xmath0o ) or completely suppressed ( in bati@xmath0bi@xmath0o ) . furthermore , doping and pressure dependent measurements show an anticorrelation between t@xmath25 and t@xmath22 , t@xmath25 is enhanced when t@xmath22 decreases . on a microscopic level , there are several parameters that may affect the two ordered phases . with the increase of the @xmath5 ionic size , the @xmath10- and @xmath11-axis lattice parameters increase in both , na@xmath0ti@xmath4o and in bati@xmath4o . comparing na@xmath0ti@xmath4o and bati@xmath4o , @xmath10 and @xmath11 are slightly smaller in the ba compounds . an increase of @xmath10 will reduce the direct overlap of the ti @xmath17 orbitals and also affect the hybridization with the @xmath5 @xmath18 states . furthermore , the calculated charge transfer from the ( na@xmath63)@xmath0 and the ba@xmath14 layers is not complete and changes with the cation ( na or ba ) and with @xmath5 . it turns out that the smallest number of electrons is transferred to the ti@xmath4o layer in bati@xmath0bi@xmath0o , the compound with no density wave phase and the highest superconducting t@xmath25 . there are a number of questions that should be addressed in future studies of pnictide oxide superconductors . the experimental verification and the theoretical understanding of the nature of the density wave phase is crucial to understand the competition and mutual interaction of this phase with the superconducting state . the solid solution bati@xmath0(sb@xmath6bi@xmath7)@xmath0o shows superconductivity only for small and large values of x. the reason for the missing superconducting state around x=0.5 has yet to be explored . the effects of electron doping , if achievable , on the density wave and superconducting states is of interest since it could help to understand the role of the electron density in the ti@xmath4o layers . the ti - based pnictide oxides are an interesting class of layered compounds and future work will certainly contribute to a deeper understanding of superconductivity and density wave instabilities and how those states interact with one another . the authors gratefully acknowledge support from the t.l.l . temple foundation , the john j. and rebecca moores endowment , the state of texas through the texas center for superconductivity at the university of houston , the us air force office of scientific research ( fa 9550 - 09 - 1 - 0656 ) , the national science foundation ( che-0616805 ) , and the r. a. welch foundation ( e-1297 ) .

the f - theory vacuum constructed from an elliptic calabi yau threefold with section yields an effective six - dimensional theory . the lie algebra of the gauge sector of this theory and its representation on the space of massless hypermultiplets are shown to be determined by the intersection theory of the homology of the calabi yau threefold . ( similar statements hold for m - theory and the type iia string compactified on the threefold , where there is also a dependence on the expectation values of the ramond ramond fields . ) we describe general rules for computing the hypermultiplet spectrum of any f - theory vacuum , including vacua with non - simply - laced gauge groups . the case of monodromy acting on a curve of singularities is shown to be particularly interesting and leads to some unexpected rules for how-branes are allowed to wrap certain-cycles . we also review the peculiar numerical predictions for the geometry of elliptic calabi yau threefolds with section which arise from anomaly cancellation in six dimensions . 0

hep-th0002012

the f - theory vacuum @xcite constructed from an elliptically fibered threefold @xmath2 with section determines an effective theory with @xmath3 supersymmetry in six dimensions . such supersymmetric theories will have fields in hypermultiplets , vector supermultiplets and tensor supermultiplets . ( see , for example , @xcite for a discussion of such theories . ) for any particular f - theory vacuum , the taxonomy of the supermultiplets may be derived from the geometry of @xmath2 as an elliptic fibration via seemingly straightforward methods in the case of the vector and tensor multiplets @xcite . the classification of the hypermultiplet content has always been a little harder to carry out . many methods have been proposed which allow the hypermultiplets to be determined from the geometry in certain cases @xcite . the purpose of this paper is to outline a systematic approach to the problem of determining the gauge symmetry and hypermultiplet content of a given six - dimensional theory obtained from f - theory . ( note that as far as the moduli space of hypermultiplets in concerned , our methods utilize the associated type iia compactification and thus also apply directly to the compactification of m - theory on @xmath2 giving an @xmath4 theory in five dimensions and to the compactification of the type iia string on @xmath2 to yield an @xmath5 theory in four dimensions , provided that the expectation values of certain ramond ramond fields have been tuned appropriately . ) the methods we employ will not be particularly new but we will see that the process of analyzing the gauge group and matter content can be quite a bit more subtle than had previously been appreciated . in particular , the case of monodromy of the fibration leading to non - simply - laced lie algebras requires some care . a particularly awkward case which has caused some confusion is when a @xmath6 monodromy acts on a curve of @xmath0 singularities , i.e. , a curve of @xmath7 fibers in f - theory language . in this paper we resolve this problem in agreement with an observation by intriligator and rajesh in @xcite concerning anomaly cancellation . in section [ s : lcy ] we will show how many features of a lie algebra structure arise naturally from an elliptically fibered threefold . this will allow us to elucidate the method for determining the gauge algebra . in section [ s : hyper ] we discuss exactly how to analyze the hypermultiplet content in the cases where the associated curves and surfaces within the threefold are smooth . we discuss the cases where these curves and surfaces are singular in section [ s : odd ] . this section includes some unexpected rules we are forced to adopt for @xmath1-brane wrapping . although the results of this section are less rigorous than the preceding section , we are able to give precise results in many instances which can be extended to the general case under the fairly conservative assumption that the relevant physics is determined locally from the geometry of the singularities . finally in section [ s : num ] we emphasize the peculiar numerical predictions which arise from anomaly cancellation in the f - theory compactification on @xmath2 . we begin with a threefold @xmath2 which admits an elliptic fibration @xmath8 , where @xmath9 is a complex surface , and also assume that this elliptic fibration has a section .- field has been turned on in the base @xcite . ] the type iia string compactified on @xmath2 yields an effective four - dimensional theory with @xmath5 supersymmetry ; its strong - coupling limit , known as `` m - theory compactified on @xmath2 , '' yields an effective five - dimensional theory . one more effective spatial dimension is obtained in a limit in which the areas of all components of the elliptic fibers shrink to zero this is the `` f - theory limit . '' see , for example , @xcite for an explanation of this . we point out that most of the following analysis does not really depend upon this elliptic fibration structure and applies to m - theory and type iia compactifications of @xmath2 . we use the f - theory language as an organizational tool to give examples later on . one also has the advantage in f - theory of being able to use anomaly cancellation as a powerful tool in checking the consistency of results concerning spectra of massless particles . in the f - theory context we can freely exchange the notion of , say , an @xmath10 fiber and an @xmath11 singularity . the former is the elliptic fibration description for the latter . recall @xcite that this is because although @xmath10 is really the _ extended _ dynkin diagram of @xmath11 and that one always ignores the components of the fiber which hit the chosen section of the elliptic fibration . thus , in the zero - area fiber limit of f - theory , a shrunken @xmath10 fiber gives the same physics as an @xmath11 singularity one dimension lower . whenever rational curves in @xmath2 are shrunk down to zero size we expect @xmath1-branes of the type iia string wrapped around these curves to contribute massless particles to the spectrum . it is precisely these massless states which are the focus of our interest in this paper . actually we need to be careful with the statement that massless states appear automatically when a brane wraps a vanishing cycle . there is always the subtlety of @xmath12-fields and r - r fields which should be tuned to the right value ( usually denoted `` zero '' by convention ) to really obtain a massless state . as emphasized in @xcite the relevant parameters to worry about in this context are the r - r fields . we may see this as follows . if one considers the type iia string compactified on @xmath2 then deformations of the khler form ( and @xmath12-field ) on @xmath2 are given by vector moduli . suppose we use these khler moduli to shrink down a holomorphic @xmath1-cycle to obtain an enhanced gauge symmetry . once we reach this point of enhanced symmetry we may have a phase transition releasing new hypermultiplet degrees of freedom . thus at the point of phase transition , these new parameters , which include r - r fields , are fixed at some value . reversing this point of view , we may tune parameters in the hypermultiplet moduli space to achieve an enhanced gauge symmetry but these parameters include r - r fields . thus we need to assume always that the r - r parameters have been tuned to the appropriate values required to obtain the enhanced gauge symmetries we discuss below . witten @xcite analyzed how to determine the massless particle content for a given configuration of rational curves . let us assume that a given rational curve lives in a family parametrized by a moduli space @xmath13 . in the simplest case one has an embedding @xmath14 . an isolated rational curve is a trivial example of this where @xmath13 is simply a point . according to witten s calculation , one half - hypermultiplet may be associated to the fact that a @xmath1-brane breaks half of the supersymmetry . this half - hypermultiplet is then tensored with the total cohomology of @xmath13 in an appropriate sense . the result is that if @xmath13 is a point , then we simply obtain a single half - hypermultiplet . if @xmath13 is an algebraic curve of genus @xmath15 then we obtain a single vector multiplet and @xmath15 hypermultiplets . this was also argued by a different method in @xcite . note that for any wrapping we may also wrap with the opposite orientation to double this spectrum . of central interest to us is the fact that compactifying a type iia string theory ( and thus m - theory and f - theory ) on a threefold @xmath2 produces a theory with a yang mills sector . the gauge fields may be viewed as arising from integrating the r - r @xmath16-form of the type iia string over @xmath1-cycles in @xmath2 to produce @xmath17-forms .- form field over @xmath1-cycles . ] these @xmath17-forms play the role of the yang - mills connection . in addition , the @xmath18-cycles in @xmath2 which are dual ( via intersection theory ) to these @xmath1-cycles will play an important role . let @xmath19 denote the @xmath18-form field strength of the r - r @xmath16-form in the type iia string . note that the @xmath1-branes of the type iia theory are _ electrically _ charged under this field that is @xmath20 for a @xmath21-dimensional shape @xmath22 ( such as a six - sphere ) enclosing the seven directions transverse to a fundamental @xmath1-brane . upon compactification we will be wrapping @xmath1-branes around a @xmath1-cycle in @xmath2 to produce a point particle in four - dimensional space - time . to find the charge of this resulting particle we may take @xmath23 , where @xmath24 is a sphere in four dimensional space - time enclosing the particle and @xmath25 is a @xmath18-cycle within @xmath2 . it follows that in the type iia compactification 1 . we have @xmath26 gauge symmetries of the type @xmath27 , each labelled by an element of @xmath28 , in addition to the @xmath29 gauge symmetry coming from the r - r @xmath17-form ( whose charge is measured using @xmath30 , the generator of @xmath31 ) . 2 . if a @xmath1-brane wraps a @xmath1-cycle @xmath32 to produce a particle then the `` electric '' charge of this particle under the @xmath27 symmetry associated to a @xmath18-cycle @xmath25 will be the intersection number @xmath33 . we thus obtain a perturbative @xmath34 gauge symmetry in type iia . in the m - theory compactification , there is no r - r @xmath17-form , and the eight transverse directions to the m - theory @xmath1-brane are enclosed by @xmath35 , so the total `` perturbative '' gauge symmetry is given by @xmath36 . in the f - theory limit , the only @xmath18-cycles which contribute gauge fields are those with intersection number zero with the elliptic fiber ; moreover , @xmath18-cycles which are the inverse images of @xmath1-cycles in the base @xmath9 are associated to tensor multiplets rather than gauge fields . thus , in the f - theory limit , we get a `` perturbative '' gauge symmetry group of @xmath37 . as is now well - known , and as we will discuss , the wrapped @xmath1-branes will elevate this @xmath38 gauge symmetry to a non - abelian lie group ( since certain wrapped branes include vector multiplets in their spectra ) , where @xmath39 , @xmath40 , or @xmath41 for iia , m - theory or f - theory , respectively . from now on we will concern ourselves only with the lie _ algebra _ of the gauge symmetry . it was noted in @xcite that , at least in f - theory , the global structure of the gauge group may be recovered from the mordell - weil group of @xmath2 as an elliptic fibration . of the gauge group is equal to the mordell - weil group ( including both the free and torsion parts ) . ] if this @xmath42 appears as the cartan subalgebra of our gauge algebra then the discussion above implies that we may make the following identifications . let @xmath43 be the ( real ) cartan subalgebra , let @xmath44 be the dual space , and let @xmath45 be the coroot lattice and @xmath46 be the weight lattice so that the cartan subgroup @xmath38 is naturally identified with @xmath47 . for the iia compactification , we take @xmath48 and @xmath49 , and in m - theory , we take @xmath50 and @xmath51 . in f - theory , we begin with the orthogonal complement within @xmath28 of the elliptic fiber @xmath52 , and then we mod out by @xmath53 ( that is , @xmath54^\perp/\pi^{-1}h_2(\sigma)\subset h_4(x)/\pi^{-1}h_2(\sigma)$ ] ) ; we then take @xmath55 to be the dual lattice of @xmath56 . in each case , a @xmath1-brane wrapped around a particular @xmath1-cycle is then naturally associated with an element of the weight lattice and its charges under the cartan subalgebra are given in the standard way . we work this out in detail in several particular cases . consider first the case that @xmath2 contains a `` ruled '' complex surface @xmath57 admitting a fibration @xmath58 , for some @xmath13 , where all fibers are isomorphic to @xmath59 . the fibers will shrink down to zero size in the f - theory limit . the simplest example of this is @xmath60 where @xmath13 is a riemann surface of genus @xmath15 and that @xmath61 is in the fiber direction . that is to say , in our elliptic fibration @xmath8 we have a curve @xmath62 over which the fiber is i@xmath63 . clearly we have massless states appearing for the @xmath1-branes wrapped around @xmath64 . we also have a @xmath65 symmetry associated to @xmath66 . let us consider the normal bundle of a single @xmath64 curve . this normal bundle may be written as @xmath67 where @xmath68 by the adjunction formula and the fact that @xmath2 is a space . since this curve may be translated along the @xmath13 direction one of these line bundles must be trivial . thus the normal bundle is @xmath69 where the @xmath70 describes the normal bundle direction which is also normal to @xmath71 . therefore @xmath72 . this tells us that we have a vector supermultiplet and @xmath15 hypermultiplets from wrapping @xmath1-branes around @xmath73 all with charge @xmath74 with respect to the @xmath65 gauge symmetry associated to this divisor . similarly by wrapping with the opposite orientation we obtain a copy of this except with charge @xmath75 . these vector supermultiplets enhance the @xmath65 symmetry to @xmath76 in the usual way and we have an additional @xmath15 hypermultiplets in the adjoint representation . the key point is to notice in this construction that the condition @xmath77 has played the role of the _ cartan matrix _ of @xmath76 . the next simplest case is where we have a set of curves @xmath78 which may intersect each other and are each isomorphic to @xmath59 and lying in the fiber direction . we assume that @xmath79 embeds algebraically into @xmath2 . distinct @xmath59 fibrations over @xmath13 along appropriate disjoint sections , forming a chain . in the remainder of this paper , we will continue to explain by example and will not explicitly state the most general form of the algebraic surfaces which contract to @xmath13 in the f - theory limit . ] we now have a cartan matrix given purely by the configuration of @xmath78 . applying the above method we obtain the standard f - theory result of a simply - laced enhanced gauge symmetry as listed , for example , in table 4 of @xcite . ( 8865,6298)(511,-6347 ) ( 1201,-1861 ) ( 5 , 2)4500 ( 1201,-1861 ) ( 0,-1)3900 ( 1201,-5761 ) ( 5 , 2)4500 ( 5701,-61 ) ( 0,-1)3900 ( 3631,-511 ) ( 2,-1)4350 ( 1984,-1180 ) ( 2,-1)4350 ( 1201,-2774 ) ( 5 , 2)4500 ( 1201,-3689 ) ( 5 , 2)4500 ( 1201,-4591 ) ( 5 , 2)4500 ( 1201,-5253 ) ( 5 , 2)4500 ( 3619,-1428 ) ( 2,-1)4350 ( 3579,-2326 ) ( 2,-1)4350 ( 3469,-3190 ) ( 2,-1)4350 ( 3290,-4023 ) ( 2,-1)4350 ( 2692,-3462 ) ( 2,-1)4350 ( 2272,-2172 ) ( 2,-1)4350 ( 8401,-1861)(-1,-1)1275 ( 9301,-3961)(-1 , 0)1275 ( 3301,-5761)(-3 , 4)1125 ( 6901,-1261)(-1 , 0)1125 ( 3081,-4165 ) ( 2,-1)4350 ( 601,-5161 ) ( 0 , 1)2955 ( 2401,-1381 ) ( 1973,-1178 ) ( 6332,-3356 ) ( 8476,-1936)(0,0)[lb ] ( 9376,-4036)(0,0)[lb ] ( 6976,-1336)(0,0)[lb ] ( 3376,-5911)(0,0)[lb ] ( 411,-5561)(0,0)[lb ] as noted first in @xcite the real power of this cartan matrix approach is that it gives a clear way of describing _ non_-simply - laced gauge algebras . consider a less trivial example of ruled surfaces as shown in figure [ fig : one ] . in this example the moduli space @xmath80 of the curve @xmath64 is different from the moduli space @xmath81 the curve @xmath82 . think of @xmath80 as the vertical direction in the figure . we obtain ruled surfaces @xmath83 and @xmath84 . we have a two - fold cover @xmath85 branched at one point in figure [ fig : one ] . any other branch points are not shown in the figure . the intersection matrix of this configuration may be written @xmath86 this is the cartan matrix for @xmath87 ( or @xmath88 ) and so our enhanced gauge algebra should be @xmath87 . this phenomenon of obtaining a non - simply - laced symmetry algebra was first noted in @xcite inspired by the construction of @xcite . there it was explained by monodromy acting on the fibers as follows . let @xmath80 be embedded in @xmath9 and let @xmath81 be a two - fold cover of @xmath80 ( branched at various points ) . over a generic point in @xmath80 we see that , ignoring the component meeting the chosen section , the kodaira fiber consists of one line from @xmath71 and two lines from @xmath89 forming the ( dual ) dynkin diagram of @xmath90 . moving along a closed path in the complement of the set of branch points of @xmath85 we will exchange the two lines in @xmath89 . this action on the dynkin diagram is induced by an _ outer automorphism _ of @xmath90 and the invariant subgroup under this outer automorphism can be taken to be @xmath87 . one might therefore suspect that the effective gauge algebra is the monodromy - invariant subalgebra of the simply - laced gauge symmetry generated locally by the vanishing cycles . this was the assertion in @xcite . unfortunately it is an ambiguous statement . let us analyze carefully all possible outer automorphisms of @xmath91 . an element @xmath15 of @xmath91 satisfies @xmath92 . complex conjugation @xmath93 is an example of an outer automorphism . indeed this acts on the dynkin diagram of @xmath91 by reflection about the middle node . clearly the invariant subgroup under this outer automorphism is given by @xmath15 real . but this yields the group @xmath94not what we were expecting ! a general outer automorphism of @xmath91 can be obtained by combining complex conjugation with an arbitrary inner automorphism , yielding @xmath95 , where @xmath96 ( there are no other possibilities since that would imply further symmetries of the dynkin diagram ) . since this outer automorphism acts on the dynkin diagram as the reflection , it is also a viable candidate for the monodromy action on the gauge group . in this general situation , the invariant subalgebra satisfies @xmath97 the case @xmath98 yields @xmath94 as stated before . now if we put @xmath99 where @xmath100 is the @xmath101 identity matrix , we obtain the group @xmath102as desired . in this case , the outer automorphism is an involution , but this is not a requirement in general . we see then that the method of directly working out the cartan matrix from intersection theory is a better way to determine the effective gauge algebra in f - theory than trying to find subalgebras invariant under outer automorphisms . the latter method is ambiguous . one might try to assert that f - theory picks out the `` maximal '' invariant subalgebra under all possible outer automorphisms . indeed , @xmath103 is `` bigger '' than @xmath104 in as much as it has a larger dimension ( although in general @xmath105 ) . however , even this approach is inadequate as will be shown by examples in section [ s : odd ] . note that in the m - theory or type iia compactifications , an ambiguity of the sort we have discovered is actually to be expected . as we have already pointed out , if the ramond ramond fields have non - zero expectation values then some of the non - abelian gauge fields will become massive ; when these are integrated out , the gauge group becomes smaller . this is precisely what happens when the outer automorphism of the covering lie group is varied in the construction above . the gauge algebra which we wish to determine is the one in which these effects have been turned off so that the f - theory limit can be taken . ( a similar phenomenon of variable gauge group depending on the precise value of an outer automorphism has been observed in a closely related context by witten @xcite , and applied in @xcite ) . in the last section we described how to determine the gauge algebra in f - theory ( or m - theory or iia string theory ) by determining the cartan matrix from intersection theory . similar methods will in principle determine the hypermultiplet spectrum completely as we now discuss . first there can be the case of a family of rational curves acquiring extra rational curves at certain points in the family . in the context of elliptic fibrations this can be seen as collisions of curves in @xmath9 over which there are singular fibers . the simplest example is a transverse collision of i@xmath106 and i@xmath107 . the resolution of singularities associated to this collision was explained in @xcite , and applied to the case of an i@xmath106-i@xmath108 collision in the context of string theory in section 8.2 of @xcite . the key point is that there exist rational curves within the collision with normal bundle @xmath109 . one of these curves @xmath110 is the intersection of two ruled @xmath18-cycles , one lying over the curve of @xmath111 fibers , and the other lying over the curve of @xmath112 fibers . the normal bundle of @xmath110 in @xmath2 is naturally the direct sum of the normal bundles of @xmath110 in each of these @xmath18-cycles , and each of these is @xmath113 . thus this curve appears as ( minus ) a fundamental weight . the above rules imply that we have found a curve representing the ( lowest ) weight of the @xmath114 representation of @xmath115 . by adding other ( possibly reducible ) curves in the collision fiber we may indeed build up the full @xmath114 representation . thus _ the transverse collision of a curve of i@xmath106 and i@xmath107 fibers yields a hypermultiplet in the @xmath114 representation of @xmath115 . _ ( this same result had earlier been determined using quite different methods in @xcite . another approach which is closer to ours appeared in @xcite . ) many other `` collisions '' can be explained in similar ways . however , if the extra rational curves at the collision point have normal bundles other than @xmath109 , then witten s calculation does not directly apply . general methods for evaluating the corresponding contribution to the hypermultiplet spectrum are not known . the case of non - simply - laced symmetry algebras raises even more complicated possibilities . some of the hypermultiplet matter can appear in a somewhat `` non - local '' manner as we now explain . suppose we are in a situation analogous to figure [ fig : one ] . let us consider the example of a type @xmath116 fiber ( where we again ignore the component passing through the chosen section ) . let the middle component in the chain have a moduli space given by @xmath80 and the other components have moduli space @xmath81 where @xmath85 is a double cover . that is , we have a @xmath6-monodromy acting on the @xmath116 fiber ( in the only possible way ) . figure [ fig : one ] is the case @xmath117 . according to @xcite we should obtain @xmath118 hypermultiplets for @xmath1-branes wrapping the middle component and @xmath119 hypermultiplets for @xmath1-branes wrapping each of the other components . note that @xmath120 from the double cover . there are additional hypermultiplets arising from wrapping connected unions of these components . in fact , each of the positive roots of the covering algebra @xmath121 is represented by such a connected union , some of which are fixed by the monodromy , and others of which are exchanged in pairs under the monodromy . the ones which are fixed under monodromy have @xmath80 as moduli space , while those which are exchanged in pairs have @xmath81 as their moduli space . when we organize these weights in terms of representations of @xmath103 , we find that the invariant subspace describes the adjoint of @xmath103 while the anti - invariant subspace describes the remaining weights in the adjoint of @xmath121 . on the other hand , each invariant positive root contributes to the invariant subspace , while the roots exchanged in pairs contribute to both the invariant and anti - invariant subspaces . we conclude that the adjoint of @xmath103 occurs @xmath118 times while the weights in the anti - invariant subspace each occur @xmath122 times . ( 6024,6924)(2389,-6673 ) ( 3001,-661 ) ( 5401,-661 ) ( 7801,-661 ) ( 3001,-3061 ) ( 7801,-3061 ) ( 3001,-5461 ) ( 5401,-5461 ) ( 7801,-5461 ) ( 5401,-3061 ) ( 5401,-661 ) ( 3001,-3061 ) ( 5401,-3061 ) ( 7801,-3061 ) ( 5401,-5461 ) ( 5401,239 ) ( 0,-1)6900 ( 7801,239 ) ( 0,-1)6900 ( 3001,239 ) ( 0,-1)6900 ( 2401,-3061 ) ( 1 , 0)6000 ( 2401,-5461 ) ( 1 , 0)6000 ( 2401,-661 ) ( 1 , 0)6000 ( 5401,-3061 ) ( 1 , 0)2400 ( 5401,-3061)(-1 , 1)2400 ( 6776,-2986)(0,0)[lb ] ( 4501,-2086)(0,0)[lb ] ( 5551,-511)(0,0)[lb ] ( 7951,-511)(0,0)[lb ] we demonstrate which weights appear in the example of @xmath117 in figure [ fig : asym ] . we show the weights of the adjoint representation . the dots represent weights associated to @xmath80 , i.e. , from @xmath64 . the circles represent weights associated to @xmath81 , i.e. , from @xmath82 . it is important to note that _ reducible _ curves may also be wrapped by @xmath1-branes . that is , two rational curves intersecting transversely at a point may be viewed together as a nodal rational curve . these wrappings of reducible curves are required to obtain all the adjoint weights of the vector multiplets of the previous section . the reducible curve @xmath123 has moduli space given by @xmath81since @xmath82 has a moduli space given by @xmath81 . looking at figure [ fig : one ] we see that there is also a chain of rational curves in the class @xmath124 but note this this combination is invariant under the @xmath6 monodromy and so has moduli space given by @xmath80 . these circles form the weights of @xmath125 of @xmath87 . ignoring the zero weights for now we see that the adjoint appears @xmath118 times and the @xmath125 of @xmath87 appears @xmath122 times . indeed the zero weights also work out correctly . a zero weight must represent an uncharged hypermultiplet and therefore a modulus . we may use the work of wilson @xcite to demonstrate this . wilson showed that a threefold containing a ruled surface @xmath73 has a moduli space which preserves this ruled surface only in codimension @xmath126 . that is , there are @xmath126 deformations of the threefold which destroy this ruled surface . applying this to both ruled surfaces , we get @xmath127 deformations . on the other hand , each @xmath87 adjoint contains a two - dimensional weight zero eigenspace while each @xmath125 contains a one - dimensional weight zero eigenspace . thus the dimension of the weight zero eigenspace is @xmath128 , which simplifies to @xmath127 , as claimed . the above construction may be easily generalized to @xmath103 : [ ieven ] let @xmath6 monodromy act on an i@xmath129 fiber in f - theory so that the central component of the fiber has moduli space @xmath80 and the outer components have moduli space @xmath81 . thus @xmath85 is a double cover . then the resulting gauge algebra is @xmath103 and we have @xmath118 hypermultiplets in the adjoint representation and @xmath122 hypermultiplets in the @xmath130 representation ( which has dimension @xmath131 ) . similarly @xmath132 with @xmath6 monodromy will yield an @xmath133 gauge algebra with @xmath122 hypermultiplets in the @xmath134 representation ( in addition to the usual @xmath118 adjoints ) . also @xmath104 with @xmath6 monodromy will yield an @xmath135 gauge algebra with @xmath122 hypermultiplets in the vector @xmath136 representation . in the case of @xmath137 with @xmath138 or @xmath139 monodromy , a similar analysis yields @xmath122 hypermultiplets in the @xmath140 representation of @xmath141 . this agrees with the various computations in @xcite where @xmath142 . let @xmath81 be the double cover of @xmath80 branched at @xmath143 points . thus @xmath144 . then , for example , in the @xmath133 case we should have @xmath145 @xmath134 s . this agrees with section 4.3 of @xcite by identifying the branch points with the @xmath146 zeroes of @xmath147 . one might note that the above cases with @xmath6 monodromy may be combined into a simple rule as follows . let @xmath148 be the simply - laced lie algebra which contains the actual gauge symmetry algebra @xmath149 as a subalgebra invariant under an outer automorphism given by the monodromy action . ( in fact , in each of the above cases , the outer automorphism which we use actually has order @xmath1 as an automorphism of @xmath148 , not merely order @xmath1 as an automorphism of the dynkin diagram . ) we may then decompose the adjoint representation of @xmath148 as follows @xmath150 where @xmath151 is a ( possibly reducible ) representation of @xmath149 on which the generator of the @xmath6 outer automorphism acts as @xmath152 . the above rules may be combined to say that we obtain @xmath122 hypermultiplets in the @xmath151 representation . as we have already noted above in the case @xmath153 , the rule will be different if the monodromy group is not @xmath6 . in fact , we will see more generally in section [ ss : small ] that if the outer automorphism representing the monodromy has higher order , the simple rule expressed in equation must be modified . in addition to these `` non - local '' hypermultiplets coming from rational curves moving in families one may also obtain further hypermultiplets from collisions of curves of reducible fibers as in the i@xmath106-i@xmath107 collision discussed above . note that some simple collisions may just induce monodromy without further contributions ( that is , their contributions are completely accounted for by the representation @xmath154 obtained in eq . ( [ eq : rul1 ] ) ) . as an example we show in figure [ fig : so9 ] the generic case of a spin(9 ) gauge symmetry in f - theory . ( 7512,3045)(2389,-2803 ) ( 2401,-1561 ) ( 1 , 0)6000 ( 6901,-61 ) ( 0,-1)2700 ( 3301,-2761 ) ( 8551,-1636)(0,0)[lb ] ( 9901,-1561)(0,0)[lb ] ( 3451,-2761)(0,0)[lb ] ( 7051,-2761)(0,0)[lb ] ( 4501 , 14)(0,0)[lb ] ( 6751 , 14)(0,0)[lb ] this figure shows an @xmath155 fiber along a section of the hirzebruch surface @xmath156 . this section has self - intersection @xmath157 and is denoted @xmath158 in the notation of @xcite . in the most generic situation , the rest of the discriminant locus of the elliptic fibration will consist of @xmath159 fibers along curves which intersect @xmath158 as shown in figure [ fig : so9 ] . generically there are two types of collisions occurring with the frequencies shown . a lengthy computation shows that the @xmath160 cubic collisions where @xmath161 and @xmath162 are affine coordinates in @xmath163 . ] produce extra rational curves in the fiber but no monodromy while the @xmath164 transverse collisions produce monodromy but no extra rational curves . thus the @xmath164 collisions produce @xmath165 of the vector @xmath166 s of @xmath167 . an analysis of the rational curves in the cubic collision shows that we have @xmath160 spinor @xmath168 s . assuming @xmath169 , the existence of these spinors shows that the gauge group must be spin(9 ) . this agrees perfectly with section 4.6 of @xcite . similarly all the other results of @xcite may be confirmed . finally in this section let us return to the case of @xmath6 monodromy acting on a curve of @xmath121 singularities to give an effective @xmath103 gauge symmetry . we will consider the higgs branch in which we give expectation values to the hypermultiplets so as to break completely this @xmath103 gauge symmetry . recall that the geometry of moduli spaces of supersymmetric field theories in question imply that the dimension of this higgs branch should equal the total dimension of the representations of charged hypermultiplets minus the dimension of the gauge group which is broken . we will observe that the geometry is in accord with theorem [ ieven ] . we do this by describing the deformations after shrinking all of the curves in the fibers to zero volume . in section [ ss : below ] we will use the ideas introduced here towards the justification of our main assertion stated in the next section , which states that the gauge algebra in the case of @xmath170 with @xmath6 monodromy is @xmath103 . we let @xmath171 be an unramified ( for simplicity ) double cover of @xmath80 . in addition , we denote by @xmath172 the involution which exchanges sheets of the double cover . we now describe a local threefold @xmath2 containing the geometry of @xmath121 with @xmath6 monodromy over @xmath80 . first we construct a threefold @xmath173 with an @xmath121 fibration over @xmath81 without monodromy . then @xmath2 will be constructed as a @xmath6 quotient of @xmath173 . we construct a singular threefold inside the bundle @xmath174 as the variety defined by the equation @xmath175 where @xmath176 and @xmath177 are in @xmath178 and @xmath179 is in @xmath180 . this threefold has an @xmath181 singularity along @xmath81 , which is identified with the zero section of @xmath182 . it has trivial canonical bundle by the adjunction formula . in a moment we will construct a nowhere vanishing holomorphic @xmath16-form on it , giving independent verification of this fact . the desired threefold @xmath173 is obtained by blowing up the singular locus @xmath183 times in the usual way to obtain a chain of @xmath184 ruled surfaces over @xmath81 . to obtain the desired geometry , we take the quotient of @xmath173 by the fixed point free involution obtained by using @xmath185 on @xmath81 while sending @xmath186 to @xmath187 . note that the fibers of @xmath180 over @xmath188 and @xmath189 are canonically identified with the fiber of @xmath190 over @xmath191 , so this map makes sense . using the explicit description of the blowup and the fact that @xmath176 and @xmath177 are interchanged , it follows that there is @xmath6 monodromy . to show that the quotient @xmath2 by this involution has trivial canonical bundle , it suffices to show that the involution preserves the holomorphic @xmath16-form on @xmath173 . it suffices for our purposes to compute on the singular model . let @xmath192 be any holomorphic @xmath17-form on @xmath81 . then @xmath193 is a holomorphic @xmath18-form on @xmath182 with values in @xmath194 . now , thinking of @xmath192 as a section of @xmath180 , we divide by @xmath192 to obtain the nowhere vanishing @xmath18-form @xmath195 on @xmath182 with values in @xmath196 which is independent of @xmath192 . finally , the residue @xmath197 is the holomorphic @xmath16-form on the singular model of @xmath173 . it is clearly invariant under the involution . the deformations of @xmath2 may be described as the deformations of @xmath173 in @xmath182 which preserve the involution . the general deformation of @xmath173 ( up to change of coordinates ) is given by @xmath198 where the @xmath199 are sections of @xmath200 . note that we are implicitly assuming @xmath201 to construct these deformations . the invariance condition is that @xmath199 lies in the @xmath202-eigenspace of @xmath185 . we now count parameters . the @xmath203-eigenspace of @xmath204 has dimension @xmath205 , while the @xmath152-eigenspace has dimension @xmath206 . thus the dimension of the higgs branch is @xmath207 this is exactly the number of parameters freed up from the @xmath118 adjoints and @xmath122 copies of the @xmath130 representation by higgsing an @xmath103 , as expected . we have implicitly assumed that there are no global obstructions to the local deformations that we have constructed above , and our parameter count is consistent with this assumption . while we appear to have given fairly general rules in the previous section for computing the massless particle spectrum of an f - theory compactification , there are actually many cases where the rules we have given so far become difficult to apply . in particular , witten s analysis of the moduli space of rational curves in @xcite assumes that everything is smooth ( and reduced ) . this need not be the case . we will discuss some awkward cases which appear quite commonly in f - theory . we begin with a discussion of a case which has caused some confusion in the literature that of @xmath6 monodromy acting on a gauge algebra of @xmath209 . the approach of asking for the largest subalgebra invariant under an outer automorphism is not that helpful in this case . putting @xmath98 in eq . ( [ eq : inv ] ) shows that @xmath210 is one possibility . decomposing @xmath211 as @xmath212 and putting @xmath213 ( where @xmath100 is the @xmath101 identity matrix ) shows that @xmath103 is another possibility . ( this form of @xmath214 is nicely adapted to the action on the dynkin diagram . ) now it so happens that @xmath215 ( and that this is the largest dimension which can occur ) . so which is the gauge algebra that f - theory actually wants ? using the approach of section [ s : lcy ] we immediately run into a problem . one of the ruled surfaces , which we will denote @xmath71 , swept out by the reducible components of the fibers will look inevitably locally like the surface @xmath216 in @xmath217 . we show a sketch of ( the real version of ) this surface in figure [ fig : nd ] . each line @xmath64 in this surface crosses another line @xmath218 in the same class . in the case of @xmath170 for @xmath219 there will be other smooth surfaces . this case is a little hard to visualize . in figure [ fig : nd2 ] we show the case of monodromy acting on @xmath220 . ( in this case @xmath89 is the surface @xmath221 . ) the thick lines at the bottom of this sketch show the fiber over a branch point of @xmath85 . the problem is that it is not clear what value we should give to @xmath222 since @xmath64 meets the singular line in @xmath71 . the most nave interpretation of figure [ fig : nd2 ] is to completely ignore the fact that @xmath71 is singular and from the figure read off the intersection matrix @xmath223 this would imply that the gauge algebra is @xmath224 . in general , according to this argument , @xmath6 monodromy acting on @xmath170 would produce @xmath225neither of the possibilities suggested above ! it would be the most obvious algebra suggested by `` folding the dynkin diagram up '' by the outer automorphism . we could get an @xmath103 cartan matrix from the case in question if we could somehow tie @xmath64 and @xmath218 in figure [ fig : nd ] together . that is , somehow the rules of @xmath1-brane wrapping would have to assert that @xmath64 may not be wrapped alone one must also wrap the intersecting curve @xmath218 simultaneously . since @xmath71 is singular , there is no known reason for ruling such a possibility out . by considering the reducible curve @xmath226 as a single curve , we effective replace @xmath71 by a simple ruled surface . thus we would reduce figure [ fig : nd2 ] to figure [ fig : one ] . that is , the case @xmath170 is reduced to @xmath121 and so we get @xmath103 under monodromy . at this point therefore we do not really seem to know what the gauge algebra is . the geometry seems to suggest @xmath225 or @xmath103 while the outer automorphism argument suggests @xmath103 or @xmath227 . we will now give various arguments in support of the following assertion . [ conj : x ] for an f - theory compactification on an elliptic threefold with a curve of @xmath228 fibers ( which locally suggests a symmetry of @xmath170 ) with @xmath6 monodromy , the resulting gauge symmetry is @xmath103 ( provided that the r - r fields are set to `` zero '' ) . this assertion corrects some statements which appeared in earlier literature where it had been assumed the resulting gauge symmetry was @xmath227 for reasons we discussed above . as mentioned in the introduction one can show that the spectrum of various f - theory models , such as point - like instantons on a @xmath229 singularity , is anomaly free for @xmath103 but inevitably would have anomalies in some cases had the gauge algebra contained @xmath227 . note that the outer automorphism of @xmath230 which yields @xmath103 actually has order @xmath18 , since its square is conjugation by the matrix @xmath231 where @xmath100 is the @xmath101 identity matrix ; @xmath232 is not a central element of @xmath230 . this outer automorphism of course still induces the required reflection of the dynkin diagram , as we explained near the end of section [ s : lcy ] . this modifies the analysis which led to eq . as follows . we let the outer automorphism of order @xmath18 act on @xmath148 and decompose into eigenspaces : @xmath233 each of which will be a representation of @xmath149 ( possibly reducible ) . as before , the eigenspaces for eigenvalues @xmath234 can be accounted for by certain positive roots which are left invariant under the involution and by certain pairs of positive roots which are exchanged under that involution . the moduli space for the former is @xmath80 and for the latter is @xmath81 ; when we consider the quantization of the d2-branes wrapped on the corresponding curves , we find @xmath235 half - hypermultiplets for each of the invariant roots , and @xmath236 half - hypermultiplets for each of the pairs . since each pair contributes to both the @xmath203 and @xmath152 eigenspace , this adds up to a total of @xmath235 half - hypermultiplets in the adjoint representation of @xmath149 , and @xmath237 half - hypermultiplets in the representation @xmath154 . we have yet to account for the representations @xmath238 and @xmath239 . in fact , these are the roots which contain either @xmath110 or @xmath240 , which as we have argued above cannot occur as wrapped d2-branes at the generic point of the parameter curve @xmath80 if we are to reproduce the cartan matrix compatible with our main assertion ( or at least , such wrapped branes can not produce vector multiplets ) . however , as we will see in section [ ss : small ] , the representations @xmath238 and @xmath239 _ do _ occur in the hypermultiplet spectrum perhaps because at the branch points of the map @xmath85 , @xmath110 and @xmath240 are identified and there is no apparent obstruction to wrapping the d2-brane there . to be more concrete concerning the case at hand , with @xmath241 , @xmath242 , and the outer automorphism determined by the @xmath214 in eq . , we have @xmath243 the second exterior power of the fundamental of @xmath103 ( which has a trivial one - dimensional summand ) , and @xmath244 the fundamental representation of @xmath103 . thus , the predicted spectrum is : * @xmath118 hypermultiplets in the adjoint representation * @xmath122 hypermultiplets in the second exterior power representation ( including its trivial summand ) , and * additional hypermultiplets in the fundamental representation . in fact , an anomaly calculation @xcite predicts that there will be precisely @xmath245 such hypermultiplets . one possible interpretation of this formula is that there are two fundamentals ( @xmath238 and @xmath239 ) associated to the parameter curve @xmath81 and an additional half - fundamental at each branch point . of @xmath80 which corresponds to the order four element of @xmath246 , and express things in terms of the genus of @xmath247 . ] a kodaira type iv fiber would intrinsically produce an @xmath224 gauge symmetry but monodromy may act on this fiber producing the case of interest . at first sight this might not look like such a good candidate for examination since @xmath249 ! however , the hypermultiplet spectrum will allow us to distinguish the cases . consider the case of amassing point - like @xmath250-instantons on an orbifold point of a k3 surface along the lines analyzed in @xcite . we will be interested in the case of four instantons and six instantons on a @xmath251 quotient singularity . from result 3 and figure 7 of @xcite we may deduce the spectrum without encountering any difficulties . in the case of four instantons , the @xmath138 singularity may actually be partially resolved to a @xmath6 singularity without affecting the particle spectrum . this @xmath6 singularity may then be effected by a `` vertical '' line of @xmath252 fibers ( in the notation of @xcite ) . the six instanton case is effected by a vertical line of @xmath253 fibers . the results are * for four point - like @xmath250-instantons on a @xmath138 singularity we have a nonperturbative enhanced gauge algebra of @xmath76 with hypermultiplets in four @xmath254 representations . * for six point - like @xmath250-instantons on a @xmath138 singularity we have a nonperturbative enhanced gauge algebra of @xmath255 with hypermultiplets as @xmath256 . we may also produce exactly the same physics by using a vertical line of type iv fibers . the configurations of curves of kodaira fibers in the base of the elliptic fibration is shown in figure [ fig : iv ] for the cases of four and six instantons respectively . these diagrams are again similar to those presented in @xcite and represent the situation after the base has been blown up the requisite number of times . the short curved lines represent fragments of the the curve of @xmath159 fibers . ( 10148,2967)(365,-2773 ) ( 377,-61 ) ( 1 , 0)3824 ( 1577,-661 ) ( 1 , 0)1500 ( 2777,-361 ) ( 0,-1)2100 ( 1577,-2161 ) ( 1 , 0)1500 ( 377,-2761 ) ( 1 , 0)3824 ( 5701,-61 ) ( 1 , 0)4800 ( 5701,-2761 ) ( 1 , 0)4800 ( 6301,-511 ) ( 1 , 0)1800 ( 6301,-2311 ) ( 1 , 0)1800 ( 7651,-2536 ) ( 4 , 3)900 ( 8326,-586 ) ( 0,-1)1650 ( 7651,-286 ) ( 4,-3)900 ( 2552,-811 ) ( 2552,-1111 ) ( 2552,-1711 ) ( 2552,-2011 ) ( 7976,-698 ) ( 8126,-751 ) ( 8149,-1988 ) ( 8044,-2086 ) ( 1352,-2236)(0,0)[lb ] ( 1352,-736)(0,0)[lb ] ( 2852,-2611)(0,0)[lb ] ( 3901 , 14)(0,0)[lb ] ( 3901,-2686)(0,0)[lb ] ( 10201,-2686)(0,0)[lb ] ( 10276 , 14)(0,0)[lb ] ( 6601,-436)(0,0)[lb ] ( 6601,-2236)(0,0)[lb ] ( 8401,-1486)(0,0)[lb ] ( 8626,-1936)(0,0)[lb ] ( 8626,-1036)(0,0)[lb ] let us begin with the case of six instantons on the right of figure [ fig : iv ] . the lines of type ii fibers produce no gauge symmetry enhancement . the upper and lower diagonal lines of type iv fibers each collide once transversely with a line of type ii fibers and once non - transversely with the curve of @xmath159 fibers . actually these collisions are very similar . fibers can be turned into a line of type ii fibers intersecting the line of type iv fibers transversely . ] each of these collisions produces @xmath6 monodromy in the type iv fiber producing the geometry of figure [ fig : nd ] . thus each of these diagonal lines of type iv fibers produce an @xmath76 ( or @xmath257 ) gauge symmetry . the remaining vertical line of type iv fibers collides with the other two lines of type iv fibers . resolving this collision shows that no monodromy is induced . thus this vertical line represents an @xmath224 gauge symmetry . an analysis of the collisions shows that there would be an induced hypermultiplet in the @xmath258 of @xmath259 for each collision _ if there were no monodromy_. clearly from the desired spectrum above , this @xmath258 must break up as @xmath260 of @xmath261 . this tells us immediately that the inclusion @xmath262 produced by the action of the monodromy produces a decomposition of the fundamental of @xmath224 via @xmath263 . _ this rules out the natural embedding @xmath264 _ for which @xmath265 . we are left with having to account for a hypermultiplet @xmath254 in each of the @xmath76 s . this must come from the monodromy - inducing collisions of the diagonal lines of type iv fibers with the lines of type ii and type @xmath159 fibers . as these collisions are all the isomorphic locally , each collision must produce a half - hypermultiplet @xmath254 . this is in agreement with our comments concerning the @xmath238 and @xmath239 representations at the end of the previous subsection . the collision point is the point around which the monodromy acts and so it associated with the location of the curve denoted @xmath266 in figure [ fig : nd ] . the choice of associating this @xmath76 as the @xmath267 case of @xmath103 or @xmath225 differs as explained above by whether we view the positive root of @xmath76 as being associated to @xmath64 or to @xmath226 . clearly in the latter case we have @xmath268 as divisor classes and so @xmath266 naturally generates the @xmath254 as required . if only @xmath64 were identified as the positive root then @xmath266 would produce nothing new . therefore we can only correctly identify the spectrum f - theory in the case of the geometry on the right - hand side of figure [ fig : iv ] if we take one of roots of the gauge algebra to be @xmath226 . that is , there really does appear to be a rule in string theory which allows @xmath1-branes to wrap @xmath226 together but not @xmath64 or @xmath218 individually . we can further verify our picture by considering the spectrum for four instantons on the left of figure [ fig : iv ] . there are four collisions with the vertical line of type iv fibers , each producing monodromy . thus , @xmath269 and there are @xmath270 branch points . following the arguments at the end of the previous subsection , we thus predict a spectrum consisting of one hypermultiplet in the @xmath271 representation ( from @xmath154 ) , and @xmath272 hypermultiplets in the fundamental representation ( i.e. @xmath273 ) . this precisely agrees with the spectrum found above : there are four @xmath254 s of @xmath76 . even the @xmath154 representation `` @xmath274 '' occurs correctly : it is the deformation _ la _ wilson @xcite of @xmath80 , or in physical terms of the heterotic string , it is the deformation of the @xmath138 singularity to a @xmath6 singularity which does not affect the spectrum as noted earlier . the rules of @xmath1-brane wrapping are therefore rather unusual for the curves @xmath64 and @xmath218 . as observed previously , away from the branch points , a @xmath1-brane can _ never _ wrap @xmath64 or @xmath218 individually . however , as we have just seen , at the branch points where @xmath64 and @xmath218 coincide , the @xmath1-brane is allowed to wrap the curve . in fact , when this wrapping is taken with both orientations , a hypermultiplet in the @xmath254 of @xmath76 is produced for each branch point . since @xmath64 lies in the singular surface @xmath71 it is perhaps not surprising that the usual rules of @xmath1-brane wrapping appear to break down . anyway , since this same @xmath71 appears as the `` end '' component for the higher rank gauge groups of this type , assuming string theory wraps branes around curves in @xmath71 in a similar way in that context , we arrive at our main assertion . we can also give a different argument in favor of the @xmath103 gauge group in case @xmath80 has positive genus . let us start with the case of @xmath276 fibers with @xmath6 monodromy . by theorem [ ieven ] , this leads to a @xmath277 gauge group and at least one adjoint hypermultiplet . we will show in a moment that the corresponding @xmath278 singularity can be smoothed to an @xmath279 singularity with @xmath6 monodromy . this corresponds to giving a nonzero vev to a semisimple element of the adjoint hypermultiplet , and the @xmath277 gauge group gets higgsed to some rank @xmath183 subgroup @xmath280 . we still have to determine what @xmath149 without knowledge of the @xmath281 that acquires a vev . clearly @xmath282 is possible , so we could have @xmath242 . we now argue that @xmath283 is impossible . there is no embedding of @xmath227 in @xmath277 for @xmath219 . * proof * this argument is due to r. zierau . suppose that there were an embedding of @xmath227 in @xmath277 . then the fundamental @xmath284 dimensional representation @xmath285 of @xmath277 would restrict to a representation of @xmath227 , which necessarily decomposes as a fundamental representation @xmath286 of @xmath227 plus a trivial representation . the alternating form on @xmath285 restricts to a alternating form on @xmath286 . since @xmath286 is odd dimensional , this form is degenerate . nullspace @xmath287 is invariant under @xmath227 , and is a proper subspace since the alternating form on @xmath286 is not identically zero . this is a contradiction . the singular surface in question given by eq . ( [ eq : prob ] ) may be written as @xmath288 and thought of as a double cover of the @xmath289-plane branched along @xmath290 and doubly along @xmath291 . it is the double branching that makes the surface singular . we may smooth the surface by deforming to @xmath292 now the double branching has been split to @xmath291 and @xmath293 . for a fixed value of @xmath179 this process replaces a nodal rational curve by a smooth rational curve , where the nodal rational curve can be viewed as two rational curves intersecting transversely at a point . that is , each pair of intersecting lines in figure [ fig : nd ] is replaced by a single line and the surface is smoothed . this smoothing process is remarkably benign at the level of global geometry . it is often possible to perform it even when the geometry of the ambient threefold , @xmath2 , is completely smooth at all times . we can then derive the @xmath103 gauge symmetry indirectly as follows . the existence of the deformation shows that @xmath1-branes are not allowed to wrap the individual lines of figure [ fig : nd ] . the deformation converts each pair of lines into a single new line . thus if physics is not discontinuously affected by the deformation , the @xmath1-branes contributing to vector particles must only be allowed to wrap the pairs of line in figure [ fig : nd ] together . as we have observed in the discussion immediately preceding the main assertion , we can now conclude that we do indeed obtain @xmath103.-branes around the individual lines is not compromised by this argument . when we deform the curve of @xmath279 singularities to a curve of @xmath181 singularities we may affect the geometry of some points on this curve . thus hypermultiplets which were `` spread '' over the whole curve of singularities may be localized to isolated rational curves by this deformation process . massless vectors can not come from such isolated curves . ] there is of course a problem with the proof of our main assertion by this argument there may be global obstructions to such a deformation . wilson s criterion suggests that such deformations only occur when @xmath294 , where @xmath80 is the base of the fibration of figure [ fig : nd ] as a ruled surface . we address this in part by giving an example of a deformation when @xmath294 . we return to the setup introduced at the end of section [ s : hyper ] . using the notation leading to eq . ( [ eq : su2kdef ] ) , we write the equation @xmath295 this gives @xmath170 with @xmath6 monodromy at the generic point of @xmath80 . the deformation is simply @xmath296 to make sense of this , we have to say a little more about the blowup . the @xmath279 blowups are determined by a procedure given in @xcite after choosing an ordering of the @xmath211 factors of @xmath297 . choosing the @xmath298 factor to be in the middle , we obtain the desired geometry . the last blowup creates a single ruled surface , which smooths out the singular component . it is immediate to see from the description in @xcite that for generic @xmath199 this is a smooth deformation of the desired type . in this model , we have placed a restriction on the genus and have introduced localized matter at the zeros of @xmath299 . however , if we are willing to accept that the process of gauge symmetry enhancement is dictated by local geometry then this example is enough to justify our main assertion . finally we close with a note on the peculiar numerical predictions dictated by anomaly cancellation in the six - dimensional physics produced by f - theory compactified on @xmath2 . this has been discussed in many places before ( for example @xcite ) and is often used as a method of enumerating the spectrum of hypermultiplets . here we have outlined a systematic way of constructing the hypermultiplet spectrum and so the anomaly constraint becomes a peculiar numerical property of the geometry of an elliptic threefold . for completeness we will repeat the anomaly condition here . we consider an elliptic fibration @xmath300 with a section . let @xmath301 be the gauge group ( or algebra ) in six dimensions and @xmath302 the picard number of @xmath9 . then anomaly cancellation along the lines of @xcite yields the following @xmath303 where the hypermultiplets fall into representations @xmath304 of @xmath301 and @xmath305 is equal to @xmath17 if the representation is real or @xmath306 if the representation is complex or quaternionic ( pseudoreal ) . note that the trivial representations also contribute to the sum . these can be determined from the fact that the number of neutral hypermultiplet moduli are equal to @xmath307 . as an example consider the extreme case of @xmath308 corresponding to @xmath309 point - like @xmath250-instantons on a binary icosahedral quotient singularity in the heterotic string @xcite . the threefold for the f - theory description of this has @xmath310 . applying the methods of sections [ s : hyper ] and [ ss : small ] to this threefold we also arrive at a spectrum of hypermultiplets of a @xmath311 for each of the @xmath312 copies of @xmath313 . these representations are quaternionic . equation ( [ eq : anom ] ) then reads @xmath314 the anomaly condition in eq . ( [ eq : anom ] ) has been verified in situations illustrating theorem [ ieven ] and our main assertion ( see the footnote in section 4 of @xcite ) . note that one may obtain further conditions from the anomaly cancellation condition . for example one may require the vanishing of coefficient of each `` @xmath315 '' term in the anomaly . see for example @xcite . it would be very satisfying to give a purely geometric proof of eq . ( [ eq : anom ] ) and the other anomaly conditions . ( a geometric proof which covers a wide variety of cases has recently been constructed @xcite . ) sadly at present the origin of this formula without using string theory is something of a mystery . note that the existence of a section in the fibration @xmath300 is necessary for this to work . if this requirement is not satisfied then there is no six - dimensional physics and the condition need not be satisfied @xcite ( for an example , see @xcite ) . it is a pleasure to thank a. grassi , k. intriligator , j. morgan , r. plesser , g. rajesh , and r. zierau for useful conversations and insights . is supported in part by a research fellowship from the alfred p. sloan foundation . the work of d.r.m . is supported in part by by nsf grant dms-9401447 . the work of s.k . is supported by nsa grants mda904 - 96 - 1 - 0021 and mda904 - 98 - 1 - 0009 , and nsf grant dms-9311386 . s.k . also thanks the mittag - leffler institute for support during the early stages of this project .

we present and compare in this paper new photometric redshift catalogs of the galaxies in three public fields : the ntt deep field , the hdf n and the hdf s . in the case of the ntt deep field , we present here a new photometric catalog , obtained by combining the existing bvri and jk with new deep u observations acquired with ntt - susi2 , and which includes also the contiguous field centered on the quasar br1202 - 07 . photometric redshifts have been obtained for the whole sample ( nttdf + hdf n + hdf s ) , by adopting a minimization technique on a spectral library drawn from the bruzual and charlot synthesis models , with the addition of dust and intergalactic absorption . the accuracy , determined from 125 galaxies with known spectroscopic redshifts , is in the redshift intervals . the global redshift distribution of i selected galaxies shows a distinct peak at intermediate redshifts , at and at followed by a tail extending to . systematic differences exist among the fields , most notably the hdf s which contains a much smaller number of galaxies at and at than the hdf n . we also present for the first time the redshift distribution of the total ir - selected sample to faint limits ( and ) . it is found that the number density of galaxies at isarcmin at andarcmin at , and drops toarcmin ( at ) at . the hdfs data sets are used to compare the different results from color selection criteria and photometric redshifts in detecting galaxies in the redshift range . photometric redshifts predict a number of high candidates in both the hdf n and hdf s that is nearly 2 times larger than color selection criteria , and it is shown that this is primarily due to the inclusion of dusty models that were discarded in the original color selection criteria by madau et al 1998 . in several cases , the selection of these objects is made possible by the additional constraints from the ir bands . this effect partially reflect the poor spectral sampling of the hdf filter set , and is not present in ground based observations where a color selection criteria may be applied . finally , it is shown that galactic m stars may mimic candidates in the hdf filter set and that the 4 brightest candidates at in the hdf - s are indeed most likely m stars . the data and photometric redshift catalogs presented here are available on line at http://www.mporzio.astro.it/highz . 2cm hsd

astro-ph0009158

starting with the pioneering studies of kron ( 1980 ) and tyson ( 1988 ) , one of the main targets of deep imaging surveys has been the observation and study of galaxies at high redshift . the combination of deep imaging and systematic spectroscopic follow up has led to the discovery that the bulk of galaxies at the spectroscopic limit is made of blue star forming galaxies at @xmath27 , where most of the stars of the present day universe appear to form ( see koo and kron 1992 , ellis 1998 for extensive reviews ) . a new approach recently developed to access the earlier epochs of galaxy formation relies on deep multicolor surveys , where multiband images are taken with a complete set of standard broad band filters in order to cover the overall spectrum of the galaxy and to discriminate the populations at different redshifts . in this case , the multiband catalogs are used to define sharp color criteria that select high redshift galaxy candidates ( steidel et al 1995 , madau et al 1996 , madau et al 1998 , m98 hereafter , fontana et al . 1996 , giallongo et al 1998 , g98 hereafter ) . the very successful results of the spectroscopic follow up of color selected samples ( steidel et al 1996 , steidel et al 1999 ) witness the reliability of the method . however , this technique does nt exploit all the information contained in multiband catalogs , that allow to sample the spectral shape of each galaxy over a very extended wavelength range . with this aim , the `` photometric redshift '' approach is rapidly becoming common practice . the concept that the redshift of a galaxy is reflected in its broad band colors dates back to the early 60 s ( baum 1962 ) , and the existence of a definite relation between @xmath22 and magnitude has been demonstrated by koo ( 1985 ) . however , extensive applications of this method have been developed only recently in order to analyze new sets of deep multicolor images . following the original approach of koo , methods have been applied to the hdf - n that make use of the observed redshift magnitude relation ( connolly et al 1997 , wang et al 1998 ) derived from the spectroscopic subsample of the hdf n . a different class of methods must be used when no large spectroscopic follow up has been obtained yet , or when the technique is to be extended beyond the redshift range covered by the spectroscopic sample . in these cases , a spectral library is used to compute galaxy colors at any redshift , and a matching technique is used to obtain the best fitting redshift . with different implementations , this method has been used on hdf n ( sawicky 1997 , fernandez soto et al 1999 , benitez et al 1999 , arnouts et al 1999a ) and ground based data ( g98 , fontana et al 1999a , pello et al 1999 ) . of paramount importance for the development of these applications have been the public surveys carried out over the last few years , primarily the hdf n ( williams et al 1996 ) and hdf s ( casertano et al 2000 ) , since the superb photometric quality of the data and the extensive follow up observations enable a large number of scientific issues to be addressed . large ground based telescopes have also been used to provide multicolor data with an extended wavelength coverage , often to support deep observations from space facilities . examples are the vlt observations of the hdfs nicmos field ( fontana et al 1999a ) and the ntt observations of the whole hdf s ( da costa et al 1998 ) and of the axaf deep field ( rengelink 1998 ) . in this context , the field of the @xmath28 quasar br1202 - 07 was one of the first where deep multicolor observations have been collected from ground to obtain catalogs suitable for determining photometric redshifts . first observations in the bvri bands were used to search for galaxies at @xmath29 in the qso surroundings , and led to the identification of a star forming companion to the qso ( fontana et al 1996 , see also hu et al 1996 , petitjean et al 1996 , fontana et al 1998 ) , providing the first successful example of the `` drop - out '' technique at @xmath29 . from the same set of data , a photometric redhift distribution was obtained and discussed in g98 . a much deeper field in the same bvri bands was obtained in 1997 with the susi1 imager at the ntt , and named `` ntt deep field '' ( arnout et al 1999b ) . wider images covering both fields have been obtained in j and k@xmath0 during the commissioning phase of the sofi infrared spectro imager at the ntt ( saracco et al . 1999 ) , and also made publicly available . in this paper , we present new u band observations of the br1202 and ntt deep fields and describe the procedures used to produce the final ubvrijk@xmath0 catalog in both fields ( sect . the photometric redshift technique is described in sect . 3 and applied to the ntt catalogs and to the public catalogs of the hdf n and hdf s . the main properties of the resulting photometric redshift catalogs are described in sect.4 . in sect . 5 , we focus on the selection of high redshift galaxies to discuss the differences of this approach with respect to the lyman lymit selection and the possible contamination by foreground m stars . the photometric redshift catalogs presented here have been used to address several scientific issues in the high redshift universe , ranging from the evolution of the luminosity density and the number of massive galaxies already assembled at early epochs ( fontana et al 1999b ) to the evolution of galaxy sizes ( poli et al 1999 , giallongo et al 2000 ) . the data and photometric redshift catalogs are also available on line at the web site http://www.mporzio.astro.it/highz . the new data used in this paper are deep ubvrijk@xmath0 images taken with the eso ntt telescope on the field of the high redshift quasar br1202 - 07 ( @xmath1 ) . the b , v , gunn @xmath30 and i observations were taken with the ntt susi imager during two separate observing runs , on 2326 apr 1995 and during the spring of 1997 respectively . the pointings were chosen in order to obtain two slightly overlapping fields , with the first field ( hereafter named br1202 ) centered on the quasar and the second ( ntt deep field , nttdf hereafter ) 100 arcsec south of it , resulting in a total field of 2.3@xmath314. since observations were obtained with two slightly different instrumental setups , the two fields have been analyzed independently and the objects in the overlapping region have been used as a final test of the photometric catalog . the bvri observations and data reduction of the br1202 field are discussed in fontana et al 1996 and g98 . the nttdf was observed in service mode by the ntt team in february through april 1997 , and the data obtained are publicly available at http://www.hq.eso.org/science/ndf . details of the observations , the data reduction and the photometric catalog are discussed in arnouts et al 1999b . image quality in both fields is quite good , with sub arcsec seeing in all the final images . an image of the final field in the @xmath30 band is shown in fig [ image ] . the j and k@xmath0 observations were obtained with the sofi spectro imager at the eso ntt ( moorwood et al 1998a ) in march 1998 . the field of view of this instrument is considerably larger ( 5@xmath315 ) than the susi one so both fields could be observed simultaneously . the data reduction and calibration of the images are described in saracco et al 1999 and publicly available at the eso website . we have extracted from the full format j and k@xmath0 frames two sub images corresponding to the two susi fields described above , by locating a number of corresponding bright sources and applying a linear resampling to the images . since this operation added a further rebin to the original pixels , altering the noise statistics , we estimated the final limiting magnitude of the ir images by computing the @xmath32clipped standard deviation of the intensity of the sky background as measured in random positions of the rebinned image with a 1@xmath33 aperture ( we have verified that the same procedure on the original image provides the same standard deviation ) . the resulting @xmath34 limiting johnson magnitudes are @xmath35 and @xmath36 respectively . these data have been complemented with deep u images obtained with the new susi2 imager at eso ntt . this instrument is equipped with a mosaic of two @xmath37 ccd chips with a field of view of @xmath38 that matches the sofi one . observations were done with the same pointing as the sofi observations , with the two former susi fields both falling on the same chip . data were obtained during two clear nights on 21 and 24 april 1998 . a total of 25200 second of integration time were obtained , split into individual exposures of 1200 - 1500 s each . observations were done in dark sky under moderate seeing conditions ( about 1.4@xmath33 on average ) . the instrument performance was somewhat reduced with respect to the nominal one because of residual parasitic light entering the detector , that produced an additional background pattern with an intensity of about 40% of the sky background in the u band . several exposures of this scattered light were taken during a bad weather night ( 25 april 1998 ) . the pattern proved to be constant and could therefore be subtracted from the individual raw images before further processing . aside from this aspect , the data reduction followed the usual steps for deep images acquired in dither mode . it has been performed with the automated dither pipeline developed for the susi2 images at the rome observatory . it makes use of standard midas commands and of sextractor ( bertin and arnouts 1998 ) to flat - field , align and coadd the images to produce a final image together with its absolute variance map ( i.e. the variance in adu of each pixel in the frame ) . a set of spectrophotometric standard stars have been observed at several airmasses during both nights to calibrate the u images in the ab system . we estimate that the final accuracy of this procedure is of the order of 0.1 _ mags_. again , two smaller images rebinned and aligned to the bv@xmath30i frames were extracted , and their zeropoints were computed in order to match the u magnitude of the brightest objects . for the reader s convenience , we summarize in table1 the main characteristics of these observations and compare them with those of the hdf n and hdf s , which cover similar areas and will also be used in the following sections . object detection and flux measurement have been performed with the sextractor software . in the nttdf we have repeated the procedure used in arnouts et al 1999b by using the sum of the bv@xmath30i frames , that have comparable seeing . the numbering , position and bvri magnitudes of the nttdf provided here are therefore identical to those published in arnouts et al 1999b . in the case of the br1202 field we have used the @xmath30 image , that has the best seeing , to detect and deblend the objects , partly for consistency with g98 and partly because the addition of the bvi images - that have either worse seeing or less depth - did not result in an improved object detection . since the aim of this work is to derive photometric redshifts of objects that are significantly above the detection threshold , the differences in the procedures adopted for object detection are not important , although they must be kept in mind when using the data . in particular , only the nttdf has been used in fontana et al 1999b to measure the evolution of the rest frame uv luminosity density , using the i selected sample . we have also performed an independent object detection on the k@xmath0 images alone to ensure that all the galaxies at @xmath39 were included in the catalogs . the only object detected in the ir images but not in the optical is the methane brown dwarfs already discussed in a separate paper ( cuby et al 1999 ) . we have first obtained a reference magnitude @xmath40 in the @xmath30 band for each object in both fields by obtaining both isophotal and aperture magnitudes in a 2x fwhm aperture . the isophotal magnitude was used for the brighter objects , i.e. for those objects where the isophotal radius is larger than the aperture one . for fainter objects , an aperture correction to @xmath41 has been estimated on bright stars and applied to correct the 2x fwhm aperture magnitude . although this procedure is strictly valid for star - like objects only , it has been shown to be a good approximation for faint galaxies ( smail et al . 1995 ) . to obtain total magnitudes in the other bands we have first measured aperture magnitudes in a fixed circular area corresponding to 2x fwhm of the detection frame . total magnitudes in each band @xmath42 have then been computed as @xmath43 , i.e. the observed aperture colors with respect to the reference @xmath30 band have been normalized to the total @xmath30 magnitude @xmath40 as obtained above . this procedure in principle ensures that colors - that are the quantities used in estimating the redshifts - are measured in the same physical region at any wavelength . in ground based images this procedure is complicated by the different seeing of the images in the various bands . to compensate for this , we have degraded the @xmath30 reference frame to the seeing of each band and computed the aperture colors @xmath44 on the images with the same seeing . to give an idea of the importance of this effect , we have found roughly that each 0.1@xmath33 of difference in seeing translates to about a 0.1 _ mags _ difference in the aperture color , although the exact value depends obviously on the object morphology . the zeropoints were corrected for galactic absorption assuming @xmath45 ( burstein & heiles 1982 ) with @xmath46 , @xmath47 , @xmath48 , @xmath49 , @xmath50 . we summarize in table 1 the zeropoints of the photometric systems adopted for the catalogs , namely ab for all the hdf catalogs and for the @xmath51 band of the ntt fields , and johnson for all the other bands of the ntt fields . since we have not applied any color term to the johnson magnitudes of the catalogs , , the resulting magnitudes are calibrated in the `` natural '' system defined by our instrumental passbands ( see fontana et al 1996 ) . we have used the overlapping section of the two fields to compare the two photometric catalogs - that have been taken with slightly different instrumental setups and at different epochs . the differences found ( @xmath52 , @xmath53 , @xmath54 , @xmath55 ) are close to the limit of the quoted uncertainties on the zeropoints . since the br1202 field has been observed under more controlled photometric conditions , we have applied this set of small corrections to the nttdf catalog to make the two catalogs fully consistent . however , the differences may be indicative of the actual uncertainties associated with the photometry on deep exposures from ground based telescopes , that typically result from the stacking of several exposures obtained on different nights . we will discuss in sect3.2 the possible effects of these uncertainties on the obtained . the final catalogs produced ( along with the photometric redshifts , see sect . 3 ) are reported in table 2 and are available on line at http://www.mporzio.astro.it/highz . this table contains also the half light radius measured with the morphological analysis described in poli et al 1999 . to produce the final catalogs on which photometric redshifts are computed , we have identified and removed from the final galaxy sample obvious bright stars using the class_star parameter provided by sextractor . a threshold class_star@xmath56 has been set ( at @xmath57 in the nttdf and @xmath58 in the br1202 field ) , on the basis of the comparison between ground based and hst data ( arnouts et al 1999b ) . we have also made use of the more detailed morphological analysis by poli et al ( 1999 ) to identify stars down to @xmath59 on the nttdf only . we have applied our photometric redshift code also to the hdf - n and hdf s with the ir observations obtained at kitt peak ( dickinson 1998 ) and at ntt sofi ( da costa et al , 1998 ) , respectively . for the hdf n we have used the multicolor catalog published by fernandez soto et al ( 1999 ) , which uses an optimal technique to match the optical and ir images that have a quite different seeing . a similar catalog for the hdf s has been provided by the same authors and is available at the stony brook web site http://www.ess.sunysb.edu/astro/hfds/home.html . the stony brook catalog of the hdf s contains also the ntt optical magnitudes ( ubvri ) for each object , obtained from the public data of the eis deep survey . very little difference on the photometric redshift is found when the ntt data are used or not , which is due to the lower statistical weight ( compared to wfpc data ) that is associated with them . for this reason , we have used only wfpc bands in the optical and ntt sofi in the ir , for consistency with the hdf n . in both cases we have clipped the outer regions of the frame with lower s / n . obvious stars have been excluded at @xmath60 in the hdf - s again on the basis of the sextractor class_star@xmath61 morphological classification . also the complete set of hdf photometric redshift catalogs is available at http://www.mporzio.astro.it/highz much emphasis has been recently placed on the choice of the spectral library used to compute galaxy colors as a function of redshift . several authors ( e.g. fernandez soto et al 1999 , benitez 1999 ) use observed spectral templates ( e.g. coleman et al 1980 ) . this has the obvious advantage of only relying on a compact set of empirically determined data , but these must be extended in the uv and ir regions using synthetic models and augmented with bluer templates to represent the vast fraction of blue galaxies that dominate the counts at faint magnitudes . an even more empirical approach is to determine a set of autofunctions from the spectroscopic control sample ( csabai et al 2000 ) , from which a set of template spectra can be derived . our recipe for photometric redshifts is based on the use of spectral synthesis models , at present the gissel library by bruzual & charlot . at the cost of an additional computational effort , these models allow - at least in principle - to take into account the spectral evolution of galaxies at high redshift . we will show below that these models are able to provide a redshift accuracy which is at least equivalent to more empirical choices . however , the main reason for following this approach is the gain in physical information on the properties of high redshift galaxies that can be obtained by comparing their observed spectral energy distributions with the model predictions . from the input parameters used to draw the best fitting solution for each galaxy we can indeed obtain estimates of the main physical quantities of each galaxy in the sample : age of the last major starburst , mass already assembled in stars and dust content . even taking into account the known degeneracies among some of the input parameters , one may use this information in a statistical way to further constrain galaxy evolution . applications of this technique have already been presented in g98 and we refer to forthcoming papers for further applications of this method . in practice , we have used a miller - scalo initial mass function ( imf ) including stars in the @xmath62 m@xmath63 range ( the effect of different choices of the imf has been shown in g98 ) . we have considered the cases of metallicity @xmath64 , @xmath65 and as low as @xmath66 with a weak correlation between age and metallicity . for these models , we span over a grid of @xmath67 star formation time - scales @xmath68 ( from @xmath69 gyrs to constant ) and ages to produce the evolution of different spectral types ( the complete grid adopted is listed in g98 ) . since we are aware that the smoothly declining star formation histories that we have used are a simplified rendition of the actual histories experienced by real galaxies , we have added a few models with additional bursts of star formation . in practice , we have overimposed short - lived burst ( 0.3 gyr of timescale ) to two evolving models ( 0.3 and 5 gyr of timescale ) , starting at an age of 3 gyr . in each model , we let a substantial fraction of the final mass in the galaxy to be assembled during the burst : 10% , 30% or 50% in the former and 10% , 20% or 30% in the latter ( for higher fractions the light from the starburst would dominate that from the underlying galaxy at any wavelength ) . _ a posteriori _ , we have verified that `` multiple burst '' models account for about 16% of the best fitting models . in 30% of them , removing the multiple bursts from the library changes the best fit by more than @xmath70 , and in 14 % of the cases by more than @xmath71 , but always less than @xmath72 . in all these cases the burst models are selected to match the contemporary presence of a flat star forming uv continuum and a large 4000break with a significant slope of the ir continuum , that is difficult to match with the simple exponential laws ( see also the discussion on the scalo @xmath73 parameter in g98 ) . although the spectroscopic sample is too small to provide a meaningful statistics , we note that a marginal improvement of the photometric redshift accuracy is usually obtained with the `` multiple burst '' models . more important is the effect that these models have on the parameters estimated for the best fitting spectrum . for instance , the stellar mass in each galaxy is found lower by typically 20% ( but up to 60% ) at fixed redhift , because of the contribution of post agb stars ( that have a low m / l ) to the near ir flux . we have also added to all spectral synthesis model the reddening produced by internal dust ( smc , pei 1992 and calzetti , 1997 ) and lyman series absorption produced by the intergalactic medium ( madau 1995 ) . as a further improvement with respect to g98 , we have added [ oii ] and h@xmath74 emission lines by inverting the kennicutt 1994 relations for the relevant imf . the emission is less straightforward to implement , since the spectral properties at the frequency observed in spectroscopic samples range from strong emission lines to damped absorption systems . as a tradeoff , we have removed from the unattenuated spectrum the stellar absorption features at the frequency . at any redshift , galaxies are allowed to have any age smaller than the hubble time at that redshift ( @xmath75 and @xmath76 km s@xmath77mpc@xmath77 have been adopted throughout the paper ) . the choice of the cosmological parameters are not important , since only colors are effectively used in deriving the photometric redshifts . from the large dataset ( @xmath78 ) of `` simulated galaxies '' thus produced we compute the expected fluxes @xmath79 in any filter @xmath42 by convolving the theoretical spectrum with the normalized transmission curve of the telescope+detector+filter system ( see pickles 1998 for the formulae adopted ) . finally , a @xmath80minimization procedure has been applied to find the best fitting spectral template to the observed fluxes . for each template @xmath81 at any redshift @xmath22 we first minimize @xmath82 ^ 2\ ] ] with respect to the scaling factor @xmath0 where @xmath83 is the flux observed in a given filter @xmath42 , @xmath84 is its uncertainty and the sum is over the filters used , and then we identify the best fitting solution with the lowest @xmath85 . the scaling factor @xmath0 is applied to the input parameters of the input spectrum to compute all the rest frame quantities , such as absolute magnitudes or stellar masses . the stony brook catalogs of the hdf n and hdf s provide @xmath83 and @xmath86 already in physical units ( jy ) , and are therefore directly usable in eq . 1 . in the two ntt fields , where catalogs are given in standard magnitudes @xmath87 ( zeropointed to the johnson system ) , @xmath83 is given by @xmath88 , where @xmath89 is the conversion factor from the ab to the johnson systems , found by convolving an a0v stellar spectrum of the pickels 1998 library with the telescope+detector+filter efficiency . as is well known , the conversion between fluxes and magnitude is not well behaved as the noise level is approached , which raises the question of how non - detections are handled in eq . 1 . in the cases of the two hdfs , the stony brook catalogs provides very small or even negative fluxes for objects that are non detected ( like `` u dropout '' or faint galaxies with no ir detection ) , so that eq . 1 can be directly applied . in the ntt fields , we have followed the common practice of defining a magnitude limit at a given @xmath32 level , and computed @xmath85 according to the following recipe : _ a ) _ we have discarded from the sum in eq.1 the bandpasses where _ both _ the observed galaxy flux and the models fall below the limiting magnitude ; _ b ) _ we have included in the sum the bandpasses where the models lay above the limit : to weight properly the non detection , the object has assigned zero flux and noise equal to the @xmath90 magnitude limit . the former approach is closer to the physical definition of `` measure '' and statistically cleaner , but relies on an accurate estimate of the noise level at very low fluxes , which in real images may be dominated by non - poissonian effects ( such as residuals of flat fielding , pixel correlation and accuracy of sky subtraction ) that may be difficult to quantify . the latter may be useful either on public data with conventional formats or when conservative upper limits on fluxes are more appropriate to the data available . it is therefore interesting to investigate the effects of the different criteria on the final photometric redshifts . to this purpose , we have converted the stony brook hdf n catalog into magnitudes , and defined the @xmath91 limits in the various bands by locating the magnitude at which @xmath92 . we compare the photometric redshifts obtained in the original `` flux '' format and in the conventional `` magnitude '' one in fig [ uplim ] , differentiating galaxies in the three samples used in fontana et al 1999b : @xmath93 it is seen that very modest differences arise only in the faintest sample , with a few galaxies migrating from @xmath94 to @xmath95 and even less in the opposite way ( which makes the choice of the `` flux '' format more conservative ) , and with some scatter in the @xmath96 region , where the different treatment of the ir band plays a role . this procedure has been tested on a sample of 125 galaxies with spectroscopic redshifts and multicolor photometry . of these , 112 come from the cohen et al ( 2000 ) compilation of the hdf n ( we have included only objects that fall in the inner region of the frame , as for the whole optical catalog , see sect . 2.3 ) , 3 are taken from the stony brook public catalogs of the hdf s , while the remaining 10 are taken from the vlt spectroscopic follow up of `` dropout '' galaxies in the hdf s and axaf deep fields ( cristiani et al 2000 ) . the results are shown in fig . [ zspe ] , where we differentiate between objects with secure or uncertain redshift , following the classification by cohen et al ( 2000 ) . if we consider only the objects with secure redshift , and remove the object at @xmath97 with @xmath98 , for a total 101 objects , the accuracy is @xmath99 at @xmath100 and @xmath101 at @xmath102 , that we keep as reference value . if we take all the galaxies in the sample , and remove only the three objects with clearly discrepant redshift , the accuracy raise to @xmath103 at @xmath100 and @xmath104 at @xmath102 . if we compute a 3-sigma clipped rms we obtain an accuracy @xmath105 over the whole redshift range , with 111 objects out of 125 ( i.e. 89% ) , and @xmath106 in the @xmath100 interval ( 85 objects out of 91 , i.e. 93% ) . in any case , the relative precision @xmath107 $ ] is always of the order of 0.05 in the @xmath108 sample , as in most of similar analysis ( cohen et al 2000 ) . a 5% systematic underestimate @xmath109 appears to exist at @xmath102 in the spectroscopic sample , similar to that found by a recent revised version of the photometric redshifts based on the coleman spectra ( @xmath110 , lanzetta 2000 ) , but with opposite sign . we will explore in future works how this underestimate can be reduced with a different choice of the gissel grid . in particular , we are collecting a much wider spectroscopic sample in different fields ( dodorico et al , in preparation , and within the `` k20 project '' , p.i . cimatti ) and with a variety of bandwidths to perform a more accurate calibration of the spectral library , especially in the range @xmath111 , where most of the observed data come from the rest frame uv spectra . undoubtedly , the availability of only one major spectroscopic control sample ( the hdf - n ) is a major limitation that prevents to some extent a full evaluation of all the systematics involved in . it is worth reminding that at least 4 different ingredients are needed to pin down an accurate recipe for photometric redshifts : i ) the photometric catalog ( which depends on the accuracy of data processing and calibration and on the procedure used to measure colors ) ; ii ) the knowledge of the response curve of the telescope+filters+detector system ; iii ) the spectral library used and iv ) the matching algorithm used . despite this , most of the current debate is focused on the best choice of the spectral library and the relative influence of the different components in a typical survey is far from being assessed , especially in the case of ground based observations . some hints on the expected biases may be obtained by monte carlo simulations . however , these typically assume errors to be gaussian distributed which may be an oversimplification since , at the very bright and very faint limits , the uncertainties involved in background estimation and flat fielding may dominate the errors . in fernandez soto et al 1999 and in arnouts et al 1999a monte carlo simulations have been used to show that photometric redshifts in the hdf n catalogs are stable with respect to noise fluctuation down to about @xmath112 . arnouts et al 1999a showed that the maximum number of misidentifications ( i.e. shift in @xmath22 exceeding @xmath113 ) is below 30% in the @xmath112 catalogs . we adopt here a brighter limit @xmath114 ( much brighter than the @xmath115 limiting magnitude ) , that is ultimately set by the depth of the bluest images ( @xmath116 ) , so that a minimum @xmath117 is still measurable . at this magnitude the misidentification are obviously much lower ( see fig 4 of arnouts et al 1999a ) . by analogy , we set an upper limit to @xmath118 in the nttdf . as a first test of the effect of systematic differences , we have compared the obtained in the nttdf with the two different photometric calibrations ( see sect . [ zp ] ( upper panel ) shows the histogram of the scatter between the @xmath119 obtained with the `` original '' calibration and the @xmath120 resulting with the adopted calibration . we first note that , for most of the objects , the difference found is within the expected accuracy of . objects with clearly discrepant redshifts are typically faint and amount to only @xmath121 of the total , a result that is consistent with the monte carlo simulation of arnouts et al 1999a , when scaled do the depth of the nttdf . a close scrutiny of the @xmath122 relation ( fig . [ zp ] , lower panel ) shows that the systematic changes to the zeropoints induce systematic offsets in the that vary with redshift as the spectral breaks pass across the various filters . these changes marginally affect the selection of high redshift galaxies in the nttdf : two objects at @xmath123 in the final catalog are assigned at @xmath124 with the original calibration ( out of 40 in the @xmath125 bin ) , compensated by two objects that are moved from @xmath126 to much lower @xmath22 . one galaxy ( out of 14 ) is added to the @xmath127 bin with the adopted calibration . the major statistical effect is due to 5 objects that move from @xmath128 to @xmath129 , slightly changing the predicted values in the uv luminosity density in the bins adopted in fontana et al 1999b . in the long run , we expect that extended spectroscopic follow - up of multicolor deep fields with different filter sets and in different regions of the sky will allow a detailed evaluation of the different systematics involved . we discuss in this section the fundamental properties of the redshift distributions in the catalogs . beyond being the basis for several scientific studies - as discussed in the introduction - these distributions can be used to compare the predictions of photometric redshift analyses in different fields and with different instrumental setups . given the limited size of the fields ( the hdfs and the nttdf have areas of 4 - 4.8 arcmin@xmath130 , respectively ) small scale fluctuations are expected to lead to discrepancies in the redshift distributions . at the same time , photometric redshifts may still be regarded as an experimental technique and small uncertainities in the photometric calibration or inadequacy of the spectral templates in the different filter sets may also lead to small but systematic effects in the resulting redshifts . at the present stage , it is not posssible to fully disentangle these effects , pending full spectroscopic follow - ups and independent checks of the image calibrations . however , the comparison among the results in different fields may be regarded as an estimate of the total uncertainities in this kind of analysis . for an evaluation of its effects on some scientific topics see fontana et al 1999b . fig [ nz_br1202 ] shows the redshift distribution in the catalogs of the two fields around , separately ( we remind the reader that these catalogs have different depths ) . the distribution of fig [ nz_br1202]_b _ of the br1202 field is similar to the one published in giallongo et al 98 , where no u and j band and poorer k data were available . we find here the typical characteristics of redshift distributions at faint limits i.e most of the galaxies are located at @xmath131 with a tail extending to greater @xmath22 . although the comparison between the @xmath30selected distribution of the br1202 field and the i selected distribution of the nttdf is not straightforward , we note that the tail of high redshift galaxies is - as expected - more pronounced in the deeper sample . we also note that the position of the peak in the redshift distribution that is skewed to lower redshifts in the nttdf compared to the br1202 field , opposite to what is naively expected in a deeper sample . this is , however , consistent with the observed number densities in the two fields - the nttdf being richer in relatively bright galaxies - and emphasizes the need for much wider surveys . the average redshift distributions at faint limits can be obtained by merging the nttdf and the two hdfs . fig [ nz_i ] . shows the i - limited distributions , at @xmath132 ( for a total of 973 objects ) , resulting from the combination of nttdf and the two hdfs , and at @xmath133 ( 1361 objects ) , from the two hdfs alone . the global redshift distribution shows again the well known features of a distinct peak at intermediate redshifts , @xmath5 for the brighter and @xmath134 for the fainter followed by a tail extending to the highest redshifts compatible with the i selection criteria . a close scrutiny of the individual distributions reveals systematic differences amongst the fields . at both magnitude limits , the hdf s has a much smaller number of galaxies at @xmath10 than the hdf n . since the filter set used in both observations is identical , this effect is not likely to be an artefact of the photometric redshifts . finally , we use the combined data sets of nttdf and hdfs to build the redshift distrbution in ir selected samples shown in fig [ nz_ir ] . the basic scientific output of these distributions has been discussed in fontana et al 1999b , where it is shown that the redshift distribution of k band selected galaxies provides a direct estimate of the number of massive galaxies already assembled at high redshift . we provide here a more detailed view of the redshift distribution in j and k selected samples at different magnitude limits . these distributions may be useful for tuning extended spectroscopic surveys with the new ir spectrographs that are being implemented at 8m telescopes . these surveys are designed to improve the knowledge of the large scale structures and emission line features of galaxies in the range @xmath135 , where spectroscopic measurements with optical spectrographs are difficult because of the lack of suitable features . we detail the observed number densities at different j and ks limits in table [ ircounts ] . we find that the number density of galaxies at @xmath136 is @xmath137arcmin@xmath16 at @xmath17 and @xmath138arcmin@xmath16 at @xmath13 , and drops to @xmath19arcmin@xmath16 ( at @xmath13 ) at @xmath20 . the search for extremely high redshift galaxies is one of the most interesting applications of deep multicolor imaging . the so called `` dropout '' technique has gained widespread popularity due to its simplicity and high success rate in detecting galaxies at @xmath139 . however , it is worth remembering that this method relies on the contemporary presence of _ two _ distinct features : the sharp cutoff provided by the intrinsic lyman limit ( at @xmath140 ) or by the igm lyman@xmath74 absorption ( at higher @xmath22 ) _ and _ the flat star forming continuum at longer wavelengths , that samples the rest frame extreme uv spectral region . the latter criterium is necessary to discriminate against interlopers ( e.g. intermediate @xmath22 elliptical or dusty galaxies , or even late type stars ) that may also show a dramatic drop in their rest frame near uv . in a nutshell , high redshift galaxies appear as objects that are `` blue in the red and red in the blue '' . an important consequence of this approach is that the `` dropout '' technique is essentially a highly conservative technique that is designed to minimize the fraction of interlopers , but that is biased against even moderately dust reddened galaxies : as shown in pettini et al 1997 , even a relatively weak dust absorption ( @xmath141 with an smc extinction law ) would prevent the detection of galaxies at @xmath142 in their well - studied @xmath143 color plane . given the strength of the igm color feature , color selection criteria and photometric redshifts provide comparable results when the filter set ensures an appropriate coverage of both features . this is illustrated at @xmath144 in the vlt observations of the hdfs nicmos field ( fontana et al 1999a ) and at higher @xmath22 in the previous version of the br1202 catalog ( g98 ) , where a @xmath145 requirement was explicitly chosen to detect galaxies at @xmath146 . a similar requirement @xmath147 has been adopted by steidel et al 1999 to improve the rejection of low @xmath22 interlopers in their first systematic spectroscopic follow up of @xmath29 candidates . unfortunately , a similar agreement between photometric redshifts and color selection criteria does not hold in the case of galaxies at @xmath148 in the hdf n and hdf s . as a result , the estimate of the uv luminosity density obtained with color selection criteria ( @xmath149 erg s@xmath77 hz@xmath77 mpc@xmath150 , m98 ) is significantly different from that obtained with photometric redshifts ( ( @xmath151 erg s@xmath77 hz@xmath77 mpc@xmath150 , pascarelle et al 1998 , fontana et al 1999b ) . this is primarily due to the much larger number of candidates selected by the latter technique ( roughly 2 times larger ) . this difference appears rather puzzling at first glance , because both methods have been applied to the same data set ( albeit on different catalogs ) and the m98 selection criterium has been defined using a synthesis spectral library similar to the one used here . to investigate the origin of this discrepancy , we reproduce in fig [ madau ] the @xmath152 vs @xmath153 plane adopted by m98 , whose selection criteria requires galaxy candidates to fall within the polygonal region shown in fig [ madau ] ( solid line ) , _ and _ to have @xmath154 . the latter requirement ensures a proper detection of the `` @xmath155dropout '' and is therefore ultimately set by the depth of the @xmath156 frame . we also plot in fig [ madau ] the observed galaxy colors of the joined hdf n + hdf s catalogs , differentiating galaxies at different photometric redshifts . at variance with the m98 catalog , our data are limited at @xmath157 , that corresponds to a rest frame wavelength of @xmath158 . the following conclusions result from this comparison : - most of the galaxies lying within the m98 polygonal region have a photometric redshift falling in the @xmath159 range . the few exception are in any case at @xmath160 , in agreement with the expected scatter of photometric redshifts . this confirms our statement that the strengths of the `` dropout '' signature is strong enough to provide a unique redshift assignment . conversely , about 50% ( i.e. 29 out of 54 ) of the galaxies with @xmath161 fall within the m98 area ( we include also the two objects at the left of the m98 area ) . - of the remaining fraction ( 25 out of 54 ) of the @xmath162 candidates , most are distributed in a `` belt '' around the m98 area , that is indeed occupied by high @xmath22 theoretical models ( see fig . 5 of m98 ) but was excluded to minimize the contamination by lower redshift interlopers . - finally , other objects lay progressively apart from the m98 region , filling the bridge toward the color position of galaxies at @xmath163 , with progressively larger @xmath153 and smaller @xmath152 with respect to what is required by the m98 approach . these `` @xmath164 outliers '' are typically faint in @xmath165 and have a very large @xmath166 . the brightest among these `` @xmath164outliers '' is hdf n3259652 , that is marked by a large circle in fig [ madau ] and whose spectral distribution and best fitting spectrum is shown in fig [ hdf173 ] . this object has a large @xmath152 because it is undetected in @xmath156 and is the brightest object among this class ( these objects are undetected in @xmath156 , so that the @xmath152 should be read as lower limits ) , and has a spectroscopic redshift of 4.58 and a photometric redshift @xmath167 . it is fitted by the spectrum of a star forming galaxy with a significant amount of dust ( @xmath141 with an smc extinction law ) to account for the large @xmath152 and @xmath153 observed colors . as can be seen from fig [ hdf173 ] , the unique identification of this galaxy is made possible by the additional constraints provided by the ir bands : the best - fitting solution when the j , h and k@xmath0 band are not used corresponds indeed to an evolved galaxy at @xmath168 , consistent with the m98 plot ( as for all the objects in the spectroscopic sample , the photometric redshift does not depend on how non - detection in the b band is treated ) . to evaluate the effect of the ir bands on the photometric redshift selection of high @xmath22 galaxies , and in particular on these `` z-4 outliers '' , we have obtained a fit on the hdf n and hdf s samples after removal of the ir bands . the comparison between the 7bands and the 4bands photometric redshifts is shown in fig [ noir ] . it is shown that , while all the objects selected with the m98 criteria are assigned at @xmath127 also with only 4 optical bands , a substantial number of the `` @xmath164 outliers '' is placed at @xmath169 when the ir bands are removed . these objects have redshifts @xmath170 , and are typically fitted with significant amounts of dust [ @xmath171 . in all these cases the role of ir bands is to constrain the fit to relatively small values of @xmath172 , even in the cases where only upper limits exist in the ir , with the result of discarding models of evolved or dusty galaxies at lower @xmath22 that could easily mimic the red @xmath152 and @xmath153 optical colors . also the `` @xmath164 outliers '' that are placed at @xmath170 irrespective of the inclusion of ir bands ( a total of 6 out of 54 ) are selected among models with @xmath173 , that were present also in the m98 synthetic grid but not included in the selection criteria . respect to the m98 approach , the faint objects with large @xmath174 and @xmath175 were also discarded by not only because of their colors but also because they fade beyond the @xmath176 limit applied by m98 . not suprisingly , the situation is different in our ground based images . if we remove the ir bands from the nttdf sample , all the @xmath177 candidate are found as well , and all the @xmath178 candidates by g98 are found by photometric redshifts , and viceversa . the reason lays in the different spectral sampling of the two filter sets . the spectral region covered by the three filters @xmath156 , @xmath165 and @xmath179 , where lyman@xmath74 blanketing starts to affect the hdf filter set already at @xmath180 , is sampled by 4 different filters ( @xmath181 ) in the ground based observations used here or by other groups . _ summarizing , the @xmath182 candidates that are selected with photometric redshifts in the hdfn+s data set and that lay far from the m98 region are typically predicted to be objects at @xmath170 with substantial amount of dust , i.e. a class of objects which multicolor criteria are biased against_. the ir bands are often required with the hdf filter set to include these objects in the sample . a correction of similar amplitude to the m98 criteria for dusty objects has been derived by steidel et al 1999 using the observed color distribution of galaxies in their spectroscopic survey at @xmath183 . _ it is also worth emphasizing that the differences between the methods have not been tested with spectroscopic follow up in a systematic way_. it is certainly reassuring that hdf n3259652 , that lays outside the m98 area albeit it is at @xmath184 , is recovered with the photometric redshift analysis , but the major differences arise from the fainter sample , where the constraints from the ir bands are less stringent and the noise on the observed magnitudes is larger . another issue that should not be underestimated concerns the differences among different algorithms used for source detection and measure , even on the same data . based on the same set of hdf n and hdf s images , casertano et al 2000 find a larger number of @xmath185 candidates than are found in the stony brook catalogs , at the same magnitudes and adopting the same m98 criteria , and find that the hdf n shows less candidates respect to the hdf s , contrary again to what it is found in the stony brook catalogs . we conclude with a comment on fig [ noir ] : we note that , as expected , the lack of ir bands causes a relatively large scatter in the @xmath186 region , the region where the major spectral breaks fall in the ir bands . however , the number of completely discrepant redshifts ( @xmath187 ) is limited . to some extent , this reflects the fact that for a typical object the s / n ratio of the wfpc bands is always much higher than that of the ir bands , so that they dominate the @xmath2 even when the ir bands are included . it is well known that galactic stars are one of the main sources of foreground interlopers that may be identified as high redshift galaxies . as discussed in sect . 2.3 , we have applied a pure morphological criterium to select bright stars in the hdf s and in the nttdf , where spectroscopic follow up is still lacking . it is interesting to note that 8 ( out of 32 ) of these `` morphological '' stars are assigned at @xmath170 by our photometric redshift code ( see also lanzetta 2000 ) . most of these objects are very bright ( @xmath188 ) so that the morphological classification is expected to be very robust , since high redshift galaxies are compact but resolved in high quality images , as shown by giavalisco et al 1996 and giallongo et al 2000 . however , we have complemented the morphological selection with a comparison with the stellar library by pickles 1998 , that we have used as input to our photometric redshift code . we have therefore compared the @xmath189 obtained for each object in the hdf s with the use of the pickles 1998 library with the @xmath190 obtained by the ordinary procedure . we find that , while all bright galaxies are very poorly fitted by any stellar template , with @xmath191 ( typically by a factor of 100 ) , 85% of the stars selected by morphological criteria have @xmath192 and , in particular , five of the @xmath170 candidates have @xmath193 . two typical examples are shown in fig [ mstar ] . it is shown that the high redshift identification comes from the flat continuum at long wavelengths , coupled with the large spectral break in the bluer bands - all features typical of high redshift galaxies ( see previous section ) . however , they are better fitted by an m star spectral template because of the relatively gentle fading - when compared to the abrupt drop due to the igm at @xmath163 - at blue wavelengths and the negative slope at @xmath194 that are both typical of m stars . we conclude that also the color properties of these objects support the morphological classification of faint halo stars . it is important to emphasize that , once bright m stars ( i.e. objects that correspond to the morphological criteria discussed above ) are removed , _ no other convincing candidate at @xmath24 is found in the hdf s sample_. in addition to the photometric redshift predictions , this can be tested directly by looking at the multicolor data : no object in the stony brook catalog of the hdf s is detected in @xmath115 ( at @xmath195 ) but not in @xmath196 , contrary to what is observed in the objects detected so far at @xmath197 in the hdf n ( lanzetta et al 1996 , spinrad et al 1998 ) . as a result , the uv luminosity density @xmath198 at @xmath199 is still poorly determined , since it changes by a factor of @xmath200 between the hdf n and the hdf s at @xmath199 ( fontana et al 1999b ) . this paper presents the photometric redshift catalogs that have been used in previous papers to investigate different aspects of the evolution of galaxies in the high redshift universe , such as the history of the uv luminosity density and the number of massive galaxies already assembled at early epochs ( fontana et al 1999b ) and the evolution of galaxy sizes ( poli et al 1999 , giallongo et al 2000 ) . a new deep multicolor ( ubvrijk@xmath0 ) photometric catalog has been produced of the galaxies in the ntt deep field and in the slightly overlapping field centered on the @xmath1 quasar br1202 - 07 . this has been obtained by combining the existing bvri and jk@xmath0 images with new , deep , u band observations of both fields acquired with ntt - susi2 . we have further presented in this paper the photometric redshift catalog drawn from this galaxy sample , using a @xmath2 minimization technique based on the bruzual and charlot spectral library , with the addition of dust and intergalactic absorption . we have also presented the results of applying the same photometric redshift techniqe to public catalogs of the hdf n and hdf s , where a simlar optical - infrared coverage is available . the global redshift distribution of i selected galaxies shows a distinct peak at intermediate redshifts , @xmath5 at @xmath6 and @xmath7 at @xmath8 followed by a tail extending to @xmath9 . systematic differences exist amongst the fields , most notably the hdf s contains a much smaller number of galaxies at @xmath10 and at @xmath11 than the hdf we have also presented the redshift distribution of the total ir - selected sample , which may be useful to tailor the planned surveys with ir spectrographs at large telescopes that will target the redshift range @xmath201 . we find that the number density of galaxies in the redshift range is @xmath202arcmin@xmath16 at @xmath17 and @xmath203arcmin@xmath16 at @xmath13 . we have also discussed the different results from applying color selection criteria and photometric redshifts for detecting galaxies in the redshift range @xmath21 using the hdfs data sets . we find that photometric redshifts predict a two times larger number of high @xmath22 candidates in both the hdf n and hdf s and show that this is primarily due to the inclusion of slightly dusty ( @xmath204 with smc extinction law ) models that were discarded in the original color selection criteria conservatively applied by madau et al 1998 . in several cases , the selection of these objects is made possible by the additional constraints from the ir bands . this effect partially reflect the poor spectral sampling of the hdf filter set , and is not present in ground based observations where a @xmath23 color selection criteria may be applied . finally , we show that galactic m stars may mimic @xmath24 candidates in the hdf filter set and that the 4 brightest candidates at @xmath24 in the hdf - s are , indeed , most likely to be m stars . the estimates of the uv luminosity density @xmath198 at @xmath199 from these data , when the selection against halo stars is applied , show that @xmath198 changes by a factor of @xmath200 between the hdf n and the hdf s ( fontana et al 1999b ) . * acknowledgments * + the paper is based on observations made with : the eso new technology telescope at the la silla observatory ( some under the eis programs 59.a-9005(a ) , 60.a-9005(a ) ) , the nasa / esa hubble space telescope and the kitt peak national observatory . the ultraviolet observations of the nttdf were performed in susi-2 guaranteed time of the observatory of rome in the framework of the eso - rome observatory agreement for this instrument . arnouts , s. , cristiani , s. , moscardini , l. , matarrese , s.,lucchin , f. , fontana , a. , giallongo , e. , 1999a , 310 , 540 arnouts , dodorico , s. , cristiani , s. , zaggia , s. , fontana , a. , giallongo , e. , 1999b , a&a 341 , 641 baum , w.a . , 1962 , iau symp 15 , 390 benitez , n. , 1999 apj subm . , astro - ph/9811189 cohen , j. , hogg , d. , blandford , r. , cowie , l. , hu , e. , songaila , a. , shopbell , p. , richberg , k. , 2000 , apj in press , astro - ph/9912048 connolly , a. j. , szalay , a. s. , dickinson , m. , subbarao , m. u. , & brunner , r. j. 1997 , apj , 486 , l11 cristiani , s. , appenzeller , i. , arnouts , s. , nonino , m. , aragn salamanca , a. , benoist , c. , da costa , l. , dennefield , m. , rengelink , r. , renzini , a. , szeifert , t. , white , s. , 2000 , a&a in press , astro ph/0004213 fontana a. , dodorico s. , fosbury , r. , giallongo , e. , hook , i. , poli , f. , renzini , a. , viezzer r. , , 1999a , a&a 343 , l19 fontana , a. , menci , n. , dodorico , s. , giallongo , e. , poli , f. , cristiani , s. , moorwood , a. , saracco , p. , 1999b , mnras , 310 l27 giallongo , e. , dodorico , s. , fontana , a. , cristiani , s. , egami , e. , hu , e. , mcmahon , r. g. , 1998 , aj 115 , 2169 ( g98 ) lanzetta , k. m. , chen , h. , , fernandez soto , a. , pascarelle , s. , yahata , n. , yahil , a. , 2000 , in : the hy - redshift universe : galaxy formation and evolution at high redshift , eds bunker , a and van breugel , w.,j.,m .. , astro - ph/9910554 madau , p. , 1995 , apj , 441 , 18 madau , p. , pozzetti , l. , dickinson , m. , 1998 , apj , 498 , 106 moorwood , a. , cuby , j. g. , lidman , c. , 1998a , the messenger , 91 , 9 moorwood , a. , cuby , j. g. , devillard , n. , lidman , c. , saracco , p. , 1998b , http://www.hq.eso.org/science/sofi_deep/ pascarelle , s. m. , lanzetta , k. m. , fernandez soto , a. , 1998 , apj , 501 , l1 pei , y.c . , 1992 , apj , 395 , 130 pello , r , kneib , j. p. , le borgne , j. f. , bezecourt , j. , ebbels , t. m. , tijera , i. , bruzual , g. , miralles , j. m. , smail , i. , soucail , g. , bridges , t. j. , 1999 , a&a 346 , 359 pickles , a. j. , 1998 , pasp , 110 , 863 pei , y. c. 1992 , apj , 395 , 130 petitjean , p. , pcontal , e. valls - gabaud , d. , charlot , s. 1996 , nature , 380 , 411 pettini , m. , steidel , c. c. , adelberger , k. l. , kellogg , m. , dickinson , m. , giavalisco , m. , 1997 , in cosmic origins : evolution of galaxies , stars , planet and life , shull , woodward and thronson eds . , asp series poli , f. , giallongo , e. , menci , n. , dodorico , s. , fontana , 1999 , apj , 527 , 662 br1202 & 4 & 26.9 & 26.5 & 26 & 25.6 & 25 & 23.4 & & 21.7 + nttdf & 4.8 & 26.9 & 27.5 & 26.85 & 26.5 & 26.4 & 23.4 & & 21.7 + hdf - n & 4.2 & 28.4 & 29.2 & 29.5 & & 29 & 23.8 & 22.9 & 22.4 + hdf - s & 3.9 & 28.2 & 29 & 29.3 & & 28.7 & 24 & 22.1 & 22 + rcccccccllll id&@xmath74 [ h]&@xmath205 [ deg]&u@xmath206&b&v&r&i & j & k&@xmath120&@xmath207 + n0001&12:05:18.01&-7:44:40.98&24.74&24.28&23.33&22.85&22.63&21.92&21.01&0.34&0.25 + n0002&12:05:18.43&-7:44:40.43&24.72&24.46&23.83&23.06&22.12&20.93&19.43&0.80&0.63 + n0003&12:05:18.21&-7:44:39.53&23.35&23.20&22.42&21.79&21.61&21.14&20.28&0.26&0.66 + n0005&12:05:17.81&-7:43:33.34&25.19&24.86&24.27&23.96&23.54&22.28&21.76&0.10 & + n0006&12:05:17.79&-7:43:26.36&25.80&25.88&24.43&24.12&23.97&22.49&21.63&0.29 & + c|cc|cc|cc @xmath22 & gal & gal / arcmin@xmath130 & gal & gal / arcmin@xmath130 & gal & gal / arcmin@xmath130 + & & & + & 43 @xmath208 & 3.08 & 96 @xmath209 & 6.88 & 196 @xmath210 & 14.06 + @xmath211 & 1 @xmath212 & 0.07 & 12 @xmath213 & 0.86 & 38 @xmath214 & 2.73 + @xmath215 & 0 @xmath216 & 0 . & 1 @xmath217 & 0.07 & 13 @xmath218 & 0.93 + @xmath219 & 1 @xmath220 & 0.07 & 1 @xmath220 & 0.07 & 5 @xmath221 & 0.36 + @xmath102 & 0 @xmath216 & 0 & 0 @xmath216 & 0 & 2 @xmath222 & 0.14 + & & & + & 69 @xmath223 & 4.95 & 125 @xmath224 & 8.98 & 247 @xmath225 & 17.72 + @xmath211 & 6 @xmath226 & 0.43 & 24 @xmath227 & 1.72 & 48 @xmath228 & 3.44 + @xmath215 & 1 @xmath217 & 0.07 & 5 @xmath229 & 0.36 & 15 @xmath230 & 1.08 + @xmath219 & 1 @xmath220 & 0.07 & 5 @xmath231 & 0.36 & 17 @xmath232 & 1.21 + @xmath102 & 0 @xmath216 & 0 & 0 @xmath216 & 0 & 13 @xmath233 & 0.93 +

the dynamics of symmetry - breaking after a quench is numerically simulated on a lattice for the ( 2 + 1)-dimensional model . in addition to the standard sigma - model with temperature - dependent-potential the energy functional includes a four - derivative current - current coupling to stabilize the size of the emerging extended topological textures . the total winding number can be conserved by constraint . as a model for the chiral phase transition during the cooling phase after a hadronic collision this allows to investigate the interference of baryon - antibaryon production with the developing disoriented aligned domains . the growth of angular correlations , condensate , average orientation is studied in dependence of texture size , quench rate , symmetry breaking . the classical dissipative dynamics determines the rate of energy emitted from the relaxing source for each component of the 3-vector field which provides a possible signature for domains of disoriented chiral condensate . we find that the pions are emitted in two distinct pulses ; for sufficiently small lattice size the second one carries the dcc signal , but it is strongly suppressed as compared to simultaneous sigma-meson emission . we compare the resulting anomalies in the distributions of dcc pions with probabilities derived within the commonly used coherent state formalism .

hep-ph0107034

the observation of clear signatures for the chiral phase transition still remains one of the outstanding challenges of hadron physics . there is hope that the extreme energy densities located inside the spatial region between receding baryonic slabs shortly after ultrarelativistic baryon or heavy - ion collision may provide physical conditions for such a phase transition to take place . from the theoretical point of view we are dealing with a complex dynamical system , strongly interacting quantum fields with many different fermionic and bosonic degrees of freedom , far from equilibrium , creating and emitting baryons and mesons from a rapidly expanding and cooling spatial volume . correspondingly , there is a vast theoretical literature ranging from cascade models in the colored partonic degrees of freedom to hydrodynamic flow models for hadronic currents , which might be applicable at different stages and shed light on different aspects during the time evolution of such events . if within the hot fireball chiral symmetry indeed is restored then the dynamics of its spontaneous breaking and formation of a chiral condensate is one important aspect of the cooling process in which the highly excited spatial volume returns back to normal physical vacuum . a possible strategy which has proven useful for many other cases of phase transitions in microscopically very complex systems is to comprise the essential phenomenology into an effective action for an ordering parameter field . for the hadron phenomenology near @xmath2 a chiral field with spontaneously broken @xmath3-symmetry has become a standard and successful tool , so it is natural to try to extend and apply this concept to the temperature region where the chiral phase transition is expected . in its most simple version the effective action then consists of the ( 3 + 1)dimensional @xmath3 linear @xmath4-model , with a suitable temperature - dependent @xmath1-potential for the spontaneous symmetry breaking . depending on the ratio between relaxation and cooling times , the dynamics of the ordering field then follows the cooling quench imposed on the temperature dependence of the potential . as in other cases of multi - component field ordering , it is characterized by formation and growth of misaligned domains separated by domain walls or other topologically nontrivial structures , depending on field and space dimensionality @xcite . numerical simulations in this classical framework have been performed and they confirm the transient formation of domains of false vacuum , in which the condensate is approaching its vacuum value while the direction of the aligned field still deviates from the surrounding vacuum @xcite . the possible influence of quantum- and thermal effects has been investigated , mostly in mean - field approximation @xcite . it has been suggested that pions emitted from these domains of disoriented chiral condensate ( dcc ) carry the misalignment in their isospin multiplicities , and in this way may provide a signature of the phase transition @xcite . a peculiar feature of the chiral @xmath3-field in 3 spatial dimensions is that its winding density may be identified with baryon density @xcite . this has led to a most interesting and remarkably successful model for baryon structure and dynamics at @xmath2 . the corresponding topological arguments @xcite have been used @xcite to estimate nascent baryon - antibaryon multiplicities in random chiral field configurations at high temperature ; of course , the first few time steps in an ordering evolution lead to an almost instantaneous reduction of the initially large and rapidly fluctuating winding densities . strictly , the topological conservation of the total winding number is limited to a field manifold where the point @xmath5 is excluded . in numerical simulations of field evolutions on a discrete spatial lattice , however , even topological conservation laws have to be enforced by constraint , because arguments based on continuity can not be applied . therefore there is no principal difference for implementing on a lattice the conservation of winding number in the nonlinear or linear realization of the @xmath3-symmetry . this allows to study formation of condensate and disoriented domains during the phase transition with simultaneously forming baryons plus antibaryons such that the total baryon number is fixed to any desired value . it is well known that topological textures play an important role in ordering transitions , so it is of special interest to study how the stabilization of extended baryonic structures and baryon number conservation may affect the growth laws of dcc domains , and possible dcc signatures in the emitted radiation . for this purpose it is necessary to include in the effective action terms which stabilize the spatial size of the baryons generated in the course of the ordering evolution . scaling arguments show that an additional four - derivative term is sufficient for soliton stabilization in 2- and 3-dimensional @xmath6 models . a different extension , the inclusion of the anomaly term , has been advocated to increase the probability of dcc - formation @xcite . the linear @xmath4-model ( even with explicite inclusion of quarks ) sometimes has been used @xcite to investigate the possibility of dcc - signals by considering spatially averaged fields , only . naturally , within such an approximation the effects we are investigating here , are lost . on the other hand , there is no immediate need to explicitly include fermionic degrees of freedom in our present approach . before embarking on a full - scale simulation of the ( 3 + 1)-dimensional @xmath3 case it is helpful to analyse the relevant questions in the ( 2 + 1)-dimensional @xmath0 model @xcite . it shares all essentials with the higher - dimensional case . although it represents an efficient tool for the description of 2-dimensional spin systems in its own right @xcite , we here discuss it mainly in view of its hadron physics generalization . that is , we denote the order parameter field as chiral meson field , we call the fluctuations pions with internal isospin components , and @xmath4-mesons , which aquire mass through spontaneous breaking of the chiral @xmath0-symmetry , and we call the domains with small values of the condensate and unbroken symmetry bags , in which the baryonic winding density is confined . although there is no compelling connection to fix the parameters of the model from its higher - dimensional analogue , we choose them in correspondence to simulate relative sizes and masses for baryons and mesons . various features of the static solutions of this model , and their formation in relaxation processes after a quench have been studied in @xcite . in the present investigation we discuss in more detail the different stages which characterize the long - time ordering evolution under dissipative dynamics , the dependence of the relevant times on model parameters , different types of symmetry breaking , and quench rates . our main aim , however , is an attempt to extract from the dynamics of the evolution the possibility of a dcc - signal . generally , the emission of radiation from a moving source is determined by quantum - field - theoretic amplitudes which require a separation of the time - dependent field into a classically moving part and quantum fluctuations . the result is commonly put into the statement that the intensity of pion radiation is proportional to the square of the classical pion field @xcite . this argument is the basis of the interest in dcc - pions . we analyse the validity of this statement under the assumption of dissipative dynamics , i.e. assuming that the energy loss of the relaxing and ordering classical configurations is carried away by propagating fluctuations . under this assumption we obtain the timing and strength and isospin distribution of dcc - pions from purely classical considerations . this general dynamical scheme is outlined in sect.2 , while the specifics of the model and the observables of interest are defined in sect.3 . the essential features of configurations during the course of the transition , the growth of angular correlations , decrease of numbers of textures , saturation of condensate and average orientation , formation of baryonic clusters , are presented in chpts.4,5 , and 6 , in dependence on parameters of the model , symmetry breaking , and cooling rates . finally , in sect.7 , we analyse the rates of energy loss with respect to possible dcc pion pulses , and extract the isospin distributions from large ensembles of events . conclusions are drawn in sect.8 . we consider an effective action for an @xmath0 vector field @xmath7 [ ] = ( t[]-u[])d^2xdt where the kinetic energy density @xmath8 $ ] comprises all terms containing time - derivatives of @xmath7 . we consider the evolution of field configurations , which at some initial time deviate from the global equilibrium configuration ( the true vacuum ) within some finite spatial region @xmath9 . the equations of motion @xmath10 which govern the classical evolution of the initial non - equilibrium configuration describe both , the approach towards the vacuum configuration in the interior of that spatial region and the propagation of outgoing ( distorted ) waves into the ( initially undisturbed ) surrounding vacuum . although , of course , the classical equations of motion conserve the total energy @xmath11 , the outgoing waves carry away energy from the interior of the spatial region in which the total energy @xmath12 initially was located . in quantum field theory the quantization of the propagating fluctuations describes multiparticle production of pions emitted from the relaxing field configurations . in a separation [ sep ] ( , t)=_cl(,t)+(,t ) where @xmath13 comprises all of the large amplitude motion of @xmath7 , and @xmath14 contains only small fluctuations around @xmath13 , expansion of @xmath15 $ ] to second order in @xmath14 @xmath16 = { \cal s}[\bphi_{cl } ] + \int \frac{\delta{\cal s}}{\delta\bphi(\bx',t')}|_{[\bphi_{cl } ] } \delta\bphi(\bx',t')d^2x'dt'\ ] ] [ exp ] + ( ,t)|_[_cl ] ( , t)d^2xdtd^2xdt + o(^3 ) and variation with respect to @xmath14 provides the classical expression for the fluctuating part [ fluc ] ( , t)=^(0)(,t ) + ( , t,,t)|_[_cl ] d^2xdt. the homogeneous part @xmath17 , after quantization , represents the scattering of field quanta off the classical configuration @xmath13 ( i.e. , these are on - shell distorted waves ) . the green s function @xmath18 of the operator @xmath19}$ ] relates emitted ( i.e on - shell ) radiation to the source term @xmath20}$ ] . there is , however , no unique way to separate the propagating fluctuations @xmath14 from a more or less smoothly evolving classical configuration @xmath13 because both , @xmath13 and @xmath14 , are integral parts of one and the same evolving order - parameter field @xmath7 . conclusions drawn from one part only are subject to the arbitrariness of the chosen separation . in any case , in an equation of motion that separately describes the evolution of @xmath13 we expect a dissipative term to account for the loss of energy through the emitted radiation : [ deom ] _ cl=/|_[_cl ] . for sufficiently small values of the relaxation constant @xmath21 the dissipative term dominates the time evolution of @xmath13 , propagating parts get damped away and we can replace ( [ deom ] ) by the time - dependent ginzburg - landau ( tdgl ) equation [ tdgl ] _ cl =- u/|_[_cl ] . the potential energy functional @xmath22 contains no time derivatives of @xmath7 , therefore @xmath13 has no propagating parts , i.e. it does not pick up field momentum , and it provides a very slowly moving adiabatically evolving classical background field . back - reaction of the fluctuations on the moving classical field can re - enter in different ways : as loop corrections they can lead to renormalization of the coupling constants in the functional @xmath22 which thus may become time- or temperature - dependent and drive the spontaneous symmetry breaking ; an appropriate noise term to be added on the right - hand side of ( [ tdgl ] ) accounts for stochastic interactions of the fluctuations with @xmath13 at finite temperature . ideally , through that procedure one hopes to achieve a separation ( [ sep ] ) such that the fluctuating parts @xmath14 contain no exponentially increasing amplitudes during the relaxation process . in the following we assume that the coupling constants which appear in the tdgl - equations have their renormalized physical values with an appropriate time- or temperature dependence and we will restrict the discussion to the consequences of those equations . evolutions via ( [ tdgl ] ) of individual initial @xmath13 configurations into final quasi - stable configurations are denoted as events , and we shall consider ensembles of events sampled with different initial configurations . the individual events show a variety of physically interesting features , textures , walls , disoriented domains , growth laws , which form the main body of the subsequent discussions . the lack of second - order time derivatives in ( [ tdgl ] ) means that also localized stable structures created in @xmath13 will not propagate but only drift in response to their mutual interaction . this is as expected because uniform motion of textures represents zero modes which may still be contained in @xmath23 . with ( [ tdgl ] ) the rate of decrease of the energy @xmath24 $ ] stored in the classical configuration @xmath13 then is [ udot ] _ cl(t)=-|_cl(,t)|^2 d^2x . if particle emission from the evolving source is the only mechanism for energy nonconservation this rate of energy loss has to match the energy flow carried away by the emitted radiation . this requirement determines the relaxation constant @xmath21 in ( [ tdgl ] ) which fixes the time scale for the approach of @xmath13 towards equilibrium . ( of course , external interference with the system like bjorken rescaling of the spatial metric may provide other sources of energy nonconservation . ) in quantum field theory , for field fluctuations coupled to a time - dependent classical source @xmath25 , the probability density for the emission of @xmath26@xmath27 field quanta with momentum @xmath28 and mass @xmath29 is given by n_=|s(,_k)|^2/(2_k ) where @xmath30 is the energy - momentum - space transform of the source , expanded in terms of the complete set of distorted scattering states with asymptotic momentum @xmath28 , and taken at the on - shell energy @xmath31 . in our present case according to ( [ fluc ] ) the time - dependent source is given by @xmath32}$ ] . consistent with our assumptions about the slow motion of the classical field @xmath13 we again may approximate here @xmath33 by @xmath34 . with @xmath13 obeying the tdgl equations ( [ tdgl ] ) the source field then is just the time - derivative of the classical field @xmath35 and we obtain for the average number of particles emitted with momentum @xmath28 and isospin component @xmath36 ( omitting the index @xmath37 ) [ nios ] n_i,=. with particle emission as the only mechanism for energy nonconservation the total energy @xmath38@xmath39@xmath40 carried away by radiation during the ordering process must balance the loss of energy stored in the classical field . with ( [ udot ] ) this requirement leads to [ econs ] |_cl(,_k)|^2 d^2k = |_cl(,t)|^2 d^2x dt . note that on the left - hand side of this equation only on - shell frequencies appear , while the right - hand side ( when written in fourier space ) contains an independent integral over all fourier components of the moving classical field . of course , together with ( [ tdgl ] ) , this relation is an implicit definition of the relaxation constant @xmath21 . for the adiabatically evolving process we consider the time @xmath41 as parameter such that we may define [ epsi ] _ i(t)=_i(,t)^2 d^2x = _ i^2(t ) as the energy carried away per time interval by particles emitted with field orientation @xmath36 . here `` @xmath42 '' denotes the lattice average over an @xmath43 lattice . due to the slow motion of the source we expect the particles to be mainly emitted with low energies , i.e. with low momenta @xmath44 and @xmath45 . in our present context we choose the @xmath0-symmetry to be spontaneously broken in 3-direction ( selected by the surrounding true vacuum boundary condition or by small explicit symmetry breaking , or by a small bias in the initial configuration ) . then the i=1,2 ( isospin ) components of the @xmath0 chiral field constitute two pionic fluctuational fields @xmath46 with small mass @xmath47 due to explicit symmetry breaking , while the @xmath4-fluctuation @xmath48 acquires a large mass @xmath49 due to the spontaneously broken symmetry . so we expect the emission rate for pions with isospin component @xmath50 [ ni ] n_i(t ) = during the early stages of the ordering evolution the spatial averages @xmath51 will be similar for both isospin directions , but at late times with the formation of larger disoriented domains they might differ appreciably . the relative abundancies [ fi ] f_i(t)= are free of unknown constants and could serve as dcc signature , if it is possible to separate in each individual event the small number of late time pions from the background of those produced during earlier stages , from decaying @xmath4 s , and from other sources . the features which are of physical interest in the chiral phase transition for a cooling hadronic gas in the 3 + 1 dimensional @xmath3 model can be explored and demonstrated in the lower - dimensional @xmath0 model . it is defined in terms of the dimensionless @xmath52-component field @xmath53 with @xmath54 , with the following lagrangian density in @xmath55 dimensions [ lag ] = _ ^- ( ^4 - 2 f^2 ^2 + 1 ) - ^2 _ ^ + m_^2 _ 3 . in addition to the usual linear sigma - model with standard @xmath1 potential for the modulus field @xmath56 this lagrangian contains a four - derivative current - current coupling @xmath57 for the conserved topological current [ top ] ^= ^ ( _ _ ) , which satisfies @xmath58 . in the present @xmath0 model this nonlinear coupling comprises the four - derivative skyrme-term and the sixth - order current - current coupling of effective chiral @xmath3-models . the independent strengths of the @xmath1- and the skyrme couplings are written in terms of one common dimensionless parameter @xmath59 , and a length @xmath60 which may be absorbed into the space - time coordinates . so , for @xmath59 fixed , @xmath60 sets the scale for the spatial extent of localized static solutions , in lattice implementations relative to the size of the basic lattice unit cell . the static solutions have been investigated and described in @xcite . for continuous coordinates ( or , on a lattice , for @xmath61 ) the total energy ( apart from the explicitly symmetry - breaking @xmath62 contribution ) of such localized solutions is independent of @xmath60 ( as long as @xmath60 is small as compared to the total lattice size @xmath63 ) . in the following we put the lattice constant to unity . if we consider the field @xmath7 as the basic order parameter to reflect the chiral phase transition as function of the temperature @xmath64 , then the factor @xmath65 which multiplies the square of the field vector in ( [ lag ] ) drives the spontaneous symmetry breaking . as the temperature goes to zero it is supposed to converge towards its vacuum value @xmath66 , and it should become very small @xmath67 as @xmath64 approaches the critical temperature @xmath68 . for very high temperature @xmath69 is supposed to become negative . a small symmetry - breaking term with strength @xmath62 has also been added to the otherwise o(3)-symmetric lagrangian ( [ lag ] ) . this explicit symmetry breaking is applied in @xmath52-direction of the internal space . if unequal zero it singles out the 3-component @xmath70 of the @xmath7-field as that field which aquires the nonvanishing condensate at @xmath2 , and we will denote this component as `` @xmath4''-field , while the @xmath71- and @xmath72-components represent the `` pionic '' ( goldstone- ) degrees of freedom of the remaining unbroken symmetry . it is convenient to have also for nonvanishing @xmath62 the minimum of the ( @xmath2)-potential located at the vacuum field configuration @xmath73 , such that the length @xmath56 of the field vectors approaches unity for @xmath2 also for nonzero symmetry breaking . this is achieved by subtracting @xmath74 from @xmath65 in the coefficient of the second order term in the potential @xmath75 in ( [ lag ] ) [ f2 ] f^2=f^2(t)- . this form shows directly how the explicit symmetry breaking supplies the pionic goldstone bosons with mass @xmath76 . for @xmath77 the temperature dependence of @xmath62 can be ignored . on the other hand , the mass of the @xmath4-field is given by [ sigmass ] m_^2=2f^2(t)+m_^2 which is mainly determined by the temperature dependence of @xmath78 . evolution of @xmath13 proceeds through the tdgl equations of motion as they result from the potential part of ( [ lag ] ) , ( i.e. also from the current - current coupling only the density term @xmath79 is kept ) . as no other time - derivative terms appear on the right - hand side , the constant @xmath21 which multiplies the left - hand side of ( [ tdgl ] ) can be set to one , i.e. in the following the relaxation time @xmath21 serves as time unit . omitting for convenience the index @xmath80 we have [ eom ] = _ i - ( ^2-f^2 ) _ i -^2 + m_^2 _ i3 + _ i. as far as the dissipative term originates from elimination of fluctuational modes which simulate a heat bath for the evolving field configuration , addition of fluctuational noise @xmath81 is required with a strength dictated by the dissipation - fluctuation theorem . such a term has also been added in ( [ eom ] ) . in lattice implementation , we impose periodic boundary conditions on the field configurations . this implies compactification of coordinate space to a torus @xmath82 . these boundary conditions preserve the global @xmath0 symmetry of the model ( for vanishing @xmath62 ) . having singled out the 3-direction as the @xmath4-direction in which the @xmath2 condensate is formed we can impose the stronger vacuum boundary conditions @xmath73 for all points on the lattice boundary throughout the whole evolution . this implies compactification of coordinate space to the two - sphere @xmath83 . in both cases the winding density @xmath84 satisfies @xmath85 with integer winding number @xmath86 . of course , these vacuum boundary conditions represent another way of explicit symmetry breaking which eventually drives the evolution of the condensate into the 3-direction . it is , however , a very weak form which is only active at the surface of the spatial region , unlike a nonvanishing @xmath62-term which is locally active at every point in space . with the equations of motion ( [ eom ] ) being purely first order in time derivatives it is sufficient to specify for initial conditions at @xmath87 only the field configurations themselves . as the time - dependent fluctuations are not part of @xmath13 the high - temperature initial configuration is , ( apart from a small optional bias in 3-direction due to an explicit symmetry breaker ) , @xmath88 in the interior of the spatial region in which the hot chiral field is located and takes the true vacuum values on its boundary . the first step in the time evolution is therefore governed by the stochastic force @xmath81 alone . but this first step will immediately create a configuration @xmath13 which is of stochastic nature itself . therefore , choosing each of the three cartesian field components at each interior point of the lattice from a random gaussian deviate , and fixing @xmath89 on the boundary of the lattice , provides convenient initial configurations with initial spatial correlation length less than one lattice unit . from these random configurations , those with the desired total winding number @xmath86 are selected and propagated in time . during the evolution of each individual event the fluctuational noise @xmath81 at every time step and every lattice point is picked from an appropriate gaussian deviate . a large number of events generated in this way constitute the statistical ensemble from which ensemble averages can be obtained at some point in time during the evolution . for configurations where the orientation of the vector field @xmath7 shows strong local variations ( as it happens near localized textures or throughout random initial distributions ) the implementation of the winding density @xmath84 on the lattice in the differential form ( [ top ] ) is not sufficiently accurate to add up to integer total winding number @xmath86 . therefore @xmath84 and its functional derivatives have to be evaluated according to the geometrical meaning of @xmath84 as surface area of spherical triangles cut out on the two - sphere @xmath83 by the image of elementary triangles in coordinate space . details of this procedure have been discussed in @xcite . it allows to assign a value @xmath90 of the winding density @xmath84 to each point @xmath91 of the square lattice ( or , more precisely , to each lattice cell with lowerleft corner @xmath91 ) such that the total winding number b=_i , j=0^l-1 ( i , j ) summed over the whole lattice is integer and initial configurations can be selected with some desired integer value of @xmath86 . we also define [ d ] d=_i , j=0^l-1 |(i , j)| by summing up the absolute values of the local winding densities . of course , for random or slowly varying smooth configurations @xmath92 generally is not an integer . however , if a configuration describes a distribution of localized textures ( and antitextures ) which are sufficiently well separated from each other , then @xmath92 is close to an integer and counts the number of these textures ( plus antitextures ) . in that case we can define the numbers @xmath93 of `` particles '' and `` antiparticles '' through b = n_+ - n_- d = n_++n_- . in the following we will briefly call @xmath92 the `` particle number '' , even if it is not integer and comprises textures and antitextures . in a lattice implementation the local update of @xmath94 at some timestep at some lattice point @xmath91 will occasionally lead to a discrete jump in the total winding number @xmath86 . for well - developed localized structures this corresponds to unwinding textures or antitextures independently , such that @xmath86 decreases or increases by one or more units . this will eventually happen even for implementations of the nonlinear @xmath0-model where the length @xmath56 is constrained to unity and @xmath86 is topologically conserved , because the topological arguments based on continuity do not apply to the discrete lattice configurations . it can be prevented by an ( optional ) @xmath86 filter which rejects such a @xmath86-violating local field update at that lattice point and timestep . this eliminates all independent unwinding processes . only simultaneous annihilation of texture and antitexture in the same time step remains possible , and , as the update proceeds locally at each lattice vertex it can happen only if texture and antitexture overlap . this @xmath86-conserving evolution is characteristic for the nontrivial topology of the nonlinear @xmath0-model and in this way can be implemented as an optional constraint also into the linear @xmath0-model . in the continuum limit it implies that a vanishing modulus @xmath5 of the field vector is excluded . in order to get a feeling for a reasonable choice of the two essential parameters @xmath60 and @xmath59 in ( [ lag ] ) we study at first the simple case of a sudden quench without any symmetry breaking , neither explicit nor via boundary condition , the so - called critical quench . that is , we put @xmath95 and impose only periodicity for otherwise arbitrary field vectors at the boundary of a sufficiently large @xmath43 lattice . the sudden quench is realized by exposing the initially ( at @xmath87 ) random configuration to the @xmath2-value of @xmath96 for all times @xmath97 . for this zero - temperature evolution the noise term @xmath81 is also not present . as an example fig.[ell2 ] shows one individual evolution of a configuration with fixed winding number @xmath98 , on a lattice with @xmath99 , with @xmath100 and @xmath101 . as significant observables we show the time - dependence of total energy @xmath102 , particle number @xmath92 , field modulus @xmath103 averaged over the whole lattice , and the angular correlation length @xmath104 . the latter is defined as the ( interpolated ) half - maximum distance @xmath105 of the equal - time angular correlation function [ corr ] c(r)=_i , j=0^l_k , l=0^l ( i , j)(k , l ) /_i , j=0^l_k , l=0^l 1 ( with the @xmath106 sum restricted such that the distance @xmath107 between lattice vertices @xmath91 and @xmath108 lies inside bins of unit size around fixed positive integers @xmath109 ) . note that the definition ( [ corr ] ) contains only the unit vectors @xmath110 , and not the full field vectors @xmath7 , i.e. @xmath111 measures only the angular correlations in a given configuration . as discussed in @xcite this definition of @xmath104 serves well to provide an average radius of oriented domains . fig.[ell2 ] follows the evolution over a very long time ( up to @xmath112 relaxation - time units ) and clearly shows the different phases of the ordering process . up to @xmath113 the average @xmath103 stays ( or becomes ) small ( @xmath114 ) , while the particle number @xmath92 decreases by almost one order of magnitude due to local angular reorientation , also reflected in @xmath104 which increases approximately as @xmath115 . once the oriented domains have reached sizes which cover several lattice units the roll - down phase sets in where @xmath56 approaches unity within these domains ( while remaining small at their boundaries where strong angular variations prevail ) . this roll - down phase takes a few tens of time units and it is accompanied by a significant energy reduction and a more rapid increase of @xmath104 , such that by the end of this roll - down the oriented domains cover areas with typical radii of 10 lattice units . during the end of , and after this phase the isolated textures assume their stable shapes with the appropriate bag and angular field profiles . this is accompanied by further angular ordering of the topologically trivial field configurations in which these textures are embedded , which proceeds by moving boundaries between differently oriented domains . it leads to further decrease of @xmath102 and increase of @xmath104 , which follows very nicely the classical @xmath116 allen - cahn law @xcite . this phase lasts up to several hundreds of time units , without further reduction of the particle number @xmath92 . due to the existence of the remaining stable bags the average @xmath56 does not reach the vacuum limit @xmath117 and due to the floating angular textures the increase in @xmath104 begins to saturate after @xmath118 . on a timescale which is still larger by another order of magnitude the stable isolated textures may meet and annihilate as can be observed in fig.[ell2 ] near @xmath112 . fig.[ell2 ] is obtained for @xmath86 constrained to @xmath98 and for scale @xmath119 , such that resulting bag structures extend only over a few lattice cells . the long - time behaviour after the roll down phase is higly sensitive to the size of the developing stable textures , i.e. to the fraction of lattice space occupied by the bags because , evidently , overlapping textures quickly combine or annihilate . their size is governed by @xmath60 , and in the range up to @xmath120 the structures created are reasonably isolated . we shall see later that for the point of view of dcc signatures the long - time behaviour after the roll - down is irrelevant . defining the typical roll - down time @xmath121 as that point in time when the lattice averaged @xmath103 has reached a value of 0.8 , we plot in fig.[trdfig ] @xmath121 as function of @xmath60 and @xmath59 , respectively , for 15 events each . for @xmath86 constrained to @xmath98 the average values follow power laws @xmath122 and @xmath123 with good accuracy . these numerical results show that the roll - down time @xmath121 scales as @xmath124 which means that it is essentially determined by the potential term @xmath125 alone . it is almost unaffected by the fourth - order derivative term which scales with the product @xmath126 , while the second - order gradient terms in @xmath22 cause only a mild increase of the scaling exponent to the observed value of 1.2 . we interpret the stable textures as baryons floating in the effective meson pion field . fixing their spatial extent in terms of lattice units through the choice of @xmath60 provides the physical scale for the lattice constant . we shall typically choose @xmath119 such that baryons carrying one unit of winding number ( = baryon number ) acquire a radius for their topological density of about two lattice constants . with a typical nucleon radius of 0.6 - 1.0 fm , for @xmath119 , the lattice constant then corresponds to @xmath127 fm . for an accurate description of the texture and bag profiles of individual baryons such a small value @xmath119 is not really sufficient , as has been discussed in detail in the investigation of static solutions in @xcite . but within the present context this aspect is not essential . for physical vacuum boundary conditions this choice also suggests the typical lattice size @xmath63 , which should be sufficiently large to cover the spatial extent of the high - temperature area after a typical hadronic ( heavy - ion ) collision . so values of @xmath128 seem appropriate , and they are technically easy to handle . through ( [ sigmass ] ) the value of @xmath59 is directly tied to the @xmath4-mass . for @xmath119 the choice @xmath101 leads to @xmath129 ( in inverse lattice units ) , which appears quite reasonable , being of the order of a typical baryon mass . in the discussion in @xcite we have seen that for @xmath101 the stable textures are confined into the interior of well - developed bag structures which we consider an essential feature of physical baryons . for all these reasons we shall in the following present results mostly for @xmath119 and @xmath101 . in order to obtain the roll - down times for that case we extract from fig.[trdfig ] the values of @xmath121 for this two - dimensional @xmath0-model [ trd ] t_rd6 ( ) ^1.2 . for @xmath119 , @xmath101 this is about 14 relaxation time units . to guide the eye for the increase of the correlation length @xmath104 we have included in fig.[ell2 ] the power law @xmath130 as straight dashed line . although there are typical deviations prior and during the roll - down phase it is remarkable that this classical scaling expectation serves well to describe the growth of oriented domains for this rather complicated interplay of forming and stabilizing extended textures and aligning domains . apparently , the use of this growth law for estimating domain sizes seems justified throughout the whole time period relevant for dcc signatures . it is interesting to compare these results with the plain linear @xmath4-model . for that purpose we omit in ( [ lag ] ) the current - current coupling and follow the evolution without putting a constraint on the winding number @xmath86 . fig.[linsig ] shows two events evolving from identical initial configurations on a 100@xmath131100 grid with vacuum boundary conditions . after about 200 relaxation time units the unconstrained linear sigma model has lost all particles and antiparticles ( @xmath132 ) , and the configuration tends towards global alignment in 3-direction . on the other hand , the full model with fourth - order skyrme term included and winding number constrained to @xmath98 , finally stabilizes solitons with @xmath92 of the order of 100 particles plus antiparticles . comparing the growth rates of the angular correlation lengths @xmath104 for both evolutions it may be seen that there is no dramatic difference in the overall growth exponent . still , it is evident that especially during the roll - down phase the emerging extended stabilizing textures do present an obstacle against the free and unconstrained ordering process such that the average size of the ordered domains in the full model is noticeably smaller than in the plain linear @xmath4-model evolution . at this point we may add a few remarks concerning the relaxation constant @xmath21 which in the tdgl - equation ( [ eom ] ) we have simply used as a time unit , because no other propagating terms are present . of course , we could use ( [ deom ] ) instead of ( [ tdgl ] ) for simulating evolutions of the classical field . keeping the propagation velocity at @xmath133 the time unit then is set by the lattice constant . a trivial consequence is that all features which reflect the relaxation , like growth rates or saturation times , are scaled with powers of @xmath21 . e.g. , for sufficiently small @xmath21 the roll - down time @xmath121 scales like @xmath134 . it turns out that for @xmath135 the evolutions resulting from ( [ deom ] ) and ( [ tdgl ] ) essentially coincide . for @xmath136 the propagating waves start showing up in the field configurations and cause oscillatory features in ( lattice- ) averaged quantities without strongly affecting the late - time evolution . for @xmath137 oscillations dominate and due to the nonlinear terms stabilization becomes difficult . so , to be on the safe side for use of the tdgl equation ( [ tdgl ] ) , @xmath21 should be less than 1 . looking once more at fig.[linsig ] , @xmath138 ( in units of @xmath21 ) implies that for @xmath135 the roll - down is completed after travelling waves have passed about 100 lattice units . the assumption of a sudden quench which underlies the ordering features shown in the previous chapter represents that limit for the actual cooling process where the cooling time @xmath139 ( during which the temperature drops from @xmath140 to @xmath2 ) is much smaller than the relaxation time which governs the evolution of the field configurations . this instantaneous cooling , where the hot initial configuration for @xmath97 is immediately exposed to the zero - temperature @xmath2 effective potential defines the most off - equilibrium scenario which we can study within this model . to weaken this rather drastic assumption we replace it with an appropriate time dependence of the temperature @xmath141 . then also the temperature dependence @xmath78 in ( [ f2 ] ) has to be specified . the function @xmath78 in principle is determined from loop contributions and their renormalization group resummation in finite - temperature field theory . perturbative evaluations are restricted to the vicinity of @xmath2 ; the complete form of @xmath78 depends on the specific model and is not accurately known . its essential features , however , are a slow decrease from its @xmath2 value @xmath142 for @xmath143 , followed by a steeper decrease @xmath144 as @xmath64 approaches @xmath68 . for the simulations these essentials are sufficient ; the specific form ( generally in form of logarithms ) is not important . a very convenient ansatz which comprises these basic features in the region @xmath145 is [ f2 t ] f^2(t)= ( 1-(t / t_c)^))^2 with @xmath146 , and @xmath147 for @xmath148 . a slow cooling scenario can be realized by externally imposing the dependence @xmath141 of temperature on time . for a linear quench we take [ quench ] t(t)/t_c= 1-t/_c with quench time @xmath139 , and @xmath2 for @xmath149 . a crucial feature of the ordering process is that the early phase prior to the onset of the roll - down is dominated by angular aligning driven by the gradient terms in the @xmath4-model part of ( [ lag ] ) while the effective potential for @xmath56 is of only minor importance . as a consequence , the angular ordering up to the onset of the roll - down proceeds quite independently of the actual cooling rate , with @xmath150 in any case this is fortunate because it renders knowledge of the precise form of @xmath141 and @xmath78 unimportant and we can safely adopt convenient ansaetze like ( [ f2 t ] ) and ( [ quench ] ) . for a very slow quench @xmath151 , after this first angular aligning phase , the roll - down then only proceeds up to the momentaneous value of @xmath65 , @xmath152 . specifically , the stable textures are being created with profiles which correspond to an instantaneous scale [ ell ] ( t)= f^-1(t(t ) ) as can be seen from rescaling the modulus of the order parameter to @xmath153 in ( [ lag ] ) . with @xmath154 the nascent stable objects in the slow quench are much larger than for the sudden quench , i.e. fewer of them survive as separate isolated textures . subsequently , further ordering proceeds essentially as equilibrium evolution where the stable objects shrink according to @xmath155 in ( [ ell ] ) until they finally assume their @xmath2 size with accompanying rearrangement of surrounding aligned domains . as intermediate case we consider evolutions where the cooling time is comparable with the roll - down time @xmath121 in ( [ trd ] ) , conveniently again as linear quench with @xmath156 for @xmath119 and @xmath101 . in the very early phase the evolutions closely follow the slow quench with very small @xmath103 and steeper decrease of the particle number @xmath92 as compared to the sudden quench , but as @xmath41 approaches @xmath139 the roll - down proceeds rapidly towards the sudden quench value @xmath157 , with corresponding pronounced loss in energy @xmath102 and typical increase of @xmath104 . for @xmath158 the number of textures with @xmath2 profiles stabilizes at values intermediate between sudden and very slow quench and the angular ordering of aligned vacuum domains proceeds along the classical @xmath116-law . fig.[tau ] shows three typical evolutions with @xmath119 and @xmath101 and total winding number constrained to @xmath98 , starting from the same initial configuration , for @xmath159 ( very slow ) , @xmath160 ( intermediate ) , and @xmath161 ( sudden ) quench . fig.[ctr ] presents bag contours at @xmath162 of the stable textures with varying individual winding numbers ( summing up to @xmath98 ) which have been established after @xmath163 for these three different cooling rates . it is quite evident from fig.[tau ] that the rate of increase of the correlation lengths @xmath104 before and after the roll - down ( @xmath164 ) is unaffected by the cooling rate . differences in the angular ordering , however , show up in the evolving particle numbers which reflect the changing effective potential through the size of nascent textures . for these slow quenches which proceed with finite cooling rate evolutions are subject to fluctuational noise given by the last term in ( [ eom ] ) . its variance depends on the physical nature of the dissipative @xmath165 term in ( [ eom ] ) , i.e. , to which extent it originates purely from particle emission , or from kinematical expansion of the spatial volume , or from a heat bath of eliminated high - frequency modes . in any case , its strength according to the fluctuation - dissipation theorem decreases linearly with temperature @xmath64 . its general influence on the evolving configurations is very transparent : for finite @xmath64 it will tend to disturb the ordering process at each time step . this results in a slight decrease of the slopes of those observables which reflect the ordering , specifically the correlation length @xmath104 , the average @xmath103 , the particle numbers @xmath92 , the total energy @xmath102 . on the other hand , fluctuations support ( brownian ) motion of nascent textures which combine more easily into larger structures with increased individual winding numbers . so , evolutions subject to noise finally produce increased values of @xmath92 distributed into larger clusters of positive or negative winding number . naturally , only very slow quenches are noticeably affected by the noise . for definiteness , the evolutions shown in figs.[tau ] are obtained from ( [ eom ] ) with the last term @xmath81 omitted completely . effectively , inclusion of noise produces evolutions which resemble those for quenches with slightly reduced quench time . as we have seen , the physically most relevant characteristic of the ordering evolution is the roll - down time where the bag field @xmath56 approaches within aligned ( but still mostly disoriented ) domains the momentaneous value of @xmath166 and isolated textures are being established near their borders . this important feature is most clearly brought forth for @xmath167 therefore in the following we preferably will consider ordering processes after instantaneous cooling . explicit symmetry breaking of the global @xmath0-symmetry is of crucial importance for the ordering process . its weakest form is imposed through boundary conditions where a specific vacuum orientation , e.g. @xmath73 , along the lattice boundary is enforced . for physical processes this , apparently , is also a most natural condition . with increasing time after a sudden quench the influence of this surrounding vacuum will grow deeper and deeper into the randomly ordering interior and eventually reorient aligned domains along the chosen vacuum direction . clearly , the strength of this aligning effect depends on the finite size @xmath63 of the lattice . for @xmath63 of the order of 100 it does not change the essential features of the evolution . there is only one ordering parameter , the absolute value of the averaged direction = _ i , j=0^l-1 ( i , j ) /l^2 . which is sensitively affected . for arbitrary periodic boundary conditions this observable remains very small with erratic variations for times long after the roll - down , while for the symmetry - breaking vacuum boundary conditions , it continually grows in accordance with the growth of the angular correlation length . it is saturated by the averaged 3-component @xmath168 which carries all of the total @xmath169 strength . so , as expected , the oriented domains growing in the vicinity of the boundary contribute coherently to the magnetization and align in vacuum direction . explicit symmetry breaking through a nonvanishing @xmath62-term in ( @xmath170 ) exerts much stronger influence on the ordering process than imposing symmetry - breaking boundary conditions only . we compare in fig.[csb ] evolutions under vacuum boundary conditions for values of @xmath171 , @xmath172 , and @xmath95 . this appears as a reasonable range for pions with mass of about 10% of the @xmath4-mass . for @xmath171 the roll - down of @xmath103 is almost unaffected as compared to @xmath95 , while @xmath173 accelerates the roll - down by a factor of two ( with a correspondingly larger number of stable textures created ) . both evolutions with @xmath174 show the exponential approach of @xmath175 to saturation expected for the off - critical quench , in contrast to the unbroken case . ( for infinite systems the exponential approach to saturation is governed by @xmath176 @xcite ) . the saturation value is smaller than one due to the presence of stable textures . in both cases @xmath169 is completely saturated by @xmath168 . it may also be noted from fig.[csb ] that for @xmath173 saturation of @xmath168 is achieved long before the roll - down of @xmath103 is completed , while for @xmath177 the order is reversed . it should be noted that the angular correlation function @xmath111 in the off - critical system approaches @xmath178 for @xmath179 . this shows up as an abrupt increase of the half - maximum distance @xmath104 , as soon as the average @xmath180 has grown beyond @xmath181 of its saturation value . before that sudden increase , @xmath104 already starts to deviate from the growth law of the critical system , because it reflects also the increasing global alignment in 3-direction . so , in the off - critical system , it appears more appropriate to consider only the angular correlations of the transverse ( @xmath50 ) components as a measure for the size of disoriented domains . additional breaking of the @xmath0-symmetry occurs through an initial bias in the starting configurations . it originates from the asymmetry introduced into the initial @xmath182 ensemble through the @xmath183 symmetry - breaking term in ( [ lag ] ) . it will not alter the exponent of the exponential approach to saturation , but it will reduce the factor in front of the exponential . this causes another considerable reduction of saturation times . as we discussed in sect.2 signatures of the evolving disoriented domains may be carried in the fractions @xmath184 of pion multiplicities of kind @xmath36 emitted at late time during the ordering process . we will investigate these signatures for an ensemble of events evolving after a sudden quench . we compare results for the plain linear @xmath4-model , ( i.e. omitting the four - derivative current - current coupling in ( [ lag ] ) , without constraint on winding number @xmath86 ) , with the corresponding evolutions for the full lagrangian ( [ lag ] ) , with @xmath86 constrained to @xmath98 , which lead to stabilized solitons floating in the surrounding vacuum . for the parameters in the energy functional we choose @xmath119 and @xmath101 , such that the emerging solitons develop deep bags and extend over several lattice cells . with this choice the mass of the ( t=0 ) @xmath4-fluctuations according to ( [ sigmass ] ) is @xmath185 . let us at first look at the rates @xmath186 ( [ epsi ] ) of energy carried away by the field components @xmath187 for the pure linear @xmath4-model with no explicit symmetry breaking , @xmath188 . the direction of spontaneous symmetry breaking then is fixed by the surrounding vacuum alone , which is imposed through the condition that the chiral field is kept at @xmath73 at all times on the boundary of the lattice . for these events the roll - down phase where the average bag field @xmath189 grows from a small value to saturation @xmath190 covers the time intervall from @xmath191 to @xmath192 in units of the relaxation time @xmath21 . during this time interval the correlation lengths @xmath104 grow from values near @xmath193 to @xmath194 . these growth rates are quite independent of the lattice size @xmath63 , such that the average number of disordered domains by the end of the roll - down phase is of the order of @xmath195 . in figs.[eps ] we show for some arbitrarily selected individual event the rates @xmath186 of emitted energy for two lattice sizes , @xmath196 and @xmath197 . due to the symmetry breaking effect of the surrounding vacuum , emission into the @xmath198 @xmath4-direction dominates the pionic @xmath50 emission rates during most of the roll - down phase by one to two orders of magnitude . naturally , this feature is more pronounced in the smaller lattice . only after completion of the roll down , when the absolute emission rates are getting very small , the average rates become again comparable for @xmath187 . while @xmath199 is essentially monotonously decreasing during the whole ordering process , the pionic emission occurs in two well - separated pulses . an early , short pulse with high intensity is emitted for times @xmath200 and it is characterized by equipartition @xmath201 . this is as expected because at those times the angular correlation lengths @xmath104 are near or less than 1 . the second pulse appears during the second half of the roll - down phase ; its maximum is at least one order of magnitude smaller than that of the first pulse and it is reached near @xmath202 . for the larger lattice also this second pulse is characterized by equipartition for both isospin components . however , for the smaller lattice , a significant difference between @xmath203 and @xmath204 can be observed . it develops with the onset of the roll down and extends over the whole second pulse of pion emission and thus constitutes a possible dcc signature . unfortunately , it is just in the small lattice where the pionic signal is strongly dominated by @xmath4-emission . ( counting particles instead of energy , this is , however , compensated by the mass ratio ) . it is instructive to analyse the two pion pulses in terms of the tdgl - equation ( [ eom ] ) which relates ( in the pure linear @xmath4-model with @xmath188 ) @xmath205 to @xmath206 . as expected , the first pulse is mainly due to the angular ordering effectuated by @xmath207 , while the second pulse is dominated by the potential term . so , the simple approximation @xmath208 provides the essential features of the second maximum , ( the dcc pions ) . we may even go further and replace @xmath209 by its lattice average @xmath210 , such that the abundance ratios ( [ fi ] ) for dcc pions are approximated by [ fpi ] f(_i ) . in this approximation we recover the commonly used prescription @xcite to take the strength of pion emission proportional to the square of the classical field components . ( evidently , this works only for ratios , and only for pions in the second peak ) . based on this approximation , we may derive simple expressions for the probabilities @xmath211 , to observe in a large ensemble of configurations with @xmath212 disoriented domains the ratios @xmath184 of dcc pion multiplicities ( see appendix ) . as discussed in the previous chapter , addition of explicit symmetry breaking through a small pion mass like @xmath213 speeds up the roll - down phase and the aligning of @xmath7 into the 3-direction and thus shifts the dcc pions peak to slightly smaller times without affecting the ratios . however , the important difference is that for @xmath214 the emission rate for the 3-component is strongly enhanced relative to the dcc pions , such that also for a large lattice ( @xmath197 ) during and beyond the whole roll - down phase @xmath199 dominates the dcc pions by more than an order of magnitude ( for @xmath213 ) . ( for the small lattice ( @xmath196 ) @xmath199 dominates by two orders of magnitude ) . the fourth - order skyrme term plays a minor role during these early stages of the ordering process . its size stabilizing effect for the emerging solitons affects the late stages after completion of the roll down . still , as we have discussed at the end of sect.4 , already the emergence of stable extended structures during the roll - down phase presents an obstacle against the growth of large aligned domains and thus it counteracts large differences in the ratios @xmath215 and @xmath216 of dcc pions . this becomes apparent by comparing the probability distributions @xmath217 obtained from a large ensemble of 3000 events on a 30@xmath13130 lattice ( fig.[pf30 ] ) , for the plain unconstrained linear @xmath4-model , and the full model ( [ lag ] ) with constraint @xmath98 . we sample into 30 bins the ratios @xmath218 extracted from 3000 evolutions at the time when @xmath219 grows beyond 0.8 , which approximately marks the peak in the dcc - pions pulse . comparing with linearly mixed @xmath211 taken from ( [ pnu ] ) for two neighbouring values of @xmath212 , we may conclude that the presence of stable textures increases the average number of disoriented domains present at that point in time ( for the fixed lattice size @xmath196 ) from 2 - 3 to 4 - 5 . as we have discussed above , this average number of @xmath212 increases with @xmath220 . fig.[pf60 ] shows the corresponding distributions for 3000 events on a @xmath221 lattice . the distribution obtained from the linear @xmath4-model is very nicely reproduced by @xmath211 with @xmath222 , while the full and constrained to @xmath98 evolutions are compatible with @xmath223 , which again shows the strong obstruction of domain growth through the creation of stable textures . still , it should be remembered that the @xmath211 of ( [ pnu ] ) which we use for comparison are derived under the rather poor approximation ( [ fpi ] ) and the assumption that all domains are of equal size . to illustrate the influence of the cooling rate for slow quenches figs.[slow ] show two evolutions with cooling time @xmath224 as discussed in sect.5 . with this choice the cooling time is large as compared to the roll - down time after a sudden quench . the decisive feature of the evolution on a small 30@xmath13130 lattice is that the saturation of @xmath225 is almost completed before the begin of the roll - down of @xmath219 . consequently , the dcc pion pulse which still occurs during the late roll - down phase , is suppressed by almost four orders of magnitude as compared to @xmath4-emission @xmath199 . on this small lattice the disoriented domains show up in the pion emission ratios , while on a larger ( 80@xmath13180 ) grid , the saturation of @xmath225 is delayed , the height of the dcc pion pulse increased , but it is characterized by equipartition between @xmath203 and @xmath204 . so , also in this respect the sudden quench provides the better conditions for a dcc signal . in this paper we have investigated the phase ordering of a 2d-@xmath0 model under dissipative dynamics . the peculiarity of the underlying lagrangian is that it stabilizes extended textures through a fourth - order current - current coupling . this is introduced to get an idea how emerging topological solitons with stable sizes interact with the growth of aligned domains . the winding number can be kept fixed to simulate in this model the analogy of baryon number conservation which is an important feature of models of hadronic systems . two parameters characterize the model : one fixes the spatial size scale of the stabilized baryons , the other determines the depth of the bag profile , into which the winding ( baryon- ) density is confined . shallow bags correspond basically to the nonlinear version of the model , deep bags realize the unbroken chiral symmetry in the interior of baryons , as it would be expected in hadronic systems . apart from studying the influence of different ways of symmetry breaking , and the effects of different quench rates on the ordering observables , we are especially interested in the possibility to find dcc signatures in the emitted particle radiation . all results are based on the assumption that the relaxation proceeds on a sufficiently slow time scale , such that it provides an adiabatically changing source for the emitted radiation . therefore its motion is treated classically , and the amount of radiated energy is simply obtained from the classical energy loss of the source . quantum field - theoretic amplitudes enter only into the relaxation constant @xmath21 which , however , has not been determined , but simply used as time unit for the ordering dynamics . from the results obtained in this framework we may draw the following conclusions : the classical allen - cahn growth law @xmath226 for disoriented domains is not seriously disturbed by the stabilizing terms in the lagrangian and the emerging stable textures , although there are characteristic deviations from it during the roll - down phase where the separation into domains of false vacua and deep bags mainly takes place . the factor in front of the @xmath116 power law , however , is decreased as compared to the pure unconstrained linear @xmath4-model , such that the average size of disoriented domains after completion of the roll - down is reduced . the roll - down time varies with the two parameters of the model essentially as expected from simple scaling arguments . it represents the crucial characteristic of the ordering process , it is not very sensitive to small explicit symmetry breaking and varies with the quench rate in the expected way . explicit symmetry breaking strongly affects the saturation time of the average orientation , the factor in front of the exponential approach to saturation depends severely on the ( random ) initial conditions of each individual event . the number of baryons ( plus antibaryons ) finally surviving the ordering process is determined by the time of the onset of the roll down . the angular correlation length at that time measures the average distance of regions with high winding density which develop into separate stable structures with integer winding number during and after the roll down phase . therefore slow quenches lead to low baryon density . the intensities of emitted radiation in our approach is determined by the square of the time - derivative of the classical field , averaged over the spatial volume ( or area ) , and not by the square of the classical field itself . this allows to identify two well - separated radiation pulses for the transverse ( pionic ) components of the chiral field . the later one , emitted towards the end of the roll - down phase carries the signature of the disoriented domains established by that time , so it constitutes the dcc signal . with the growth law of the angular correlations fixed , the average size of disoriented domains at the time of completion of the roll down is also fixed . therefore the average number of disoriented domains on an @xmath227 lattice which contribute to the dcc signal increases like @xmath220 , and deviations from equipartition in the intensities carried by the different isospin directions quickly vanish with increasing @xmath63 . on the other hand , for small @xmath63 , the pion signal is buried under a large pulse of emission in longitudinal ( @xmath4-)direction , because the explicit symmetry breaking effect of the surrounding true vacuum is strong for small lattice sizes . the broadening of the probability distributions for the pionic intensity ratios constitutes the dcc signature . we have extracted it from large ensembles of events under the assumption that these pions , emitted at a specific point in time ( or averaged over a short time interval near the roll - down time ) can be observed separately from the bulk of other pions and @xmath4 s emitted . we find that in those cases where a noticeable broadening occurs , the relative amount of energy carried away in the dcc pulse , is suppressed by orders of magnitude . of course , quantitative statements require knowledge of the parameters in the model , cooling rates , physical size of the lattice as compared to correlation length , baryon size . in the present model , all these conditions remain more or less arbitrary , and we have chosen them at will to illustrate the main features . however , for application to the ordering dynamics of a hadronic chiral @xmath3-field , we have quite specific ideas about the relevant parameters in the effective actions . the exponent in the classical allen - cahn scaling law is the same for ( 2 + 1)d-@xmath0 and ( 3 + 1)d-@xmath3 models ; the homotopy groups @xmath228 and @xmath229 are isomorph , both characterized by integer winding numbers . so , apart from differences in dimension - dependent cooling rates and temperature - dependence of the condensate @xmath166 , we do not expect dramatic qualitative differences in the ordering features , and we hope that extending the present considerations to the ( 3 + 1)d-@xmath3 model will serve to sharpen our expectations with respect to the observability of dcc signals in hadronic processes . the author appreciates helpful discussions with h.walliser . consider a configuration where the classical field is uniformly aligned into some direction , thus forming one single disoriented domain , with the field components ( @xmath187 ) parametrized as [ orient ] _ i=1,2,3= ( , , ) . then , within the approximation ( [ fpi ] ) , we find for the fraction of pions emitted with isospin component @xmath230 , relative to the number of all pions @xmath231 [ fp1 ] f(_1 ) = ^2 . in an ensemble of events where all orientations of @xmath232 are equally probable the ensemble average @xmath233 of @xmath218 is , of course , @xmath234 . the probability @xmath217 to find in one event of that ensemble the fraction @xmath235 ( for pions of either kind @xmath36 ) then is obtained from f = _ 0 ^ 2f d=_0 ^ 1 f p(f ) df as [ pf ] p(f)=. ( note the difference to the standard chiral @xmath3-model , where @xmath236 . ) for @xmath212 such domains of size @xmath237 the probability @xmath211 to find the fraction @xmath235 in a given event then is [ pnu ] p_(f)=_0 ^ 1 ... _0 ^ 1 ( f- ( f_1+ ... +f _ ) ) p(f_1) ... p(f _ ) df_1 ... df_with @xmath217 given by ( [ pf ] ) . some of these functions , for values from @xmath238 to @xmath239 , are shown in fig.[pfig ] . for increasing @xmath212 they are , of course , more and more sharply peaked around @xmath240 . k. rajagopal and f. wilczek , _ nucl.phys . _ * b404 * ( 1993 ) 577 ; s. gavin , a. gocksch , and r.d . pisarski , _ phys.rev.lett . _ * 72 * ( 1994 ) 2143 ; g. amelino - camelia , j.d . bjorken , and s.e . larsson , _ phys.rev . _ * d56 * ( 1997 ) 6942 ; m. asakawa , z. huang , and x .- n . wang , _ phys.rev.lett . _ * 74 * ( 1995 ) 3126 ; a.k . chaudhuri , hep - ph/0007332 ; j. serreau , _ phys.rev . _ * d63 * ( 2001 ) 054003 . d. boyanovsky , d .- s . lee , and a. singh , _ phys.rev . _ * d48 * ( 1993 ) 800 ; d. boyanovsky , h.j . de vega , and r. holman , _ phys.rev . _ * d51 * ( 1995 ) 734 ; f. cooper , y. kluger , e. mottola , and j.p . paz , _ phys.rev . _ * d51 * ( 1995 ) 2377 ; m.a . lampert , j.f . dawson , and f. cooper , _ phys.rev . _ * d54 * ( 1996 ) 2213 ; j. randrup , _ phys.rev.lett . _ * 77 * ( 1996 ) 1226 ; _ phys.rev . _ * d55 * ( 1997 ) 1188 ; _ heavy ion phys . _ * 9 * ( 1999 ) 289 ; _ phys.rev . _ * c62 * ( 2000 ) 064905 . phys.lett . _ * b217 * ( 1989 ) 169 ; a.a . anselm and m.g . ryskin , _ phys.lett . _ * b266 * ( 1991 ) 482 ; j.p . blaizot and a. krzywicki , _ phys.rev . _ * d46 * ( 1992 ) 246 ; _ phys.rev . _ * d50 * ( 1994 ) 442 ; _ acta phys.polon._ * b27 * ( 1996 ) 1687 ; j.d . bjorken , _ int.j.mod.phys . _ * a7 * ( 1992 ) 4189 ; _ acta physica polonica _ * b23 * ( 1992 ) 561 , * b28 * ( 1997 ) 2773 ; k.l . kowalski and c.c . taylor , slac - pub-92 - 6 ; k.l . kowalski , j.d . bjorken , and c.c . taylor , slac - pub-6109 ( 1993 ) ; s. gavin , _ nucl.phys . _ * a590 * ( 1995 ) 163c . degrand , _ phys.rev . _ * d30 * ( 1984 ) 2001 ; + j. ellis and h. kowalski , _ _ * b214 * ( 1988 ) 161 ; j. ellis , u. heinz , and h. kowalski , _ phys.lett . _ * b233 * ( 1989 ) 223 ; j. ellis , m. karliner , and h. kowalski , _ phys.lett . _ * b235 * ( 1990 ) 341 ; j. dziarmaga and m. sadzikowski , _ phys.rev.lett . _ * 82 * ( 1999 ) 4192 . rutenberg , _ phys.rev . _ * e51 * ( 1995 ) r2715 ; + m. zapotocky and w.j . zakrzewski , _ _ * e51 * ( 1995 ) r5189 ; a.d . rutenberg , w.j . zakrzewski , and m. zapotocky , _ europhys.lett . _ * 39 * ( 1997 ) 49 ; + g.j . stephens , _ phys.rev . _ * d61 * ( 2000 ) 085002 .

in this paper we introduce a construction of directed strongly regular graphs from smaller ones using equitable partitions . each equitable partition of a single dsrg satisfying several conditions leads to an infinite family of directed strongly regular graphs . we construct in this way dozens of infinite families . for order at most 110 , we confirm the existence of dsrgs for 30 previously open parameter sets .

1504.00161

the main subject of this paper is directed strongly regular graph ( briefly dsrg ) , a possible generalisation of the well - known ( undirected ) strongly regular graphs for the directed case introduced by duval in @xcite . the undirected version plays a central role in algebraic graph theory , while the directed version received less attention . we show how to construct larger dsrgs from smaller ones with the aid of suitably defined equitable partitions for directed graphs . according to our best knowledge , equitable partitions for directed graphs have not been defined in exact terms , though similar concept can be found for association schemes in @xcite with an application for ( undirected ) strongly regular graphs in @xcite . the main results of the introduced idea involve constructions of many infinite families of dsrgs . several families have new parameter sets , others generalise and connect constructions which have been previously known from various ideas . for the small parameter sets there is an evidence of the current state - of - the - art on the webpage of a.brouwer and s.hobart ( see @xcite ) . according to their catalogue of parameter sets with order at most 110 , we have confirmed the existence of dsrgs for 30 open parameter sets . the paper is organised as follows : in section 2 the necessary and most significant terms are repeated . in section 3 a brief summary is given on the used computer algebra packages . in section 4 we define a new construction called @xmath0-join . it serves as a key concept for our project . we derive the necessary and sufficient conditions , when this construction gives us a dsrg if the basic graph is a smaller dsrg . it turns out that the crucial moment is finding a certain partition in the basic graph . in section 5 we present good partitions for several well - known families of dsrgs and in section 6 for dsrgs with some sporadic parameter sets . in this section we present the basic and most significant concepts used in this paper . all graphs considered in this paper are finite , simple , directed or undirected graphs , thus no loops or multiple edges are allowed . if there is a dart from vertex @xmath1 to vertex @xmath2 in a directed graph ( digraph ) , then we say that @xmath2 is an _ outneighbour _ of @xmath1 , and @xmath1 is an _ inneighbour _ of @xmath2 . the number of outneighbours ( inneighbours ) of a vertex @xmath1 is called _ outdegree ( indegree ) _ of @xmath1 . if the in- and outdegrees of all vertices in a digraph are equal to @xmath3 , then we say that the digraph is _ @xmath3-regular _ , or _ regular of degree ( valency ) @xmath3_. let @xmath4 be a group . suppose that @xmath5 and @xmath6 , where @xmath7 is the identity element of the group @xmath4 . then the digraph @xmath8 with vertex set @xmath4 and dart set @xmath9 is called the _ cayley digraph over @xmath4 with respect to @xmath10 . _ let @xmath11 be a simple undirected graph . assume that @xmath12 is a partition of @xmath13 . it is called _ equitable _ if for all @xmath14 the amount of neighbours of a vertex @xmath15 in the set @xmath16 is equal to a constant @xmath17 which does not depend on the choice of @xmath15 . thus , to an equitable partition we can associate a matrix @xmath18 which is called _ quotient matrix_. an equitable partition @xmath0 is called _ homogeneous of degree @xmath19 _ , if each cell has size @xmath19 . equitable partitions of graphs can be defined also using their adjacency matrix . we say that a partition @xmath0 of a matrix @xmath20 is _ column equitable _ , if there exists such an @xmath21 matrix @xmath22 , usually called _ quotient matrix _ , for which @xmath23 , where @xmath4 is an @xmath24 matrix , the _ characteristic matrix _ of @xmath0 , whose columns are the characteristic vectors of @xmath0 . in other words , @xmath25 similarly , a partition @xmath0 of a matrix @xmath20 is _ row equitable _ , if @xmath0 is column equitable for @xmath26 . clearly , if @xmath0 is both row and column equitable for an adjacency matrix @xmath20 , then @xmath0 defines an equitable partition of the corresponding graph . following this idea one can define a kind of equitable partitions also for directed graphs . we will say that @xmath0 is a _ column equitable partition _ of a directed graph @xmath27 , if it is a column equitable partition of its adjacency matrix @xmath28 . let @xmath20 be an adjacency matrix of a ( di)graph @xmath27 of order @xmath29 , and @xmath0 a partition of @xmath30 into @xmath31 cells @xmath32 . we say that adjacency matrix @xmath20 _ respects _ the partition @xmath0 , if the corresponding ordering @xmath33 of the vertices satisfies : @xmath34 for all vertices @xmath35 , @xmath36 . we assume that the reader is familiar with the terms like _ adjacency matrix _ , _ kronecker product of matrices _ , _ group of automorphisms of a graph _ , _ 1-factor ( perfect matching ) _ , _ clique , coclique _ and _ line graphs_. otherwise , we refer the reader to @xcite . we will use also terms like _ blocks of imprimitivity , rank of a permutation group _ or _ transitivity _ from permutation group theory . for definitions see @xcite . an undirected graph @xmath27 is called _ regular _ of valency @xmath3 , or _ @xmath3-regular _ , if each vertex is incident to the same number @xmath3 of edges . in the terms of adjacency matrix of a graph we can equivalently say that @xmath27 is @xmath3-regular if and only if for its adjacency matrix @xmath37 the equation @xmath38 holds , where @xmath39 is the all - one matrix . a simple regular graph with valency @xmath3 is said to be _ strongly regular _ ( srg , for short ) if there exist integers @xmath40 and @xmath41 such that for each edge @xmath42 the number of common neighbours of @xmath43 and @xmath44 is exactly @xmath40 ; while for each non - edge @xmath42 the number of common neighbours of @xmath43 and @xmath44 is equal to @xmath41 . previous condition can be equivalently rewritten into the equation @xmath45 using the adjacency matrix of @xmath27 . the quadruple @xmath46 is called the _ parameter set _ of an srg @xmath27 . strongly regular graphs play a central importance in algebraic graph theory . there are many dozens of papers with interesting constructions and results . the most significant ones altogether with crucial information on their properties are collected on the webpage of a. brouwer @xcite . a database of small srgs is located on the webpage of e. spence @xcite . a possible generalisation of the notion of srgs for directed graphs was given by duval @xcite . while the family of srgs has been well - studied , the directed version has received less attention . a _ directed strongly regular graph _ ( dsrg ) with parameters @xmath47 is a regular directed graph on @xmath29 vertices with valency @xmath3 , such that every vertex is incident with @xmath48 undirected edges , and the number of directed paths of length 2 directed from a vertex @xmath1 to another vertex @xmath2 is @xmath40 , if there is an arc from @xmath1 to @xmath2 , and @xmath41 otherwise . in particular , a dsrg with @xmath49 is an srg , and a dsrg with @xmath50 is a doubly regular tournament . * in this paper we prefer to use the ordering @xmath47 for the 5-tuple of parameters of a dsrg , however in several other papers the ordering @xmath51 is used . the adjacency matrix @xmath37 of a dsrg with parameters @xmath47 , satisfies @xmath52 oppositely , if an @xmath53 zero - one matrix @xmath20 with zeros on the diagonal satisfies ( [ am ] ) for some @xmath54 and @xmath41 , then it is an adjacency matrix of a dsrg with parameters @xmath47 . if @xmath27 is a dsrg with parameter set @xmath47 and adjacency matrix @xmath20 , then the complementary graph @xmath55 is a dsrg with parameter set @xmath56 with adjacency matrix @xmath57 , where @xmath58 [ prop2 ] for a directed graph @xmath27 let @xmath59 denote the digraph obtained by reversing all the darts in @xmath27 . then @xmath59 is called the _ reverse _ of @xmath27 . in other words , if @xmath20 is the adjacency matrix of @xmath27 , then @xmath26 is the adjacency matrix of @xmath59 . the following proposition was observed by pech , and presented in @xcite : let @xmath27 be a dsrg . then the graph @xmath59 is a dsrg with the same parameter set . [ prop1 ] we say that two dsrgs @xmath60 and @xmath61 are _ equivalent _ , if @xmath62 , or @xmath63 , or @xmath64 , or @xmath65 ; otherwise they are called _ non - equivalent_. ( in other words , @xmath60 is equivalent to @xmath61 if and only if @xmath60 is isomorphic to @xmath61 or to a graph , obtained from @xmath61 via reverse and complementation . ) from our point of view the interesting dsrgs are those which are non - equivalent . under _ eigenvalues _ of a dsrg we mean the eigenvalues of its adjacency matrix . it is known , that there are precisely three different eigenvalues , all of them are integers , and their values together with their multiplicities are uniquely determined by the parameter set . more explicitly , a dsrg with parameter set @xmath47 has eigenvalues : @xmath66 with respective multiplicities : @xmath67 the parameters @xmath68 are not independent . relations , which have to be satisfied for such parameter sets are usually called _ feasibility conditions_. most important , and , in a sense , basic conditions are the following ( for their proof see @xcite ) : @xmath69 there exists a positive integer @xmath19 such that : @xmath70 further : @xmath71 we have to mention that for a feasible parameter set it is not guaranteed that a dsrg does exist . a feasible parameter set for which at least one dsrg @xmath27 exists is called _ realizable _ , otherwise _ non - realizable_. the smallest example of a feasible , but non - realizable parameter set is @xmath72 . the non - existence of such graph was shown in @xcite . there are known several general constructions of dsrgs . the most common constructions are based on algebraic or combinatorial approaches , or even on their combination . they include approaches using block matrices @xcite , kronecker product of matrices @xcite , finite geometries @xcite , combinatorial block designs @xcite , finite incidence structures @xcite , coherent algebras @xcite , association schemes @xcite , regular tournaments @xcite , partial sum families @xcite , cayley graphs @xcite or computer algebra experimentation @xcite . in this paper we extend this list by a construction which is on the border of matrix construction , equitable partitions and computer algebra experimentation . this project has been supported by an essential amount of results which have been obtained with the aid of a computer . all such computations have been run in gap ( groups , algorithms , programming @xcite ) jointly with using its share packages : grape a package for computation with graphs @xcite together with nauty for testing automorphisms of graphs @xcite ; coco ii an unreleased package for computation with coherent configurations and association schemes , written by reichard @xcite ; and the package setorbit for finding representatives of orbits of group actions on sets of various size , written by pech and reichard @xcite . in this section we introduce a new construction , which serves as a basic idea for constructing new graphs from old ones . we will investigate when this construction gives us a larger dsrg from smaller ones and derive its most important properties . starting from this section it often happens that we do not need to distinguish between directed and undirected strongly regular graphs . in such a case we use abbreviation ( d)srg . let us consider a directed or an undirected strongly regular graph @xmath27 of order @xmath29 . let us call it _ basic graph_. suppose that @xmath73 is a homogeneous partition of its vertex set into @xmath31 cells of size @xmath74 , where @xmath75 . consider a positive integer @xmath76 and define a digraph @xmath77 as following . create @xmath78 disjoint copies @xmath79 of the graph @xmath27 , i.e. for its vertex set we have @xmath80 in each of the copies we can distinguish the images of the cells of the partition @xmath0 . let us denote @xmath81 the image of the cell @xmath82 in the @xmath83-th copy of @xmath27 . we define the adjacency in @xmath77 between vertices by giving the set of outneighbours of vertices . for fixed @xmath83 there is a dart from vertex @xmath84 to vertex @xmath85 whenever there is a dart from @xmath43 to @xmath44 in @xmath27 . further , let @xmath86 for some @xmath87 and @xmath83 . there is a dart from @xmath84 to all of the vertices in cells @xmath88 , @xmath89 , where the superscripts are taken modulo @xmath90 . let us call this construction _ @xmath0-join of @xmath27 in power @xmath76 , _ or just shortly _ @xmath0-join _ when @xmath27 and/or @xmath76 are either clear from the context or irrelevant . according to our best knowledge this construction is new . however , it can be regarded as a mixture of the lexicographic product of graphs @xcite and the x - join of graphs @xcite . in particular , if @xmath91 then our construction coincides with the lexicographic product of @xmath27 with the complete graph @xmath92 . the similarity with x - joins can be seen after collapsing each cell in @xmath77 into a single vertex . from this follows that the only difference between our construction and x - join is , that we define differently adjacencies between vertices in the cells which do belong to the same copy of @xmath27 , and those which do not . now , we translate the @xmath0-join construction into the language of matrices , since it will be easier to prove statements using adjacency matrices of graphs . let @xmath27 be a ( di)graph and @xmath73 be a homogeneous partition of its vertices into @xmath31 cells of size @xmath74 , for suitable positive integers @xmath31 and @xmath74 . reorder the vertices of @xmath27 according to @xmath0 and create adjacency matrix @xmath20 of @xmath27 which respects @xmath0 . then we may consider @xmath20 as a block matrix with blocks of size @xmath93 . let us define for @xmath94 matrix @xmath95 as a matrix whose all entries are ones in the columns with indices @xmath96 , and all other entries are zeros , i.e. @xmath97 , where @xmath39 is the all - ones matrix of size @xmath98 , and @xmath99 stands for the kronecker product . further , for an arbitrary integer @xmath76 let us create a circulant block matrix @xmath100 whose first row of blocks is @xmath101 and all other rows of blocks are obtained by shifting the blocks to right by one position . the matrix @xmath100 is the adjacency matrix of the @xmath0-join of @xmath27 in power @xmath76 , defined above . for @xmath102 , @xmath103 and @xmath104 then we have the following matrices : @xmath105 while for @xmath106 and @xmath107 we have : @xmath108 the motivation behind the @xmath0-join construction defined in the previous subsection comes from an analysis of previously known dsrgs . it showed that a significant amount of them can be constructed as a @xmath0-join of certain smaller dsrg @xmath27 , its partition @xmath0 and a suitable integer @xmath76 . inspired by this observation we started to investigate the following problem : * problem : * _ for a given ( d)srg @xmath27 , what are the necessary and sufficient conditions for a homogeneous partition @xmath0 in order to get a dsrg from the @xmath76-th power of its @xmath0-join ? _ we will solve this problem in two steps . first , we will consider the smallest case @xmath106 to derive some necessary conditions on @xmath27 and @xmath0 . after that we will show that if these conditions are satisfied , then for arbitrary integer @xmath76 the @xmath76-th power of the @xmath0-join produces again a dsrg . if a @xmath0-join of a ( d)srg @xmath27 produces a larger dsrg , then we say that @xmath0 is a _ good partition _ for @xmath27 . here we derive the necessary conditions for a homogeneous partition @xmath0 in a ( d)srg @xmath27 in order to be good for the @xmath0-join construction . let us start with an easy observation about matrices @xmath95 . computational arguments lead to the following properties : * observation 1 : * let @xmath109 are as in the definition of the @xmath0-join construction . then for any @xmath94 and @xmath110 we have : * @xmath111 , * @xmath112 , * @xmath113 , * @xmath114 has constant columns , and this constant in an arbitrary column represents the number of darts starting from cell @xmath115 and terminating in the vertex corresponding to the given column . the following theorem is very important , since it has many consequences , which significantly restricts our search space for possibly good partitions . let @xmath20 be an adjacency matrix of a ( d)srg @xmath27 with parameters @xmath47 , which respects the homogeneous partition @xmath116 of degree @xmath74 . suppose that the @xmath0-join @xmath117 for @xmath106 is a dsrg with parameters @xmath118 . then * @xmath119 * for arbitrary @xmath120 the number @xmath121 of darts starting in @xmath115 and terminating in @xmath122 does not depend on the concrete choice of @xmath44 , just on @xmath123 and @xmath124 . moreover , @xmath125 * proof . * denote @xmath126 the adjacency matrix of @xmath117 coming from the construction . clearly , @xmath127 will be a circulant block matrix which is completely determined by its first row of blocks , so let us say that its first row is @xmath128 . * the equations for @xmath129 and @xmath130 follow immediately from the method of construction . for other parameters let us compute : @xmath131 @xmath132 and the claim follows . * again compute : @xmath133 but from strong regularity we have also @xmath134 altogether this means that @xmath135 . now using part @xmath136 from _ observation 1 _ we get , that the number of darts starting in @xmath137 and terminating in a concrete vertex in @xmath137 is precisely @xmath138 ; while the number of those starting from @xmath137 and terminating in a concrete vertex in @xmath115 , where @xmath139 , is precisely @xmath41 . analogous computations with @xmath140 prove the claim . * corollary 1 . * let @xmath126 and @xmath20 are as above . then if @xmath126 is an adjacency matrix of a dsrg then necessarily @xmath141 * proof . * counting the number of incoming darts to a fixed vertex in @xmath142 in @xmath27 we get @xmath143 * corollary 2 . * necessarily : @xmath144 * corollary 3 . * if @xmath126 is an adjacency matrix of a dsrg , then @xmath0 is a homogeneous column - equitable partition with quotient matrix @xmath145 . * immediately follows from corollary 1 . * corollary 4 . * if @xmath20 is an adjacency matrix with respect to a partition @xmath0 of an undirected strongly regular graph , then necessarily @xmath0 is an equitable partition and the cells are regular graphs of valency @xmath138 . * proof . * in an undirected graph indegree and outdegree of a vertex are equal , therefore the partition @xmath0 satisfies all conditions for being equitable . * corollary 5 . * for given ( d)srg with parameter set @xmath47 the number of solutions of the equation is at most two . * proof . * since @xmath75 , replacing @xmath31 by @xmath146 results in a quadratic equation in @xmath74 . * corollary 6 . * if @xmath91 , then @xmath147 , and the potential parameter set of the @xmath0-join is @xmath148 . feasibility conditions for parameter sets imply that @xmath149 , which is not an interesting case . * proof . * if @xmath148 is a parameter set of a dsrg , then from ( [ c1 ] ) we have @xmath150 after simple algebraic manipulations , using equations @xmath151 and we obtain that @xmath149 . * corollary 7 . * if @xmath152 , then the initial graph @xmath27 is necessarily complete . * proof . * the cells are single vertices , therefore part ( b ) of lemma 1 implies that the initial graph is necessarily complete . in this section we state that the conditions derived in the previous subsection are , in fact , sufficient let @xmath27 be a ( d)srg with parameter set @xmath47 . let @xmath31 and @xmath74 are positive integers such that @xmath75 and there exists a homogeneous column equitable partition @xmath116 of vertices with quotient matrix @xmath153 . let @xmath20 be an adjacency matrix respecting @xmath0 , and let us define matrix @xmath100 in accordance with our @xmath0-join construction for an arbitrary positive integer @xmath76 . then @xmath100 is an adjacency matrix of a dsrg with parameter set @xmath154 * proof . * the necessity of conditions has been shown above . by direct computations one can show that if @xmath20 satisfies all the posed conditions , then they are also sufficient , therefore @xmath100 corresponds to a dsrg for any positive integer @xmath76 . this theorem provides an algorithm how to proceed when we are looking for a @xmath0-join of a dsrg : 1 . take a directed or an undirected srg @xmath27 ; 2 . solve for the parameter set of @xmath27 ; 3 . for the solutions of satisfying count the quotient matrix and find a column equitable partition @xmath0 with this quotient matrix . 4 . for @xmath0 , @xmath27 and an arbitrary integer @xmath76 create the @xmath0-join in power @xmath76 for @xmath27 according to the construction . of course , the key moment of the construction is the finding of a good column equitable partition . in the last sections we try to provide such partitions ( if they do exist ) at least for the small cases of dsrgs and for some general families . as we will see , in some cases the method of construction of the initial dsrg immediately provides us such a partition . in other cases some combinatorial tricks help to find it ; and it happens sometimes that we are forced to leave this job to a computer . here we investigate two properties of the @xmath0-joins : behaviour under the complementation , and the set of eigenvalues . in this subsection we investigate when does there exist a partition @xmath0 for which both ( d)srg and its complementary ( di)graph can be @xmath0-joined into a new dsrg , simultaneously . after that we derive the sufficient condition when we can get two dsrgs with complementary parameter sets as a result of the @xmath0-join construction . suppose that @xmath27 is a ( d)srg with parameters @xmath47 , @xmath155 its complementary digraph with parameters @xmath156 and @xmath0 a homogeneous column equitable partition for both @xmath27 and @xmath55 . then @xmath157 and @xmath0 has precisely two cells . [ thm2 ] * proof . * from the necessary conditions on the quotient matrix follow that @xmath158 and @xmath159 . using the identities for the parameters of the complementary digraph of a dsrg we get : @xmath160 from the first equation . after this the second equation gives us : @xmath161 @xmath162 hence @xmath163 and the claim follows . from the proof and the assumptions of the previous theorem we can derive the following corollary : * corollary 8 . * let @xmath27 be a dsrg with parameters @xmath47 , @xmath55 its complement with parameters @xmath156 . suppose that @xmath164 and @xmath0 is a good homogeneous column - equitable partition with two cells for @xmath27 . then it is a good partition for @xmath55 as well . let @xmath27 be a dsrg with parameters @xmath47 , @xmath55 its complement with parameters @xmath156 . suppose that they can be simultaneously @xmath0-joined via partition @xmath0 to a dsrg . then @xmath165 and @xmath166 have complementary parameter sets . [ thm3 ] * proof . * we already know from theorem [ thm2 ] that under the posed assumptions the partition @xmath0 has two cells . now putting @xmath167 into the formulas for parameter sets of the @xmath0-join and using relations between parameter sets of mutually complementary digraphs we get that the parameter sets for @xmath168 and @xmath166 coincide . it is well - known that a dsrg has three different eigenvalues and all of them are integers . these eigenvalues are uniquely determined by the parameter set . it is not hard to see that if the basic graph @xmath27 with parameter set @xmath47 has eigenvalues @xmath169 then its @xmath0-join has eigenvalues @xmath170 . in this section we try to describe some good equitable partitions for well - known families of ( d)srgs . consequently , the @xmath0-join construction gives us many infinite families of dsrgs , which are in many cases new . for fixed parameter set we consider just one graph and we focus on finding at least one good equitable partition . the complete graph @xmath171 can be regarded as an srg with parameters @xmath172 , where the parameter @xmath41 is undefined . thus , we can not apply our previous theory . however , if we apply the @xmath0-join construction for the complete graph @xmath173 using the homogeneous equitable partition with @xmath31 cells of size @xmath74 , then it is easy to show that the result is again a dsrg . its parameter set is @xmath174 this parameter set is known for example from @xcite , section 4 . duval ( in @xcite , lemma 8.2 ) constructed an infinite family of dsrgs with parameter set @xmath177 by giving its adjacency matrix . it is an easy exercise to check that his matrix defines a good partition for both feasible integer solutions of : @xmath178 and @xmath179 . more precisely , if the vertices @xmath180 of the graph are in accordance with the order of rows / columns of the adjacency matrix given by duval , then @xmath181 and @xmath182 are the corresponding good partitions . this proves the existence of dsrgs with the following parameter sets : @xmath183 and @xmath184 where @xmath76 and @xmath3 are arbitrary positive integers . among these parameter sets there are several new : ( 60,15,4,3,4 ) , ( 60,20,7,6,7 ) , ( 78,26,9,8,9 ) , ( 96,32,11,10,11 ) , and ( 108,27,7,6,7 ) for @xmath185 . in similar manner one can show that from duval s generalized construction ( @xcite , theorem 8.3 ) , in conjunction with our construction it is possible to obtain dsrgs with parameter sets @xmath186 @xmath187 for arbitrary positive integers @xmath188 . for any positive integers @xmath3 and @xmath41 such that @xmath189 , jrgensen in @xcite constructed a dsrg @xmath27 with parameter set @xmath190 . the vertices of his graphs are integers modulo @xmath191 , and there is a dart from @xmath1 to @xmath2 if and only if @xmath192 . ( operations are also taken modulo @xmath29 . ) in other words , a fixed vertex @xmath1 has outneighbours of form @xmath193 and inneighbours of form @xmath194 for @xmath195 . for this parameter set has always a solution in integers : @xmath196 . the possibly second solution is @xmath197 , but since @xmath189 an easy number theoretical argument shows that @xmath198 is integer if and only if @xmath199 . let us define for @xmath202 the following sets : @xmath203 i.e. @xmath115 contains the number @xmath204 and its outneighbours . it is not hard to show that if from vertex @xmath204 there is a dart to some @xmath2 , then there is no dart from @xmath205 to @xmath2 for @xmath206 , and there is no dart between @xmath204 and @xmath205 . therefore the sets @xmath115 are cells of a partition of @xmath207 which was identified with the vertex set of @xmath27 . pick a vertex @xmath208 in @xmath115 . then , clearly , @xmath209 is a dart in @xmath27 , and we know that there are precisely @xmath40 directed two - paths from @xmath204 to @xmath83 in @xmath27 . however , all these two - paths are inside the cell @xmath115 , because all the outneighbours of @xmath204 belong to @xmath115 . hence , the number of inneighbours of @xmath83 starting from @xmath115 is precisely @xmath210 , and the number of inneighbours of @xmath204 in @xmath115 is also @xmath210 , which follows from the property that @xmath27 is a dsrg with parameter @xmath211 . altogether this means that for each vertex in @xmath115 there are precisely @xmath210 inneighbours in @xmath115 , thus if our partition is column equitable , then @xmath212 . now take @xmath213 , an arbitrary vertex @xmath214 , and count the number of inneighbours of @xmath83 in @xmath115 . clearly , @xmath204 can not be an inneighbour of @xmath83 , since @xmath215 . therefore there are precisely @xmath41 oriented paths of length two from @xmath204 to @xmath83 , and we know that the midvertices on these paths lie in @xmath115 . hence , @xmath216 . in fact , we proved the following lemma : for arbitrary @xmath3 and @xmath41 satisfying @xmath189 let @xmath217 , @xmath201 and @xmath27 be the corresponding dsrg from jrgensen s construction . for @xmath202 the sets @xmath218 form a column equitable partition of @xmath27 with quotient matrix @xmath219 . therefore for an arbitrary positive integer @xmath76 there exists a dsrg with parameter set @xmath220 , where : @xmath221 calculations show that @xmath222 does hold , therefore this parameter set is not new , since it is covered by the construction given by jrgensen . however , the resulting graphs from different constructions are not isomorphic , in general . for example , when @xmath106 , @xmath223 and @xmath224 then @xmath225 and @xmath226 , but the resulting dsrgs from our and jrgensen s construction with parameters @xmath227 are non - equivalent . recall that in this case the assumption @xmath189 on the initial parameters imply that the only possibilities are @xmath199 , otherwise @xmath31 can not be an integer . now let us consider a partition of the set @xmath230 into cells @xmath115 as equivalence classes modulo @xmath31 , i.e. @xmath231 we show that @xmath232 is a good equitable partition . let us select any @xmath233 and count the number of @xmath234 such that @xmath235 is a dart . we repeat that if @xmath1 is an inneighbour of @xmath2 , then for some @xmath236 we have @xmath237 , therefore necessarily @xmath238 . equivalently , @xmath239 , because @xmath240 is a multiple of @xmath31 . * if @xmath241 , then @xmath242 has no solution in @xmath236 . * if @xmath243 , then @xmath244 has a unique solution @xmath245 . now let us consider the number of inneighbours in @xmath16 of a fixed vertex @xmath233 , where @xmath213 . similarly as above , for an inneighbour @xmath246 there is @xmath236 such that @xmath237 , and @xmath247 . after simplification : @xmath248 . * if @xmath241 , then @xmath249 has a unique solution @xmath250 ( taken modulo @xmath31 ) . * if @xmath243 , then @xmath251 has exactly two solutions . we proved the following : for @xmath199 and arbitrary @xmath3 satisfying @xmath189 let @xmath252 , @xmath229 and @xmath27 be the corresponding dsrg from jrgensen s construction . for @xmath202 the sets : @xmath253 form a column equitable partition of @xmath27 with quotient matrix @xmath254 . therefore for arbitrary positive integer @xmath76 there exists a dsrg with parameter set @xmath220 , where : @xmath255 in the case , when @xmath256 the resulting parameter set can be also obtained from jrgensen s construction . also this is the case , when @xmath257 , @xmath258 . in all other cases @xmath259 does not divide @xmath260 . according to @xcite we constructed first dsrg with parameters @xmath261 , @xmath262 , @xmath263 and @xmath264 , if we consider just @xmath265 . the _ triangular graph @xmath266 _ is defined as the line graph of the complete graph @xmath171 . for @xmath267 it is an srg with parameters @xmath268 . moreover , graphs with these parameter set are unique , when @xmath269 . for @xmath270 there are three more graphs , the so - called _ chang graphs _ ( see @xcite ) , apart from @xmath271 . by solving @xmath272 we get that for @xmath273 there is a unique solution @xmath274 ; for @xmath275 also a unique solution @xmath276 , while for @xmath277 there is no integer solution . for odd @xmath29 we have @xmath278 and @xmath147 . so the cells of the potential equitable partitions are regular of valency @xmath279 . in other words , each cell is a collection of cycles . let us consider the vertex set of the triangular graph as 2-subsets of the set @xmath280 and there is an edge between two different 2-subsets if and only if they are not disjoint . for @xmath281 let us define sets @xmath282 . the partition @xmath283 is equitable with quotient matrix @xmath284 , thus the existence of dsrgs with parameter set @xmath285 is proved , where @xmath286 , @xmath287 , @xmath288 , @xmath289 , @xmath290 and @xmath291 are arbitrary integers . in all the cases when @xmath265 the parameter set has been known previously . here the cells are of order @xmath292 and valency @xmath293 . again , equivalently we can consider the graph @xmath171 and correspondingly an edge partition in it . the partition determines an edge - coloring ( not proper ) in the sense , that the same color is assigned to edges from the same cell and different colors to edges from different cells . each color defines a subgraph of @xmath171 . we can assume that the edge @xmath294 is black and we further consider the `` black subgraph '' of @xmath171 . this edge is incident to @xmath29 black edges , therefore @xmath295 , where @xmath296 stands for `` black - degree '' , i.e. the number of black edges incident to @xmath43 . by pigeonhole principle there exists a vertex @xmath297 such that @xmath298 and @xmath299 are black edges . thus , @xmath300 , hence @xmath301 , and necessarily all the @xmath302 neighbours of @xmath303 have black - degree @xmath302 . so the number of black edges is at least @xmath304 . however , for @xmath305 this number is greater than @xmath292 , thus for @xmath306 there is no good partition . for @xmath307 the partition is trivial . for the three exceptional cases , chang graphs , we executed an exhaustive computer search for finding induced regular subgraphs of size 14 and valency 8 . since the number of subsets of size 14 in a set of size 28 is quite large , we used the setorbit package written by pech and reichard @xcite . with its aid we can generate just one representative for each orbit of the group of automorphisms on subsets of size 14 . this tool reduced the amount of subsets being considered for a few hundreds of thousands , and therefore it took just a few hours of computation . however , this exhaustive search showed that there is no such induced subgraph , and therefore there is no good partition for our construction . triangular graph @xmath266 has been described above . let us now consider its complement @xmath308 . again , to avoid trivial cases we suppose that @xmath309 . the parameter set of @xmath308 is @xmath310 . equation has integer solution just for @xmath311 . for @xmath312 see the section about the petersen graph . for @xmath102 exhaustive computer search showed that there is no good partition for the only solution @xmath313 . finally , for @xmath270 we have just one solution @xmath314 , but @xmath315 and theorem [ thm2 ] applies . another famous family of strongly regular graphs is the so - called _ lattice square graphs_. the lattice square graph @xmath316 has vertex set @xmath317 and two vertices are adjacent if their coordinates agree in exactly one position . the parameter set of @xmath316 is @xmath318 . for each positive integer @xmath29 the equation has at least one solution @xmath319 . for odd @xmath29 this is the only solution , but for even @xmath29 there is a second solution @xmath320 . conditions posed on the quotient matrix @xmath22 of a good partition say that in the first case @xmath321 , and in the second @xmath322 . first , consider partition @xmath323 of vertices of @xmath316 such that @xmath324 it is not hard to see that this partition is equitable with quotient matrix @xmath321 , and it guarantees the existence of a dsrg with parameters @xmath325 for arbitrary positive integers @xmath76 and @xmath29 . the resulting parameter set is new at least in the case when @xmath106 and @xmath307 . when @xmath29 is even , we can consider the partition @xmath326 such that for all @xmath327 the cell @xmath328 is : @xmath329 it is an equitable partition with quotient matrix @xmath322 . the corresponding @xmath0-join of the lattice square graph @xmath316 is a dsrg for each integer @xmath76 with parameter set @xmath330 klin et al . in @xcite constructed a family of dsrgs with parameter set @xmath331 , where @xmath29 is odd , as cayley graphs over dihedral group of order @xmath332 . so let @xmath333 be the dihedral group of order @xmath332 . let @xmath334 . the cayley graph with connection set @xmath10 over @xmath335 is the requested dsrg . for this parameter set has always a unique solution , when @xmath336 . this is @xmath337 . for a good partition the quotient matrix is @xmath338 , and it is not hard to see that the partition @xmath339 satisfies this condition , therefore @xmath0 is a good partition . hence the existence of dsrgs with parameter set @xmath340 is proved for arbitrary @xmath76 and odd @xmath29 . duval and iourinski in @xcite gave a sufficient condition for construction dsrgs as cayley graphs over metacirculant groups . for our construction it is sufficient to choose their orbit partition satisfying the @xmath341-orbit condition in order to get a good partition for @xmath342 . for their graphs equation has a second solution just in the case , when @xmath343 . this case coincides with the construction leading to dsrgs with @xmath344 , @xmath345 . in this section we discuss the @xmath0-joins of small ( d)srgs with parameter sets which have not been included in the previous section . we show just a single good partition for each parameter set , usually . in fact , in the smallest cases ( @xmath346 ) we run an exhaustive search with the aid of a computer for the values @xmath347 . in many cases we have found dozens of good partitions and non - isomorphic @xmath0-joins . these are available upon request from the author . exceptionally , we will consider more graphs sharing the same parameter set or a graph of larger order . it happens since we want to include into consideration some famous graphs , as well . for example , the petersen graph , the shrikhande graph , the schlfli graph or the hoffman - singleton graph . on the other hand , we will not consider parameter sets @xmath47 , where @xmath348 and the only solution of the is the pair @xmath349 , since in these cases we can apply theorem [ thm2 ] and theorem [ thm3 ] . a summary of the all new parameter sets of dsrgs is given in appendix . the smallest undirected srg is a 4-cycle . the only good solution of is @xmath350 . though there are two equitable partitions into two cells of size 2 , just one of them , the pairs of non - adjacent vertices , satisfies the condition posed on the quotient matrix . our construction creates dsrgs with parameter set @xmath351 . for this parameter set and its complementary parameter set there are known several constructions , see @xcite , @xcite , @xcite and @xcite . one of the most famous graphs in graph theory , petersen graph ( see fig . [ pic4 ] ) , is srg with parameter set @xmath352 . it is the complementary graph to the triangular graph @xmath353 and it is known to be unique with this parameter set . all of its equitable partitions have been enumerated in @xcite . equation has two solutions : @xmath354 and @xmath355 , but the second is ruled out by . for @xmath354 , a good equitable partition is determined by a partition into two disjoint 5-cycles . the corresponding dsrgs from the @xmath0-join have parameter set @xmath356 . for @xmath106 the parameter set @xmath357 is not new , since in @xcite the authors constructed a dsrg with this parameter set . their digraph is not vertex - transitive , while ours is , thus these graphs are non - isomorphic . the first dsrg with parameter set @xmath358 has been just recently constructed in @xcite . for @xmath359 the resulting parameter set is new , according to our best knowledge . @xmath360 there are 1292 non - isomorphic dsrgs sharing this parameter set , all with very small group of automorphisms ( orders less than 6 ) . we investigated the dsrg corresponding to the adjacency matrix @xmath361 . equation has unique solution : @xmath313 . a good partition is @xmath362 the parameter set of the @xmath0-join is @xmath363 , which is new . there is a unique srg with parameters @xmath364 , it is the clebsch graph ( see fig . equation has two solutions : @xmath365 and @xmath366 . the latter solution is ruled out by . a good partition is @xmath367 the parameter set of the @xmath0-join is @xmath368 which also comes from @xcite . there are precisely two non - isomorphic graphs sharing this parameter set : the shkrikhande graph ( see fig . [ pic7 ] ) and the lattice square graph @xmath369 ( considered in section [ ss7.6 ] ) . using the labeling from the figure , the partitions @xmath370 and @xmath371 are good partitions of the shrikhande graph for @xmath365 and @xmath366 , respectively . first case lead to the parameter set @xmath372 , which is known from @xcite , and the second case to the parameter set @xmath373 , which is new . there is a unique dsrg on this parameter set , its adjacency matrix is @xmath374 . equation has two solutions : @xmath365 and @xmath366 . in the case when @xmath365 a good partition is : @xmath375 the resulting parameter sets is @xmath376 , which is known from @xcite . in the case when @xmath366 a good partition @xmath377 the resulting parameter set @xmath378 is new . @xmath379 duval in his seminal paper presented also a dsrg with parameter set @xmath380 . this graph has been known already from other problems in graph theory , and often is referred to as _ bosk graph _ @xcite . it is known to be unique on this parameter set . though has two solutions , one of them is ruled out by condition . for @xmath381 a good partition is given by the partition into three vertex - disjoint induced copies of the subgraph isomorphic to the unique dsrg(6,2,1,0,1 ) . hence the existence of a dsrg with parameters @xmath382 is proven . for @xmath106 a dsrg with parameter set @xmath383 has been recently found in @xcite . for a dsrg(18,7,5,2,3 ) with adjacency matrix @xmath384 we have to consider two possibilities : @xmath385 and @xmath381 . a good partition is @xmath386 in the first case , and @xmath387 in the second case . the resulting parameter sets @xmath388 and @xmath389 are both new , in general . jrgensen in @xcite besides a remarkable amount of non - existence results also constructed a few dsrgs as cayley graphs . let us consider his dsrg(21,6,2,1,2 ) , which was constructed as a cayley graph over the group @xmath390 with connection set @xmath391 . after solving the equation and counting the quotient matrix we figure out that @xmath392 is a solution and @xmath393 . therefore the partition @xmath394 where @xmath395 is a good partition leading to the existence of a family of dsrgs with parameter set @xmath396 . jrgensen ( @xcite , theorem 14 ) constructed three dsrgs on 24 vertices as cayley graphs of @xmath397 . their connection sets are @xmath398 @xmath399 and @xmath400 the corresponding cayley graphs are dsrgs with parameters @xmath401 , @xmath402 , and @xmath403 , respectively . for dsrg(24,8,3,2,3 ) the equation has only one solution @xmath404 , while for ( 24,9,7,2,4 ) and ( 24,10,8,4,4 ) there are two solutions : @xmath405 and @xmath406 . in order to get a good partition for the dsrg(24,8,3,2,3 ) we run an exhaustive computer search . it returned us hundreds of such partitions . one of them has cells : @xmath407 , @xmath408 , @xmath409 . thus the existence of dsrgs with parameter set @xmath410 is proved . for the remaining two dsrgs we have found good partitions combinatorially . using conditions on the quotient matrix we immediately get that the partition of @xmath397 into even an odd permutations gives us a good partition in the case of dsrg(24,9,7,2,4 ) . thus the existence of a dsrg with parameters @xmath411 is proved . for @xmath404 the situation is a little bit more complicated . after some computer algebra experimentation we concluded , that considering the index 3 subgroup @xmath412 the partition @xmath413 is good . hence , the existence of dsrgs with parameter set @xmath414 is proved . let us put permutation @xmath415 into @xmath137 if @xmath416 , otherwise put @xmath417 into @xmath418 . partition @xmath419 is a good partition for dsrg(24,10,8,4,4 ) with @xmath420 . this proves the existence of a dsrg with parameter set @xmath421 . considering the index 3 subgroup @xmath422 the partition @xmath423 is good for dsrg(24,10,8,4,4 ) for @xmath404 , therefore the existence of dsrg with parameters @xmath424 is proved . in all the cases , the resulting parameter sets for @xmath106 has been displayed in @xcite as open . jrgensen ( @xcite , theorem 15 ) constructed a dsrg(26,11,7,4,5 ) . he prescribed adjacencies between elements of two sets of size 13 . for this case says that @xmath425 and from the quotient matrix it is easy to check that the two original sets define a good partition . this confirms the existence of dsrgs with parameters @xmath426 . for @xmath106 the parameter set @xmath427 is new according to @xcite . according to paulus @xcite there are precisely 10 srgs with parameters ( 26,10,3,4 ) . the second most symmetric of them has group of automorphisms of order 39 , which has two orbits of size 13 on the vertex set . this orbit - partition is good for the unique solution @xmath425 of the equation , hence the existence of dsrgs with parameters @xmath428 is proved for all positive integer @xmath76 . for @xmath106 the resulting parameter set ( 78,36,23,16,17 ) is new according to @xcite . the equation gives us a unique solution @xmath429 . for diagonal elements of the quotient matrix of a good partition we get @xmath430 , what means that in the basic graph we have to find a coclique of size 9 , or equivalently , a clique of size 9 in the schlfli graph . this graph has been well - studied , and we know that the maximal clique is of size 6 . therefore there is no good partition for our construction . martinez and araluze @xcite constructed a handful of dsrgs with previously unknown parameter sets with order at most 45 using partial sum families over cyclic groups . from their construction immediately follows that the equivalence classes modulo @xmath31 ( where @xmath31 is the number of blocks ) form a column equitable partition , and these partitions are not excluded neither by condition nor by . moreover , from the blocks it is easy to count the quotient matrix , since it is enough to count the amount of numbers modulo @xmath31 in each block for each possibility . in this way we checked all their 14 graphs constructed from cyclic groups , and figured out that six of them have quotient matrix satisfying part ( b ) in theorem 1 . these correspond to parameter sets : ( 27,10,6,3,4 ) , ( 30,13,8,5,6 ) , ( 34,14,12,5,6 ) , ( 34,15,9,6,7 ) , ( 39,10,6,1,3 ) and ( 40,17,11,6,8 ) . consequently , using our @xmath0-join construction we can create for any positive integer @xmath76 a dsrg with parameter set : @xmath431 , @xmath432 , @xmath433 , @xmath434 , @xmath435 and @xmath436 . according to @xcite we constructed the first dsrgs with parameters @xmath437 , @xmath438 , @xmath439 , while the parameter set @xmath440 appears only in this paper , but here already for the third time . the first dsrg with parameter set @xmath441 has been recently constructed in @xcite as a union of classes in a non - commutative association scheme . the equation has two solutions @xmath442 and @xmath443 . the full group of automorphisms of the graph is transitive of order 144 , therefore we can consider its system of imprimitivity . one can obtain a good partition for the first solution as a union of a block of imprimitivity of size 12 , and a suitable block of size 6 . hence the existence of dsrgs with parameter sets @xmath444 is confirmed for all positive integer @xmath76 . for @xmath106 the parameter set @xmath445 is new according to @xcite . for the case @xmath446 a good partition is given by the system of imprimivity with 3 cells of size 12 , where each cell is inducing a regular graph of valency 1 . this confirms the existence of dsrgs with parameters @xmath447 for all positive integer @xmath76 . there are precisely 180 srgs ( for more information see @xcite ) with parameters ( 36,14,4,6 ) . for this parameter set has two solutions : @xmath442 and @xmath443 . let us consider the second most symmetric graph of them , i.e. which has group of automorphisms of order 432 . the imprimitivity system of the group of automorphisms has blocks of size 18 . the corresponding partition fits the conditions for being good for our construction , therefore we can construct for every positive integer @xmath76 a dsrg with parameter set @xmath448 for @xmath106 the resulting parameter set @xmath449 is new according to our best knowledge . if we consider the only imprimitivity system of the group of automorphisms with blocks of size 6 , then after some experimentation we can obtain 3 pairs of blocks whose unions results in a 3-cell partition with cells of size 12 and quotient matrix @xmath450 . this proves the existence of a dsrg with parameter set @xmath451 for arbitrary positive integer @xmath76 . it is known that the exact number of srg sharing the parameter set ( 36,15,6,6 ) is 32548 ( for more information see @xcite ) . we considered a highly symmetric one of them with automorphism group of order 648 . ( in the catalogue of e. spence it is the last graph on this parameter set . ) it has a unique system of imprimitivity with blocks of size 9 . union of any two of these blocks and the remaining two blocks gives us a suitable partition for @xmath442 . hence , the existence of a dsrg with parameter set @xmath452 is proved for every integer @xmath76 . for @xmath106 the parameter set @xmath453 is new according to @xcite . in a similar fashion , using unions of 4 blocks of imprimitivity of size 3 we have found a good partition into 3 cells of size 12 , therefore the existence of a dsrg with parameter set @xmath454 is proved for all integer @xmath76 . just for curiosity we considered one of the most famous srgs , the hoffman - singleton graph . solutions of are @xmath455 and @xmath456 , however the latter is excluded by . the first solution challenges us to consider the partition consisting from 5 vertex - disjoint copies of the petersen graph . by counting the possible quotient matrix it is not hard to see that this is really a good partition , therefore there exists a dsrg @xmath457 for each positive integer @xmath76 . using our @xmath0-join construction introduced in section [ spi ] we constructed dozens of infinite families of dsrgs from smaller ones . among the constructed graphs with order less than 110 there are 30 with new parameter sets according to @xcite . the new parameter sets are displayed in the appendix . the author gratefully acknowledges the contribution of the scientific grant agency of the slovak republic under the grant 1/0151/15 , as well as the constribution of the slovak research and development agency under the project apvv 0136 - 12 . this research was also supported by the project : mobility - enhancing research , science and education at the matej bel university , itms code : 26110230082 , under the operational program education cofinanced by the european social fund . 99 f. adams , a. gendreau , o. olmez , s.y . song , _ construction of directed strongly regular graphs using block matrices _ , preprint , arxiv : 1311.0494 . a. araluze , i. kovcs , k. kutnar , l. martinez , d. marui , _ partial sum quadruples and bi - abelian digraphs _ , j. combin . a * 119*(2012 ) , 18111831 . j. bosk , _ partially directed moore graphs _ , math . 29*(1979 ) , 181196 . a.e . brouwer , s. hobart , _ tables of directed strongly regular graphs _ , ( november 2014 ) , http://homepages.cwi.nl/~aeb/. a.e . brouwer , o. olmez , s.y . song , _ directed strongly regular graphs from @xmath458-designs _ , europ . j. combin . * 33*(2012 ) , 11741177 . chang , _ the uniqueness and nonuniqueness of triangular association schemes _ , sci . record * 3*(1959 ) 604613 . chang , _ association schemes of partially balanced block designs with parameters @xmath459 , @xmath460 , @xmath461 and @xmath462 _ , sci . record * 4*(1960 ) , 1218 . g. chartrand , l. lesniak , _ graphs & digraphs ( 4th edition ) _ , chapman & hall/ crc , boca raton , 2005 . dixon , b. mortimer , _ permutation groups . _ graduate texts in mathematics , 163 . springer - verlag , new york , 1996 . xii+346 pp . a.m. duval , _ a directed graph version of strongly regular graphs _ , j. combin . a * 47*(1988 ) , 71100 . a.m. duval , d. iourinski , _ semidirect product construction of directed strongly regular graphs _ , j. combin . a * 104*(2003 ) , 157167 . f. fiedler , m.h . klin , m. muzychuk , _ small vertex - transitive directed strongly regular graphs _ , discrete math . * 255*(2002 ) , 87115 . f. fiedler , m. klin , ch . pech , _ directed strongly regular graphs as elements of coherent algebras . _ in : k. denecke , h .- j . vogel , ( eds . ) general algebra and discrete mathematics : proc . conf . on general algebra and discrete mathematics , potsdam 1998 , pp . shaker verlag , aachen ( 1999 ) . groups , algorithms , programming a system for computational discrete algebra , www.gap-system.org . godsil , _ algebraic combinatorics _ , chapman and hall mathematics series . chapman & hall , new york , 1993 . xvi+362 pp . godsil , s.a . hobart , w.j . martin , _ representations of directed strongly regular graphs _ , europ . j. combin . * 28*(2007 ) , 19801993 . gyrki , m. klin , _ sporadic examples of directed strongly regular graphs obtained by computer algebra experimentation _ , in gerdt , v.p . et al . , computer algebra in scientific computing . casc 2014 , proceedings . berlin : springer . lecture notes in computer science * 8660 * ( 2014 ) , 155170 . hammersley , _ the friendship theorem and the love problem _ , pp . 3154 in : surveys in combinatorics ( e. keith lloyd , ed . ) , cambridge university press 1983 . m. hirasaka , h. kang , k. kim , _ characterization of association schemes by equitable partitions _ , j. combin . * 27*(2006 ) , 139152 . hobart , t.j . shaw , _ a note on a family of directed strongly regular graphs _ , europ . j. combin . * 20*(1999 ) , 819820 . jrgensen , _ search for directed strongly regular graphs _ , aalborg university preprint r-99 - 2016 , october 1999 . jrgensen , _ non - existence of directed strongly regular graphs _ , discrete math . * 264 * ( 2003 ) , 111126 . jrgensen , _ directed strongly regular graphs with @xmath463 _ , discrete math . * 231 * ( 2001 ) , 289293 . jrgensen , _ new mixed moore graphs and directed strongly regular graphs _ , discrete math . * 338 * ( 2015 ) , 10111016 . m. klin , a. munemasa , m. muzychuk , p.h . zieschang , _ directed strongly regular graphs obtained from coherent algebras , _ lin . appl . , * 377 * ( 2004 ) , 83109 . m. klin , ch . pech , p.h . zieschang , _ flag algebras of block designs : i. initial notions , steiner 2-designs and generalized quadrangles . _ preprint , math - al-10 - 1998 , technische universitt dresden ( 1998 ) l. martinez , a. araluze , _ new tools for construction of directed strongly regular digraphs : difference digraphs and partial sum families _ , j. combin . b , * 100*(2010 ) , 720728 . b. mckay , nauty user s guide , ver . 1.5 , technical report tr - cs-90 - 02 . computer science department , australian national univ . ( 1990 ) . b. mckay , e. spence , _ classification of regular two - graphs on 36 and 38 vertices _ , australasian j. of combin . , * 24*(2001 ) , p.293 . o. olmez , s.y . song , _ some families of directed strongly regular graphs obtained from certain finite incidence structures _ , graphs and combinatorics , * 30 * ( 2014 ) , 15291549 . paulus , _ conference matrices and graphs of order 26 _ , technische hogeschool eindhoven , report wsk 73/06 , eindhoven , 1983 , 89 pp . pech , s. reichard , _ the setorbit package for gap _ , . pech , s. reichard , _ enumerating set orbits_. in : m. klin et al . , algorithmic algebraic combinatorics and grbner bases , pp . 137150 . springer , berlin , heidelberg ( 2009 ) . s. reichard , coco ii , ( personal communication ) . soicher , grape : a system for computing with graphs and groups , groups and computation ( new brunswick , 1991 ) , dimacs ser . discrete math . 287291 , amer . soc . , providence , ri ( 1993 ) g. sabidussi , _ graph derivatives _ , math . z. * 76*(1961 ) , 385401 . g. sabidussi , _ the lexicographic product of graphs _ , duke math . j. * 28*(1961 ) , 573578 . e. spence , _ regular two - graphs on 36 vertices _ appl . * 226228*(1995 ) , 459-497 . e. spence , _ strongly regular graphs on at most 64 vertices _ , m. ziv - av , _ interactions between coherent configurations and some classes of objects in extremal combinatorics _ , phd thesis , ben - gurion university of the negev , 2014 , .new parameter sets up to @xmath185 appearing in this paper . [ cols="^,^,^,^,^ , > " , ]

single crystals of the organic charge - transfer salts-(bedt - ttf)hg(scn) have been studied using hall - potential measurements (k ) and magnetization experiments ( k , rb ) . the data show that two types of screening currents occur within the high - field , low - temperature cdw phases of these salts in response to time - dependent magnetic fields . the first , which gives rise to the induced hall potential , is a free current ( ) , present at the surface of the sample . the time constant for the decay of these currents is much longer than that expected from the sample resistivity . the second component of the current appears to be magnetic ( ) , in that it is a microscopic , quasi - orbital effect ; it is evenly distributed within the bulk of the sample upon saturation . to explain these data , we propose a simple model invoking a new type of quantum fluid comprising a cdw coexisting with a two - dimensional fermi - surface pocket which describes the two types of current . the model and data are able to account for the body of previous experimental data which had generated apparently contradictory interpretations in terms of the quantum hall effect or superconductivity .

cond-mat0311362

in their rudimentary form , charge - density waves ( cdws ) consitute a simple one - dimensional spin singlet groundstate in which the collective mode consists of a charge density modulated at a characteristic wavevector @xmath7 . here , @xmath8 is the fermi wavevector of the fermi - surface section responsible for cdw formation @xcite . compared to superconductors , their behavior in a magnetic field is relatively easy to predict , providing perhaps the simplest example of a singlet ground state reaching the pauli limit @xcite . this limit is defined as the magnetic field required to cause the energy of the partially spin - polarized normal state to become lower than that of the condensate . in spite of the simplicity of cdw systems , the large values of the cdw condensation temperatures @xmath9 , typically of order 100 k @xcite , make the pauli limit inaccessible in standard laboratory magnetic fields for the vast majority of cdw materials @xcite . our reason for studying the title compounds , @xmath0-(bedt - ttf)@xmath1hg(scn)@xmath2 , is that they lie at the far lower end of the spectrum of transition temperatures , with @xmath10 k @xcite for @xmath3 k and @xmath11 k for @xmath3 rb , making them perhaps the most marginal of cdw systems . consequently , they are the only cdw compounds in which the gap is sufficiently low for the primary cdw phase , cdw@xmath12 , to have been shown to be pauli limited by a relatively modest magnetic field of @xmath13 t for @xmath3 k and @xmath14 t for @xmath3 rb @xcite . it must be stated , however , that the marginal cdw properties of @xmath0-(bedt - ttf)@xmath1hg(scn)@xmath2 render it somewhat beyond the predictive power of the standard theory @xcite . for instance , the low transition temperature results in a cdw with a weak charge modulation that is vulnerable to fluctuations and sample - dependent effects @xcite . furthermore , the presence of large sections of the fermi surface left ungapped by the cdw order @xcite ( see inset to fig . [ fermisurface ] ) causes the energy associated with the coupling of the magnetic field to orbital degrees of freedom of the itinerant electrons to rival the condensation energy . this leads to a modification of the quantum oscillatory effects @xcite and to the possibility of field - induced cdw sub - phases @xcite . at magnetic fields above the pauli paramagnetic limit , @xmath0-(bedt - ttf)@xmath1hg(scn)@xmath2 then crosses over into a new regime in which the transition temparature becomes even lower @xcite , falling to between 2 and 4 k : a schematic of the phase diagram is shown in fig . [ fermisurface ] . since the conventional cdw phase , cdw@xmath12 , is no longer stable at high magnetic fields , the spin - up and spin - down electrons must instead form independent charge and spin modulations with momentum vectors @xmath15 and @xmath16 @xcite , leading to a different type of cdw phase . experimental studies of this high - magnetic - field phase , denoted cdw@xmath4 , lead one to question whether it is a cdw at all . while conventional wisdom has it that cdws constitute a class of narrow - gap insulators @xcite , experimental studies find behavior that is reminiscent either of the quantum hall effect @xcite or superconductivity @xcite . these experimental findings include a sharp drop in the electrical resistivity at low temperatures , persistent currents and unusual hall - voltage phenomena . problems arise because the quantum hall effect @xcite and superconductivity @xcite have fundamentally different origins . if one attempts to categorize interpretations of experimental data in terms of the quantum hall effect @xcite or superconductivity @xcite , one obtains two sets of conclusions that are in contradiction . the present paper seeks to reconcile the body of experimental evidence @xcite in exploring the predicted properties of a new type of quantum fluid , consisting of a cdw coexisting with a 2d fermi surface . a simple model shows that the exchange of quasiparticles between these subsystems results in the sample partially screening time- and spatially - varying electromagnetic fields via two types of current flow ; this result is supported by the experimental data in this work , which show that the screening currents observed in earlier magnetic measurements @xcite comprise two parts . the first is weighted mostly towards the sample edges and can be considered a free current @xmath5 , furnishing a hall - potential gradient . the second contribution is analogous to the `` magnetic currents '' @xmath6 used to represent the effects of localised orbital moments @xcite in producing magnetization . these currents , which are essentially of infinite duration @xcite , represent a local , microscopic current flow within the bulk of the sample ; they occur if the surface scrrening mechanism fails , causing the cdw , instead , to be coerced by an external change in magnetic field into a non - equilibrium state . in this case , bulk currents are supported by cdw pinning forces and are not free . the remainder of this paper is organized as follows . experimental details are given in section [ experiment ] , whilst section [ honoldstuff ] describes the hall potential measurements which characterize the surface current density @xmath5 . the `` magnetic '' current density @xmath6 is studied using torque magnetometry in section [ torquetorque ] . conclusions are given in section [ discussion ] ; for ease of reference , the model describing the microscopic mechanism that leads to @xmath5 and @xmath6 is described in the appendix . -(bedt - ttf)@xmath1hg(scn)@xmath2 salts . a schematic of the phase diagram as a function of reduced temperature @xmath17 and reduced magnetic field @xmath18 ( based on ref . the second order transition temperature @xmath9 at @xmath19 0 is approximately 8 k in the @xmath3 k salt and 12 k in the @xmath3 rb salt . likewise , the first order `` kink '' transition field @xmath20 at @xmath21 0 ( roughly corresponding to the pauli paramagnetic limit ) , is 23 t for @xmath3k and 32 t for @xmath3rb . ` cdw@xmath12 ' refers to the proposed conventional cdw phase while ` cdw@xmath4 ' refers to an incommensurate phase that comes into existence at high magnetic fields . the inset shows a simplified schematic of the fermi - surface in the repeated brillouin zone representation ( after ref . @xcite ) which consists of open one - dimensional ( 1d ) sections which run along @xmath22 and a closed two - dimensional ( 2d ) section . ( an accurate representation of the actual fermi surface topolgy is given in ref . cdw formation is thought to occur for a characteristic `` nesting vector '' @xmath23 @xcite ( with a component @xmath24 along @xmath25 ) , causing the 1d section to become gapped . , width=302 ] the free electrical current distribution , @xmath5 , was determined using the variant of the corbino geometry described in ref . @xcite in which pairs of small graphite - paint contacts are placed on the outer edge and upper surface of a sample . currents were induced in the sample by applying a small sinusoidal oscillatory field of amplitude @xmath26 and angular frequency @xmath27 superimposed on the quasistatic field @xmath20 provided by a 33 t bitter coil at nhmfl , tallahassee . alternatively , currents were induced by sweeping the quasistatic field at a rate @xmath28 . provided that the sample dimensions are much smaller than its skin depth ( so that it is entirely penetrated by the changing field ) , the resulting hall potential between the edge of the sample , of area @xmath29 , and its geometrical centre is @xcite @xmath30 and @xmath31 where @xmath32 is an appropriate average of @xmath33 and @xmath34 and standard symbols for the components of the resistivity tensor are used . under the conditions of almost complete penetration by the oscillatory field , the phase of @xmath35 in eqn . [ hallpotential1 ] with respect to @xmath36 is the same as that of a voltage induced in an open loop ( i.e. @xmath37 ) . however , if significant screening occurs , @xmath35 will also contain a component in quadrature to this . using standard complex notation , we write @xmath38 where @xmath39 and @xmath40 are the dissipative and reactive ( quadrature ) components of @xmath35 respectively . in our experiment , phase - sensitive detection techniques allowed @xmath39 and @xmath40 to be simultaneously recorded for analysis purposes ; the signal from a multiturn pick - up coil mounted close to the sample was used to define the _ real _ phase . the approximate distribution of the current was further determined by placing pairs of contacts arranged between the outer and upper surfaces of the sample of of volume @xmath41 mm@xmath42 shown in fig . [ samplephoto ] . while the finite size of 100 @xmath43 50 @xmath44 m of the electrical contacts applied using graphite paint limited the spatial resolution of the experiment , the contacts do enable one to distinguish between different models . during these experiments , temperatures down to 0.5 k were achieved using a plastic @xmath42he refrigerator . the orbital magnetic currents , @xmath6 , were determined by measuring the magnetic torque of a different @xmath3 k sample to that shown in fig . [ samplephoto ] as well as a @xmath3 rb sample . in each case , @xmath45 , the angle between the magnetic field @xmath46 and the normal to the sample s conducting planes was restricted to small angles ( @xmath47 ) ; this avoided complications due to magnetic torque interaction @xcite . the torque was measured capacitively by attaching the sample to a @xmath48m - thick phosphor - bronze cantilever that forms one of the plates of a capacitor . electrical contacts applied to the sample confirmed that @xmath6 continued to persist after the electrical currents dissipate . in these experiments , temperatures down to 50 mk were provided by a dilution refrigerator . fig . [ hallpotential ] shows typical measured hall potentials @xmath39 and @xmath40 due to an oscillatory field ; in this case @xmath49 mt , @xmath50 hz and the temperature of the sample ( @xmath51 mm@xmath52 ) was 0.5 k. the @xmath35 data in fig . [ hallpotential ] correspond to the contact labelled 3 in fig . [ samplephoto ] ; as the contact on the upper face of the sample was very near to its geometrical center , we expect that the hall potential will be close to the maximum value possible . note that there was a small field - independent inductive pick - up contribution @xmath53 to @xmath54 due to the open - loop area ( @xmath55 mm@xmath52 ) between the contacts . by comparing measured @xmath39 values at several different fields , it is possible to infer that @xmath56v . and @xmath40 measured with the third pair of contacts on the sample shown in fig . [ samplephoto ] . an oscillatory applied field with @xmath57 mt rms and @xmath58 hz was used ; @xmath59 0.5 k. arrows indicate the axis corresponding to each data set.,width=302 ] in the discussion that follows , we first attempt to analyse the data of fig [ hallpotential ] in terms of conventional bulk currents within the _ whole _ of the sample ; the analysis uses conductivity and resistivity tensor elements which do not vary spatially . we shall find that whilst this conventional analysis describes the behavior of @xmath60 , albeit with some remarkably low values of @xmath32 , it is unable to account for the variation of @xmath40 with @xmath61 . in a reasonably isotropic quasi - two - dimensional metal , the components of the resistivity @xmath62 and conductivity @xmath63 tensors are related by the expressions @xmath64 thus , the ratio @xmath65 in eqn . [ hallpotential1 ] is equivalent to @xmath66 ; it is therefore likely that the asymptotic variation of @xmath39 at low fields is due to the divergence of @xmath67 where @xmath68 is the charge density of the 2d holes , as @xmath69 . as @xmath61 increases , @xmath65 becomes smaller , leaving @xmath53 as the dominant contribution to the measured @xmath39 . however , after the `` kink '' transition at @xmath70 t , @xmath39 undergoes a resurgence and the quadrature component @xmath71 becomes significant for the first time . the maxima in @xmath60 occur at integral landau level filling factors @xmath72 , where @xmath73 is the magnetic quantum oscillation frequency ; at such fields , the chemical potential @xmath44 is situated in a landau gap . the peak values @xmath74v ( at @xmath75 t ) correspond to @xmath76 ; the fact that @xmath77 allows us to use @xmath78 ( see eqn . [ oldhall ] ) with @xmath79 m@xmath80 @xcite to extract a minimum sample resistivity of @xmath81 m at the maxima close to @xmath82 t. such resistivity values are characteristic of a good metal at room temperature and are not too dissimilar from the results of ac susceptibility experiments on @xcite . to examine the apparent bulk resistivity components in more detail , we rearrange eqs . [ tensor ] in terms of @xmath83 and @xmath84 to yield @xmath85}\ ] ] and @xmath86}.\ ] ] these equations can then be used along with eqns . [ hallpotential1 ] and [ oldhall ] to convert the experimental values of @xmath60 to apparent bulk in - plane resistivity components . the results of this procedure are shown in fig . [ resistivity ] ; the linear increase of @xmath87 with @xmath61 and quadratic increase of @xmath88 with @xmath61 in at low magnetic fields show that the method gives a behavior in accord with expectations . at higher fields , the deduced values of @xmath87 behave in a similar manner to earlier direct measurements of oscillations in the hall resistivity @xcite . however , it is interesting to note that the in - plane resistivity is lower in fig . [ resistivity ] than that obtained from four - terminal methods @xcite . owing to non - uniform current flow associated with sample imperfections , direct measurements of the in - plane resistivity are often contaminated by the inter - plane component @xmath89 , which is several orders larger in magnitude @xcite . and @xmath88 estimated from @xmath39 using eqs . [ rhoxy ] and [ rhoxx ] ( @xmath90 k ) . arrows indicate the axis corresponding to each data set.,width=302 ] measured for the third pair of contacts in a slowly varying magnetic field ( sweep rate @xmath91 0.5 t@xmath92 ; @xmath93 k ) . ( a ) shows raw data for a single sweep , with arrows indicating the sweep direction . no filtering has been applied . the background non oscillatory component could be caused by variations in the contact potentials with magnetic field . ( b ) the difference @xmath94 between rising- and falling - magnetic - field data averaged over 5 sweeps.,width=302 ] the unusually large value of @xmath95 42 and consequent low value of @xmath88 estimated from the oscillatory field measurements are confirmed by setting @xmath96 and by sweeping the quasistatic field . data obtained using @xmath97ts@xmath92 are shown in figure [ dchallpotential ] for both rising and falling magnetic fields . insertion of this rate of change of field into eqn . [ hallpotential2 ] , along with the value @xmath98 obtained in the oscillatory - field experiments ( see above ) , yields @xmath99v at @xmath100 t , in good agreement with the data shown in fig . [ dchallpotential ] . however , the magnitude of the hall potential continues to increase with increasing field , rising to a value of @xmath101v at around 32 t ( fig . [ dchallpotential ] ) . this yields @xmath102 , corresponding to @xmath103 m , which is several times lower than the resistivity of room - temperature copper . such values for these parameters would not be unusual for a two - dimensional electron gas exhibiting the quantum hall effect @xcite . they are unusual for a bulk organic metal far from the quantum limit ; for example @xmath104 at 32 t. thus far , a conventional treatment employing uniform conductivity tensor elements @xmath105 and @xmath84 has given an _ apparently _ satisfactory explanation of the hall potential measured in swept fields and the real component of the hall potential @xmath39 observed in oscillatory fields , albeit with extremely low values of the bulk resistivity . however , this simple model fails on considering the imaginary component @xmath40 which appears within the high magnetic field regime in fig . [ hallpotential ] . were a conventional metal , then in order to have a quadrature component of order 30 % of the in - phase component at @xmath106 hz , the skin - depth must be comparable to or shorter than the effective radius @xmath107 590 @xmath44 m of the sample @xcite . if we estimate the skin - depth @xmath108 from the above resistivity value ( @xmath109 2.4 @xmath110 10@xmath111 @xmath112 m ) , however , we obtain a value @xmath113 3.9 mm that is significantly greater than the sample size . the presence of a large finite @xmath40 term , therefore , can not be easily explained by a simple conductivity tensor model . a further indication of the fact that the induced currents do not obey a conventional bulk mechanism is given by using the contact arrangments shown in fig . [ samplephoto ] to explore the dependence of @xmath39 and @xmath40 on the distance @xmath114 of the contact on the upper surface of the sample from the edge . for a spatially homogeneous metallic system in a slowly time - varying magnetic field , it is trivial to show that the current density and concomitant hall potential should vary quadratically ; @xmath115 where @xmath116 is the total potential difference between the outer surface and geometric centre of the sample . this form occurs because the size of the area loop element susceptible to induced voltages increases parabolically with the distance from the geometric centre of the sample . on comparing the detected hall potentials for the three different contact arrangements shown in fig . [ samplephoto ] , we find that @xmath39 at @xmath75 t in fig . [ halldistribution ] varies more slowly with @xmath114 than predicted by eq . [ halldistribution1 ] . in the case of the reactive component , no discernable dependence of @xmath40 on @xmath114 is seen . this type of behavior shows that the hall potential difference ( and therefore also the current ) is heavily weighted towards the sample edge , with the effect being particularly pronounced for the imaginary component . inside the upper surface where @xmath117 100 , 310 and 400 @xmath44 m for contacts 1 , 2 and 3 respectively , with @xmath21 0.5 k , @xmath118 409.6 hz and @xmath119 2.6 mt . ( a ) shows the normal hall voltage @xmath39 . the dotted line shows the voltage distribution expected according to eq . [ halldistribution1 ] while the dashed line shows the voltage distribution expected for an exponential variation of the hall potential , @xmath120 . ( b ) shows the reactive hall voltage @xmath40 , with the solid line depicting @xmath120 with @xmath121 100 @xmath44m.,width=302 ] and @xmath122 . all estimates were made for @xmath123 t and @xmath124 k , with the data points corresponding to excitation frequencies @xmath125 , 208.6 , 409.6 and 8809 hz and oscillating field amplitudes @xmath126 , 3.90 , 2.62 and 3.75 mt.,width=302 ] currents carried at or close to the edge are expected to occur in quantum - hall - effect systems for which @xmath127 0 at integral landau - level filling factors . these currents are either carried within edge channels @xcite , giving rise to a kind of chiral fermi liquid @xcite , or are bulk states near the edge in which electrostatic forces locally deplete ( or enhance ) the charge carrier density @xcite . were a situation where @xmath127 0 realised in , perfect screening would give rise to a hall potential that is entirely reactive , even in the limit as @xmath128 . experimentally , @xmath129 , implying two things : first , @xmath130 within the bulk , and second , the partial screening that takes place has to occur within a region around the perimeter of the sample that is thinner than the bulk skin - depth . in the appendix , we show that a system comprising a cdw coexisting with quasi - two - dimensional landau levels can lead to very effective screening currents within a distance @xmath131 of the sample surface . although the finite size of the electrical contacts in fig . [ samplephoto ] ( @xmath132 m in diameter ) , limits the resolution to which we can analyse the hall voltage distribution in fig . [ halldistribution ] , we can , nevertheless , perform some simple consistency checks on this model . we propose that all of the current flows within a distance @xmath133 of the surface of the sample , and that this current flow encounters a characteristic ( very low ) resistivity @xmath134 . the system is like an @xmath135 ( inductance - resistance ) circuit , with @xmath136 and @xmath137 where @xmath138 is the sample height , @xmath139 is its effective radius and @xmath0 is a so - called nagaoka parameter , a numerical factor @xmath140 that depends on @xmath141 @xcite . for our sample , @xmath142 . in a magnetic field swept at a slow uniform rate , this treatment yields @xmath143 ( c.f . eqn . [ hallpotential2 ] ) whereas for a sinusoidally - oscillating field we obtain @xmath144 here @xmath145 is a characteristic inductive decay time and @xmath146 represents transient solutions due to rapid changes in @xmath147 and/or @xmath148 . we first compare eqn . [ sweptedge ] with the hall potential of 8 @xmath44v observed in fig . [ dchallpotential ] at @xmath149 t , obtaining @xmath150 , or @xmath151 m , using @xmath152 nm ( see appendix ) . inserting this into the semiclassical drude expression @xmath153 , we obtain a scattering time of @xmath154 ns , or , a mean free path @xmath155 mm ( where we have obtained the fermi velocity from @xmath156 ) . the fact that @xmath157 is comparable to the sample perimeter of 4 mm could be consistent with ballistic transport . the ballistic transport regime , for which finite values of @xmath105 can no longer be locally defined , is one of the essential preconditions of the edge - weighted current model ( see appendix ) . two separate estimates of @xmath158 can be made . first , we can rearrange eqns . [ oscedge ] and [ tdefn ] to yield @xmath159 . in the limit of low frequency , @xmath160 , so that this can be written @xmath161 where the superscript @xmath162 denotes that this estimate of the characteristic time is derived from the dissipative ( i.e. resistive ) component of @xmath35 . an alternative estimate can be made by comparing the resistive and reactive parts of @xmath35 : @xmath163 where the superscript @xmath164 denotes that this is analogous to the technique used to find the decay time of conventional @xmath135 circuits . [ times ] shows @xmath165 and @xmath166 deduced from experimental data ( @xmath75 t , temperature = 0.5 k ) . if the sample surface layer functioned as a conventional ( but high - conductivity ) metal , we should expect the two times to be the same . however , the divergence between @xmath166 and @xmath165 at low frequencies , or low values of @xmath35 , is unconventional . this suggests that the sample behaves resistively when attempting to drive a current through it , but once established , the currents have a tendency to last longer than expected for a conventional metal . their lifetime @xmath166 becomes particularly long at low values of the hall potential @xmath35 observed in slowly - varying magnetic fields . in rapidly - varying magnetic fields , @xmath166 converges with @xmath165 at @xmath167 mv . this is similar to the saturation value of @xmath35 observed in ms - duration pulsed magnetic fields @xcite . attempts to explain this saturation in terms of the breakdown of the quantum hall effect have been largely unsuccessful owing to the enormous magnitude of the electric field that is required . an electric field that is concentrated towards the sample edge within a distance @xmath133 ( see appendix ) , however , increases the probability of zener tunneling . moreover , the increased size of the unit cell in the cdw phase compared to the normal metallic phase makes zener tunneling across the cdw gap more likely than landau level tunneling . in such a model for the magnitude of the threshold electric field @xmath168 ( where @xmath169 is the cdw gap and @xmath170 is the fermi energy ) , the lattice periodicity @xmath171 occurs in the denominator . assuming an exponential variation of the electric field where @xmath172 , we obtain @xmath173 mev for @xmath174 mev , which is comparable to the saturation observed in pulsed magnetic field studies @xcite . should normal scattering processes be inhibited according to the screening model described in the appendix , zener tunneling would provide a natural means for the decay of the currents , given the potentially large values of the electric field concentrated close to the sample edge . the hall potential associated with free currents @xmath5 becomes difficult to detect for magnetic field sweep rates much less than 0.5 ts@xmath92 . in contrast , magnetic currents @xmath6 continue to be present for an arbitrarily long period of time after the magnetic field sweep is stopped @xcite . [ torque ] shows examples of the magnetization of @xmath0-(bedt - ttf)@xmath1hg(scn)@xmath2 ( for @xmath3 k and rb ) estimated from the magnetic torque . magnetic fields of @xmath70 t and @xmath175 t are required to access the cdw@xmath4 phase for @xmath3 k and rb respectively @xcite ; in this phase , hysteresis similar to that originally obtained by christ _ et al . _ @xcite is observed . of samples of the @xmath3 k ( @xmath176 ) and rb ( @xmath177 ) salts measured on both rising and falling fields at @xmath21 50 mk for @xmath3 k and @xmath21 0.45 k for @xmath3 rb made using the magnetic torque method . arrows indicate the direction of sweep . for the @xmath3 rb salt , at fields below approximately 29 t , the cdw@xmath12 phase is stable . the region between 29 t and 33 t appears to be disturbed since @xmath178 is different between rising and falling fields . at fields above approximately 33 t , the transition into the cdw@xmath4 phase is complete . the de haas - van alphen oscillations are the same for rising and falling fields , apart from an relative offset 2@xmath179 . for the @xmath3 k salt , the magnetic torque shows similar features to that measured in reference @xcite . the inset shows the magnetization measured after a field cool ( as indicated in fig . 1 ) . the magnetization starts out at the value indicated by the large center - dot circle , after which the field is swept down to @xmath180 29 t followed by an upsweep to @xmath181 30 t.,width=302 ] measured as the field is swept up to 29.33 t then momentarily down to 29.24 t before being resumed . arrows indicate the change in the locus of @xmath179 versus @xmath61 . the solid lines show the results of fits of @xmath179 to eq . [ beanman ] . ( b ) a model hysteresis loop calculated according to eq . [ beanman ] ( i.e. the bean model for a long cylinder with its axis parallel to @xmath61 @xcite ) , showing the theoretical saturation value of @xmath182 and the reversal state . ( c ) a series of @xmath179 versus @xmath61 loops like that in ( a ) measured over an extended interval of field ( top ) , and ( bottom ) , a similar measurement but where the field interval over which the sweep direction is reversed is increased . arrows indicate the appropriate axes . ( d ) estimates of @xmath183 from fits of eq . [ beanman ] to the many loops shown in ( c ) ( full circles ) together with those ( large full squares ) calculated according to eq . [ irrev ] ( see also ref . estimates of the coercion field @xmath184 ( x - symbols ) are also shown . arrows indicate the appropriate axes.,width=302 ] the magnetization @xmath185 of quasi - two - dimensional metals such as is largely directed normal to the conducting layers . a torque @xmath186 per unit volume therefore occurs when there is a finite angle @xmath45 between @xmath185 and and the applied field @xmath46 @xcite . sawtooth - like de haas - van alphen oscillations ( originating from the closed section of fermi surface that survives the cdw formation @xcite ) account for the larger , reversible contribution to @xmath185 . these oscillations occur in all metals with closed sections of fermi surface at low temperatures and result from changes in the equilibrium populations of orbitally quantized landau levels as @xmath61 is swept @xcite . the waveform of the oscillations in the @xmath3 k salt at high magnetic fields has received much theoretical attention and is well understood @xcite . by contrast , the contribution which we attribute to @xmath6 is irreversible , and depends on how the sample is driven to a particular field and temperature . an example of this is shown in the inset to fig . [ torque ] ; if the sample is cooled to 0.5 k at fixed field , the magnetization attains the value surrounded by the circle . on sweeping the field down for the first time , the magnetization rises to meet the path which will be followed by subsequent downsweeps . moreover , it proves to be impossible to go back to the encircled value of @xmath185 by merely changing the field ; data recorded on subsequent upsweeps fall below it , whilst downsweep data lie above it ( fig . [ torque ] , inset ) . one can only recapture the field - cooled value by warming the sample well above the cdw@xmath4 phase and cooling again in the presence of a steady field . thus , sweeping the field produces an irreversible contribution to the magnetization that manifests itself as an offset @xmath187 between rising and falling field data in fig [ torque ] . [ loops ] shows @xmath179 for a sample ; values have been extracted by subtracting the reversible de haas - van alphen effect contribution ( obtained by averaging full up and down sweeps ) from the raw magnetization data to leave just the irreversible component . the magnetic field in fig . [ loops]a is first swept slowly up to 29.34 t , then down to 29.24 t , after which the up sweep is resumed . each time the magnetic field sweep direction is reversed , a finite interval in field @xmath188 ( where @xmath184 is the coercion field ) is required in order for @xmath179 to reach a saturation value that opposes the direction of sweep . the appendix describes how the cdw system produces a diamagnetic contribution to the magnetization that acts to partially screen the applied magnetic field ( see eq . [ irrev ] and the discussion following it ) . however , this contribution saturates when the force on the cdw , @xmath189 , resulting from a spatially - varying free energy , exceeds the pinning force , @xmath190 . elastic energy considerations ( see appendix ) show that the region in which @xmath191 propagates inwards from the sample surface as the magnetic field changes , so that the volume - fraction of the sample able to contribute to the diamagnetic screening decreases . this is analogous to the scenario in the bean model of type - ii superconductors @xcite , yielding , under the approximation of a cylindrical sample , an irreversible susceptibility @xmath192 ^ 2 . \label{beanbean}\ ] ] and an irreversible magnetization @xmath193 ^ 3 - 1 ) . \label{beanman}\ ] ] fig . [ loops]b shows the predicted behavior of the irreversible component of @xmath194 according to eq . [ beanman ] , and fig . [ loops]a shows a fit of the model to the experimental data . the fact that eq . [ beanman ] reproduces the experimental result indicates that the volume fraction of the sample @xmath195 that is able to respond with an irreversible susceptibility @xmath196 declines in a simple quadratic manner with @xmath197 ( see eq . [ beanbean ] ) , behaviour typical of critical - state models such as that of bean @xcite ; here , @xmath195 is the cross - sectional area of the remaining non - critical portion of the sample . as the magnetization varies throughout the sample , maxwell s relation @xmath198 @xcite implies that a magnetic - field - induced critical state always involves currents . however , although a critical - state model provides an excellent fit to the experimental data in fig . [ loops]a , it is important to note that there are two essential properties of the currents giving rise to the loops in fig . [ loops ] that distinguish them from those in superconductors . first , the magnitude of the non - equilibrium susceptibility @xmath199 ( obtained from the fits ) departs significantly from the ideal diamagnetic value @xmath200 observed in superconductors . this implies that the magnetic currents in @xmath0-(bedt - ttf)@xmath1hg(scn)@xmath2 can not be considered as screening currents in the same sense as those in superconductors . because the magnetic currents screen only a part of the magnetic field , the effective coercion field @xmath201 is rather large . second , the value of @xmath196 varies strongly as a function of @xmath61 in figs . [ loops]c and [ loops]d , a property that would be very difficult to explain in terms of superconductivity , but which is an intrinsic feature of the model of a cdw coexisting with well - defined landau levels discussed in the appendix . [ loops]c shows a series of loops in @xmath182 versus @xmath61 like that in fig . [ loops]a , from which @xmath196 is extracted at various values of @xmath61 in fig . [ loops]d . the magnitude of @xmath196 is low and of order 5@xmath11010@xmath202 when @xmath44 is situated in the middle of a landau level ( at half - integral filling factors ) , but increases almost ten fold when @xmath44 is between landau levels ( at integral filling factors ) . the theoretical limit for @xmath196 becomes equivalent to that of a superconductor ( i.e. @xmath203 ) only in that case of an ideal sample at integral filling factors in which the quantum lifetime is infinite ( see eq . [ irrev ] in the appendix , remembering that demagnetising factors @xcite must be taken into account once the susceptibility becomes very large ) . when pinning of the cdw occurs , the non - equilibrium susceptibility @xmath196 provides a measure of how quickly the cdw departs from equilibrium as the magnetic field is swept . this occurs more rapidly at integral landau level filling factors where the density of states is lowest , enabling @xmath44 to jump quickly between landau levels @xcite . pinning of the cdw prevents the equilibrium redistribution of carriers between the bands , causing this jump to become much more abrupt . this can account the origin of the strong variations in @xmath196 in fig . [ loops]d , as shown by the model calculations ( see appendix ) . -(bedt - ttf)@xmath204khg(scn)@xmath2 using different methods . in ( a ) this is determined by fitting eq . [ beanman ] to many hysteresis loops over an extended interval in magnetic field . the imaginery component is determined from the area of the loop divided by its width . in ( b ) , the effective susceptibility is obtained by equating @xmath205 , where @xmath206 is the reactive component of the hall potential . the imaginary susceptibility is given by @xmath207 . here , the susceptibility is estimated from the data in fig . ( c ) is the published ac susceptibility measured on another sample @xcite.,width=302 ] before concluding this paper , it is intructive to compare effective ac susceptibilities of @xmath0-(bedt - ttf)@xmath204khg(scn)@xmath2 determined using different methods in fig . [ comparison ] . in fig [ comparison]a , @xmath208 is obtained by fitting the bean model ( as described above- see eq . [ beanman ] ) to many hysteresis loops over an extended region of magnetic field . the imaginary component @xmath209 , which accounts for losses , is obtained from the loop areas . in fig . [ comparison]b , the susceptibility is estimated using @xmath210 the most clear aspect of fig . [ comparison ] is that the magnitudes of the real and imaginary susceptibilitiies due to free currents @xmath5 , estimated from the ac hall potential , agree more closely with the high frequency ac susceptibility measurements made in ref . @xcite ( shown in fig . [ comparison]c ) than with the magnetic torque measurements . this indicates that the ac susceptibility made at frequencies @xmath211 hz consists mostly of free currents @xmath5 , which are shown ( above , and in the appendix ) to be confined to the sample edges . the data in fig . [ times ] shows that @xmath5 can persist for times as long as @xmath212 ms for sufficiently small values of the hall potential @xmath213v in a sample of @xmath140 mm@xmath52 cross - section at @xmath123 t and @xmath214 k. in spite of the fact that the corresponding susceptibility in fig . [ comparison]b is much larger than that for the torque measurements in fig . [ comparison]a , the magnetization @xmath215 am@xmath92 due to free currents is actually smaller . free currents therefore have insufficient duration to contribute a significant amount to the irreversible steady magnetic torque in fig . [ loops ] . by contrast , magnetic currents observed in the torque experiments last for an indefinite period of time @xcite . the combined induction of free and magnetic currents @xmath216 can therefore be understood as follows : for small changes in magnetic field @xmath217 mt , the greatest contribution to the susceptibility initially originates from free currents at the surface that attempt to screen changes in magnetic flux density from within the bulk . these currents quickly saturate , however , as dissipation sets in , possibly assisted by zener tunneling . at that point , changes in magnetic flux density enter the bulk which cause the landau level structure and chemical potential to change , causing the cdw to depart from equilibrium , as described in the appendix . the stored energy increases quadratically with @xmath197 , eventually causing the cdw to collapse at the edges , initiating the critical state . in conclusion , hall potential and magnetic torque measurements on @xmath0-(bedt - ttf)@xmath1hg(scn)@xmath2 ( @xmath3 k , rb ) show that two types of screening currents occur within the high - field , low - temperature cdw@xmath4 phase in response to changing magnetic fields . the first , which gives rise to the induced hall potential , is a free current ( @xmath5 ) , weighted mostly towards the edge of the sample . the time constant for the decay of these currents is longer than that expected from the sample resistivity . the second component of the current appears to be magnetic ( @xmath6 ) , in that it is a microscopic , quasi - orbital bulk effect ; it is evenly distributed within the sample upon saturation . a simple model ( appendix ) , describing a new type of quantum fluid comprising a cdw coexisting with a two - dimensional fermi - surface pocket , is able to account for the origins of the currents . taken together , these findings are able to reconcile the body of experimental evidence @xcite which had previously been interpreted in terms of the quantum hall effect @xcite or superconductivity @xcite . this work is supported by us department of energy ( doe ) under grant ldrd - dr 20030084 and by the uk engineering and physical sciences research council ( epsrc ) . the national high magnetic field laboratory is supported by doe , the national science foundation ( nsf ) and the state of florida . we thank stan tozer and alessandro narduzzo for experimental assistance , and acknowledge stimulating discussions with james brooks , jason lashley and albert migliori . we treat a simplified version of the bandstructure @xcite , comprising a 1d electron band and a two - dimensional ( 2d ) hole band with dispersions @xmath218 and @xmath219 respectively ; here @xmath220 is the effective mass , @xmath221 is the fermi velocity of the 1d band , and @xmath222 and @xmath223 define the centre of the 2d hole pocket . each of these bands intersects the fermi energy ( @xmath170 ) , giving rise to the simplified fermi surface shown in the inset to fig . 1 . the simplest scenario to consider is that where only the 1d fermi - surface section is subject to cdw formation , giving rise to a gap @xmath224 in its density of electronic states , while the 2d hole section remains ungapped . under equilibrium conditions , the average volume charge densities @xmath225 and @xmath226 associated with the 1d and 2d fermi - surface sections respectively would be subject to the conservation equation @xmath227 , where @xmath228 is the density of charge due to the ionic cores . however , below we consider slight local deviations @xmath229 and @xmath230 in the charge density from the equilibrium values ( i.e. the total local charge densities associated with the 1d and 2d fermi - surface sections become @xmath231 and @xmath232 respectively ) ; because of the presence of the cdw , the overall conservation equation need no longer hold _ locally_. however , the charge densities must obey poisson s equation @xmath233 where @xmath234 is the electrostatic potential and @xmath235 is the permittivity . we are at liberty to set the origin of potential ; we choose @xmath236 in the absence of spatial charge variations . equation [ monsieurfish ] implies that the presence of local charge - density variations will lead to an in - plane , spatially - varying electric field @xmath237 . we now introduce the magnetic flux density @xmath238 applied along @xmath239 ( i.e. @xmath240 , @xmath241 to the 2d @xmath242 planes ) . this has two effects ; first , the crossed @xmath243 and @xmath244 fields force the 2d hole wavefunction centres to drift at a velocity @xmath245 . this leads to a spatially - varying in - plane current density , which , provided that the scattering rate is rather small ( e.g. under conditions of ballistic or quasi - ballistic transport ) , can be written as @xmath246 ; note that the relationship becomes exact in the total absence of scattering . integration of maxwell s fourth equation ( @xmath247 @xcite ) shows that this current produces an additional contribution @xmath248 to the magnetic flux density parallel to @xmath239 , which then becomes @xmath249 . a second effect of @xmath244 is to produce landau quantisation of the 2d holes ; this is usually dealt with using the landau gauge @xmath250 @xcite . after some manipulation @xcite , the schrdinger equation for the in - plane wavefunctions @xmath251 and eigenenergies @xmath252 may be written as @xmath253 . here , the effect of the crystalline potential has been taken into account by the effective mass @xmath220 ; the cyclotron frequency is @xmath254 , with @xmath255 the charge ; @xmath256 represents the wavefunction guiding - centre coordinate . the in - plane electric field also modifies the landau - level energies @xcite . to quantify this , we consider the case @xmath257 , _ i.e. _ the potential varies only in the @xmath258 direction ( variation in an arbitrary direction in the @xmath259 plane is reintroduced below ) . if @xmath260 varies slowly over lengthscales @xmath261 , it can be expanded about @xmath256 in a taylor s series @xmath262 , where the primes indicate differentiation with respect to @xmath258 . after a little algebra , the schrdinger equation becomes @xmath263 where @xmath264 is a constant ( absorbing terms in @xmath256 , @xmath265 and @xmath266 ) . the chief effect of @xmath260 comes from @xmath267 ; this gives a modified landau - level spectrum , @xmath268 , with @xmath269 an integer and @xmath270 where @xmath271 and only terms to leading order in @xmath272 and @xmath273 are retained in the right - hand bracket . reintroducing a potential variation in an arbitrary intraplane direction merely changes @xmath273 in eqn . [ maddog3 ] to @xmath274 . in the absence of potential variations , the number of states per unit volume per landau level is @xmath275 , where @xmath276 is the layer separation in the @xmath239 direction ; the introduction of varying @xmath234 ( and consequent @xmath277 ) causes @xmath278 to change to @xmath27 ( eqn . [ maddog3 ] ) , modifying this to @xmath279 , where @xmath280 the change in landau - level degeneracy results in a change to the local 2d hole density , @xmath281 it is worth taking time to understand this equation fully , since it is the key to understanding the screening effects that are the point of this paper . the first term of the right - hand side results from the change in degeneracy of _ all _ of the occupied landau levels ; positive @xmath282 or @xmath277 gives a greater landau - level degeneracy ( eqn . [ maddog4 ] ) , producing an _ increase _ in the number of holes . the second term results from the energy shift of the @xmath283th landau level closest to the chemical potential @xmath44 ; in the high - field limit of well - resolved levels , the contribution from the tails of the landau levels remote from @xmath44 can be ignored . as an increase in @xmath282 causes @xmath27 to increase ( eqn . [ maddog3 ] ) , this results in a slight _ depopulation _ of the landau levels closest to @xmath44 , acting to reduce @xmath68 . the dimensionless factor @xmath284 therefore depends critically on the position of @xmath44 amongst the landau levels ; i.e. it oscillates as a function of @xmath285 . if @xmath44 is in the middle of the highest occupied landau level ( at half - integral landau - level filling , @xmath286 ) , the density of states is large , so that the shift in energy of the landau level has a relatively large effect on @xmath68 ; in this situation , @xmath287 . on the other hand , if @xmath44 is directly between two landau levels ( at integral landau - level filling @xmath286 ) then the density of states is small and @xmath288 . this tends to zero in the case of a sample of pristine purity for which the quantum lifetime @xmath289 the final term accounts for possible spatial variations @xmath290 in the chemical potential which are sustained by the cdw ( see below ) ; here , @xmath291 is the mean value of the 2d density of states . a shift in the chemical potential also affects the 1d carrier density ; @xmath292 where @xmath293 is the 1d density of states . we can now reformulate eqn . [ gerrya ] by substituting from eqns . [ monsieurfish ] , [ fromage ] , [ maddog4 ] and [ davidt ] to yield @xmath294 we have retained terms up to first order in small quantities , omitted @xmath295 from eqn . [ monsieurfish ] because it is @xmath296 times smaller than the other term in @xmath274 and substituted @xmath297 for the hole charge ; , and so is independent of whether we consider the 2d fermi surface as a hole pocket with positive @xmath220 and @xmath255 or an electron pocket with negative @xmath220 and @xmath255 . ] here , @xmath298 . in the chemical potential @xmath44 on the 2d hole landau levels and the 1d electron dispersions ( in this case at half - integer filling ) , giving rise to a direct relationship between @xmath299 , @xmath300 and @xmath301 and consequently between @xmath301 and @xmath61.,width=302 ] as long as the 1d band remains metallic , @xmath290 is unconstrained . scattering facilitates the transfer of current between positive and negative variations in the charge density , causing them to dissipate rapidly . the stable result is when @xmath302 , causing eqn . [ everything ] to become a simple proportionality between @xmath290 and @xmath272 ; i.e. a variation in the chemical potential due to a change in magnetic field @xcite . by forming a cdw on the 1d bands , the system can benefit in two ways . first , by the opening of a cdw gap @xmath224 , the system has the potential to become partly gapped at @xmath44 . second , by forming a quantum - coherent state , the 1d band resists the spontaneous exchange of charge between the two bands . as with a superconductor , the quantum state of a cdw is locally defined by a phase @xmath303 , which defines the phase of the charge modulation that oscillates in one direction ( @xmath258 ) on a lengthscale @xmath304 much shorter than the cyclotron length . any shift in the chemical potential in eqn . [ everything ] will correspond to a gradient in phase ( see fig . [ mechanism ] ) ; @xmath305 the proportionality between @xmath306 and @xmath290 leads to two qualitatively different solutions to eqn . [ everything ] . in the bulk of the sample , the transport is dissipative ( @xmath307 ) and the cdw is able to adjust its phase within certain limits to accommodate currents caused by a non - uniform potential . [ everything ] therefore becomes @xmath308 thus , there is a direct proportionality between local variations in the magnetic field and @xmath290 . as noted above ( eq . [ phase ] ) , a change @xmath290 results in a shift in cdw phase , and , via @xcite @xmath309 to local adjustments of @xmath24 away from its equilibrium value , @xmath310 ( see fig . [ mechanism ] ) . these changes result in a concomitant shift in the free energy @xmath311 of the system , which , following ref . @xcite , can be written as @xmath312 here , @xmath313 is the free energy of the landau - level system ( which depends at fixed field on the number of quasiparticles in the 2d hole band , and hence on @xmath314 ) and @xmath315 is a dimensionless parameter depending on the position of the chemical potential amongst the landau - levels ; it takes the limiting values @xcite @xmath316 for integer landau - level filling factors and @xmath317 for half - integer filling factors . the effects described by eqs . [ dong1]-[dong3 ] result in a relationship between @xmath311 and @xmath272 , implying that there is a contribution @xmath318 from the cdw to the overall susceptibility @xmath319 , where @xmath320 . using the identity @xmath321 in conjunction with eqs . [ dong1 ] and [ dong3 ] yields @xmath322 ^ 2 . \label{irrev}\ ] ] thus , the non - equilibrium ( i.e. spatially - varying ) state of the cdw thus provides a diamagnetic contribution to the overall susceptibility , which resists changes in the magnetic field in the bulk of the sample ; note that the presence of the factors @xmath284 and @xmath315 indicate that the effectiveness of this screening will oscillate as a function of landau - level filling factor @xmath283 . however , as we discuss in the next paragraph , eq . [ irrev ] represents the response of the sample only to infinitessimal changes in magnetic field ; the build up ( and release ) of elastic energy as the field develops further leads to an irreversible change in the overall sample susceptibility which eventually saturates . the spatially - varying free energy associated with the mechanism causing eq . [ irrev ] corresponds to a force @xmath323 ; the non - equilibrium excited state of the cdw will only persist if @xmath189 remains less than the cdw s pinning force @xmath190 @xcite . this produces two restrictions ; first , because of the differential relationship between @xmath189 and @xmath290 , @xmath290 must tend to zero at the sample surface , or else @xmath189 would be singular ( and hence exceed @xmath190 ) . second , once @xmath189 develops with changing field to such an extent that the cdw depins , the non - equilibrium build up of @xmath290 must stop , preventing any further contribution from @xmath318 . these combine to ensure that the region in which @xmath191 propagates inwards from the sample surface as the magnetic field changes , so that the volume - fraction of the sample able to contribute @xmath318 ( eq . [ irrev ] ) to the overall volume - averaged sample susceptibility @xmath319 becomes progressively smaller . this is very similar to the bean critical - state model @xcite ; indeed , a common feature of _ all _ critical - state models is that motion initiates at the surface because it is there that the potential energy can be dissipated with the least amount of action . the cdw begins to slide at the sample surface so that the restoring force does not exceed the cdw depinning force . approximately linear gradients in @xmath229 and @xmath301 occur because the work done by @xmath189 is a linear function of the departure @xmath301 of the cdw from equilibrium @xcite . currents result from changes in the charge density @xmath68 associated with the two - dimensional fermi surface section as the quasiparticles redistribute themselves so as to screen the changes in the open fermi surface section ( see eq . [ monsieurfish ] ) . the variation in the orbital magnetization of the closed fermi - surface pocket @xmath179 with @xmath230 then leads to linear gradients in @xmath179 . maxwell s equations imply that a linear gradient in @xmath179 , sustained by pinning , corresponds to a critical current density @xmath324 , in the region where the cdw is able to slide . therefore , although the mechanisms are very different , there are close analogies between the situation here and flux - pinning in superconductors . it is therefore not surprising that in section [ torquetorque ] , we shall see that a simple bean model is able to fit the variation of sample susceptibility with applied magnetic field . we have already seen that @xmath325 as one approaches the sample surface because the elastic energy cost of shifting the phase would otherwise become too great . hence , eq . [ everything ] assumes the form @xmath326^{\frac{1}{2 } } , \label{gits}\ ] ] where we have substituted for @xmath272 using eq . [ fromage ] . [ gits ] defines a penetration depth @xmath133 analogous to the london penetration depth in superconductivity ; it implies that deviations @xmath234 from the equilibrium value of the electrostatic potential will be screened from the bulk of the sample . owing to the proportionality between @xmath234 and @xmath272 in eqn . [ fromage ] , spatial variations in the magnetic field will also be screened . substituting values for taken from ref . @xcite yields @xmath327 nm . the existence of a finite @xmath328 close to the surface implies quasi - ballistic transport . by carrying the current close to the edges , the total system can save energy ( in this case , elastic energy due to the non - equilibrium arrangement of the cdw within the bulk of the sample ) , and any process that saves energy also protects the quasipartices from ordinary scattering events that would otherwise give rise to dissipative bulk currents . this type of explanation for the variation in the current density , electrostatic potential , electric field and perturbed charge carrier density follows on from a model originally proposed by macdonald _ et al . _ @xcite for two - dimensional electron gas systems . whereas for a single two - dimensional layer an exact solution can only be obtained numerically , in bulk crystalline systems like , the above treatment shows that the variation in all of these quantities becomes a simple exponential function . 99 g. grner , rev . phys . * 60 * , 1129 ( 1988 ) . g. grner , _ density waves in solids ( frontiers in physics v. 89 ) _ , ( addison - wesley publishing company , 1994 ) . j. slyom , adv . phys . * 28 * , 201 ( 1979 ) . w. dietrich and p. fulde , z. phys . * 265 * , 239 ( 1973 ) . d. zanchi , a. bjelis and g. montambaux , phys . b * 53 * , 1240 ( 1996 ) . r. h. mckenzie , cond - mat/9706235 ( unpublished ) . n. harrison , phys . * 83 * , 1395 ( 1999 ) . this situation contrasts with that in superconductors , where phenomena associated with the pauli limit are accessible to laboratory fields : see e.g. j. singleton , j.a . symington , m .- s . nam , a. ardavan , m. kurmoo and p. day , j. phys . : condens . matter , * 12 * , l641 ( 2000 ) ; h. won , k. maki , s. haas , n. oeschler , f. weickert and p. gegenwart , preprint cond - mat0306548 ( 2003 ) . in superconductors , the characteristic energy scale that defines the transition temperature is debye frequency of the phonons , whereas in cdw systems it is the total electronic bandwith @xcite . foury - leylekian , s. ravy , j. p. pouget and h. muller , synth . met . * 137 * , 1271 ( 2003 ) . n. biskup , j. a. a. j. perenboom , j. s. brooks , and j. s. qualls , solid state commun . * 107 * , 503 ( 1998 ) . p. christ , w. biberacher , m. v. kartsovnik , e. steep , e. balthes , h. weiss , and h. mller , jetp lett . * 71 * , 303 ( 2000 ) . n. harrison , l. balicas , j. s. brooks , and m. tokumoto , phys . b * 62 * , 14212 ( 2000 ) . n. harrison , e. rzepniewski , j. singleton , p. j. gee , m. m. honold , p. day , and m. kurmoo , j. phys . : condens . matter * 11 * , 7227 ( 1999 ) . n. harrison , l. balicas , j. s. brooks , j. sarrao , and z. fisk , phys . b * 61 * , 14299 ( 2000 ) . n. harrison , n. biskup , j. s. brooks , l. balicas , and m. tokumoto , phys . b * 63 * , 195102 ( 2001 ) . d. andres , m. v. kartsovnik , w. biberacher , h. weiss , e. balthes , h. mller , n. kushch , phys . b * 64 * , 161104 ( 2001 ) . n. harrison , a. house , m. v. kartsovnik , a. v. polisski , j. singleton , f. herlach , w. hayes , n. d. kushch , phys . lett . * 77 * , 1576 ( 1996 ) . s. hill , p. s. sandhu , j. s. qualls , j. s. brooks , m. tokumoto , n. kinoshita , t. kinoshita , and y. tanaka , phys . b * 55 * , r4891 ( 1997 ) . n. harrison , m. m. honold , m. v. kartsovnik , j. singleton , s. t. hannahs , d. g. rickel , and n. d. kushch , phys . rev . b * 55 * , 16005 ( 1997 ) . m. m. honold , n. harrison , j. singleton , h. yaguchi , c. h. mielke , d. rickel , i. deckers , p. h. p. reinders , f. herlach , m. kurmoo , and p. day , j. phys . : condens . matter * 9 * , l533 ( 1997 ) . m. m. honold , n. harrison , j. singleton , m .- s . nam , s. j. blundell , c. h. mielke , m. v. kartsovnik , and n. d. kushch , phys . b * 59 * , r10417 ( 1999 ) . n. harrison , c. h. mielke , j. singleton , j. s. brooks , and m. tokumoto , m , j. phys . : condens . matter * 13 * , l389 ( 2001 ) . n. harrison , int . b * 16 * , 3072 ( 2002 ) . c. h. mielke , n. harrison , a. ardavan , p. goddard , j. singleton , a. narduzzo , l. k. montgomery , l. balicas , j. s. brooks , and m. tokumoto , synth 133 * , 99 ( 2003 ) . t. chakraborty and p. pietilinen , _ the quantum hall effects : integral and fractional , second edition , springer series in solid - state sciences _ * 85 * , ( springer verlag , berlin , 1995 ) . m. tinkham , _ introduction to superconductivity , second edition _ , ( mcgraw - hill , 1996 ) . j. singleton , reports on progress in physics , * 63 * , 1111 ( 2000 ) . b. i. bleaney and b. bleaney , _ electricity and magnetism _ ( oxford university press , oxford , revised third edition , 1989 ) . n. harrison , phys . rev . b * 66 * , 121101(r ) ( 2002 ) . n. harrison , c. h. mielke , a. d. christianson , j. s. brooks and m. tokumoto , phys . lett . * 86 * , 1586 ( 2001 ) . a. a. abrikosov , _ fundamentals of the theory of metals _ ( elsevier science publishers , amsterdam , 1988 ) . l. balents and m. p. a. fisher , phys . lett . * 76 * , 2782 ( 1996 ) . a. h. macdonald , t. m. rice and w. f. brinkman , phys . b * 28 * , 3648 ( 1983 ) . see e.g. leigh page and n.i . adams , _ principles of electricity _ ( van nostrand , princeton , nj , fourth edition , 1969 ) , p 319 . m. m. honold , n. harrison , m. v. kartsovnik , h. yaguchi , j. singleton , c. h. mielke , n. d. kushch , m. kurmoo , and p. day , phys . b * 62 * , 7908 ( 2000 ) . p. christ , w. biberacher , h. mller , k. andres , e. steep and a. g. m. jansen , physica b * 204 * , 153 ( 1995 ) . d. shoenberg , _ magnetic oscillations in metals _ ( cambridge university press , cambridge 1984 ) pp . m. a. itskovsky , t. maniv and i. d. vagner , phys . b * 61 * , 14616 ( 2000 ) . n. harrison , r. bogaerts , p. reinders , j. singleton , s.j . blundell and f. herlach , phys . b , * 54 * , 9977 ( 1996 ) . c. p. bean , rev . mod . phys . * 36 * , 31 ( 1964 ) . poole , h.a . farach , r.j . creswick , _ superconductivity _ ( academic press , sandiego , ca , 1995 ) , pp 371 - 390 . see e.g. , j. singleton , _ band theory and electronic properties of solids _ ( oxford university press , oxford , uk , 2001 ) , pp 87 - 89 .

in this paper , we consider the generalized hodgkin - huxley model introduced by austin in . this model describes the propagation of an action potential along the axon of a neuron at the scale of ion channels . mathematically , this model is a fully - coupled piecewise deterministic markov process ( pdmp ) in infinite dimensions . we introduce two time scales in this model in considering that some ion channels open and close at faster jump rates than others . we perform a slow - fast analysis of this model and prove that asymptotically this two - time - scales model reduces to the so called averaged model which is still a pdmp in infinite dimensions for which we provide effective evolution equations and jump rates .

1109.6581

the hodgkin - huxley model is one of the most studied models in neuroscience since its creation in the early 50 s by hodgkin and huxley @xcite . this deterministic model was first created in order to describe the propagation of an action potential or nerve impulse along the axon of a neuron of the giant squid . in our paper , we will consider a mathematical generalization of the hodgkin - huxley model investigated numerically in @xcite and analytically in @xcite . this model , called the spatial stochastic hodgkin - huxley model in the sequel , is made of a partial differential equation ( pde ) describing the evolution of the variation of the potential across the membrane , coupled with a continuous time markov chain which describes the dynamics of ion channels which are present all along the axon . it is a piecewise deterministic markov process ( pdmp ) in infinite dimensions ( see @xcite for pdmp in infinite dimensions and also @xcite and references therein for pdmp in finite dimension ) . moreover it is fully - coupled in the sense that the evolution of the potential depends on the kinetics of ion channels and vice - versa . the role of ion channels is fundamental because they allow and amplify the propagation of an action potential . we introduce different time scales in this model considering that some ion channels open and close at faster rates than others . we perform a slow - fast analysis of this model and prove that asymptotically it reduces to the so called averaged model which is still a pdmp in infinite dimensions for which we provide effective evolution equations and jump rates . we thus reduce the complexity of the original model by simplifying the kinetics of ion channels . to the best of our knowledge no averaging results are available for pdmp in infinite dimensions . even if the hodgkin - huxley model was first introduced in neurophysiology , its paradigm is commonly used to describe the behavior of cardiac cells , see for instance @xcite . it was at first deterministic . in order to include more variability , better describe the axon itself and a certain class of phenomena , stochastic versions of this model have been proposed since the 90 s , see for instance @xcite , introducing in the model two kinds of randomness . one is to include in the hodgkin - huxley equations an external noise modeling in this way the multitude of impulses received by one single neuron through its connections to the extraordinary complex system which is the brain . the other is to introduce in the hodgkin - huxley equations the intrinsic noise of the neuron due to the presence of ion channels which are intrinsically stochastic entities . the spatial stochastic hodgkin - huxley model which deals with the intrinsic noise of neurons pertains to the latter class . the generalized hodgkin - huxley equations of @xcite belong also to a larger class of stochastic systems describing the evolution of a macroscopic variable , here the membrane potential , coupled with a markovian dynamic modeling the kinetics of smaller entities , here ion channels . between two successive jumps of the markovian kinetics , the macroscopic variable follows a deterministic evolution . these models are known mathematically under the name of piecewise deterministic markov processes ( pdmp ) and are often called stochastic hybrid systems in the applications . the theory of pdmp in finite dimension ( dealing with ordinary differential equations coupled with markovian kinetics ) has been developed in @xcite . the extension to infinite dimensions ( dealing with pdes coupled with markovian kinetics ) has been performed in @xcite . the spatial stochastic hodgkin - huxley model belongs to the class of pdmp in infinite dimensions since the equation describing the evolution of the membrane potential is a pde . reducing the complexity of a model is of first importance since it allows to perform more tractable mathematics and numerical simulations and in this way , to go further in the understanding of the model . the reduction can be achieved by two different means . the first is to consider that the number or the population of ion channels is large enough to apply a law of large numbers argument so that we can replace the empirical proportion of ion channels in a given state by the averaged theoretical proportion of ion channels in this state . this is achieved by letting the number of ion channels go to infinity , the corresponding law of large numbers is proved in @xcite and in the context of pdmp in @xcite . the second way to reduce the complexity of the model , the one we will be interested in in this paper , is to consider that the model has intrinsically two time scales . it is a quite natural approach in this context , see for instance @xcite . in the present paper the number of ion channels is fixed and we consider that some ion channels have a faster dynamic than others . we perform then a slow - fast mathematical analysis of this two - time - scales system . we obtain a reduced system in which the intrinsic variability of ion channels is still taken into account but with a simplified kinetic . unlike the main result of @xcite , after reduction , we remain here at a stochastic level . the methods used in this paper are known mathematically under the name of stochastic averaging . as far as we know , these methods have never been applied to a pdmp in infinite dimensions which we are dealing with in the present paper . for pdmps in finite dimension , a theory of averaging has been developed in @xcite and central limit theorem has been proved in @xcite . in the infinite dimensional framework which we consider in this paper we have been able to reduce the complexity of a fully - coupled pdmp in infinite dimensions by using averaging methods . the natural step after this averaging result is to prove the central limit theorem associated to it . we wish to investigate this in a future work . we conclude this introduction with the plan of the paper . in section [ intro_math ] we introduce the spatial stochastic hodgkin - huxley model and the formalism of pdmp in infinite dimensions . then we introduce the two time scales in the model and prove a crucial result for the sequel . we finish this section by presenting the main assumptions on the model and our main result . section [ section_proof ] is devoted to the proof of our main result . in section [ section_ex ] we apply our result to a hodgkin - huxley type model as an example . at the end of the paper , an appendix provides an important lemma proved in @xcite , two results about tightness and the parameter values used in the simulation presented in section [ section_ex ] . we set @xmath0 $ ] . let us define : @xmath1 in this paper we will work with the following triplet of banach spaces @xmath2 . @xmath3 will be denoted by @xmath4 , it denotes the completion of the set of @xmath5 functions with compact support on @xmath6 with respect to the norm @xmath7 . @xmath8 endowed with the norm @xmath9 is the usual space of measurable and square integrable functions with respect to the lebesgue measure on @xmath6 . @xmath10 is the dual space of @xmath4 and will be denoted by @xmath11 . we recall here a few basic results about these three spaces . @xmath8 and @xmath4 are two separable hilbert spaces endowed with their usual scalar products denoted respectively by @xmath12 and @xmath13 . we denote by @xmath14 the duality pairing between @xmath4 and @xmath11 . for an integer @xmath15 we define the following functions on @xmath6 : @xmath16 the family @xmath17 ( resp . @xmath18 ) is a hilbert basis of @xmath8 ( resp . @xmath4 ) . in @xmath8 ( resp . @xmath4 ) , the laplacian with zero dirichlet boundary conditions has the following spectral decomposition : @xmath19 for @xmath20 in the domain @xmath21 ( resp . @xmath22 ) . we refer the reader to chapter 1 , section 1.3 of @xcite for more details . we denote by @xmath23 the ( poincar ) constant such that , for all @xmath24 we have : @xmath25 the embeddings @xmath26 are continuous and dense . moreover , for any @xmath27 and any @xmath24 : @xmath28 . we say that a function @xmath29 has a frchet derivative in @xmath24 if there exists a bounded linear operator @xmath30 such that : @xmath31 we then write @xmath32 $ ] for the operator @xmath33 . for example , the square of the @xmath7 norm and the dirac distribution in @xmath34 are frchet differentiable on @xmath4 . for all @xmath24 : @xmath35(h)=2(u , h),\quad \frac{d\delta_x}{du}[u](h)=h(x)\ ] ] for all @xmath36 . frchet differentiation is stable by sum and product . we introduce here the stochastic hodgkin - huxley equations considered in @xcite . basically this spatial stochastic hodgkin - huxley model describes the propagation of an action potential along an axon . the axon is the part of a neuron which transmits the information received from the soma to another neuron on long distances : the length of the axon is big relative to the size of the soma . along the axon are the ion channels which amplify and allow the propagation of the received impulse . for mathematical convenience , we assume that the axon is a segment of length one and we denote it by @xmath0 $ ] . the ion channels are assumed to be in number @xmath37 and regularly placed along the axon at the place @xmath38 for @xmath39 . this distribution of the channels is certainly unrealistic but we assume it to fix the ideas and in accordance with @xcite . however the mathematics are the same if we consider any finite subset of @xmath6 instead of @xmath40 . each ion channel can be in a state @xmath41 where @xmath42 is a finite states space . for example , for the hodgkin - huxley model , a state can be : `` receptive to sodium ions and open '' ( see @xcite ) . the ion channels switch between states according to a continuous time markov chain whose jump intensities depend on the local potential of the membrane , that is why the model is said to be fully - coupled . for any states @xmath43 we define by @xmath44 the jump intensity ( or rate function ) from the state @xmath45 to the state @xmath46 . it is a real valued function of a real variable and supposed to be , as is its derivative , lipschitz - continuous . we assume moreover that : @xmath47 for any @xmath43 and either @xmath44 is constant equal to zero or is strictly positive bounded below by a strictly positive constant @xmath48 . that is , the non - zero rate functions are bounded below and above by strictly positive constants . for a given channel , the rate function describes the rate at which it switches from one state to another . a possible configuration of all the @xmath49 ion channels is denoted by an element @xmath50 of @xmath51 : @xmath52 is the state of the channel which is at the position @xmath38 for @xmath39 . the channels , or stochastic processes @xmath52 , are supposed to evolve independently over infinitesimal time - scales . denoting by @xmath53 the local potential at point @xmath38 at time @xmath54 , we have : @xmath55 for any @xmath41 we also define the maximal conductance @xmath56 and the steady state potentials @xmath57 of a channel in state @xmath45 which are both constants , the first being positive . the membrane potential @xmath58 along the axon evolves according to the following pde : @xmath59 we will assume zero dirichlet boundary conditions for this pde ( clamped axon ) . we are interested in the process @xmath60}$ ] . we recall here the result of @xcite which states that there exists a stochastic process satisfying equations ( [ dyn ] ) and ( [ saut ] ) . let @xmath61 be in @xmath62 such that @xmath63 , the initial potential of the axon . let @xmath64 be the initial configuration of ion channels . [ prop_a1 ] fix @xmath65 and let @xmath66 be a filtered probability space satisfying the usual conditions . there exist a pair @xmath67 of cdlg adapted stochastic processes such that each sample path of @xmath20 is a continuous map from @xmath68 $ ] to @xmath4 and @xmath69 is in @xmath70 for all @xmath71 $ ] and satisfying : * ( regularity ) the map @xmath72 lies in @xmath73,h^*)$ ] almost surely . * ( dynamic : pde ) @xmath74 @xmath75 $ ] , @xmath76-almost surely . * ( dynamic : jump ) @xmath77 * ( initial condition : pde ) @xmath61 given . * ( initial condition : jump ) @xmath78 given . * ( boundary conditions : pde only ) @xmath79 , @xmath75 $ ] . moreover there exists a constant @xmath80 such that @xmath81 for all @xmath71 $ ] and @xmath82 . in @xcite the author proves that when the number of ion channels increases to infinity , the above model converges , in a sense , toward a deterministic model . in our paper , unlike in @xcite we will work with a fixed number of ion channels but introduce two time scales in the model in order to perform a slow - fast analysis . the paper @xcite extends the theory of finite dimensional piecewise deterministic markov processes ( pdmp ) introduced in @xcite and @xcite to pdmp in infinite dimensions . the results stated here come from section 2.3 and 3.1 of @xcite adapted to our particular notations . for all @xmath83 we define the following function on @xmath4 : @xmath84 which is @xmath11 valued . we do not write the dependence in @xmath49 in the notation of @xmath85 since , unlike in @xcite , @xmath49 is a fixed parameter here . the stochastic process @xmath60}$ ] can be described in the following way . we start with a given initial potential @xmath86 and ion channels configuration @xmath78 . then the @xmath20 component evolves according to the evolution equation : @xmath87 until the first jump of the @xmath88 component at time @xmath89 which occurs according to the transition rate function @xmath90 given for @xmath91 by : @xmath92 according to ( [ saut ] ) . a new value @xmath93 for @xmath88 is then chosen according to a transition measure @xmath94 from @xmath95 to @xmath96 ( the set of all probability measures on @xmath70 ) . this transition measure is also given by the jump distribution of the ion channels . for @xmath91 and @xmath97 which differs from @xmath88 only by the component @xmath98 : @xmath99 if @xmath100 differs from @xmath88 by two or more components then @xmath101 . then @xmath20 evolves according to the `` updated '' evolution equation : @xmath102 with initial condition @xmath103 and dirichlet boundary conditions , until the second jump of the @xmath88 component . and so on . this description justifies the term `` piecewise deterministic process '' : the dynamic of the process is indeed purely deterministic between the jumps . as shown in @xcite in theorem 6 , the equations of proposition [ prop_a1 ] define a standard piecewise deterministic process ( pdp ) in the sense of definition 3 of @xcite . we recall briefly here what this means technically : * for a given ion channel configuration @xmath83 , the pde : @xmath104 with zero dirichlet boundary conditions admits a unique global ( weak ) solution continuous on @xmath4 for every initial value @xmath105 denoted by @xmath106 . + * the number of state switches in the @xmath107}$ ] component during any finite time interval is finite almost surely . + * the transition rate @xmath108 from one state of @xmath70 to another is a measurable function from @xmath95 to @xmath109 and for all @xmath91 , the transition rate as a function of time is integrable over every finite time interval but diverges to @xmath110 when it is integrated over @xmath109 ( the expected number of jumps tends to @xmath110 when the time horizon increases ) . + * the transition measure @xmath94 from one state of @xmath70 to another is such that the mapping @xmath111 is measurable for every @xmath112 and @xmath113 for all @xmath91 . in fact the stochastic process @xmath60}$ ] is more than a piecewise deterministic process , it is a piecewise deterministic _ markov _ process . [ th_br ] 1 . there exists a filtered probability space satisfying the usual conditions such that our infinite - dimensional standard pdp is a homogeneous markov process on @xmath95 . + 2 . locally bounded measurable functions on @xmath95 which are absolutely continuous as maps @xmath114 for all ( x , r ) are in the domain @xmath115 of the extended generator . the extended generator is given for almost all @xmath54 by : @xmath116 where @xmath117\alpha_{r_t(i),\zeta}(u_t(\frac{i}{n}))\ ] ] with @xmath118 is the component of @xmath70 with @xmath119 equals to @xmath120 if @xmath121 and to @xmath46 if @xmath122 . the notation @xmath123 means that the function @xmath124 is differentiated at @xmath125 , where @xmath126 is the solution of the pde of proposition [ prop_a1 ] such that @xmath127 and with the channel state @xmath88 held fixed equal to @xmath69 . it is not usual to write a generator only along paths @xmath128 as we do . we would expect for a generator an analytical expression of the form @xmath129 for any @xmath91 . it is in fact possible to obtain such an expression for more regular function @xmath130 . the part of the generator related to the continuous markov chain @xmath88 takes the form @xmath131 with no changes but the derivative @xmath132 can then be expressed thanks to the frchet derivative of @xmath130 , see @xcite for more details . we will not use this refinement in the sequel . knowing that a stochastic process is markovian is crucial for its study since a lot of mathematical tools have been developed for markovian processes , see for example @xcite . we introduce now two time scales in the model . indeed , we consider that the hodgkin - huxley model is a slow - fast system : some states of the ion channels communicate faster between them than others , see for example @xcite . mathematically , this leads to introduce a small parameter @xmath133 in our equation . for the states which communicate at a faster rate , we say that they communicate at the usual rate divided by @xmath134 . we can then consider different classes of states or partition the state space @xmath42 in states which communicate at a high rate . this kind of description is very classical , see for example @xcite . we regroup our states in classes making a partition of the state space @xmath42 into : @xmath135 where @xmath136 is the number of classes . inside a class @xmath137 , the states communicate faster at jump rates of order @xmath138 . states of different classes communicate at the usual rate of order @xmath139 . for @xmath133 fixed , we denote by @xmath140 the modification of the pdmp introduced in the previous section with now two time scales . its generator is , for @xmath141 : @xmath142 @xmath143 is the component of the generator related to the continuous time markov chain @xmath144 . according to our slow - fast description we have the two time scales decomposition of this generator : @xmath145 where the `` fast '' generator @xmath146 is given by : @xmath147\alpha_{r^{\varepsilon}_t(i),\zeta}(u^{\varepsilon}_t(\frac{i}{n}))\ ] ] and the `` slow '' generator is given by : @xmath148\alpha_{r^{\varepsilon}_t(i),\zeta}(u^{\varepsilon}_t(\frac{i}{n}))\ ] ] for @xmath149 fixed and @xmath150 , we denote by @xmath151 , @xmath152 the following generator : @xmath153\alpha_{\xi,\zeta}(y)\ ] ] which will be of interest in the sequel . here is a crucial result for the proof of our main result . its proof takes back the argument developed in @xcite but for the sake of completeness and given the intensive use of this proposition in the sequel , we provide a short proof . [ prop_bound ] for any @xmath154 , there is a deterministic constant @xmath155 independent of @xmath1560,1]$ ] such that : @xmath157 } \|u^{\varepsilon}_t\|_h\leq c\ ] ] almost surely . in the proof of this proposition , we will use the following representation of a solution of the pde part of our pdmp . we say that @xmath158 is a mild solution to ( [ dyn ] ) if : @xmath159 almost surely where @xmath160 is the semi - group associated to @xmath161 in @xmath4 . for any @xmath24 : @xmath162 let us mention that the mild formulation ( [ mild ] ) holds also up to a bounded stopping time @xmath163 , this will be useful later in the paper . we work @xmath82 by @xmath82 . using the mild formulation ( [ mild ] ) , we see that : @xmath164 by lemma [ aulem ] of appendix [ app_lem ] the first of the above sums is bounded by a fixed constant which can be chosen independent of @xmath71 $ ] . for the second sum , we break the integral in two pieces . for all @xmath165 by lemma [ aulem ] , we can choose @xmath166 so small that : @xmath167}|c_\xi u^{\varepsilon}_s(\frac in)|\ ] ] since @xmath168 , choosing @xmath169 such that @xmath170 we have : @xmath171}\|u^{\varepsilon}_s\|_h\ ] ] now the second estimates of lemma [ aulem ] gives : @xmath172}\|u_\sigma^{\varepsilon}\|_h ds\end{aligned}\ ] ] reassembling the above inequalities , the end of the proof is a classical application of the gronwall s lemma . we find finally that @xmath173}\|u_t^{\varepsilon}\|_h$ ] is bounded by a fixed constant independent of @xmath71 $ ] , @xmath82 and @xmath1560,1]$ ] . we make the following assumptions . for any @xmath149 fixed , and any @xmath152 , the fast generator @xmath151 is weakly irreducible on @xmath137 , i.e has a unique quasi - stationary distribution denoted by @xmath174 . this quasi - stationary distribution is supposed to be , as is its derivative , lipschitz - continuous in its argument @xmath175 ( see section [ campaign ] for more details ) . following @xcite , the states in @xmath137 can be considered as equivalent . for any @xmath39 we define the new stochastic process @xmath176 by @xmath177 when @xmath178 and abbreviate @xmath137 by @xmath179 . we then have an aggregate process @xmath180 with values in @xmath181 . it is not a markov process but we have the following proposition : [ yz ] for any @xmath182 , @xmath39 and @xmath183 , @xmath180 converges weakly when @xmath134 goes to @xmath184 to a markov process @xmath185 generated by : @xmath186 we then average the function @xmath85 against the quasi - invariant distributions . that is we consider that each pack of states @xmath137 has reached its stationary behavior . for any @xmath187 we define the averaged function by : @xmath188 therefore , we call averaged equation of ( [ dyn ] ) the following pde : @xmath189 with zero dirichlet boundary condition and initial condition @xmath61 and @xmath190 ( where @xmath190 is the aggregation of the initial channel configuration @xmath78 ) . in equation ( [ au_moy ] ) , the process @xmath191}$ ] evolves , each coordinate independently over infinitesimal time - scales , according to the averaged jump - rates between the subsets @xmath137 of @xmath42 . for any @xmath154 there exists a probability space satisfying the usual conditions such that equation ( [ au_moy ] ) defines a pdmp @xmath192}$ ] in infinite dimensions in the sense of @xcite . moreover , there is a constant @xmath80 such that : @xmath157}\|u_t\|_h\leq c\ ] ] and @xmath193,h)$ ] almost - surely . the equation ( [ au_moy ] ) is of the form ( [ dyn ] ) except that the indicator function is replaced by the probabilities @xmath194 . this changes nothing to the mathematics and the arguments developed in @xcite or @xcite still apply . we refer the reader to @xcite and @xcite for more details . we can now state our main result which states that our approximation ( [ au_moy ] ) is valid . [ apde ] the stochastic process @xmath158 solution of ( [ dyn ] ) converges in law to a solution of ( [ au_moy ] ) when @xmath134 goes to @xmath184 . more precisely , we will prove the following proposition : [ prop_main ] @xmath1950,1])$ ] is tight in @xmath196,h)$ ] and any accumulation point @xmath20 verifies : @xmath197 for all @xmath198 in @xmath199 . moreover the accumulation point is unique up to indistinguishability . to make the proof of our main result easier to read , we will consider in a first step that all the states communicate at a fast rate . this is called the all - fast case . we will prove our main result in detail in this case and then give the key points for the validity of our proof in the general case . in the all - fast case there is a unique class @xmath42 and the generator has the simplest form : @xmath200\alpha_{r(i),\zeta}(u(\frac{i}{n}))\ ] ] we make the following assumptions : for any @xmath39 with @xmath24 fixed , the markov process @xmath52 has a unique stationary distribution @xmath201 . then the process @xmath202 has the following stationary distribution : @xmath203 we assume that the rate functions are lipschitz continuous . the average ( generalized ) function is then : @xmath204 if @xmath20 is held fixed . is this hypothesis reasonable ? for the most classical model in neurosciences , the rate functions are indeed , as well as their derivatives , lipschitz continuous . see for example the rate functions in classical hodgkin - huxley model @xcite ( see also the appendix [ app ] ) . + + let us show briefly that in this case the ergodic measure @xmath205 is lipschitz continuous in its argument . let @xmath206 be in @xmath207 and the membrane potential at @xmath38 be fixed equal to @xmath208 , the state space of the continuous time markov chain @xmath52 is @xmath42 which is a finite set . we set @xmath209 . we can then enumerate the elements of @xmath42 : @xmath210 . the generator of @xmath52 is then the matrix : @xmath211 with @xmath212 . therefore , if we assumed that our continuous time markov chain is ergodic , it has an invariant measure which is of the form : @xmath213 the sums and products involved here are calculated over subsets of @xmath42 which are given by resolving the equation @xmath214 . @xmath205 is then lipschitz continuous in @xmath215 since each @xmath44 is lipschitz continuous in @xmath215 and bounded . the assumption of independence of each coordinate of the process @xmath88 over infinitesimal time - scales is important to have a simple expression of the invariant measure of the process @xmath88 and simplify the mathematics . it is possibly unrealistic from the biological point of view but is often assumed in mathematical neurosciences . the plan of the campaign to prove our main result ( [ apde ] ) in the all - fast case is the following . we want to prove the convergence in law of a family of cdlg stochastic processes with values in a hilbert space towards another , here deterministic , process . there is a well known strategy to do that : * prove the tightness of the family * identify the limit in our case , tightness ( see subsection [ main_proof ] ) will follow roughly from the uniform boundedness of the family @xmath1950,1])$ ] obtained in proposition [ prop_bound ] . the identification of the limit will follow on one side from tightness and on the other side from some classical argument in the theory of averaging for stochastic processes , see for example @xcite . we will introduce a poisson equation which will enable us to control some problematic terms ( see proposition [ prop_poi ] ) and identify the limit ( subsection [ main_proof ] ) . we will then move to the general case with multiple classes ( in subsection [ cas_g ] ) with an emphasis on each point which could give an issue and show how things work in this case . we want to prove proposition [ prop_main ] . let us recall that in the all - fast case @xmath216 in ( [ id_lim ] ) reduces to @xmath217 in ( [ f ] ) . let us outline the strategy of the proof following the plan of the campaign announced in section [ campaign ] . * tightness . 1 . thanks to proposition [ prop_bound ] , use the markov inequality to prove that aldous s condition and property ( [ th1 ] ) of theorems [ gct ] and [ th ] in appendix [ met ] hold . 2 . for @xmath71 $ ] fixed , find the finite dimensional space @xmath218 of ( [ th2 ] ) , theorem [ th ] appendix [ met ] using a truncation of the solution @xmath219 in the mild formulation . bound @xmath220 and use the markov inequality . 3 . obtain tightness in @xmath221,h)$ ] by theorems [ gct ] and [ th ] appendix [ met ] and conclude that tightness in @xmath196,h)$ ] holds as well thanks to the regularity of the solution @xmath158 for any @xmath1560,1]$ ] . * identification of the limit . 1 . begin by showing that the following property is sufficient : @xmath222 to prove this property , use the tightness of the family @xmath1950,1])$ ] . 2 . replace the term @xmath223 by @xmath224 where @xmath130 is the solution of a poisson equation . show that this solution exists and has some nice properties . 3 . write the semi - martingale expansion of @xmath225 . bound its martingale part thanks to the bound for the bracket of the martingale . 4 . bound its finite variation part thanks to the nice properties of @xmath130 and the uniform bound from proposition [ prop_bound ] . aggregate all these bounds together to obtain that @xmath226 is bounded by a constant times @xmath134 . conclude that ( [ id_lim ] ) holds and implies uniqueness up to indistinguishability . step 1 ( property ( [ th1 ] ) and aldous condition from appendix [ met ] ) . we begin with the aldous condition . let @xmath163 be a stopping time and @xmath227 such that @xmath228 . we need to control the size of @xmath229 . at first we work deterministically . we break the difference @xmath230 into three pieces using the mild formulation of @xmath158 in ( [ mild ] ) : @xmath231 let us show now that the supremum over @xmath2320,\eta[$ ] of each term on the right hand side of the above inequality goes to @xmath184 with @xmath233 ( we take @xmath166 ) . we start with the third term : @xmath234 notice that : @xmath235 where we have used the fact that @xmath236 ( recall that @xmath237 ) . thus we obtain : @xmath238 where @xmath80 is the constant of proposition [ prop_bound ] . by dominated convergence , we see that : @xmath2390,\eta[}\sum_{k\geq1}\left(e^{-(k\pi)^2\theta}-1\right)^2\frac{1+(k\pi)^2}{(k\pi)^4}\leq\lim_{\eta\to0}\sum_{k\geq1}\left(e^{-(k\pi)^2\eta}-1\right)^2\frac{1+(k\pi)^2}{(k\pi)^4}=0\ ] ] for the second term as before , we can show that : @xmath240 by dominated convergence , we see that : @xmath2390,\eta[}\sum_{k\geq1}(1+(k\pi)^2)\frac{\left(1-e^{-(k\pi)^2\theta}\right)^2}{(k\pi)^4}\leq\lim_{\eta\to0}\sum_{k\geq1}(1+(k\pi)^2)\frac{\left(1-e^{-(k\pi)^2\eta}\right)^2}{(k\pi)^4}=0\ ] ] for the first term we have , by the bessel - parseval equality : @xmath241 by dominated convergence , we see that : @xmath2390,\eta[}\sum_{k\geq1}(e^{-(k\pi)^2\theta}-1)^2(u_0,e_k)^2\leq\lim_{\eta\to0}\sum_{k\geq1}(e^{-(k\pi)^2\eta}-1)^2(u_0,e_k)^2=0\ ] ] combining the results obtained for the three terms we have : @xmath2390,\eta[}\|u^{\varepsilon}_{\tau+\theta}-u^{\varepsilon}_\tau\|^2_h=0\ ] ] uniformly in @xmath1560,1]$ ] . therefore , for all @xmath242 , using the chebyshev inequality , we can choose @xmath233 so small that : @xmath2430,1]}\sup_{\theta\in]0,\eta[}{\mathbb{p}}(\|u^{\varepsilon}_{\tau+\theta}-u^{\varepsilon}_\tau\|_h\geq m)\leq\sup_{{\varepsilon}\in]0,1]}\sup_{\theta\in]0,\eta[}\frac{{\mathbb{e}}(\|u^{\varepsilon}_{\tau+\theta}-u^{\varepsilon}_\tau\|^2_h)}{m^2}<\delta\ ] ] the first condition ( [ th1 ] ) , theorem [ th ] appendix [ met ] is verified by the application of the markov inequality . indeed , for any @xmath71 $ ] and @xmath244 by proposition [ prop_bound ] there exists a constant @xmath155 independent of @xmath1560,1]$ ] and @xmath71 $ ] such that : @xmath2430,1 ] } { \mathbb{p}}(\|u^{\varepsilon}_t\|_{h}>\rho)\leq \frac{1}{\rho}\sup_{{\varepsilon}\in]0,1]}{\mathbb{e}}\|u^{\varepsilon}_t\|_{h}\leq\frac{c}{\rho}<\delta\ ] ] for any @xmath245 large enough . step 2 ( truncation ) . we have to show that for any @xmath246 we can find @xmath247 and a space @xmath218 such that : @xmath2430,{\varepsilon}_0 ] } { \mathbb{p}}(d(u^{\varepsilon}_t , l_{\delta,\eta})>\eta)\leq\delta\ ] ] where @xmath248 . we fix @xmath71 $ ] . recalling the mild representation ( [ mild ] ) , we have : @xmath249 where , using the explicit expression of the semi - group @xmath250 , @xmath251 is equal to @xmath252 . we define , for @xmath253 and @xmath34 : @xmath254 for any @xmath246 we can find @xmath255 independent of @xmath134 such that @xmath256 with the constant @xmath257 deterministic and independent of @xmath258 and @xmath1560,1]$ ] . indeed we can easily show , as in step 1 , that : @xmath259 which is independent of @xmath1560,1]$ ] . the convergence of each series , uniformly in @xmath134 , enables us to choose a suitable @xmath255 independent of @xmath134 . let us denote @xmath260 . we choose @xmath261 . since @xmath262 @xmath263 markov s inequality gives us : @xmath264 with @xmath265 independent of @xmath1560,1]$ ] . step 3 ( tightness ) . therefore using theorems [ gct ] and [ th ] of appendix [ met ] , @xmath1950,1])$ ] is tight in @xmath221,h)$ ] . since we know that for each @xmath1560,1]$ ] , @xmath158 is in @xmath196,h)$ ] we have in fact that @xmath1950,1])$ ] is tight in @xmath196,h)$ ] . step 1 ( reduction of the problem ) . we know that the family @xmath1950,1])$ ] is tight in @xmath196,h)$ ] . we still denote by @xmath158 a converging sub - sequence of @xmath158 and we denote by @xmath20 the corresponding accumulation point . we want to show that for all @xmath198 in @xmath199 : @xmath266 almost surely . we will show that : @xmath267 notice that : @xmath268 since the function : @xmath269 is continuous on @xmath196,h)$ ] and the arguments @xmath1950,1])$ ] and @xmath20 are uniformly bounded in @xmath134 in @xmath196,h)$ ] , by convergence in law we have that : @xmath270 therefore ( [ id_weak ] ) will follow if @xmath271 . we will show more precisely that : @xmath272 step 2 ( poisson equation ) . we introduce the following poisson equation on @xmath273 : @xmath274 with @xmath275 the generator given by equation ( [ gene ] ) : @xmath276\alpha_{r(i),\zeta}(u(\frac{i}{n}))\ ] ] [ prop_poi ] the poisson equation ( [ poisson ] ) has a unique solution @xmath130 which is measurable and locally lipschitz continuous in its first variable with respect to the @xmath7 norm . for all fixed @xmath83 , the map @xmath277 has a frchet derivative denoted by @xmath278 $ ] . moreover @xmath279}|f(u^{\varepsilon}_s , r^{\varepsilon}_s)|$ ] and @xmath279}\left\|\frac{df}{du}[u^{\varepsilon}_s , r^{\varepsilon}_s]\right\|_{h^*}$ ] are both bounded almost surely by constants independent of @xmath134 . we begin the proof of the above proposition by the following lemma . [ lem_poi ] let @xmath280 be a bounded domain and @xmath88 be in @xmath70 . the map @xmath281 is : 1 . bounded on @xmath282 by a constant independent of @xmath88 . 2 . lipschitz continuous on @xmath282 with lipschitz constant independent of @xmath88 . frchet differentiable on @xmath282 with frchet derivative in @xmath24 denoted by @xmath278 $ ] . moreover the frchet derivative is bounded in @xmath283 norm uniformly in @xmath284 and @xmath83 . recall that : @xmath285 since for @xmath41 , @xmath286 is bounded and lipschitz continuous on @xmath287 , the two first points follow . for the third point we note that the map @xmath281 is a linear combination of the frchet differentiable functions @xmath288 , @xmath289 and of the product of these two functions . the frchet derivative in @xmath24 against @xmath36 is given by : @xmath290\phi(\frac{i}{n})h(\frac in)\ ] ] which gives us the boundedness property uniformly in @xmath83 and @xmath284 . for @xmath24 fixed , @xmath291 is an operator on @xmath292 which is a space of finite dimension . the fredholm alternative in such a space is @xmath293 . moreover , by the ergodicity assumption , for any fixed @xmath24 , @xmath294 . therefore equation ( [ poisson ] ) with @xmath24 fixed has a solution if and only if : @xmath295 this latter equality holds by the definition of @xmath217 and the bi - linearity of the duality product @xmath14 . moreover , we can then always choose @xmath130 satisfying @xmath296 by taking its projection on @xmath297 . thus we have a solution @xmath298 for any @xmath24 fixed . uniqueness of @xmath130 follows easily from the condition @xmath296 . + recall that for all @xmath1560,1]$ ] , @xmath299 where @xmath300 where @xmath80 is the deterministic constant given by proposition [ prop_bound ] . then the desired properties of @xmath130 follow from lemma [ lem_poi ] and the fact that for a bounded domain @xmath301 , each function @xmath302 is bounded below and above by strictly positive constants , lipschitz continuous and frchet differentiable with frchet derivative uniformly bounded in @xmath20 , @xmath46 and @xmath88 . step 3 ( bound the martingale part ) . our previous result implies that @xmath141 and : @xmath303 defines a square - integrable martingale , see for example ethier and kurtz @xcite , chapter 4 proposition 1.7 . [ prop_var ] we have : @xmath304 ^ 2\alpha_{r_s(i),\zeta}(u^{\varepsilon}_s(\frac{i}{n}))ds\ ] ] it is a classical proof in stochastic calculus ( see for example @xcite , chapter 1 , problem 29 ) . the it and the dynkin formulas give two distinct decompositions of the squared process @xmath305 in a semi - martingale . the uniqueness of the doob - meyer decomposition of a semi - martingale enables us to identify the bracket of our martingale . we have the following semi - martingale decomposition : @xmath306 recall that we are interested in bounding @xmath307 . since @xmath279}|f(u^{\varepsilon}_s , r^{\varepsilon}_s)|$ ] is bounded independently of @xmath1560,1]$ ] , denoting by @xmath308 one of the two first terms of ( [ dec_sm ] ) we have @xmath309 where @xmath257 is a constant independent of @xmath1560,1]$ ] . for the martingale term , by the it isometry : @xmath310 since @xmath311 is bounded uniformly in @xmath71 $ ] and @xmath1560,1]$ ] thanks to the bounds on @xmath130 and each @xmath44 . more precisely we have : @xmath312 ^ 2\alpha_{r_s(i),\zeta}(u^{\varepsilon}_s(\frac{i}{n}))ds\\ & \leq&\frac{4t}{{\varepsilon}}\alpha^+n|e|\sup_{s\in[0,t]}|f(u^{\varepsilon}_s , r^{\varepsilon}_s)|^2\end{aligned}\ ] ] step 4 ( bound the finite variation part ) . it remains to bound the third term of our semi - martingale decomposition . [ prop_var ] there exists a constant @xmath80 independent of @xmath1560,1]$ ] such that : @xmath313 almost surely . recall first in theorem [ th_br ] the meaning of the notation @xmath314 . recall that the map @xmath315 for @xmath83 fixed , is frchet differentiable with frchet derivative in @xmath20 denoted by @xmath278 $ ] which is a bounded linear form on @xmath4 . by the riesz representation theorem there exists @xmath316 such that : @xmath317(h)=(f_u(u , r),h)_h\quad\forall h\in h\ ] ] moreover the correspondence is isometric : @xmath318\right\|_{h^*}=\left\|f_u(u , r)\right\|_h$ ] . we then have ( see theorem 4 of @xcite ) : @xmath319 by proposition [ prop_poi ] , there exists a constant @xmath320 independent of @xmath1560,1]$ ] and @xmath71 $ ] such that : @xmath321\right\|_{h^*}\leq c_1\ ] ] therefore : @xmath322 it remains to show that @xmath323 is bounded uniformly in @xmath134 and @xmath71 $ ] . for @xmath324 we have , denoting by @xmath325 the derivative with respect to @xmath215 : @xmath326 where @xmath80 is the deterministic constant given by proposition [ prop_bound ] . thus : @xmath327 this ends the proof . step 5 ( all bounds together ) . assembling all the bounds of the different terms we see that : @xmath328 with the constant @xmath80 independent of @xmath1560,1]$ ] . it remains to let @xmath134 go to @xmath184 to conclude the proof of the identification of the limit . step 6 ( uniqueness ) . we consider @xmath329 and @xmath330 two possible accumulation points verifying , for @xmath331 : @xmath332 for all @xmath71 $ ] almost surely . we can show easily that for any @xmath71 $ ] : @xmath333}\|u^1_s - u^2_s\|_h)\leq c\int_0^t { \mathbb{e}}(\sup_{l\in[0,s]}\|u^1_l - u^2_l\|_h)dl\ ] ] with the constant @xmath80 independent of @xmath71 $ ] . therefore , by application of the gronwall s lemma we obtain that @xmath334}\|u^1_t - u^2_t\|_h)=0 $ ] . we now consider the original case with different state classes @xmath335 . the first thing to do is to prove the tightness of the family @xmath3360,1])$ ] in @xmath221,h\times\{1,\cdots , l\}^{|{\mathbb{n}\cap{n\mathring{i}}}|})$ ] ( see section [ maae ] for the definition of @xmath337 ) . we apply a comparison argument . we notice that : @xmath338 for @xmath43 we set : @xmath339 we denote by @xmath340 the associated jump process with constant rates @xmath341 . we have the following stochastic domination : @xmath342 we construct the process @xmath343 as @xmath337 by aggregation . the sequence @xmath3440,1])$ ] is tight in @xmath221,\{1,\cdots , l\}^{|{\mathbb{n}\cap{n\mathring{i}}}|})$ ] , see for instance @xcite . therefore , by comparison , the sequence @xmath3450,1])$ ] is also tight in @xmath221,\{1,\cdots , l\}^{|{\mathbb{n}\cap{n\mathring{i}}}|})$ ] , see for instance @xcite . moreover , the sequence @xmath1950,1])$ ] is tight in @xmath221,h)$ ] by applying the arguments developed for the `` all - fast '' case . endowing @xmath346 with the product topology we see that the sequence @xmath3360,1])$ ] is tight in @xmath221,h\times\{1,\cdots , l\}^{|{\mathbb{n}\cap{n\mathring{i}}}|})$ ] . we must now deal with the identification of the limit . there are six steps in the proof of the identification of the limit , we have to check that these six steps generalize to the general case ( jumping slow - fast case ) . in step 1 , by the same arguments , we obtain that it is sufficient to show that : @xmath347 in step 2 , the poisson equation becomes : @xmath348 where @xmath146 is now the `` fast '' part of the generator ( @xmath130 does not depend on @xmath349 explicitly because @xmath349 is constructed from @xmath88 ) . a configuration @xmath88 for the ion channels is now : @xmath350 where @xmath351 ( noting that @xmath352 ) . that is each channel is in one of the class @xmath137 for @xmath152 then , since for fixed @xmath20 the quasi - stationary measure associated to the class @xmath353 for @xmath354 is @xmath355 , we have that the kernel of @xmath356 is spanned by : @xmath357 when @xmath20 is held fixed . then , by the fredholm alternative and the definition of the averaged function @xmath358 , the poisson equation ( [ poisson_g ] ) has a solution . uniqueness follows by the projection condition : @xmath359 the proof of the last three steps is then exactly the same as in the `` all fast '' case . in this section we give a concrete example where our result allows to reduce the complexity of a neuronal model staying nevertheless at a stochastic level . we consider a usual hodgkin - huxley model but , to be very straightforward in the application of our result , we only consider the sodium current i.e. all the ion channels are sodium channels . it is still a case of interest since sodium channels are involved in the increasing phase of an action potential . at a fixed potential , the kinetic of a sodium channel is described by figure [ kin_na ] which represents the states and the jump rates of a continuous time markov chain denoted by @xmath360 for the channel at position @xmath361 for @xmath39 . a sodium channel can be in @xmath362 different states denoted by @xmath363 for @xmath364 and @xmath365 , the state @xmath366 is the only `` open '' state for the sodium channel ( see @xcite for more details ) . we simply write @xmath367 and @xmath368 for @xmath369 and @xmath370 ( see appendix [ app ] for more details on the rate functions ) . @xmath371 is the fast variable and @xmath372 the slow variable for the kinetic of a sodium channel . @xmath373 each channel can therefore be in one of these eight states divided in two classes : @xmath374 where @xmath375 and @xmath376 . inside these two classes the states communicate at fast rates and transition between these two classes occurs at slow rates . a sodium channel is open if and only if it is in the state @xmath366 . we then define @xmath377 , @xmath378 and @xmath379 , @xmath380 when @xmath381 . the equation ( [ dyn ] ) describing the evolution of the potential , for a number @xmath49 of ion channels , is here given by : @xmath382 the global variable @xmath20 represents the difference of potential between the inside and the outside of the axon membrane . we see that equation ( [ ex ] ) is indeed of the form that we have studied in this paper . @xmath383 is a constant related to the radius and the internal conductivity of the axon ( see appendix [ app ] ) . let us compute the different generators and quasi - invariant measures associated to this model according to our description . the two `` fast '' generators are the same and are given for @xmath384 fixed ( again , we do not make the dependence appear ) , by : @xmath385 with @xmath386 or @xmath139 . we can compute the associated quasi - invariant measure @xmath387 ( @xmath386 or @xmath139 ) , the only term of interest for us is : @xmath388 the reader familiar with the classical hodgkin - huxley model will notice that @xmath389 is in fact equal to the steady - state function associated to the ode describing the motion of the @xmath371-gates in the classical deterministic hodgkin - huxley description , see for example @xcite . the asymptotic aggregated markov chain @xmath349 is valued in the two - state space @xmath390 . according to proposition [ yz ] its generator is : @xmath391 therefore , according to theorem [ apde ] , the reduced model is described by the following pde coupled with the continuous - time markov chain @xmath349 : @xmath392 we display a realization of the averaged piecewise deterministic markov process in figure [ fig2 ] . to perform the numerical simulation ( in c ) we extended riedler s work ( cf . @xcite ) to our particular framework . in @xcite , an algorithm ( algorithm a2 ) is proposed to simulate the trajectory of a pdmp in finite dimension . the idea is to simulate the jumping part of the pdmp and to use an accurate method to simulate the ode constituting the deterministic part of the pdmp . the kinetic of the jumping component for pdmp in finite and infinite dimensions have the same form . here we simulate the jumping part of our infinite dimensional pdmp following @xcite and simulate the pde between successive jumps by a deterministic scheme . for the pde , we used an explicit finite difference euler scheme in space and time . simulating the jumps of an inhomogeneous time markov chain is classical and there exist a lot of efficient numerical schemes for pde . therefore the described method is natural . this argument is of course not a proof but gives an heuristic interpretation of our approach . ) is displayed for a number of ion channels fixed to @xmath393 . we see what we expect from the partial differential equation : a traveling wave connecting the initial state @xmath184 to the stable state @xmath394.,width=377 ] in this paper , we apply the methods of stochastic averaging to a fully coupled pdmp in infinite dimensions . to the best of our knowledge it is the first time that such a problem is considered . the two main difficulties are due , on one hand to the infinite dimensional setting and on the other hand to the fact that the system is fully - coupled . regarding the fully - coupled property , our task was made easier by the uniform bound on the macroscopic component of the pdmp and on the jump rates of the continuous time markov chain . as an application , we have shown how to reduce the complexity of a generalized hodgkin - huxley model which can be used as a paradigm in the modelling of the propagation of an action potential for neurons but also for cardiac cells . let us make some comments about a possible extension of our work to other forms of generalized hodgkin - huxley models . firstly , to choose that the ion channels are uniformly distributed along the axon ( @xmath38 for @xmath39 ) is just a simple way to fix ideas . we could also choose any distribution in a finite subset @xmath395 of @xmath396 without changing anything to our work except the notations . secondly , in @xcite , the authors noticed that it can be of interest to replace the dirac point masses @xmath397 of the spatial stochastic hodgkin - huxley model by functions of the form @xmath398 where @xmath399 is a neighborhood of the point @xmath361 . indeed , the concentration of ions is essentially homogeneous in a spatially extended domain around an open channel . along the same line , one can argue that we could replace the dirac point masses @xmath400 by any smoother approximation @xmath401 . the mathematics become then a little easier . the methods presented in this paper apply to both cases . the next step after the present averaging result is to prove the associated central limit theorem . that is to say give more precision on how the original model converges in law to the averaged model . another line of work can be to consider that not only some ion channels but also the membrane potential have a faster kinetic . it seems that this situation is closer to the real behavior of the nerve impulse @xcite . this last question is related to the study of invariant measures for pdmp and the necessity to find explicit form or bounds for these invariant measures . these will be the object of a future work . + + * acknowledgments :* the authors warmly thank the anonymous referee for his / her helpful comments , suggestions and corrections which led to a significant improvement of the previous versions of the manuscript . we recall here a lemma in @xcite . note that the last part of conclusion 1 . of the following lemma is contained in the proof of the corresponding lemma in @xcite . [ aulem ] let @xmath175 be in the interior of @xmath6 . then @xmath402 is a smooth function on @xmath6 vanishing at the end - points for any @xmath403 . furthermore : 1 . there is some constant @xmath404 , depending on @xmath405 but otherwise not on @xmath406 $ ] , such that for any continuous function @xmath407 \to{\mathbb{r}}$ ] we have that the function @xmath408 is in @xmath4 and satisfies the estimate @xmath409 moreover for any @xmath166 we can choose @xmath133 so small that : @xmath410 2 . for any fixed @xmath411 there is some constant @xmath412 , depending on @xmath413 and @xmath405 but otherwise not on @xmath406 $ ] , such that for any continuous function @xmath407 \to { \mathbb{r}}$ ] we have @xmath414 for any @xmath406 $ ] . we recall here two important results of @xcite about tightness in hilbert spaces with notations and choice of spaces adapted to our need . [ gct ] let us assume that @xmath1950,1])$ ] satisfies aldous s condition which means that for any @xmath415 , there exist @xmath416 such that for all stopping times @xmath163 such that @xmath417 : @xmath4180,{\varepsilon}_0]}\sup_{\theta\in]0,\eta[}{\mathbb{p}}(\|u^{\varepsilon}_{\tau+\theta}-u^{\varepsilon}_\tau\|_h\geq m)\leq\delta\ ] ] if moreover , for each @xmath71 $ ] fixed the family @xmath4190,1])$ ] is tight in @xmath4 then @xmath1950,1])$ ] is tight in @xmath221,h)$ ] . [ th ] @xmath4 is a separable hilbert space endowed with a basis @xmath18 . we denote by @xmath420 for @xmath15 : @xmath421 then @xmath4190,1])$ ] is tight in @xmath4 if , and only if , for any @xmath246 there is @xmath422 and @xmath423 such that : 1 . @xmath4240,{\varepsilon}_0 ] } { \mathbb{p}}(\|u^{\varepsilon}_t\|_h>\rho)\leq\delta\ ] ] 2 . @xmath4250,{\varepsilon}_0 ] } { \mathbb{p}}(d(u^{\varepsilon}_t , l_{\delta,\eta})>\eta)\leq\delta\ ] ] where @xmath426 . for the simulation , the jump rate functions are given , for real @xmath20 , by : @xmath427 the maximal conductance associated to sodium ions is @xmath428 and the potential at rest is @xmath429 mv . the constant @xmath383 is given by @xmath430 where @xmath431 is the radius of the axon , @xmath432 cm and @xmath433 is the internal resistance of the axon , @xmath434 @xmath435.cm , these data are classical , see for example @xcite . we have used an input on the potential equal to @xmath436 all along the time on the segment @xmath437 $ ] of the axon . the emergence of the deterministic hodgkin - huxley equations as a limit from the underlying stochastic ion - channel mechanism . * 18 * , pp . 12791325 . an exact stochastic hybrid model of excitable membranes including spatio - temporal evolution . to appear in . spontaneous action potentials due to channel fluctuations . _ j. _ * 71 * , pp . 30133021 . stability and ergodicity of piecewise deterministic markov processes . _ siam j. of control and opt . _ * 47 * , pp . 10531077 . singular perturbation for the discounted continuous control of piecewise deterministic markov processes . _ appl . math . and _ , * 63 * , pp . 357384 . convergence of stochastic gene networks to hybrid piecewise deterministic processes . to appear in _ ann . . prob _ * 71*. ( 1984 ) . piecewise - deterministic markov processes : a general class of non diffusion stochastic models . . 353388 . markov models and optimization . _ monograph on stat . and appl . prob . _ * 49*. ( 1993 ) . chaotic states in a random world : relationship between the nonlinear differential equations of excitability and the stochastic properties of ion channels . _ j. of stat . phys . _ * 70 * , pp . 339354 . ( 1986 ) . _ markov processes : characterization and convergence_. john wiley , new york . ( 2009 ) . averaging and large deviation principles for fully - coupled piecewise deterministic markov processes . to appear in _ markov proc . fields _ * 137 * , pp . ion - channel noise places limits on the miniaturization of the brain s wiring . _ current bio . _ * 15 * , pp . 11431149 . thresholds and plateaus in the hodgkin - huxley nerve equations . _ * 43 * , pp . 867896 . ( 2006 ) . contributions of sustained @xmath438 and @xmath439 to transmural heterogeneity of early repolarization and arrhythmogenesis in canine left ventricular myocytes . j. physiol . heart circ . physiol . _ * 291 * , pp . quantitative estimates for the long time behavior of a pdmp describing the movement of bacteria . submitted . stochastic versions of the hodgkin - huxley equations . _ biophys . j. _ * 72 * , pp . . mechanisms of excitation - contraction coupling in an integrative model of the cardiac ventricular myocyte . _ biophys . j. _ * 90 * , pp . _ geometric theory of semilinear parabolic equations_. _ lecture notes in mathematics _ , springer verlag , berlin , heidelberg , new york . _ ionic channels of excitable membranes_. sinauer sunderland , mass . ( 1952 ) . a quantitative description of membrane current and its application to conductation and excitation in nerve . _ j. physiol . _ * 117 * , pp . ( 1987 ) . _ limit theorems for stochastic processes_. springer verlag , berlin . convergence faible et principe dinvariance pour des martingales valeurs dans des espaces de sobolev . i.h.p . _ * 20 * , pp . 329348 . asymptotic expansion and central limit theorem for multiscale piecewise deterministic markov processes . to appear in _ stoch . appl . _ . _ multiscale methods , averaging and homogenization_. springer verlag , berlin . . _ an introduction to partial differential equations_. springer verlag , berlin . almost sure convergence of numerical approximations for piecewise deterministic markov processes . _ heriot - watt math . report . _ * 10 - 34*. ( 2011 ) . limit theorems for infinite dimensional piecewise deterministic processes and applications to stochastic neuron models . _ submitted_. ( 1997 ) . _ continuous - time markov chains and applications : a singular perturbation approach_. springer verlag , berlin . _ nonlinear functional analysis and its applications ii / b , nonlinear monotone operators_. springer verlag , berlin .

i derive analytically the temporal dependence of the perpendicular transport coefficient of a charged particle in the three - dimensional anisotropic turbulence conjectured by goldreich - sridhar by implementing multi - spacecraft constraints on the turbulence power spectrum . the particle motion away from the turbulent local field line is assessed as gradient / curvature drift of the guiding - center and compared with the magnetic field line random walk . at inertial scales much smaller than the turbulence outer scale , particles decorrelate from field lines in a free - streaming motion , with no diffusion . in the solar wind at au , for energy sufficiently small ( kev protons ) , the perpendicular average displacement due to field line tangling generally dominates over two decades of turbulent scales . however , for higher energies ( mev protons ) within the range of multi - spacecraft measurements , the longitudinal spread originating from transport due to gradient / curvature drift reaches up to . this result highlights the role of the perpendicular transport in the interpretation of interplanetary and interstellar data .

1512.05352

the diffusion of charged particles in a turbulent magnetic field plays a pivotal role in the understanding of the origin of cosmic rays , more than a century since their discovery , over a broad range of particle energy , from the interplanetary solar energetic particles ( * ? ? ? * hereafter sep ) to the pev cosmic rays likely produced at individual supernova remnant shocks @xcite . in a volume of space ( for instance interplanetary region close to the sun , or interstellar medium stirred by supernova explosion ) threaded by a strong statistically uniform average magnetic field @xmath4 with a small fluctuation @xmath5 , the particle diffusion parallel to the average field has been originally modeled in @xcite . the discrepancy later found @xcite between the mean - free path of some sep events and the prediction of the quasi - linear theory ( hereafter qlt ) spurred intense theoretical work . @xcite has shown that including the measured , despite noisy , steepening of the power spectrum at high wavenumber for selected events approaches the predicted mean free path to the measurements . @xcite concluded that the geometry of the magnetic fluctuations in the solar wind must be different from the simple picture of qlt . perpendicular diffusion is usually regarded as negligible compared to the parallel diffusion . nevertheless , the interpretation of a number of heliospheric measurements and numerical simulations of energetic particles suggests a significant contribution arising from perpendicular transport . time - intensity profiles of energetic protons at spacecraft separated by more than @xmath6 in longitude at @xmath0 au provide evidence of a significant motion across the parker spiral magnetic field lines @xcite . whether the motion across the spiral is due to a pure field line meandering or , in addition to that , to the departure of protons from the actual local field lines , called here cross - field diffusion , is among the purposes of this investigation . from the @xmath7-ray variability in the solar flares observed by rhessi @xcite it has been argued that perpendicular chaotic motion of a few tens of @xmath8 electrons in flaring loops was observed in a non - diffusive phase . test - particle simulations @xcite show that even in a weak three - dimensional isotropic turbulence charged particles decorrelate from magnetic field lines on a time scale comparable to the gyroperiod . these results pinpoint to a propagation regime wherein early - time perpendicular transport can not be neglected . it is well known that the transport of charged particles in turbulence depends on the anisotropy of the power spectrum @xcite . anisotropic turbulence in the solar wind was first found by @xcite as elongation of turbulence fluctuations along the magnetic field , and confirmed by several studies later on ( * ? ? ? high - heliolatitude _ ulysses _ measurements @xcite provided a new insight : the power spectrum of the local magnetic field in the solar wind is consistent with the anisotropic incompressible mhd - scale turbulence conjectured by @xcite ( hereafter gs95 ) . such an interpretation was confirmed over a frequency interval comparable with the presumably `` entire '' inertial range of the fast solar wind turbulence @xcite . a scrutiny of the power spectrum dependence on the angle of the flow to the magnetic field and on the wave - vector anisotropy @xcite confirmed an approximate consistency with gs95 . @xcite argue that the disagreement in the outer length - scale predicted by gs95 , inferred to be @xmath9 times larger than measured , might be explained with unbalance between alfvn modes propagating in opposite directions , in contrast with the balance assumed in gs95 . however , the gs95 model does not account for the difference in power - law exponents of the power spectrum of velocity and magnetic fluctuations in the inertial range measured , e.g. , by wind throughout the solar cycle 23 @xcite . numerical simulations of mhd turbulence have pointed out that the anisotropy increases at small scales @xcite and the turbulent eddies become more elongated along the direction of the local magnetic field @xcite . in this paper i investigate the time - dependent perpendicular transport within the mhd - scale turbulence gs95 . @xcite pioneered the study of pitch - angle scattering in the gs95 turbulence within the qlt limit @xcite , finding a scattering frequency more than ten orders of magnitude smaller than the isotropic turbulence over @xmath10 decades of particle kinetic energy . extension to a regime immune to qlt divergence at small pitch - angle @xcite confirmed such a scattering inefficiency @xcite . since in the gs95 anisotropic turbulence the wave - vectors of the mhd - scale fluctuations are predominantly perpendicular to the local average field , one could expect a small ratio of the perpendicular to the parallel average displacement as compared to the 3d isotropic case . note that test - particle simulations by @xcite found that the ratio of perpendicular to parallel diffusion coefficient in a synthesised scale - dependent anisotropic turbulence devised to reproduce gs95 is a few percent , comparable to the isotropic turbulence . diffusion perpendicular to the mean magnetic field has been long known to be dominated by the meandering of the magnetic field lines @xcite . the rate of separation of turbulent field lines along the direction of the mean field was found in @xcite , by solving a fokker - planck equation , to be very fast in the 3d isotropic power - spectrum turbulence , resulting in a significant perpendicular transport solely due to field - line meandering . @xcite , arguing that field lines diverge exponentially along the mean field direction , found that even if the particle does not undergo significant pitch - angle scattering ( `` collisionless '' case ) , the jump from a field line to another warrants a strong perpendicular diffusion due to the exponential divergence . @xcite showed that an exponential divergence of field - lines reduces the electron thermal conductivity in the galaxy clusters medium . @xcite used a variety of methods , including fokker - planck equation for a static gs95 mhd turbulence and a numerical monte - carlo model , and found a field - line separation growing as a power - law . such theoretical analyses omitted gradient / curvature drift motion of the guiding - center ; this assumption is justified for the thermal electrons in the galaxy clusters medium . the present paper investigates the limits of neglecting the gradient / curvature drift in the solar wind turbulent plasma . in this paper i use the decomposition of the instantaneous perpendicular average square displacement of a charged particle in a magnetic turbulence in two distinct contributions : meandering of magnetic field lines and gradient / curvature drift of the particle guiding - center from the local field line . those have been analytically calculated for the slab- and the 3d isotropic turbulence in ( * ? ? ? * hereafter fj11 ) and the latter numerically confirmed for the 3d isotropic turbulence @xcite . the gradient / curvature drift is usually neglected in theoretical analysis or interpretation of numerical simulations as the cross - field motion is assumed to be dominated by jumps between field lines once the particle has travelled a certain distance away from the original field line . in this paper i calculate gradient / curvature drift and field line contributions to the perpendicular transport for the gs95 anisotropic turbulence and investigate the departure of particles from the field lines as a function of time . benchmark parameters are tailored to solar wind energetic particles events . this paper is organized as follows : in sect . [ anisotropy ] the power spectra of alfvn polarisation modes relevant to perpendicular transport in gs95 turbulence are described ; in sect . [ coeff_diff ] the calculation of the instantaneous transport coefficients , due to gradient / curvature drift and to magnetic field line random walk , is outlined ( details are in the appendices ) and compared with the previous finding of an exponential divergence of field lines ; in sect . [ discussion ] the two contributions to the perpendicular transport are compared and the role of gradient / curvature drift average displacement in the sep longitudinal spread is estimated ; sect . [ conclusion ] contains the conclusions . critical balance condition along with the assumption that the energy cascade rate is scale - independent ( gs95 ) yields the scaling @xmath11 , where @xmath12 are the components of the wavenumber @xmath13 parallel and perpendicular to the local average magnetic field , respectively , and @xmath14 is some outer scale which might coincide with the injection scale of the turbulence . such a scaling holds for a large - scale field alfvnic mach number @xmath15 , as originally assumed in gs95 ; a more general scaling reads @xmath16 @xcite . the previous scaling implies that the anisotropy increases at the smaller scales . the power at wave - vectors not satisfying the critical balance is exponentially suppressed . however , as well known , the functional form of the power spectrum is not predicted by the gs95 conjecture and various possibilities have been explored ( * ? ? ? * e.g. ) . in the gs95 anisotropy the scales of the turbulent eddies parallel and perpendicular are measured only with respect to the direction of the local magnetic field @xcite that varies according to the location and the scale , whereas in slab or 3d isotropic weak turbulence the direction of the global average statistically uniform field is independent on the scale . numerical simulations confirm the ratio of parallel and perpendicular scales of the eddies , as well as the relation between @xmath17 and @xmath18 , implied by the gs95 power spectrum @xcite . a spatially homogeneous , fluctuating , time - independent global magnetic field @xmath19 is considered . the total field is decomposed as @xmath20 , with an average component @xmath21 and a fluctuation @xmath22 . the unit vector @xmath23 gives the direction of the global average field component . such a time - independent decomposition of @xmath24 applies to the solar wind , where the propagation of the magnetic fluctuation is much smaller than the velocity of the bulk ionized fluid . in this paper , we assume that the inertial range extends from the outer scale @xmath14 down to some scale wherein injected energy ultimately must dissipate . we introduce a scale @xmath25 , much larger than the gyroradius @xmath26 of the particles considered here ( fig . [ cartoon ] ) . the fluctuations @xmath27 are large compared to the total magnetic field at scale @xmath28 ; however , at scales @xmath29 the magnetic fluctuations are small compared to the local average field , and the first - order orbit theory applies to eddies of scale @xmath30 . by using such assumptions , @xcite applied qlt to determine the scattering frequency in the turbulence gs95 . the resulting parallel transport depends only on the large scale @xmath14 and not on the intermediate scale @xmath31 . the most general form of the power spectrum tensor for mhd fluctuations was derived in @xcite , in their eq . ( 20 ) , and in the assumption of axisymmetric turbulence ( neglecting cross - helicity ) it reduces to @xmath32 f({\bf k } ) \label{oughton_ps}\ ] ] where @xmath33 is the unit vector in cartesian coordinates and @xmath34 the wave - number ( @xmath35 ) , where @xmath36 ( @xmath37 ) is the component of the wavenumber vector parallel ( perpendicular ) to the local average field ( the unit vector @xmath38 is not parallel to the z - axis , see fig . [ cartoon ] ) . the scalar functions @xmath39 are interpreted in terms of different linear mhd polarisation modes for an oblique orientation between @xmath34 and the local average magnetic field . in particular , @xmath40 is related to the power in the shear - alfvn modes @xmath41 , i.e. , @xmath42 , and @xmath43 to the difference between shear- and pseudo - alfvn modes power @xmath44 , i.e. , @xmath45 . the power spectrum given by eq . [ oughton_ps ] includes the shear- and pseudo - alfvn mode only , and assumes vanishing cross - helicity , i.e. , the modes @xmath46 and @xmath47 in eq.(20 ) in @xcite . fast solar wind measurements @xcite are compatible with vanishing mode @xmath48 ; here the mode @xmath49 is neglected , as expected to be comparably small . a series of incompressible mhd numerical simulations @xcite implemented the axisymmetric power spectrum tensor given by eq.[oughton_ps ] to empirically determine the functional form of @xmath50 and @xmath51 and best - fit the shear - modes power @xmath52 by using @xmath53 where @xmath54 is a normalisation constant to be determined , @xmath55 , where @xmath56 is the power spectrum cut - off where dissipative effects start to be important and mhd approximation breaks down , and @xmath15 . according to the gs95 conjecture , the pseudo - alfvn modes are carried passively by the shear - alfvn modes ( this property is an exact result for weak mhd turbulence @xcite ) with no contribution to the turbulence cascade to small scales which is seeded by collisions of shear modes only . thus , it is reasonable to assume that @xmath57 has the same functional form as @xmath58 , allowing for a relative amplitude @xmath59 ( @xmath60 ) and a different outer scale @xmath61 ( @xmath62 ) both so far inaccessible to either solar wind observations or models . likewise , the power spectrum of the pseudo - modes is cast in the form @xmath63 the constant @xmath54 is fixed by normalisation of both polarisation modes that can be extended to scales between @xmath31 and @xmath14 , as the contribution from scales between @xmath31 and @xmath14 is exponentially suppressed . within the range @xmath64 , if the largest scales @xmath65 are related to the outer scale by @xmath66 , the normalisation reads : @xmath67 which gives @xmath68 , where @xmath69 was used . we note that @xmath54 does not depend on the scale @xmath31 neither on @xmath70 , that is irrelevant as the cascade along the local average field is suppressed . recent analysis of fast solar wind data from _ ulysses _ @xcite provides us with an estimate of @xmath59 by confirming an ordering in diagonal components of the power tensor that has long been found @xcite : if @xmath38 is the unit vector in the direction of the local average magnetic field , the measurements give a power along @xmath38 smaller , but not negligible ( @xmath71 ) , than in the plane orthogonal to @xmath38 . as shown in sect . [ coeff_diff ] , such a power ordering results in a small but not negligible effect on the perpendicular transport of the pseudo modes with respect to the shear modes . the forms given by eq.s [ shear ] , [ pseudo ] are used in sect . [ coeff_diff ] to calculate the time - dependent perpendicular transport . @xcite used a power spectrum with steep cutoff beyond the critical balance scaling @xmath72 to determine the scattering frequency within the qlt approximation . in this paper i calculate the time - dependent perpendicular transport coefficient by expanding in taylor series the exponentials @xmath73 , @xmath74 in the power spectra in eqs . [ shear ] , [ pseudo ] ; such an expansion is valid as the power spectrum is a convergent quantity . only the terms growing fastest in time are retained . we note that no assumption is necessary on the explicit spatial dependence of @xmath75 , that has three non - vanishing space components , as only the fourier transform @xmath76 is used . the magnetostatic approximation has been widely used for decades @xcite to investigate the transport of charged particles in magnetic turbulence , as the speed of the particles considered is typically much greater than the average - field alfvn speed . solar wind flows outward from the sun at a speed of several hundreds of km@xmath77s ( 300 - 400 at low heliolatitudes and 700 at high heliolatitudes ) , much faster than the local alfvn speed ( tens of km@xmath77s ) and the spacecraft speed ( a few km@xmath77s ) . thus , the magnetic fluctuations can be assumed to be frozen with the plasma . in this paper i investigate the perpendicular transport at time - scales shorter than the correlation time of the perpendicular fluctuations of the local magnetic field , as seen by the particle . the transport perpendicular to the average magnetic field is dominated by guiding - center motion which includes the meandering of the magnetic field lines ( mfl ) and the gradient / curvature drift from the first - order orbit theory ( * ? ? ? * fj11 ) , in the approximation that the particle gyroradius is much smaller than the scale of magnetic field variations . the existence of a diffusion regime for the perpendicular transport is not assumed . in this section , the derivation of the gradient / curvature drift transport coefficient in fj11 is outlined and the result in the case of the gs95 turbulence is presented ( details in appendix a ) . the first - order orbit theory @xcite is used , assuming that @xmath26 is much smaller than the length - scale of any magnetic field variation : @xmath78 or , in other terms , the magnetic field varies slightly at the gyroscale so that the particle orbit can be approximated within a gyroperiod by the helicoidal trajectory with local curvature radius given by @xmath26 ; for wave - numbers @xmath18 within the inertial range , the smallness of @xmath26 reads @xmath79 ; in other terms we assume that the perturbation in the turbulent cascade at the gyroscale @xmath26 does not affect significantly the particle trajectory . at scales @xmath30 , we calculate the gradient / curvature drift from the local average field at those scales . we make use of the gyroperiod averaged guiding - center velocity transverse to the local field @xmath80 , i.e. , @xmath81 , to the first order in the fluctuation , given for a particle of speed @xmath82 , momentum @xmath83 and charge @xmath84 by @xcite @xmath85 , \label{vperp}\ ] ] where @xmath86 is the cosine of the pitch - angle with respect to the local average field . the average square transverse displacement of the particle from the direction of local field due to drift , @xmath87 , at time @xmath88 is written as @xmath89 the drift transport coefficient would be defined here as the limit @xmath90 . we assume that @xmath91 is uncorrelated at different wave - number vectors , i.e. @xmath92 for the fourier components @xmath93 ; we note that the validity of the critical balance condition scale - by - scale used here has been recently questioned in a phenomenological analysis @xcite that reports inconsistencies with ion kinetic scales for parameters in the solar wind regime . the generic non - vanishing term of eq.[dxx1 ] ( regardless of the power spectrum assumed ) is given by ( see fj11 ) : @xmath94}{k_\parallel v_\parallel } \ , , \label{fmu2}\ ] ] with indexes @xmath95 , @xmath96 , @xmath97 , @xmath98 for @xmath99 and @xmath100 , @xmath101 , @xmath102 , @xmath103 for @xmath104 . in the factor @xmath105 in eq . [ fmu2 ] we assumed that the magnitude of the local average field can be approximated by @xmath4 at those scales . here @xmath106 represents various factors depending on the particle pitch - angle @xmath86 resulting from the expansion of eq.[dxx1 ] ( fj11 ) . such a calculation is detailed in appendix [ app_drift ] . from a comparison of the terms in eqs . [ series_tot_1 ] , [ series_tot_2 ] , [ series_tot_3 ] and [ series_tot_4 ] , only the term fastest growing in time ( eq.[series_tot_1 ] ) is retained : the drift transport coefficient is dominated by the power of the pseudo - alfvn modes along the local average field @xmath107 ( term @xmath108 in appendix a ) . thus , by using eq.[fmu2 ] , the instantaneous drift transport coefficient can be cast as @xmath109\right|_{k_\perp ^{m } } \label{series_tot_drift}\end{aligned}\ ] ] where @xmath110 is the particle gyro - frequency , @xmath111 . the quantities @xmath112 depend on the magnitude of the global average field @xmath4 . the corrective factor in eq . [ series_tot_drift ] is only the ratio of the power of the global to the local average fields ( @xmath113 ) . we emphasise that eq . [ series_tot_drift ] is valid only as long as particle transport is confined within the scale @xmath31 . in fig . [ fig_time ] , panels ( b , d ) depict the time - evolution of @xmath87 as given by eq . [ series_tot_drift ] for energetic protons in the solar wind for different values of @xmath59 * ( @xmath114 ) * and @xmath115 in ( b ) and @xmath116 in ( d ) . higher power in the pseudo - alfvn modes relative to the shear- ( i.e. , higher @xmath59 ) results in an enhancement of @xmath117 ( see appendix a ) . in analogy with the 3d isotropic turbulence case ( fj11 ) , @xmath87 in the anisotropic gs95 turbulence grows linearly with @xmath118 . as compared with the bohm diffusion coefficient @xmath119 @xmath120s for @xmath121 kev , fig . [ fig_time ] ( b ) shows that @xmath122 within a very short time ( @xmath123 , @xmath124 and @xmath125 ) . we emphasise that the average square displacement due to drift ( eq . [ series_tot_drift ] ) does not depend only on large - scale properties of the turbulence ( @xmath14 ) , but also depends non trivially on the smallest scales ( @xmath126 ) . the result is independent on the scale @xmath31 @xcite as the large scales are exponentially suppressed . we can also conclude that the guiding - center is confined only for short - time to move along the field line . we compare the average square displacement from the local field - line reached by the guiding - center via gradient / curvature drift , i.e. , @xmath127 ( see fig . [ cartoon ] ) with the squared gyroscale ( @xmath128 ) . for a proton with @xmath129 kev in the solar wind at @xmath0 au ( see fig . [ fig_time ] ( b ) ) , we find that @xmath130 at @xmath131 ( @xmath132 ) is comparable with @xmath133 @xmath134 . thus , after a few gyroperiods the actual particle has an average distance from the local field greater than @xmath135 . we remind that the gradient / curvature drift transport is calculated here up to scales @xmath31 , with a given direction ( @xmath38 ) of the local average field at that scale different from the global average field direction ( @xmath136-axis ) . our finding does not rule out that at scales close to the outer scale @xmath14 the perpendicular transport can turn into diffusive regime . however , accounting for those larger scales requires a spatial dependence of the local turbulence wave - number vectors ; this is deferred to a separate work . in this section the method to compute the time - dependent particle transverse transport due to mfl meandering ( fj11 ) is outlined and the result for the gs95 turbulence is presented . previous theoretical derivations of mfl meandering in the isotropic @xcite and anisotropic gs95 turbulence @xcite focused on the relative divergence between two initially static nearby ( down to thermal electron gyroradius scale ) field lines along the direction of the average magnetic field . in contrast with this result , magnetic turbulence has been found @xcite to exhibit a field line separation in agreement with the long argued richardson scaling @xmath137 for a pair of particles in purely hydrodynamical turbulent media . numerical simulations of mhd turbulence in a partially ionised plasma @xcite show that the neutral drag has the effect of recovering the exponential field line separation argued in @xcite . in analogy with fj11 , we focus on the divergence of a single field line from the direction of the local average field . the method of two field lines separation applies to the case of zero average field ; on the other hand , being based on the solution of the fokker - planck equation , it requires several independent small displacements @xmath138 ( central limit theorem ) , thus it requires diffusive regime to be reached , whereas the method fj11 applies more generally , i.e. , prior to the diffusive regime . we assume that the correlation function of the magnetic fluctuation is homogeneous in space , i.e. , the correlation depends only on distances along the particle orbit : @xmath139 \delta b_x [ { \bf x}(x_3 ^ 0 ) ] \rangle = \langle \delta b_x [ { \bf x}(x_3 ) ] \delta b_x [ { \bf x}(0 ) ] \rangle$ ] . the mean square displacement of the mfl orthogonal to the @xmath38 vector can be defined as @xmath140 \delta b_x [ { \bf x}(0 ) ] \rangle .\ ] ] we compute the magnetic turbulence along the unperturbed trajectory of a pseudo - particle travelling with zero pitch - angle and no scattering ( fj11 ) . the mean - square transverse displacement of a mfl corresponding to a distance @xmath141 along the local uniform field travelled by a such a pseudo - particle can be written as @xmath142}{k_\parallel } \ , . \label{dmfl2}\ ] ] where @xmath91 is the magnetic turbulence power spectrum in the inertial range ( eq . [ oughton_ps ] ) . the mfl diffusion coefficient is the limit @xmath143 . as in the previous section , in eq . [ dmfl2 ] we assumed that the magnitude of the local average field can be approximated by @xmath4 . the detailed calculation for the gs95 anisotropic power spectrum ( eq.s [ oughton_ps ] , [ shear ] , [ pseudo ] ) can be found in appendix [ app_mfl ] . from eq.[shear_mfl_0 ] , the instantaneous transport coefficient due to mfl meandering is dominated by shear - modes and given by @xmath144 figure [ fig_time ] , * ( a , c ) * , depicts the time - evolution of @xmath145 as given by eq . [ shear_mfl ] for protons at various energies in the solar wind and for different value of @xmath59 * ( @xmath114 ) * and @xmath115 ( @xmath116 ) in the left ( right ) . in analogy to the mfl random walk in isotropic turbulence ( fj11 ) and the well - known qlt result , @xmath146 for anisotropic gs95 grows linearly with @xmath118 . alike the drift , the independence on @xmath31 arises from the fact that scales between @xmath31 and @xmath14 are exponentially suppressed . we find that even at the small time - scales considered here , different magnetic turbulences exhibit different scalings . in the case of gs95 anisotropy the linear growth in time of the mfl average square displacement @xmath147 is fast as compared to the 3d - isotropic turbulence , wherein a logarithmic growth at large scales and diffusion at small scales are found ( see eq . [ 49 ] in fj11 ) . with the substitution @xmath148 , eq . [ shear_mfl ] gives @xmath149 . we note that a linear dependence of @xmath150 on @xmath151 could also be derived empirically from the argument @xmath152 . by using a direct calculation we determine here the dependence of the proportionality coefficient on the turbulence parameters , otherwise to be determined empirically , for instance , through numerical simulations . we discuss the relative contribution of magnetic field line meandering and gradient / curvature drift to the perpendicular transport in the gs95 anisotropy for times shorter than the correlation time of the perpendicular fluctuations of the local magnetic field , as seen by the particle . in the guiding - center approximation used here , the ratio @xmath153 ( see eqs . [ series_tot_drift ] and [ shear_mfl ] ) is time - independent . such a ratio can be cast as a function of @xmath154 in the form @xmath155^{-1 } . \label{ratio}\ ] ] figure [ fig_ratio ] illustrates @xmath156 in eq . [ ratio ] for different values of @xmath157 . we emphasise that the ratio @xmath158 , large in the solar wind , is expected to be upper bounded : @xmath159 grows with the turbulence power along the local average field @xmath160 , that is small , although not negligible , compared to the other diagonal terms of the power spectrum tensor ( @xmath161 ) . figure [ fig_ratio ] shows that @xmath159 becomes relevant at small scales , i.e. , at large wave - numbers @xmath126 . if @xmath162 , the average perpendicular displacements due to tangled field lines and to the gradient / curvature drift become comparable : @xmath163 . [ fig_ratio ] , we conclude that for a sufficiently small energy ( @xmath1 kev ) proton at @xmath0 au in the solar wind turbulence , using an inertial range extended down to a scale @xmath164 au ( @xmath165 for @xmath166 au ) , the average perpendicular displacement due to field line meandering is generally dominant ( @xmath167 or greater ) . however , for higher energies ( @xmath168 kev ) protons ( @xmath169 ) , the ratio of the two average square displacements is smaller and within a smaller inertial range ( @xmath170 ) reaches values down to @xmath168 . the region @xmath171 is inaccessible to our model that applies only to turbulent scale wave - numbers @xmath172 such that @xmath173 . for a given scale @xmath126 , the ratio @xmath174 decreases as the particle energy increases , i.e. , for small @xmath157 , as expected . this result constrains the customary assumption that charged particles follow turbulent field lines . we note also that the result is independent of the parallel mean - free path . this is in contrast with the model in @xcite ; however , we remind that the model presented here does not account for parallel transport , that has already been determined to be irrelevant due to very small scattering frequency in gs95 within qlt limits @xcite , and is valid only for times @xmath175 . we find that at small scales the perpendicular transport in gs95 turbulence is a free - streaming motion ( @xmath176 ) , in contrast with a diffusive regime ( @xmath177 ) . theoretical evidence of perpendicular super - diffusion of the magnetic field lines alone in gs95 anisotropic turbulence ( @xmath178 , with @xmath179 ) is discussed also in @xcite , that focusses on the field line divergence ( see references therein ) , assuming no departure of the particle from the local field line . the fourth - power dependence in @xmath180 of the mfl meandering found in @xcite , in contrast with the second - power found here , is likely to originate from assumptions therein on the turbulence generation and reconnection , i.e. , normalisation factor . in addition , we find that the motion of the guiding - center away from the field line might reach , and eventually exceed , the gyroradius scale within a few hundreds @xmath181 ( as shown in sect . [ sect_drift ] ) for times smaller than the correlation time of the perpendicular fluctuations of the local magnetic field , as seen by the particle . we note that the departure from field line found here does not violate the reduced dimensionality theorem @xcite as no constraint is placed on the ignorable coordinates of @xmath182 . from eq . [ series_tot_drift ] , it is readily found ( see also fig.[fig_ratio ] ) that the drift perpendicular transport coefficient scales with the largest perpendicular wave - number @xmath126 , i.e. , with the smallest perpendicular inertial scales , as @xmath183 . in gs95 turbulence the anisotropy is known to increase at smaller scales favouring energy cascade in the perpendicular direction . thus , it is clear that the smaller the scale ( namely the larger @xmath126 ) , the larger the anisotropy in the power of the turbulence , the more significant the departure from the field lines . this is in agreement with the limit case of the slab - turbulence , i.e. , anisotropy wherein the power along the the mean field direction is completely suppressed : the gradient / curvature drift average square displacement @xmath87 is dominated by small scales ( fj11 ) , whereas the larger scales govern the mfl meandering . in this subsection we provide a crude estimate of the role of the perpendicular transport in the reconstruction of the propagation of sep from the flaring region out to the multi - spacecraft location at @xmath0 au . strong perpendicular transport has often been invoked to explain the longitudinal spread in multi - spacecraft measurements of sep events ( see @xcite and references therein ) . according to an alternative plausible mechanism , the longitudinal spread might originate from wide longitudinal extension or motion of cme - driven shocks in the interplanetary medium . we assume the large - scale ordered field @xmath19 to be approximately uniform over a distance much smaller than the putative correlation length scale of the solar wind turbulence ( @xmath184 au ) . therefore , we can constrain the average distance @xmath185 associated with gradient / curvature drift that the particles are transported in the direction perpendicular to the local field at various distances from the sun ; we emphasise that such a drift originates from the inhomogeneity of the magnetic fluctuation and not of the spiralling magnetic field itself . the effect of field line meandering is switched off in this paragraph . the field @xmath19 is assumed to be piecewise uniform ; the resulting cumulative effect might contribute to the longitudinal spread at @xmath0 au . in the sep event of january @xmath186 @xmath187 , energetic protons have been measured by the two stereos ( let and het ) and soho / ephin to spread almost over @xmath188 at @xmath0 au @xcite . let and het instruments measure energetic protons in the range from @xmath10 to @xmath189 mev and soho / ephin from @xmath10 to @xmath190 mev . a possible scenario comprises protons with kinetic energy @xmath191 equal or greater than @xmath10 mev that , while travelling along a turbulent field line of the parker spiral in the ecliptic plane and drifting by some amount away from that line , undergo scattering and back - trace a field line adjacent to the original one . the iteration of such a process over a time interval @xmath192 ( see table [ table ] ) might spread energetic protons over wide longitudinal angle over a few hours . in this crude estimate , we assume this scattering to be frequent enough that the magnitude @xmath4 can be taken as uniform . the solar wind advection , customarily taken purely radial , is neglected here ; we expect it to ultimately only enhance @xmath185 . in table [ table ] we consider two distances : @xmath0 au ( @xmath193 nt ) and @xmath194 solar radii ( @xmath195 gauss ) . the sun s spin ( @xmath196day ) is here neglected . the longitudinal spread @xmath197 is calculated in table [ table ] as @xmath185 in units of the distance from the sun , where the average displacement from the field line due to gradient / curvature drift is given by @xmath198 and @xmath199 is defined in eq . [ series_tot_drift ] . in table [ table ] we used @xmath200 au@xmath201 and , from eq . [ series_tot_drift ] , @xmath202 |_{k_\perp ^{m}l = 10 } = 149.8 $ ] . table [ table ] shows that the time elapsed since the particle injection in the turbulent region where @xmath4 can be taken as uniform is a major contributor to the spread @xmath197 , being the transport ballistic . in particular @xmath203 mev protons at @xmath204 au after @xmath0 hour cover a non - negligible longitudinal angle ( of order of @xmath205 ) . our estimate is limited by small range of fluctuations interacting with the particles : @xmath190 mev protons ( @xmath206 at @xmath0 au ) is allowed to interact only with one decade of turbulent scales , satisfying the condition @xmath173 . one can also roughly predict @xmath207 to be measured by solar probe plus mission in its approach to the sun at a distance of @xmath194 solar radii : for @xmath208 min , we find @xmath209 for a @xmath210 mev proton . .estimate of the longitudinal spread of energetic protons due to gradient / curvature drift average displacement . [ cols="^,^,^,^,^,^,^ " , ] [ table ] we emphasise that the gradient / curvature drift velocity @xmath211 scales with the particle speed @xmath82 as @xmath212 . in the unperturbed parker spiral at a distance @xmath0 au , if one uses a characteristic scale of spatial variation of @xmath213 au , by assuming that the gradient / curvature drift originates from the inhomogeneity of the parker spiral itself , and not from the intermediate scale @xmath31 ( @xmath214 au ) , @xmath207 would be suppressed by a factor @xmath215 . an extension to a turbulent parker spiral magnetic field is underway and will be presented elsewhere . i have derived analytically the time - evolution of the average transverse displacement of a charged particle in a magnetostatic turbulence based on the critical balance gs95 conjecture . the particle motion is assumed to be well - approximated by the guiding - center motion . with no assumption of diffusion , i have calculated the perpendicular transport due to the single field line meandering , customarily regarded as dominant , and the gradient / curvature drift from the local perturbed field line . the former is dominated by the shear - alfvn modes , the latter by the pseudo - alfvn modes . both contributions are found to be in free - streaming , not diffusive , regime . in particular , the departure of the guiding - center from the field line grows rapidly in time , invalidating the customary assumption that particles follow the field lines . if gs95 is the dominant mhd - scale turbulence in the solar wind , a crude estimate shows that the longitudinal spread of energetic protons measured at 1 au might be contributed to a non - negligible extent by cross - field transport due to gradient / curvature drift in the inhomogeneous turbulence , instead of the inhomogeneity of the unperturbed parker spiral itself . the measured much larger ( nearly @xmath188 ) longitudinal spread certainly includes additional factors not considered here such as : angular extension and motion of the cme - driven shock producing the seps and sun s spin dragging the parker spiral across the heliosphere . the perpendicular transport in turbulence presented here might imply for particle acceleration at quasi - perpendicular shocks an enhancement of the permanence time , hence of the maximal momentum , of energetic particles in the vicinity of the shock . such a higher energy attainable combines with the small acceleration time at perpendicular shocks , as described in @xcite . turbulent field enhancements , generated downstream by inhomogeneities of the unshocked medium @xcite , can also increase the permanence time of particles at shocks . this result has relevant implications for particle acceleration at shocks , deferred to a separate work . finally , it is relevant to note that in this paper a particular model for the anisotropic power spectrum ( eq . [ oughton_ps ] ) is assumed , that relies on the interpretation of solar wind measurements and numerical simulations . however , the dominant process in the generation of anisotropy in mhd incompressible turbulence and the resulting turbulence power spectrum are currently topic of debate ( see , e.g. , @xcite ) . moreover , a recent analysis of the magnetic field and plasma data from wind spacecraft @xcite shows that the claimed spectral anisotropy in the solar wind inertial range can be explained by localised turbulent intermittency structures with no critical balance conjecture needed . further experimental investigation is necessary to unveil the nature of the solar wind turbulence . the author acknowledges useful discussions with m. forman about the power spectrum modelling , continuing discussions with j. r. jokipii , j. kta and j. giacalone and useful suggestions from the two referees . the author benefitted from discussions at a workshop on `` first principles physics for charged particle transport in strong space and astrophysical magnetic turbulence '' held at the international space science institute ( issi ) in bern , switzerland . this work was supported by nasa under grants nnx13ag10 g and nnx11ao64 g . , 176 , 3534 ] http://adsabs.harvard.edu/abs/1994apj...420..294b[apj , 420 , 294 ] http://adsabs.harvard.edu/abs/2000phrvl..85.4656c[phys . , 85 , 4656 ] http://adsabs.harvard.edu/abs/1998phrvl..80.3077c[phys . lett . , 80 , 3787 ] http://adsabs.harvard.edu/abs/2004apj...602..170c[apj , 602 , 170 ] http://adsabs.harvard.edu/abs/2002apj...564..291c [ apj , 564 , 291 ] http://adsabs.harvard.edu/cgi-bin/bib_query?2000apj...539..273c[apj , 539 , 273 ] http://adsabs.harvard.edu/abs/2012soph..281..281d[sol . , 281 , 281 ] http://adsabs.harvard.edu/abs/1993apj...407l..95d[apj , 407 , l95 ] http://onlinelibrary.wiley.com/doi/10.1002/2014ja019933/abstract;jsessionid=1c3a51a95f4fa734ee37ba27193f6dc7.f02t04[j . res . , 119 , 6074 ] http://adsabs.harvard.edu/abs/1974apj...193..231e[apj , 193 , 231 ] http://iopscience.iop.org/0004-637x/743/1/51/[apj , 743 , 51 ] http://adsabs.harvard.edu/abs/2011apj...733...76f[apj , 733 , 76 ] http://adsabs.harvard.edu/abs/2013apj...770...84f[apj , 770 , 84 ] http://www.sciencedirect.com/science/article/pii/s0168900213016227[nuclear inst . and methods in physics research , a , 169 ] http://iopscience.iop.org/0004-637x/755/2/114/[apj , 755 , 114 ] http://iopscience.iop.org/0004-637x/734/2/83/[apj , 734 , 83 ] http://adsabs.harvard.edu/abs/2000jplph..63..447g[journal of plasma physics , 63 , 447 ] http://adsabs.harvard.edu/abs/2013ssrv..176...73g[space sci , 176 , 73 ] http://iopscience.iop.org/0004-637x/520/1/204/[apj , 520 , 204 ] http://adsabs.harvard.edu/abs/1995apj...438..763g[apj , 438 , 763 ] http://atsol.fis.ucv.cl/dariop/sites/atsol.fis.ucv.cl.dariop/files/table_of_integrals_series_and_products_tablicy_integralov_summ_rjadov_i_proizvedennij_engl._2.pdf[table of integrals , series and products , ed . 7@xmath216 , 2007 , academic press , new york ] http://adsabs.harvard.edu/abs/2010phrve..82b6406g[phys . e , 82 , 026406 ] http://adsabs.harvard.edu/abs/2008phrvl.101q5005h[phys . 101 , 17 , 175005 ] http://adsabs.harvard.edu/abs/1966apj...146..480j[apj , 146 , 480 ] http://adsabs.harvard.edu/abs/1973apj...183.1029j[apj , 183 , 1029 ] http://adsabs.harvard.edu/abs/1987apj...313..842j[apj , 313 , 842 ] http://adsabs.harvard.edu/abs/1993georl..20.1759j[geophys . 20 , 175 ] http://adsabs.harvard.edu/abs/2011apj...730l..22k[apj , 730 , l22 ] http://iopscience.iop.org/0004-637x/764/2/168/pdf/0004-637x_764_2_168.pdf[apj , 764 , 168 ] http://adsabs.harvard.edu/abs/1999apj...517..700l[apj , 517 , 700 ] http://iopscience.iop.org/0004-637x/603/1/180/pdf/56372.web.pdf[apj , 603 , 180 ] http://adsabs.harvard.edu/abs/2014apj...784...38l[apj , 784 , 38 ] http://iopscience.iop.org/0004-637x/554/2/1175/[apj , 554 , 1175 ] http://adsabs.harvard.edu/abs/2014apj...790..155m[apj , 790 , 155 ] http://adsabs.harvard.edu/abs/2000phrvl..84..475m[phys . lett . , 84 , 475 ] http://adsabs.harvard.edu/abs/2001apj...562l.129n[apj , 562 , l129 ] http://iopscience.iop.org/1538-4357/654/1/l103/pdf/21188.web.pdf[apj , 654 , l103 ] http://journals.aps.org/pre/abstract/10.1103/physreve.56.2875[phys . e , 56 , 2875 ] http://adsabs.harvard.edu/abs/1982rvgsp..20..335p[rev . geophys . and space phys . , 20 , 335 ] http://adsabs.harvard.edu/abs/2007apj...664..543p[apj , 664 , 543 ] http://adsabs.harvard.edu/abs/1978phrvl..40...38r[phys . 40 , 38 ] http://159.226.72.19/share/reference/acceleration/shock/ssnr/reynolds08.pdf [ annu . astrophys . , 46,89 ] rossi , b. & olbert , s. _ introduction to the physics of space _ 1970 , ( mcgraw - hill ) http://adsabs.harvard.edu/abs/2010apj...725.2117s[apj , 725 , 2117 ] http://journals.cambridge.org/action/displayabstract?frompage=online&aid=4733832&fileid=s0022377800000933[j . plasma phys . , 29 , 525 ] http://adsabs.harvard.edu/abs/1975rvgsp..13..547v[rev . geophys . and space phys . , 13 , 547 ] http://adsabs.harvard.edu/abs/2014apj...783l...9w[apj , 783 , l9 ] http://adsabs.harvard.edu/abs/2010mnras.407l..31w[montly not . royal astron . soc . , 407 , l31 ] http://adsabs.harvard.edu/abs/2012apj...746..103w [ apj , 746 , 103 ] http://adsabs.harvard.edu/abs/2008apj...673..942y[apj , 673 , 942 ] in this section , i illustrate the details of the calculation of the non - vanishing terms for @xmath87 in sect . [ sect_drift ] , which depends on the following components of @xmath217 : @xmath218 where @xmath219 and @xmath220 . analysis of ulysses measurements @xcite shows that only the tensor elements in eq . [ pcomp ] contribute to the solar wind turbulence . we note that @xmath221 and @xmath160 contain only the power in the pseudo - alfvn modes , whereas @xmath222 and @xmath223 are a combination of shear- and pseudo - alfvn modes . * term 3322 * - using the last relation in eq.[pcomp ] and eq.[pseudo ] , in cylindrical coordinates ( @xmath224 ) with @xmath225 , @xmath226 , @xmath227 , it is readily found @xmath228}{k_\parallel v_\parallel } = \frac{\varepsilon { \cal n } \pi b_0 ^ 2}{v_\parallel \ell^{1/3 } } \int_{k_\parallel ^0 } ^{k_\parallel ^{m } } d k_\parallel \frac{{\rm sin } [ k_\parallel v_\parallel t]}{k_\parallel } i_{5/3}(k_\parallel , k_\perp ^0 , k_\perp ^{m } ) \label{drift_gen3322}\end{aligned}\ ] ] where @xmath229 are the outer and the dissipation perpendicular wave - numbers respectively ; the boundaries @xmath230 are related to @xmath229 through critical balance scaling : @xmath66 , @xmath231 for @xmath232 . we consider here ( fj11 ) perpendicular transport on time - scale shorter than @xmath233 . in eq . [ drift_gen3322 ] the auxiliary function @xmath234 was defined . the expansion in series of the exponential factor in @xmath235 by using @xmath236 yields @xmath237 where @xmath238 is the ordinary hypergeometric function . in eq . [ iseries ] the sum is truncated at @xmath239 ( @xcite , eq . 3.194.5 , applies in eq . [ iseries ] only for @xmath240 ) ; in what follows it is shown that it suffices to consider the term @xmath241 . the right hand side of eq . [ drift_gen3322 ] can then be approximated by using the expansion in eq.[iseries ] as & _ n=0 ^3 ( ) ^2-n & + & |_k_^0^k_^m s_n(x^0 , x^m ) & [ series_tot_pre ] where @xmath242 and @xmath243 . in eq.[series_tot_pre ] the critical balance condition ( @xmath244 ) is used ; thus , the independent variable of @xmath245 in eq.[iseries ] can be approximated as @xmath246 and likewise for the boundaries of @xmath247-integration , i.e. , @xmath248 and @xmath249 . the argument for the truncation at @xmath239 in eq . [ iseries ] goes as follows . the integrals @xmath250 are extended over a time shorter than @xmath233 , i.e. , @xmath251 , and in the interval @xmath252 i have @xmath253 |_{x_0}$ ] , @xmath254 |_{x_0}$ ] , @xmath255 and @xmath256 , where @xmath257 and @xmath258 are respectively sine and cosine integrals . it is easily shown that it holds @xmath259 ( for @xmath260 ) . moreover , it is possible to show for the coefficients @xmath261 in eq.[series_tot_pre ] that @xmath262 , @xmath263 and @xmath264 are rapidly decreasing functions of @xmath265 over at least three decades in @xmath266 . finally , the @xmath267th term in eq . [ series_tot_pre ] evolves in time as @xmath268 . this justifies retaining only the term @xmath269 in eq . [ series_tot_pre ] : @xmath270 \right|_{x^0 } \left . \left[(k_\perp l)^{8/3 } { } _ { 2}f_{1 } \left(1,{4\over 3 } ; { 7 \over 3};- ( k_\perp l)^{2/3 } \right ) \right ] \right|_{k_\perp ^{m } } \nonumber.\end{aligned}\ ] ] by using the fact that @xmath271 |_{x^0 } \sim 1/x_0 $ ] for @xmath272 , the right hand side of eq . [ drift_gen3322 ] can then be approximated as @xmath273 \right|_{k_\perp ^{m } } \label{series_tot_1}\end{aligned}\ ] ] the previous calculation shows that the term in eq . [ drift_gen3322 ] grows linearly in time . the corresponding pitch - angle factor @xmath274 , averaged over an isotropic pitch - angle distribution , gives a factor @xmath275 . * term 2323 * - using the expression for @xmath221 in eq.[pcomp ] and eq.[pseudo ] , in cylindrical coordinates , it is found @xmath276}{k_\parallel v_\parallel } = - \frac{\varepsilon { \cal n}\pi b_0 ^ 2}{v_\parallel \ell^{1/3 } } \int_{k_\parallel ^0 } ^{k_\parallel ^{m } } d k_\parallel { k_\parallel } { { \rm sin } [ k_\parallel v_\parallel t ] } i_{-1/3}(k_\parallel , k_\perp ^0 , k_\perp ^{m } ) , \label{drift_gen3223}\end{aligned}\ ] ] where @xmath277 is defined in eq . [ idef ] . expanding in series @xmath278 i find that the right hand side of eq.[drift_gen3223 ] can be cast as @xmath279 \int_{k_\perp^0}^{k_\perp^m } dk_\perp \frac{k_\perp ^{-\frac{2n+1}{3}}}{k_\perp^2 + k_\parallel^2 } . \label{i'series}\ ] ] in eq . [ iseries ] the @xmath280 term decreases as @xmath281 ; moreover , the integrals @xmath282 for @xmath283 and the integrals in @xmath18 are decreasing functions of @xmath284 . therefore the series in eq . [ iseries ] can be approximated to the lowest order , @xmath269 ( @xcite , eq . 3.194.5 , valid for @xmath285 ) . using again the critical balance condition ( @xmath244 ) leads to approximate the right hand side of eq.[drift_gen3223 ] as : @xmath286 |_{x^0 } { 3\over 2 } ( k_\perp^m l)^{2/3 } { } _ { 2}f_{1 } \left.\left(1,{1\over 3 } ; { 4 \over 3};- ( k_\perp l)^{2/3 } \right ) \right|_{k_\perp ^{m } } \label{series_tot_2}\ ] ] the only time - dependence of eq . [ series_tot_2 ] is the slow variation in the factor @xmath287 |_{x^0 } $ ] . the @xmath288 coefficient in eq . [ series_tot_2 ] is much smaller than in eq . [ series_tot_1 ] . thus , the term @xmath289 in the drift transport coefficient is neglected . * term 2233 * - the last non - vanishing term for the instantaneous drift transport coefficient includes contribution from both shear- and pseudo - alfvn modes . from the expression of @xmath223 in eq.[pcomp ] i find @xmath290}{k_\parallel v_\parallel } = \frac{{\cal n}\pi b_0 ^ 2}{v_\parallel } \int_{k_\parallel ^0 } ^{k_\parallel ^{m } } d k_\parallel { k_\parallel } { { \rm sin } [ k_\parallel v_\parallel t ] } \times \nonumber\\ \left ( \underbrace{{1\over l^{1/3}}j(k_\parallel , k_\perp ^0 , k_\perp ^{m})}_{\rm shear } - \underbrace{{\varepsilon \over \ell^{1/3 } } i_{-7/3}(k_\parallel , k_\perp ^0 , k_\perp ^{m})}_{\rm pseudo } \right ) \label{drift_gen2233}\end{aligned}\ ] ] where the @xmath291-term accounts for shear - modes and @xmath292 for the pseudo- . the auxiliary function @xmath291 is given by @xmath293 where in the second equivalence the contribution from @xmath294 is neglected . expanding again in series the exponential in the last member of eq . [ jdef ] , i find that the shear term in eq . [ drift_gen2233 ] can be cast as @xmath295 \left ( \frac{v_\parallel t}{x l^{1/3 } } + \frac{1}{{k_\perp^m } ^{2/3}}\right ) \label{series_tot_3}\ ] ] where the @xmath280 term decreases as @xmath296 . thus , to the lowest order , @xmath269 , eq.[series_tot_3 ] is approximated by @xmath297 ( by using @xmath298 ) , independent on time , so negligible with respect to the term @xmath299 for the reason explained in this appendix . as for the pseudo modes , the @xmath292-term in eq.[drift_gen2233 ] can be approximated , using again the series expansion of the exponential , by @xmath300 where @xmath301 . the integrals in @xmath302 for @xmath303 over three decades ( @xmath304 ) are of order @xmath113 . the dominant term ( @xmath269 ) falls in time as @xmath305 . we conclude that this term is negligible compared to the term @xmath299 . in this section , i illustrate the details of the calculation of @xmath146 . by using the definition of @xmath222 in the first relation of eq.[pcomp ] ( the result is unchanged if @xmath223 given by the second relation in eq.[pcomp ] is used ) and eq.[dmfl2 ] it is found in cylindrical coordinates ( @xmath224 ) with @xmath225 , @xmath226 , @xmath227 @xmath306}{k_\parallel } = { \cal n}\pi \int_{k_\parallel ^0 } ^{k_\parallel ^{m } } d k_\parallel { { \rm sin } [ k_\parallel v_\parallel t ] } \times\nonumber\\ \left ( \underbrace{{1\over k_\parallel l^{1/3}}j(k_\parallel , k_\perp ^0 , k_\perp ^{m})}_{\rm shear } + \underbrace{{\varepsilon k_\parallel \over \ell^{1/3 } } i_{-7/3}(k_\parallel , k_\perp ^0 , k_\perp ^{m})}_{\rm pseudo } \right ) \label{mfl_tot}\end{aligned}\ ] ] where @xmath291 and @xmath307 are defined respectively in eq.s [ jdef ] and [ idef ] . the @xmath291-term , from the second relation in eq . [ jdef ] , is given by @xmath308}{k_\parallel ^2 } \left ( \frac{1}{k_\parallel l^{1/3 } } + \frac{1}{{k_\perp^m } ^{2/3}}\right ) { \rm exp}\left ( \frac{-l^{1/3 } k_\parallel}{{k_\perp^m}^{2/3}}\right ) .\ ] ] by using the series expansion of the exponential and the same approximation as in eq . [ series_tot_3 ] , the shear - modes contribution to mfl transport coefficient can be approximated to the lowest order by @xmath309 where the approximation @xmath271 |_{x^0 } \sim 1/x_0 $ ] for @xmath272 was used . thus , the shear term grows linearly in time . the pseudo - modes term in eq . [ mfl_tot ] is equal to the pseudo - modes term in eq . [ drift_gen2233 ] , modulo a constant factor @xmath310 ; due to its decay in time for every @xmath284 , it is neglected here with respect to the shear - modes term . at scale @xmath31 . `` gc '' labels here the trajectory ( dotted ) of the guiding center drifting away from the local field line due to gradient / curvature drift by the average displacement @xmath311 within an elapsed time @xmath192 ( see sect . [ sect_drift ] ) . , width=491 ] kev and @xmath121 kev in the solar wind at @xmath0 au with @xmath124 ( a , b ) or @xmath312 ( c , d ) ; the solid ( dashed , dotted ) line corresponds to @xmath125 ( @xmath313 , @xmath132 ) ; also @xmath193 nt , @xmath314 , @xmath315 au ( @xmath316 respectively ) and @xmath317 . the horizontal lines in ( b ) correspond to the bohm diffusion at the particle energy considered.,title="fig : " ] kev and @xmath121 kev in the solar wind at @xmath0 au with @xmath124 ( a , b ) or @xmath312 ( c , d ) ; the solid ( dashed , dotted ) line corresponds to @xmath125 ( @xmath313 , @xmath132 ) ; also @xmath193 nt , @xmath314 , @xmath315 au ( @xmath316 respectively ) and @xmath317 . the horizontal lines in ( b ) correspond to the bohm diffusion at the particle energy considered.,title="fig : " ] to the gradient / curvature drift transport coefficient @xmath318 , for @xmath319 , corresponding to proton kinetic energy at @xmath0au given by @xmath320 kev respectively , as a function of @xmath321 as determined in eq . [ ratio ] ( here we used @xmath124 and @xmath322 ) . , width=377 ]

the hyades stream has long been thought to be a dispersed vestige of the hyades cluster . however , recent analyses of the parallax distribution , of the mass function , and of the action - space distribution of stream stars have shown it to be rather composed of orbits trapped at a resonance of a density disturbance . this resonant scenario should leave a clearly different signature in the element abundances of stream stars than the dispersed cluster scenario , since the hyades cluster is chemically homogeneous . here , we study the metallicity as well as the element abundances of li , na , mg , fe , zr , ba , la , ce , nd , and eu for a random sample of stars belonging to the hyades stream , and compare them with those of stars from the hyades cluster . from this analysis : ( i ) we independently confirm that the hyades stream can not be solely composed of stars originating in the hyades cluster ; ( ii ) we show that _ some _ stars ( namely 2/21 ) from the hyades stream nevertheless have abundances compatible with an origin in the cluster ; ( iii ) we emphasize that the use of li as a chemical tag of the cluster origin of main - sequence stars is very efficient in the range , since the li sequence in the hyades cluster is very tight , while at the same time spanning a large abundance range ; ( iv ) we show that , while this evaporated population has a metallicity excess of dex w.r.t . the local thin disk population , identical to that of the hyades cluster , the remainder of the hyades stream population has still a metallicity excess of to 0.15 dex , consistent with an origin in the inner galaxy ; ( v ) we show that the hyades stream can be interpreted as an inner 4:1 resonance of the spiral pattern : this then also reproduces an orbital family compatible with the sirius stream , and places the origin of the hyades stream up to 1 kpc inwards from the solar radius , which might explain the observed metallicity excess of the stream population . [ firstpage ] galaxy : evolution - clusters and associations - disk - solar neighbourhood

1101.2583

it has been known for a long time that a spatially unbound group of stars in the solar neighbourhood is sharing the same kinematics as the hyades open cluster ( e.g. , eggen 1958 ) . assuming that it is a vestige of an initially more massive hyades cluster which dispersed with time , with the result that its distribution function is still presently evolving towards equilibrium and does not yet satisfy the jeans theorem , eggen called this kinematically cold group the hyades supercluster . more generally , it is called the hyades moving group or hyades stream , since eggen s hypothesis that kinematic groups of this type are indeed such cluster remnants has been largely debated for many years . actually , such groups may also be generated by a number of global dynamical mechanisms . most of the mechanisms able to generate such unbound groups of stars moving in a peculiar fashion are linked with the non - axisymmetry of the galaxy , namely with the presence of a rotating central bar ( e.g. , dehnen 1998 , fux 2001 , minchev et al . 2010 ) and of spiral arms ( e.g. , quillen & minchev 2005 , antoja et al . 2009 ) , or both ( see quillen 2003 , minchev & famaey 2010 ) . a way to discriminate between these two hypotheses is the chemical tagging " of stars belonging to the moving groups ( freeman & bland - hawthorn 2002 , de silva et al . 2009 ) . with this technique , a true eggen moving group ( or true supercluster ) was recently tracked down , namely the hr1614 moving group ( eggen 1978 , feltzing & holmberg 2000 ) , in which the high level of chemical homogeneity is supporting the case that it is a relic of an ancient star - forming event ( de silva et al . conversely , the hercules stream was found to be composed of stars whose abundance pattern match that of the disk stars ( bensby et al . 2007 ) , which could be interpreted as being associated with the outer lindblad resonance of the bar ( dehnen 1998 , fux 2001 ) . here , we intend to analyze the abundance trends of a sample of stars belonging to the hyades stream , in order to check whether it is compatible with eggen s scenario , with the resonant scenario , or with a mix of both . indeed , the evaporation of open clusters and the dynamical perturbations linked with density perturbers are not mutually incompatible phenomena . stars originating from a 600 myr old cluster had time to disperse over more than 500 pc in the disk ( e.g. , bland - hawthorn et al . 2010 ) , but have retained the same guiding radius and same tangential velocity @xmath3 ( woolley 1961 ) . this does not prevent an overdensity in velocity space due to orbits trapped at resonance to overlap with this cluster remnant . the full geneva - copenhagen ( gc ) sample from holmberg et al . ( 2009 ) in the @xmath4 plane . the hyades box is also indicated at @xmath5 @xmath6 and @xmath7 . , width=302 ] the study of the kinematics of k and m giant stars in the solar neighbourhood , combining coravel radial velocities with hipparcos parallaxes and proper motions ( famaey et al . 2005 ) , already indicated that the hyades stream could not be _ solely _ composed of coeval stars evaporated from the primordial hyades cluster , because the hertzsprung - russell diagram was very similar for stars in the stream and in the field , indicating a wide range of ages for stars belonging to the stream . this was later confirmed by a detailed analysis of the distribution of stars in parallax space as compared to the parallax corresponding to the hyades cluster s 600 myr isochrone ( famaey , siebert & jorissen 2008 ) . bovy & hogg ( 2010 ) recently reached the same conclusion . moreover , at about the same time , radial velocities , masses , and metallicities of more than 14000 f and g dwarfs stars were also published as the _ geneva - copenhagen _ ( gc ) _ survey _ ( nordstrm et al . 2004 , holmberg et al . 2007 , 2009 ) , supplementing the hipparcos parallaxes and proper motions . the distribution of these main - sequence stars in velocity space confirmed that the hyades stream was the most prominent feature on top of the background velocity ellipsoid ( fig . [ fig : gc ] ) . sellwood ( 2010 ) then estimated the actions of individual stars from their observed coordinates and velocities , and examined the stellar distribution in the @xmath8-@xmath9 action space . he found about 5% of the gc stars concentrated along a resonance line in action space , which was interpreted as a signature of scattering at the inner lindblad resonance of a spiral pattern . this overdensity in action space precisely corresponds to the hyades stream in velocity space , in accordance with the model of quillen & minchev ( 2005 ) . what is more , famaey et al . ( 2007 ) compared the mass distribution of stars belonging to the hyades stream with the initial mass function of the hyades cluster and with the present - day mass function of the galactic disk , showing that it was in disagreement with the former and in agreement with the latter , thus also favoring a dynamical , resonant origin for the stream . however , the latter analysis was compatible with a proportion of at least 15% of stream stars actually being past members of the hyades cluster . it was also found that , while thin disk stars have a mean metallicity [ fe / h]@xmath10 , the mean metallicity of stars moving with the hyades ( among which only @xmath1125% of field stars from the background velocity ellipsoid , the rest constituting the hyades overdensity in velocity space ) was [ fe / h ] @xmath12 . since the hyades cluster is also more metal rich than the field , with a mean [ fe / h ] @xmath13 ( cayrel de strobel et al . 1997 , perryman et al . 1998 , grenon 2000 ) , this could also argue in favor of a large proportion of stars in the hyades stream originating from the hyades cluster . however , this higher metallicity of the hyades stream could also indicate that its stars originate from the inner galaxy . in this paper , we aim at determining the chemical abundances of stars belonging to the hyades stream , in order to yield constraints on their origin ( from the cluster or from the field of the disk , and in the latter case , from which typical galactocentric radius ) . in this way we shall be able to _ independently _ confirm from chemical tagging that the hyades stream can not be solely composed of stars originating in the hyades cluster , and we shall be able to provide the first _ direct _ evidence for two distinct populations inside the stream . we also aim at making a preliminary analysis of the respective characteristics of these two populations . in sect . 2 , we present the sample of stars that we are going to analyze , and in sect . 3 we briefly describe the observations performed with the hermes / mercator spectrograph ( raskin et al . stellar parameters and abundances are respectively determined in sects . 4 and 5 . we then investigate whether stars in the stream are compatible with being evaporated from the hyades cluster in sects . 6 and 7 , and discuss these results and their consequences in sect . conclusions are drawn in sect . lrcccrlccrcccccr hd & & @xmath14 & @xmath15 & @xmath3 & & @xmath16 & @xmath17 & @xmath18 & + & & & ( km / s ) & ( km / s ) & & ( mag ) & & ( @xmath19 ) & + + & & & + + 18632 & 23 & 6.13 & -42 & -19 & -1 & 7.978 & & 0.87 & 3 + 19902 & 40 & 5.19 & -41 & -19 & -1 & 8.171 & 0.452 & 0.89 & 2 + 26756 & 46 & 5.13 & -41 & -18 & -3 & 8.457 & 0.431 & 0.93 & 5 + 26767 & 43 & 4.88 & -41 & -18 & -3 & 8.045 & 0.405 & 0.98 & 5 + hip 13806 & 39 & 5.94 & -41 & -17 & -2 & 8.90 & & 0.89 & 4 + + & & & + + 20430 & 46 & 4.07 & -42 & -23 & -2 & 7.386 & 0.364 & 1.10 & 6 + 20439 & 42 & 4.66 & -41 & -21 & -4 & 7.766 & 0.395 & 1.17 & 6 + 26257 & 58 & 3.84 & -35 & -13 & -5 & 7.639 & 0.345 & 1.25 & 8 + hip 13600 & 53 & 5.21 & -40 & -13 & -2 & 8.83 & & 1.00 & 2 + + & & & + + 25680 & 17 & 4.79 & -25 & -14 & -6 & 5.903 & 0.399 & 0.96 & 3 + 42132 & 111 & 1.52 & -43 & -16 & -23 & 6.675 & 0.525 & 0.89 & 3 + 67827 & 45 & 3.39 & -35 & -14 & -9 & 6.580 & 0.368 & 1.34 & 5 + 86165 & 51 & 4.39 & -35 & -17 & -16 & 7.926 & 0.386 & 0.97 & 2 + 89793 & 63 & 4.87 & -33 & -20 & 12 & 8.853 & 0.424 & 0.96 & 1 + 90936 & 56 & 4.62 & -42 & -18 & -14 & 8.371 & 0.389 & 0.96 & 3 + 103891 & 58 & 2.79 & -28 & -17 & 7 & 6.591 & 0.356 & 1.35 & 4 + 108351 & 88 & 3.19 & -50 & -19 & 5 & 7.905 & 0.327 & 1.40 & 9 + 133430 & 57 & 4.79 & -30 & -16 & -5 & 8.559 & 0.413 & 0.90 & 2 + 134694 & 134 & 2.87 & -43 & -20 & -17 & 8.502 & 0.336 & 1.34 & 10 + 142072 & 39 & 4.90 & -29 & -15 & -1 & 7.854 & 0.417 & 0.90 & 7 + 149028 & 48 & 5.10 & -37 & -16 & -1 & 8.530 & 0.454 & 0.91 & 2 + 149285 & 196 & 2.36 & -34 & -14 & -14 & 8.825 & 0.337 & 1.32 & 8 + 151766 & 109 & 2.42 & -35 & -13 & -20 & 7.606 & 0.348 & 1.46 & 8 + 155968 & 55 & 4.72 & -30 & -14 & 2 & 8412 & 0.423 & 0.96 & 3 + 157347 & 20 & 4.83 & -28 & -17 & -20 & 6.287 & 0.425 & 0.90 & 1 + 162808 & 63 & 4.41 & -31 & -12 & -2 & 8.423 & 0.393 & 1.00 & 6 + 171067 & 26 & 5.13 & -46 & -16 & -14 & 7.205 & 0.424 & 0.90 & 1 + 180712 & 45 & 4.73 & -25 & -12 & -6 & 7.977 & 0.395 & 0.88 & 3 + 187237 & 26 & 4.78 & -36 & -20 & 14 & 6.877 & 0.409 & 0.90 & 2 + 189087 & 27 & 5.75 & -42 & -15 & 5 & 7.886 & 0.483 & 0.81 & 2 + the present stellar sample in the @xmath20 plane . the meaning of the symbols is given in the figure ( see sect . [ sect : evap ] for more details ) . , width=302 ] fig . [ fig : gc ] presents the distribution of all stars from the gc survey in the @xmath4-plane ( @xmath15 is the velocity towards the galactic centre , @xmath3 the velocity in the direction of galactic rotation , both with respect to the sun ) , and the hyades overdensity corresponds to the box defined by @xmath21 and @xmath7 ( see famaey et al . 2007 ) : this box corresponds to a region in which the density of the gc survey in @xmath4 velocity space is higher than @xmath22 stars/(km / s)@xmath23 . a randomly chosen stellar sample within this velocity box has been selected among the gc survey stars with @xmath24 ( or equivalently , @xmath25 , according to tables 15.7 and 15.10 of drilling & landolt 1999 ) and @xmath26 @xmath27 , the latter condition in order to ease the abundance determination . members of the hyades cluster , taken from the de silva et al . ( 2006 ) sample have been added to this gc kinematical sample , in order to compare the element abundances in the stream and in the cluster . the criterion on the colour index of the stream stars is necessary to avoid any bias from the rejection of the fast rotators from the stream : since paulson et al . ( 2003 ; their fig . 5 ) have shown that in the hyades cluster , all stars ( but one ) with @xmath28 have @xmath26 @xmath27 , so that few if any stars evaporated from the cluster have been left aside by selecting only slow rotators with @xmath28 in the stream . among the hyades cluster members we have selected , 4 are controversial ( see table [ tab_param ] ) : they are spatially associated with the cluster but they might not belong to it based on kinematics ( de bruijne et al . 2001 ) . the full stellar sample is listed in table [ tab : sample ] , the distance and spatial velocities are taken from holmberg et al . ( 2009 ) , the masses from holmberg et al . ( 2007 ) , and the rotation velocity from nordstrm et al . positions in the galaxy and velocities are displayed in figs . [ fig : uv ] to [ fig : xz ] . as each star has been observed individually , the sample is relatively small , comprising 21 stars randomly chosen in the hyades box , 5 sure members of the hyades cluster , and 4 possible members . this small sample size nevertheless already allows us to draw a number of secure conclusions , as we shall see in the following sections , and paves the way towards more detailed analyses of much larger samples with future multi - fibre spectrographs such as the high resolution multi - object spectrograph of the anglo - australian telescope , which like the spectrometer used by us also is called hermes ( barden et al . 2008 ) . for the @xmath29 plane . , width=302 ] plane ( with @xmath30 pointing towards the galactic centre and @xmath31 in the direction of the galactic rotation ; distances on both axes are expressed in parsecs ) . symbols are as in fig . 2 . , width=302 ] for the @xmath32 plane.,width=302 ] .[linelist ] line list . an asterisk ( * ) after the wavelength means that hyperfine structure and isotopic shifts have been taken into account for the considered line . references to the adopted @xmath33 are given in the last column and the corresponding reference is listed at the bottom of the table . [ cols="<,^,^,^,^,^,<,^,^ , > " , ] lrrrlrrrc stream & @xmath34 & @xmath35 & [ fe i / h ] & cluster & @xmath34 & @xmath35 & [ fe i / h ] & member + & & & & hd 26767 & -1.22 & -1.24 & 0.08 & 1 + & & & & hd 19902 & -0.76 & -1.08 & 0.00 & 1 + & & & & hd 26756 & -0.57 & -0.88 & 0.00 & 1 + hd 162808 & 0.90 & 0.65 & -0.02 & hd 20430 & 0.16 & 0.54 & 0.14 & 5 + hd 149028 & 0.93 & 0.59 & 0.02 & hd 18632 & 0.50 & 0.16 & 0.04 & 1 + & & & & hip 13806 & 1.82&1.99 & -0.09 & 1 + hd 108351 & 2.00 & 2.19 & -0.10 & hip 13600 & 1.94 & 2.48 & -0.14 & 4 + hd 89793 & 2.77 & 2.53 & 0.00 + hd 155968 & 3.22 & 2.97 & 0.01 + hd 25680 & 3.64 & 3.96 & -0.14 & hd 20439 & 3.99 & 4.00 & 0.14 & 5 + hd 157347 & 3.96 & 4.27 & -0.14 + hd 90936 & 4.11 & 3.97 & -0.04 + hd 180712 & 4.12 & 4.53 & -0.16 + hd 67827 & 4.19 & 4.58 & -0.16 + hd 134694 & 4.23 & 4.14 & -0.06 + hd 171067 & 4.55 & 5.50 & -0.25 + hd 187237 & 4.91 & 4.83 & -0.06 + hd 142072 & 4.94 & 4.77 & 0.06 + hd 103891 & 4.99 & 7.27 & -0.42 + hd 133430 & 5.42 & 6.21 & -0.25 + hd 149285 & 6.55 & 7.41 & -0.28 + hd 151766 & 6.69 & 6.65 & -0.06 + hd 42132 & 7.14 & 7.53 & -0.20 + hd 189087 & 7.20 & 7.76 & -0.24 & hd 26257 & 7.22 & 7.13 & 0.06 & 4 + hd 86165 & 7.60 & 8.35 & -0.28 + + lllllllll & cluster & evaporated & possible & stream / field & hd 26756 % solar + & & candidates & cluster & population & ( asplund et al . 2005 ) + @xmath36 & 6 & 2 & 3 & 19 + @xmath37 & @xmath38 & @xmath38 & @xmath39 & @xmath40 + + % hd 26756 + @xmath41[na / fe]@xmath42 & [email protected] & [email protected] & [email protected] & @xmath44 & -0.05 + @xmath45[na / fe ] ) & 0.032 & 0.09 & 0.03 & 0.06 + @xmath46(p03 ) & 0.06 + @xmath47[na / fe ] ) & 0.06 & 0.07 & 0.06 & 0.07 + @xmath48 & @xmath49 & 0.05 & @xmath49 & @xmath50 + + @xmath41[mg / fe]@xmath42 & [email protected] & @xmath51 & [email protected] & [email protected] & -0.08 + @xmath45[mg / fe ] ) & 0.021 & 0.06 & 0.05 & 0.05 + @xmath46(p03 ) & 0.04 + @xmath47[mg / fe ] ) & 0.06 & 0.09 & 0.09 & 0.08 + @xmath48 & @xmath52 & @xmath50 & @xmath53 & @xmath53 + + @xmath41[fe i / h]@xmath42 & @xmath540.03 & @xmath55 & [email protected] & @xmath56 0.03 & 0.23 + @xmath45[fe i / h ] ) & 0.072 & 0.02 & 0.12 & 0.12 + @xmath46(p03 ) & 0.05 + @xmath47[fe i / h ] ) & 0.08 & 0.06 & 0.06 & 0.09 + @xmath48 & @xmath57 & @xmath52 & 0.10 & 0.08 + + @xmath41[fe ii / h]@xmath42 & [email protected] & @xmath58 & [email protected] & [email protected] & 0.19 + @xmath45[fe ii / h ] ) & 0.068 & 0.01 & 0.09 & 0.11 + @xmath47[fe ii / h ] ) & 0.07 & 0.08 & 0.06 & 0.08 + @xmath48 & 0.00 & @xmath59 & 0.07 & 0.07 + + @xmath41[zr / fe]@xmath42 & [email protected] & [email protected] & [email protected] & [email protected] & -0.16 + @xmath45[zr / fe ] ) & 0.046 & 0.01 & 0.09 & 0.09 + @xmath46(ds ) & 0.055 + @xmath47[zr / fe ] ) & 0.06 & 0.07 & 0.06 & 0.11 + @xmath48 & @xmath53 & @xmath59 & 0.07 & @xmath60 + + @xmath41[ba / fe]@xmath42 & [email protected] & @xmath61 & [email protected] & [email protected] & 0.12 + @xmath45[ba / fe ] ) & 0.044 & 0.08 & 0.08 & 0.13 + @xmath46(ds ) & 0.049 + @xmath47[ba / fe ] ) & 0.08 & 0.07 & 0.11 & 0.10 + @xmath48 & @xmath62 & 0.04 & @xmath63 & 0.08 + + @xmath41[la / fe]@xmath42 & [email protected] & [email protected] & [email protected] & [email protected] & -0.17 + @xmath45[la / fe ] ) & 0.033 & 0.04 & 0.11 & 0.10 + @xmath46(ds ) & 0.025 + @xmath47[la / fe ] ) & 0.06 & 0.08 & 0.07 & 0.07 + @xmath48 & @xmath49 & @xmath62 & 0.08 & 0.07 + + @xmath41[ce / fe]@xmath42 & [email protected] & [email protected] & [email protected] & [email protected] & -0.01 + @xmath45[ce / fe ] ) & 0.041 & 0.04 & 0.10 & 0.13 + @xmath46(ds ) & 0.025 + @xmath47[ce / fe ] ) & 0.10 & 0.02 & 0.07 & 0.08 + @xmath48 & @xmath62 & 0.04 & 0.08 & 0.10 + + lllllllll & cluster & evaporated & possible & stream / field & hd 26756 % solar + & & candidates & cluster & population & ( asplund et al . 2005 ) + @xmath36 & 6 & 2 & 3 & 19 + @xmath37 & @xmath38 & @xmath38 & @xmath39 & @xmath40 + + % hd 26756 + @xmath41[nd / fe]@xmath42 & [email protected] & [email protected] & [email protected] & [email protected] & -0.16 + @xmath45[nd / fe ] ) & 0.087 & 0.02 & 0.16 & 0.15 + @xmath46(ds ) & 0.032 + @xmath47[nd / fe ] ) & 0.05 & 0.07 & 0.07 & 0.07 + @xmath48 & 0.07 & @xmath52 & 0.15 & 0.14 + + @xmath41[eu / fe]@xmath42 & @xmath640.01 & [email protected] & @xmath65 & [email protected] & -0.10 + @xmath45[eu / fe ] ) & 0.028 & 0.01 & 0.13 & 0.09 + @xmath47[eu / fe ] ) & 0.06 & 0.06 & 0.06 & 0.06 + @xmath48 & 0.03 & @xmath59 & 0.11 & 0.07 + the main goal of these abundance measurements is to check whether at least a small fraction of the hyades stream is compatible with having its origin in the hyades cluster . to identify possible evaporated stars from the primordial cluster , we computed the @xmath66 statistics : @xmath67\ ; - < [ x_{i}/{\rm fe}]>_{\rm cluster}\right)/\sigma_{i,{\rm cluster * } } \right]^{2},\end{aligned}\ ] ] and @xmath68\ ; - < [ x_{i}/{\rm fe}]>_{\rm cluster}\right)/\sigma_{i,{\rm cluster * } } \right]^{2}\\ & & + \left [ \left ( [ { \rm fe i}/{\rm h}]\ ; - < [ { \rm fe i}/{\rm h}]>_{\rm cluster } \right)/\sigma_{\rm fei , cluster * } \right]^{2},\nonumber\end{aligned}\ ] ] where the average abundance for the cluster and the star - to - star scatter @xmath69 are taken over the 5 sure members ( see table [ tab_param ] ) . despite the small number of stars involved , the differential abundance scatter used in the above @xmath66 formulae agree reasonably well with the values derived by paulson et al . ( 2003 ) and de silva et al . ( 2006 ) on a much larger sample of hyades cluster stars [ see @xmath46(ds ) and @xmath46(p03 ) in table [ meanvalues_delta1 ] ] . thus the relatively small number of hyades cluster stars used in the present study does not appear to hamper our analysis . the motivation to use two statistical indices [ @xmath70 and @xmath71 based on different elemental sets is to illustrate the sensitivity to specific choices of elemental sets , as we will discuss below . we decided to leave li , nd and eu out of these indices , because * nd possibly exhibits an abundance trend with @xmath72 ( see fig . [ diff ] ) which could induce a spurious difference between the average nd abundances in the stream and cluster , depending on how stars in these two groups distribute with @xmath72 ; * eu has the largest sensitivity to errors on the stellar - model parameters ( table [ taberr ] , where @xmath73 dex for eu as compared to less than 0.10 dex for all the other elements ) ; * li is a very sensitive tagger which will be used afterwards ( sect . [ sect : li ] ) as an independent check of the quality of the tagging based on the @xmath66 statistics from fe and the heavier elements . the different number of degrees of freedom ( 4 and 5 , where we suppose that the average and @xmath46 do not act to decrease the number of degrees of freedom ; this is certainly true for the stream stars ) for the two @xmath66 statistics makes their direct comparison difficult . it is therefore useful to use the @xmath37 goodness - of - fit statistics instead , which transforms a @xmath66 statistics into a normal distribution with zero mean and unit variance , irrespective of the number of degrees of freedom . @xmath37 is defined as @xmath74,\ ] ] where @xmath75 is the number of degrees of freedom of the @xmath66 variable . the above definition corresponds to the ` cube - root transformation ' of the @xmath66 variable ( stuart & ord 1994 ) . the values of @xmath76 and @xmath77 for all the stars studied are given in table [ deviations ] . as apparent from that table , all sure hyades cluster members ( plus the possible member hd 20430 ) have @xmath78 . this is actually not really surprising , since the @xmath46 value entering the @xmath66 is derived from the cluster sample itself . the predictive power of this @xmath66 criterion therefore rather lies in its application to the stream sample . two stars from the stream ( hd 149028 and hd 162808 ) have both @xmath78 and may thus be considered as evaporated from the cluster ( the analysis of the li sequence presented in sect . [ sect : li ] will confirm that conclusion ) . all other stars , including the possible members ( hip 13600 , hd 20439 , and hd 26257 ) , have ( at least one ) @xmath79 and must not be considered as cluster members based on the fe and heavy element abundances , the latter star hd 26257 being even located way off the threshold , with @xmath37 of the order of 7 . this star was studied by de silva et al . ( 2006 ) as well , who find relative abundances @xmath80[zr / fe ] = @xmath81 , @xmath80[ba / fe ] = @xmath82 , @xmath80[ce / fe ] = @xmath83 , and @xmath80[nd / fe ] = @xmath84 not so different form ours : @xmath80[zr / fe ] = @xmath85 , @xmath80[ba / fe ] = @xmath85 , @xmath80[ce / fe ] = @xmath86 , and @xmath80[nd / fe ] = @xmath87 , the latter set being consistently underabundant , in contrast with the de silva et al . values . actually , one may wonder whether the three controversial hyades cluster stars ( hip 13600 , hd 20439 , and especially hd 26257 ) could not in fact be stream stars happening to be located in the spatial vicinity of the cluster . interestingly , we stress that doubts were raised as well by de silva et al . ( 2006 ) about the membership of hd 20439 ( also known as vb 2 ) , based on outlying abundances , in agreement with its outlying @xmath37 obtained in the present study . incidentally , these authors raise the same doubts for hd 20430 ( vb 1 ) , which our study does not tag as an outlier , however . in any case , there is a large population of stream stars which are chemically very different from the hyades cluster members , as apparent from fig . [ fig : delta ] presenting the distributions of @xmath76 , @xmath77 and @xmath80 [ fe / h ] for the hyades stream and cluster . we may thus conclude that , irrespective of the index being used , only a few stars from the hyades stream have abundances compatible with the abundance ratios of stars of the hyades cluster ( see sect . 8.1 ) . the distributions of @xmath34 ( top panel ) , @xmath35 ( middle panel ) and @xmath80 [ fe / h ] ( bottom panel ) for the hyades stream stars ( open histogram ) , cluster stars ( shaded histogram ) , and sure cluster stars ( thick histogram ) . , title="fig:",width=283 ] the distributions of @xmath34 ( top panel ) , @xmath35 ( middle panel ) and @xmath80 [ fe / h ] ( bottom panel ) for the hyades stream stars ( open histogram ) , cluster stars ( shaded histogram ) , and sure cluster stars ( thick histogram ) . , title="fig:",width=283 ] the distributions of @xmath34 ( top panel ) , @xmath35 ( middle panel ) and @xmath80 [ fe / h ] ( bottom panel ) for the hyades stream stars ( open histogram ) , cluster stars ( shaded histogram ) , and sure cluster stars ( thick histogram ) . , title="fig:",width=283 ] we now turn to lithium as a tagger that we use to check the quality of the tagging based on fe and heavier elements . boesgaard & king ( 2002 ) have provided a comparison sample of li abundances in hyades cluster stars , as shown on fig . [ fig : li ] ( crosses ) . the li abundance in the cluster stars follows a tight sequence that goes through a maximum around 6200 k , and decreases sharply on either side as li gets destroyed by diffusive or convective downward motions ( see deliyannis 2000 and sestito & randich 2005 for reviews ) . on the opposite , in field stars , the li abundance spans a substantial range ( of the order of 2 dex ) for any given effective temperature ( _ e.g. , _ israelian et al . 2009 and references therein ) , especially on the cooler side of the peak . the tight sequence observed for the hyades cluster is a direct consequence of the fact that its stars are coeval ( deliyannis 2000 ) . it offers an efficient way to tag stream stars evaporated from the cluster ( although one can not totally exclude the possibility that a stream star falls by chance along the hyades sequence ) . indices , compared to boesgaard & king ( 2002 ) li abundances for hyades cluster stars ( crosses ) . arrows mark upper limits on the li abundance . [ fig : li ] , width=302 ] fig . [ fig : li ] confirms the chemical tagging analysis performed in sect . [ sect : evap ] , since the two stars tagged as ` evaporated candidates ' indeed lie along the hyades cluster li sequence . conversely , we find three stream stars falling along the li sequence ( hd 25680 , hd 142072 , hd 149285 ) , but this can be a chance occurrence , and does not impose a revision of our former tagging ( among these three , only hd 25680 has both @xmath88 ) . finally , we note that the three ` possible cluster members ' nicely fall along the hyades li sequence . since famaey et al . ( 2007 ) estimated that the hyades overdensity in velocity space constitutes @xmath89% of the hyades box ( this frequency being almost independent of the colour range ) , we can ask ourselves whether this overdensity can _ entirely _ be due to the population of stars evaporated from the hyades cluster . however , if this were true , we would expect to have found about 15 ( or about 11 to 19 with poissonian uncertainties ) stars with abundances compatible with the hyades cluster ( @xmath90 with @xmath91 ) , instead of 2 , in the most favorable case . this difference leads to a significance level close to @xmath92 ( within @xmath93 ) for the unilateral rejection of the null hypothesis that all the stars of the velocity overdensity are originating from the hyades cluster , as obtained from the relation ( resulting from the fact that for @xmath94 and @xmath95 , the binomial distribution is already well approximated by a gaussian distribution ) @xmath96 where @xmath97 , @xmath98 and @xmath99 is the abscissa of the reduced normal distribution @xmath100 such that @xmath101 . we are confident that the sample selection criterion @xmath102 ( sect . [ sect : sample ] ) does not affect this conclusion , for the following reason : all stream stars of our sample are low - mass main sequence stars with @xmath24 ( or equivalently @xmath25 , according to tables 15.7 and 15.10 of drilling & landolt 1999 ) , and paulson et al . ( 2003 ) showed that such stars in the hyades cluster indeed have @xmath103 , so that no candidate evaporated star should have been missed in the stream by rejecting the fast rotators . this is thus an independent confirmation of the dynamical ( most probably resonant ) origin of most of the hyades stream , even though a population of stars evaporated from the hyades cluster does overlap with this dynamical stream . [ diff ] presents the differential abundances @xmath80[x / fe ] ( relative to hd 26756 ) against @xmath72 with different symbols identifying the hyades cluster members ( large filled circles ) , the possible cluster members ( small filled circles ) , evaporated candidates ( large filled squares ) , and finally stars from the stream / field ( open squares ) . in table [ meanvalues_delta1 ] , we evaluate the average abundances for each of the elements considered , separately for these four groups . we report the mean and standard deviations @xmath104 for a given chemical element separately for these four samples , and the root - mean - square @xmath105 ( taken over all stars of a given group ) of the line - to - line scatter @xmath106 ( for a given star , taken from tables [ tab_param ] and [ tabab ] ) . the @xmath105 parameter must be seen as an instrumental scatter , to which any extrinsic ( physical ) scatter ( @xmath48 ) will add quadratically to yield @xmath104 , so that we must expect @xmath107 . for the ` cluster ' and ` evaporated candidates ' samples however , @xmath108 is usually smaller than @xmath109 , which indicates that the true scatter of these samples is so small that it can not be derived accurately with the achieved line - to - line scatter . a very consistent picture thus emerges from table [ meanvalues_delta1 ] regarding the sample genuine scatter @xmath48 : for the cluster sample , it is , as expected , quite small ( @xmath110 dex ) for all elements but fe i and nd , where it reaches 0.07 dex . except for these two elements , our results agree well with the conclusion of chemical homogeneity found for 46 hyades cluster stars by paulson et al . ( 2003 ) and de silva et al . ( 2006 ; compare the entries listed as @xmath46 in table [ meanvalues_delta1 ] with @xmath46(ds ) or @xmath46(p03 ) ) . on the contrary , for the stream , the scatter is systematically much larger , ranging from 0.05 dex for mg to 0.15 dex for nd , as expected if the stream is dominated by _ field _ stars . it is not surprising either that the sample of evaporated stars has properties similar to those of the cluster , both in terms of scatter and average value , since they were actually selected to ensure such a similarity . the ` possible cluster ' stars deviate markedly from the cluster values , but as indicated in sect . [ sect : evap ] , the statistical properties of this group are dominated by the outlying star hd 26257 . there are as well noteworthy differences ( by a least 0.1 dex ) between cluster and stream stars in terms of their average abundances , especially for fe , la , nd and eu ( table [ meanvalues_delta1 ] ) . the difference in the average metallicity of the two groups ( 0.2 dex ) is very significant ( table [ meanvalues_delta1 ] and fig . [ fig : delta ] ) . as apparent from fig . [ comp ] , the difference in the average la , nd , and eu abundances between the hyades cluster and stream appears as a natural consequence of their different average metallicities and of the trend with metallicity observed for these elements in the thin disk . this finding is an important result of the present paper : the stream / field population ( thus cleaned from the evaporated metal - rich cluster stars ) remains more metal - rich ( [ fe / h ] of the order 0.04 to 0.08 , normalized to a solar fe abundance of 7.45 , or -0.01 to 0.05 with the more traditional normalization of 7.50 from grevesse & sauval ( 2000 ) , in better agreement with older studies ; table [ meanvalues_delta1 ] ) than the thin disk population in the solar neighbourhood . according to famaey et al . ( 2007 ) , the mean ( photometric ) metallicity of the whole geneva - copenhagen survey for the solar neighbourhood is [ fe / h ] @xmath111 . excluding halo stars with [ fe / h ] @xmath112 , and excluding all the stars from the hyades box , we get a mean ( photometric ) metallicity of [ fe / h ] @xmath113 . given that photometric metallicity estimates are systematically smaller than spectroscopic values by about 0.05 dex ( holmberg et al . 2009 ) , the thin disk metallicity in the solar neighbourhood may be evaluated at @xmath114 dex ( see also allende - prieto 2010 and references therein ) . comparing this thin - disk metallicity of @xmath114 dex with that of the stream / field ( of the order @xmath115 to 0.08 dex ) clearly supports an inner - disk origin for the stream . we stress that this comparison is somewhat biased by the fact that about 25% of the stars in the hyades velocity box must belong to the local background schwarzschild velocity ellipsoid ( sect . 1 and famaey et al . 2007 ) , and thus have a metallicity typical of the thin disk . one may try to clean the sample from the local background field stars to better evaluate the metallicity of the stream stars . an upper limit to the average stream metallicity may be obtained by removing from the 21 stars making up our hyades box the 25% most metal - deficient stars , believed to belong to the field . thus removing the most metal - poor stars hd 86165 , hd 103891 , hd 133430 , hd 149285 and hd 171067 from the stream / field sample , one finds an average metallicity of 0.114 dex ( with the @xmath116 normalization ) , corresponding to an excess of ( at most ) 0.18 dex with respect to the metallicity of the local thin disk . the argument of an inner - disk origin for the stream may even be made more quantitative by adopting a reasonable metallicity gradient of @xmath84 to @xmath114 dex / kpc in the galactic disk ( e.g. daflon & cunha 2004 ) : this would indicate that the hyades stream originates at @xmath110.9 to 3.0 kpc ( or even at 3.6 kpc if one adopts the upper limit to the metallicity excess , or even more taking into account the fact that the gradient for old stars may be even flatter ) inwards in the galactic disk . in sect . [ sect : ilr ] , this range will be compared with predictions made in the framework of a stream caused by an inner 4:1 resonance from a 2-armed spiral pattern ( inner ultra - harmonic resonance or iuhr ) . the large difference in the average metallicities of the hyades cluster and stream ( after cleaning the latter from the 2 evaporated candidates ) definitely and undoubtedly confirms that the stream is not solely made out of evaporated cluster stars . and concomittantly , this finding reinforces the hypothesis of the dynamical nature of the hyades stream . in this context , one may thus compare the scatter and abundance trends obtained for the hyades stream stars with those in the inner thin disk from which they originate . this is done in fig . [ comp ] using the samples of bensby et al . ( 2003 , 2005 ) and reddy et al . it is seen that the stream conforms with the properties of the thin disk : absence of metallicity trend and small scatter for mg , relatively large scatter for ba , and small scatter and well marked trend for eu ( fig . [ comp ] ) . the small shifts between our values and bensby s ones , apparent for mg and eu , may be attributed to the slightly different line list and solar abundances adopted for normalization , with @xmath117 of 7.53 , 7.45 and 0.52 adopted in this work for mg , fe , and eu respectively , as compared to 7.57 7.58 , 7.53 7.56 , and 0.47 0.56 for bensby et al . ( 2005 ; their table 5 , where the adopted abundances may depend upon the spectrograph and line used ) . with respect to a concave arm , both orbital families enter the solar neighborhood stellar velocity distribution ( black filled circle ) . the galactocentric axes are in units of @xmath118 ( the galactocentric radius of the sun ) . bottom panel the effect on the @xmath4-plane for the configuration shown in the top panel ( selecting test particles in a 200 pc circle around the sun ) . each orbital family gives rise to a stream in velocity space . we can associate the dense clump at @xmath119 km / s with the hyades and the shallow one at @xmath120 km / s with sirius . the contour levels correspond to 0.2 , 0.31 , 0.43 , 0.55 , 0.67 , and 0.8 of the maximum value at the centre of the hyades clump . , width=302 ] the question now is which dynamical effects in the galaxy can both place the hyades stream in its observed location in velocity space , while also having a metallicity typical of stars originating at 0.9 to 3.0 kpc inside the solar circle . as was shown by minchev et al . ( 2010 ) , a resonance with the galactic bar can not account for the position of the hyades stream in the @xmath4-plane ( while it actually can account for the pleiades ) , nor for an origin farther away than @xmath121 kpc in the inner disk . on the other hand , using an orbital weighting function technique , quillen & minchev ( 2005 ) showed that the 4:1 inner ultra - harmonic resonance ( iuhr ) of a 2-armed spiral structure splits the velocity distribution into two features corresponding to two orbital families , one of them consistent with the hyades stream ( see also sect . [ sec : intro ] ) . to check this possibility , we performed test - particle simulations of a stellar disk consistent with the milky way kinematics , perturbed by a 2-armed spiral pattern . details about the simulation parameters can be found in minchev & quillen ( 2007 ) . we examined a range of pattern speeds @xmath122 , and indeed found that we could reproduce the position of the hyades stream in the @xmath4-plane ( using the solar motion of schnrich et al . 2010 ) only when the solar circle is near the 4:1 iuhr . [ fig : hyades_uv ] presents our results for @xmath123 km / s / kpc for an angular speed at the solar position @xmath124 km / s / kpc ( i.e. @xmath125 ) . the top panel shows the two orbital families and the bottom panel the resulting velocity distribution in the @xmath4-plane . the beauty of this simulation is that , while reproducing the hyades stream , the other orbital family creates another remarkable feature in velocity space around @xmath120 km / s , which is consistent with the sirius moving group ( see e.g. famaey et al . 2005 , 2008 ) . it is slightly shallower than the hyades stream , exactly as observed . this simulation furthermore predicts that the hyades stream originates about 1 kpc inward from the solar radius , which lies within the error bar derived in sect . [ sect : stream ] from the observed metallicity difference ( local thin disk - hyades stream : 0.06 to 0.15 dex ) and metallicity gradient in the thin disk ( @xmath84 to @xmath114 dex / kpc ; daflon & cunha 2004 ) . this error bar is unfortunately quite large still ( 0.9 3.0 kpc ) to meaningfully constrain the simulation , and one should strive to reduce it in the future . if in the end , a metallicity difference as large as ( or larger than ) 0.1 dex were confirmed , agreement with the present simulation could only be restored at the expense of the necessity for a steeper metallicity gradient in the disk . let us however note that daflon & cunha ( 2004 ) studied ob stars , i.e. the very recent metallicity gradient , and that the gradient for older stars such as those of the hyades stream is expected to be flatter due to radial mixing , not steeper . this might cast doubt on the ability of the present dynamical model to reproduce the observed metallicity excess , should it be confirmed by further studies . on the other hand , there exists other mechanisms that could have made the hyades stream migrate from larger distances in the inner disk . it is known that stars can exchange angular momentum in galactic disks due to three distinct mechanisms : transient spirals ( sellwood & binney 2002 ) , spiral - bar resonance overlap ( minchev & famaey 2010 ) , and the effect of minor mergers on the disk ( quillen et al . one of these could be responsible for migrating the hyades stream from a distance larger than what is possible with a spiral structure model only . the `` migrating '' stream could remain coherent for a long enough time before dispersing , but more work is needed to investigate this possibility . the drawback would then be that the hyades and sirius overdensities in the @xmath4-plane would be unrelated and would need two separate explanations . from an analysis of the li , na , mg , zr , ba , la , ce , nd and eu abundances of 21 stars from the hyades stream , we performed chemical tagging to identify stream stars possibly evaporated from the cluster . the first tagging method is based on a @xmath66 test comparing the abundances of fe , zr , ba , la and ce in stream stars with the cluster average value and internal scatter . it is convenient to make use of the ` cube - root transformation ' of the @xmath66 variable to convert it into the goodness - of - fit statistics @xmath37 , which behaves as a normal distribution with zero mean and unit variance . this method relies of course on an accurate evaluation of the cluster average value and internal scatter . this tagging method works best for chemical elements exhibiting a steep slope with metallicity . any metallicity difference between the stream and the cluster is then amplified by each element involved in the @xmath66 sum and exhibiting a steep metallicity trend . the absence of any such metallicity trend among na and mg ( at least for metallicities in the range -0.1 0.3 ) makes them useless for chemical tagging . a second method uses the tight li @xmath72 sequence observed among hyades stars , a result of the slow li destruction process at work in these stars . stars of the same @xmath72 but of a different age , as will be the case in general for stream stars , will not fall along the hyades sequence . this method offers the advantage that the range spanned by li abundances is very large ( of the order of 3 dex ) , but is restricted to stars in the narrow @xmath72 range 5000 6500 k. outside this range , the li abundance is either too small to be measurable or not sensitive to age any longer . based on these methods , only two stars from the stream ( hd 149028 and hd 162808 ) appear to have abundances within ` @xmath126 ' of the cluster ( i.e. , @xmath78 , and matching the li sequence ) ; they thus very likely evaporated from the cluster . our analysis thus convincingly demonstrates that a large fraction of stars in the hyades velocity box ( @xmath127% ) can not originate from the hyades cluster . interestingly , 2 evaporated candidates have been found , grossly in line with the conclusion of famaey et al . ( 2007 ) that about 15% of the hyades stream could still originate in the hyades cluster . it is then of high interest to clean the stream population from this evaporated population and to compare its properties with the local thin disk . it is found here that these stream stars are more metal - rich than the local thin disk , with an excess of the order of @xmath128 to 0.15 dex . this metallicity excess implies an origin for the stream at about 0.9 3.0 kpc inwards from the sun , adopting a galactic metallicity gradient of @xmath84 to @xmath114 dex / kpc . predictions from a 4:1 inner resonance of a 2-armed spiral pattern locate the origin of the hyades stream at a maximum of 1 kpc inwards from the solar radius . a more precise determination of the stream metallicity is thus needed to confirm or contradict this scenario . besides reproducing the hyades overdensity in the @xmath129 plane , this scenario however makes another appealing prediction , namely the existence of the sirius moving group exactly as observed . as a conclusion , the analysis of this small subsample of 21 stars from the hyades stream has already yielded some interesting insights on its nature , and it will be of prime importance to confirm these with much larger samples , e.g. in future analyses performed with multi - fibre spectrographs such as the high resolution multi - object spectrograph of the anglo - australian telescope ( barden et al . 2008 ) . if they could derive a more accurate metallicity for the stream cleaned from the evaporated population , these large surveys could clearly provide a meaningful constraint on the distance at which the hyades stream originates , and on the metallicity gradient in the disk , thus shedding light on the dynamical process responsible for the formation of the stream . this , together with the mapping of large - scale non - axisymmetric motions ( see , e.g. , siebert et al . 2011 ) will help better constraining the non - axisymmetry of the galactic potential and its influence on the dynamical and chemical evolution of the disk . the hermes / mercator project is a collaboration between the k.u . leuven , the universit libre de bruxelles and the royal observatory of belgium with contributions from the observatoire de genve ( switzerland ) and the thringer landessternwarte tautenburg ( germany ) . it is funded by the fund for scientific research of flanders ( fwo ) under the grant g.0472.04 , from the research council of k.u . leuven under grant gst - b4443 , from the fonds national de la recherche scientifique ( f.r.s .- fnrs ) under contracts iisn4.4506.05 and frfc 2.4533.09 , and financial support from lotto ( 2004 ) assigned to the royal observatory of belgium . l.p . acknowledges financial support from the _ dpartement des relations internationales _ of the universit libre de bruxelles and thanks the f.r.s - fnrs ( belgium ) for granting her a _ bourse de sjour scientifique_. support has also been provided by the _ actions de recherche concertes ( arc ) _ ( heavy elements in the universe : stellar evolution , nucleosynthesis and abundance determinations ) from the _ direction gnrale de lenseignement non obligatoire et de la recherche scientifique - direction de la recherche scientifique - communaut franaise de belgique_. b.f . acknowledges support from the cnrs ( france ) and avh foundation ( germany ) . g.g . is a postdoctoral researcher of the fwo - vlaanderen ( belgium ) . is fnrs research associate . authors contributions : lp , tm , sve , aj and csn have conducted the spectroscopic analysis ; bf , aj , i m , as , jrdl and ob have conducted the dynamical and statistical analyses ; csi , gg , td and ep have conducted the observations ; aj , hvw , cw , gr , sp , wp , hh , yf , and ld are builders of the hermes / mercator spectrograph . allende prieto , c. 2010 , in : cunha k. , spite m. & barbuy b. , chemical abundances in the universe : connecting first stars to planets ( iau symp . 265 ) , 304 , cambridge : cambridge univ . press alvarez , r. , & plez , b. 1998 , a&a , 330 , 1109 amaral , l. h. , & lpine , j. r. d. 1997 , mnras , 286 , 885 antoja , t. , valenzuela , o. , pichardo , b. , moreno , e. , figueras , f. , & fernndez , d. 2009 , apj , 700 , l78 asplund , m. , grevesse , n. , & sauval , a. j. 2005 , cosmic abundances as records of stellar evolution and nucleosynthesis , eds . t.g . barnes iii , f. n. bash , asp conf . 336 ( san francisco ) , 25 barden , s. c. , bland - hawthorn , j. , churilov , v. , et al . 2008 . in : mclean , i. s. , & casali , ground - based and airborne instrumentation for astronomy ii , proc . spie , 7014 bensby , t. , feltzing , s. , & lundstrm , i. 2003 , a&a , 410 , 527 bensby , t. , feltzing , s. , lundstrm , i. , & ilyin , i. 2005 , a&a , 433 , 185 bensby , t. , oey , m. s. , feltzing , s. , gustafsson , b. 2007 , apj , 655 , l89 bland - hawthorn , j. , krumholz , m. r. , & freeman , k. 2010 , apj , 713 , 166 boesgaard , a. m. , & king , j. r. 2002 , apj , 565 , 587 bovy , j. , & hogg , d. w. 2010 , apj , 717 , 617 cayrel de strobel , g. , soubiran , c. , friel , e. d. , ralite , n. , & francois , p. 1997 , a&as , 124 , 299 cowan , j. j. , sneden , c. , burles , s. , et al . 2002 , apj , 572 , 861 de bruijne , j. h. j. 1999 , mnras , 306 , 381 de bruijne , j. h. j. , hoogerwerf , r. , & de zeeuw , p. t. 2001 , a&a , 367 , 111 dehnen , w. 1998 , aj , 115 , 2384 deliyannis , c. p. 2000 . in : r. pallavicini , g. micela , & s. sciortino ( eds . ) stellar clusters and associations : convection , rotation , and dynamos , asp conference ser . 198 , 235 de silva , g. m. , sneden , c. , paulson , d. b. , asplund , m. , bland - hawthorn , j. , bessell , m. s. , & freeman , k. c. 2006 , aj 131 , 455 de silva , g. m. , freeman , k. c. , bland - hawthorn , j. , asplund , m. , bessell , m. s. 2007 , aj , 133 , 694 de silva , g. m. , freeman , k. c. , & bland - hawthorn , j. 2009 , pasa , 26 , 11 den hartog , e. a. , lawler , j. e. , sneden , c. , & cowan , j. j. 2003 , apjs , 148 , 543 drilling , j. s. & landolt , a .u . , 1999 , in : allen s astrophysical quantities , cox a.n . ( ed . ) , 4th edition , springer , new york eggen , o. j. , 1958 , mnras , 118 , 65 eggen , o. j. , 1978 , apj , 222 , 203 englmaier , p. , pohl , m. , & bissantz , n. 2008 , in tumbling , twisting , and winding galaxies : pattern speeds along the hubble sequence , e. m. corsini and v. p. debattista ( eds . ) , memorie della societa astronomica italiana ( arxiv0812.3491e ) famaey , b. , jorissen , a. , luri , x. , mayor , m. , udry , s. , dejonghe , h. , & turon , c. 2005 , a&a , 430 , 165 famaey , b. , pont , f. , luri , x. , udry , s. , mayor , m. , & jorissen , a. 2007 , a&a 461 , 957 famaey , b. , siebert , a. , & jorissen , a. 2008 , a&a , 483 , 453 feltzing , s. , & holmberg , j. 2000 , a&a , 357 , 153 ford , e. b. , rasio , f. a. , & sills , a. 1999 , apj , 514 , 411 freeman , k. & bland - hawthorn , j. 2002 , ara&a , 40 , 487 fux , r. 2001 , a&a , 373 , 511 gonzalez , g. 2006 , pasp , 118 , 1494 grenon , m. 2000 , in : t. l. evans ( ed . ) luminosity calibrations with hipparcos , highlights of astronomy vol . 12 , joint discussion 13 of the xxiiiird general assembly of the iau grevesse , n. , & sauval , a. j. 1998 , space science rev . , 85 , 161 grevesse , n. , & sauval , a. j. 2000 , in : o. manuel ( ed . ) origin of elements in the solar system , implications of post-1957 observations , kluwer ( dordrecht ) , 261 gustafsson , b. , edvardsson , b. , eriksson , k. , jrgensen , u. g. , nordlund , . , & plez , b. 2008 , a&a , 486 , 951 hobbs , l. m. , thorburn , j. a. , & rebull , l. m. 1999 , apj , 523 , 797 holmberg , j. , nordstrm , b. , & andersen , j. 2007 , a&a , 475 , 519 holmberg , j. , nordstrm , b. , & andersen j. 2009 , a&a , 501 , 941 hoogerwerf , r. , & aguilar , l. a. 1999 , mnras , 306 , 394 israelian , g. , delgado mena , e. , santos , n. c. , sousa , s. g. , mayor , m. , udry , s. , domnguez cerdea , c. , rebolo , r. , & randich , s. 2009 , nature , 462 , 189 , j. e. , bonvallet , g. , & sneden , c. 2001a , apj , 556 , 452 lawler , j. e. , wickliffe , m. e. , den hartog , e. a. , & sneden , c. 2001b , apj , 563 , 1075 lawler , j. e. , sneden , c. , cowan , j. j. , ivans , i. i. , & den hartog , e. a. 2009 , apjs , 182 , 51 , t. 2006 , phd thesis , observatoire de paris , france minchev , i. & quillen , a. c. , 2007 , mnras , 377 , 1163 minchev , i. , boily , c. , siebert , a. , & bienayme , o. 2010 , mnras , 407 , 2122 minchev , i. , & famaey , b.2010 , apj , 722 , 112 murray , n. , chaboyer , b. , arras , p. , hansen , b. , & noyes , r. w. 2001 , apj , 555 , 801 neckel , h. 1999 , sol . 184 , 421 nordstrm , b. , mayor , m. , andersen , j. et al . 2004 , a&a , 418 , 989 paulson , d. b. , saar , s. h. , cochran , w. d. , & hatzes , a. p. 2002 , aj , 124 , 572 paulson , d.b . , sneden , c. , & cochran , w.d . 2003 , aj , 125 , 3185 perryman , m. a. c. , brown , a. g. a. , lebreton , y. , gomez , a. , turon , c. , cayrel de strobel , g. , mermilliod , j. c. , robichon , n. , kovalevsky , j. , & crifo , f. 1998 , a&a , 331 , 81 quillen , a. c. 2003 , aj , 125 , 785 quillen , a. c. , & minchev , i. 2005 , aj , 130 , 576 quillen , a. c. , minchev , i. , bland - hawthorn , j. , & haywood , m. 2009 , mnras , 397 , 1599 raskin , g. , van winckel , h. , hensberge , h. , et al . 2011 , a&a , 526 , a69 reddy , b. e. , tomkin , j. , lambert , d. l. , & allende prieto , c. 2003 , mnras , 340 , 304 reddy , b. e. , lambert , d. l. , & allende prieto , c. 2006 , mnras 367 , 1329 schnrich , r. , binney , j. , & dehnen , w. 2010 , mnras , 403 , 1829 sellwood , j. & binney j. , 2002 , mnras , 336 , 785 sellwood j. , 2010 , mnras , 409 , 145 sestito , p. , & randich , s. 2005 , a&a , 442 , 615 shi , j. r. , gehren , t. , & zhao , g. 2004 , a&a , 423 , 683 siebert , a. , famaey , b. , minchev , i. , et al . 2011 , arxiv:1011.4092 siess , l. , & livio , m. 1999 , mnras , 308 , 1133 , a. & ord , j. k. 1994 , kendall s advanced theory of statistics . vol.1 : distribution theory , ed . stuart , a. & ord , j. k. woolley , r. , 1961 , the observatory , 81 , 203 zhang , z. g. , svanberg , s. , jiang , z. , palmeri , p. , quinet , p. , quinet , p. , bimont , e. , & bimont , e. 2001 , phys . , 63 , 122 zhao , g. , & gehren , t. 2000 , a&a , 362 , 1077

we examine the star formation history and stellar feedback effects of dwarf galaxies under the influence of extragalactic ultraviolet radiation . previous work has indicated that the background uv flux can easily ionize the gas within typical dwarf galaxies , delaying or even preventing cooling and star formation within them . many dwarf galaxies within the local group are , however , observed to contain multiple generations of stars , the oldest of which formed in the early epochs of cosmic evolution , when the background uv flux was intense . in order to address this paradox , we consider the dynamical evolution of gas in dwarf galaxies using a one - dimensional , spherically symmetric , lagrangian numerical scheme to compute the effects of radiative transfer and photoionization . we include a physically - motivated star formation recipe and consider the effects of feedback . this scheme allows us to follow the history of the gas and of star formation within dwarf galaxies , as influenced by both external and internal uv radiation . our results indicate that star formation in the severe environment of dwarf galaxies is a difficult and inefficient process . in potentials with total mass less than a few , and velocity dispersion less than a few km s , residual gas is efficiently photoionized by cosmic background uv radiation . since the density scale height of the gas within these galaxies is comparable to their size , gas may be tidally removed from them , leaving behind starless residual dark matter clumps . for intermediate mass systems , such as the dsphs around the galaxy , star formation can proceed with in early cosmic epochs despite the intense background uv flux . triggering processes such as merger events , collisions , and tidal disturbance can lead to density enhancements , reducing the recombination timescale , allowing gas to cool and star formation to proceed . however , the star formation and gas retention efficiency may vary widely in galaxies with similar dark matter potentials , because they depend on many factors , such as the baryonic fraction , external perturbation , imf , and background uv intensity . we suggest that the presence of very old stars in these dwarf galaxies indicates that their initial baryonic to dark matter content was comparable to the cosmic value . this constraint suggests that the initial density fluctuation of baryonic matter may be correlated with that of the dark matter . for the more massive dwarf elliptical galaxies , the star formation efficiency and gas retention rate is much higher . their mass to light ratio is regulated by star formation feedback , and is expected to be nearly independent of their absolute luminosity . the results of our theoretical models reproduce the observed correlation .

astro-ph0312216

according to the popular cosmological models dominated by cold dark matter ( cdm ) , the formation of large galaxies occurs in a hierarchical fashion , through the coalescence of smaller systems ( blumenthal et al . 1984 ; navarro , frenk , & white 1997 ; klypin , nolthenius , & primack 1997 ) . during mergers , the dissipationless dark matter components combine to form the extended dark halos of larger galaxies . the gas of the dwarf systems , however , dissipates its kinetic energy during mergers , decouples from the dark matter , and falls either into the center or onto the equatorial plane , forming the bulge and disk . dwarf galaxies observed today , such as the dwarf spheroidal ( dsph ) satellites of the milky way , represent surviving members of the original population of building blocks from which the milky way was formed . along with the success of this picture , however , has come the realization of new problems , both in the theory and in comparisons with observations . three problems are of particular interest here . one difficulty is the so - called `` over - cooling problem . '' in the cdm scenario , the magnitude of linear density fluctuations , @xmath3 is a decreasing function of the mass scale , @xmath4 . such a fluctuation spectrum causes low mass systems to become nonlinear and virialize at earlier epochs , and within a more dense background , than larger systems . cooling is very efficient at early epochs , due both to the generally higher densities at high red redshift as well as the rapid increase in the rate of inverse compton cooling with redshift . the early growth of smaller systems , coupled with the more efficient cooling at early epochs , leads to the expectation of rapid condensation of small , dense gas clouds within the dark - matter halos of dwarf galaxies . but , unless star formation can be suppressed , most of the gaseous ordinary matter would be converted into stars well before these galactic building blocks can be assembled into present - day large galaxies such as the milky way ( white & rees 1978 ) . a second problem is the deficit of observed satellite dwarf galaxies relative to theoretical expectations . observationally , the cdm scenario naturally leads to a `` bottom - up '' hierarchical galaxy formation picture that is qualitatively consistent with dwarf galaxies having halos with higher phase - space densities than those around larger normal galaxies . high - resolution numerical simulations show , however , that dwarf galaxies with higher internal densities retain the integrity of their cores while their loosely bound halos merge with each other during the formation of larger galaxies ( moore et al . 1999 ; klypin et al . consequently , the number density of surviving satellite dwarf galaxies expected from the cdm scenario is much larger than that observed in the local group . early bursts of star formation could lead to the preservation of ordinary matter in a gaseous form , followed by its removal from the dark matter potential . such a process may account for the deficit of detectable satellite galaxies . the third problem involves the building up of large galaxies from the dwarf systems . under the action of dynamical friction from the surrounding sea of dark matter , the dwarf galaxy building blocks lose energy and orbital angular momentum before they eventually merge into larger disk galaxies . if the gas within the dwarf systems cools and contracts within their dark matter potentials , the mergers are delayed , leading to the formation of disk systems which are much smaller and denser than observed ( e.g. , navarro & benz 1991 ; navarro & white 1994 ) . the separation of gas from the dark matter potential of dwarf galaxies is one possible mechanism for the ordinary matter to retain sufficient angular momentum to form normal - size disk galaxies ( weil , eke & efstathiou 1998 ; navarro & steinmetz 1997 ; sommer - larsen , gelato & vedel 1999 ) . resolution of the above problems may require that the conversion of gas into stars within the dwarf systems be inefficient . recent theoretical investigations suggest that the conversion of gas into stars in dwarf galaxies may be suppressed by the background uv radiation at large redshifts ( weinberg , hernquist & katz 1997 ; kepner , babul & spergel 1997 ) . integrating the full equations of radiative transfer , heating , cooling and non - equilibrium chemistry for nine species : @xmath5 , @xmath5 , @xmath6 , @xmath7 , @xmath8 , @xmath9 , @xmath10 , @xmath11 , @xmath12 , @xmath13 , kepner et al . ( 1997 ) computed the quasi - hydrostatic equilibrium states of gas in spherically symmetric dark matter potentials roughly corresponding to those of dwarf galaxies , and found that a typical background uv radiation field can easily delay cooling and collapse of gas in halos corresponding to @xmath14 cdm perturbations with circular velocities less than @xmath15 . those results , however , introduce an inconsistency with other observed properties of low - mass dwarf galaxies . the color - magnitude diagrams of most dsphs indicate that star formation began within them at an early epoch . indeed , ursa minor and draco , whose stellar velocity dispersions are only a few @xmath16 , and whose total masses are less than @xmath17 , are composed primarily of very old stars . in some dsphs , such as carina and fornax , the coexistence of multiple generations of young and intermediate - age stars ( grebel 1997 ; grebel & stetson 1998 ) provides evidence of protracted and recent star formation episodes . this is again difficult to account for , because little atomic or molecular gas is found in them today ( knapp , kerr & bowers 1978 ; mould et al . 1990 ; bowen et al 1997 ; carignan et al . 1998 ) , though some gas has been detected around some more distant dsphs , such as sculptor . while it is natural to expect that either internal star formation or external ram - pressure stripping might easily remove any residual gas from the shallow potentials of dsphs ( noriega - crespo et al . 1989 ; mori & burkert 2000 ) , it is difficult to reconcile the lack of either cool or warm gas with the presence of young stars . a more complete picture of the early evolution of dwarf galaxies , and of their role in the formation of massive systems , therefore requires a more thorough understanding of the various physical processes , both external and internal , that act to trigger and regulate star formation within them . in this paper , we present an investigation of the hydrodynamic evolution and star formation history of dwarf galaxies . we construct a code which integrates a realistic star formation recipe , feedback , and radiative transfer in a 1d spherically symmetric lagrangian scheme . the code allows us to follow the evolution of both the state of the gas and the star formation of dwarf galaxies subjected to different background uv fluxes and different external perturbations . in examining the parameter space , we seek answers to the following questions : 1 ) . is star - formation in dwarf galaxies triggered by a decrease in the background uv radiation , or by external perturbations ? 2 ) . what is the star formation efficiency , _ i.e. _ the fraction of the initial mass of gas that is converted into stars ? 3 ) . what is the mechanism for triggering multiple bursts of star - formation , such as are observed in some dsphs ? in 2 of this paper , we describe the dwarf galaxy model , the star formation recipe , feedback , and the strmgren shell model for radiative transfer and photoionization . we combine all of these into a 1d spherically symmetric lagrangian hydrodynamical scheme . in 3 we discuss the results of our simulations , and in 4 we discuss the implications of our results . a wide range of physical processes and physical scales are associated with galaxy and star formation . including all of them would be a formidable task , beyond the capabilities of even the most powerful current generation of computers . in this work , we simplify the problem by examining the time evolution of a spherically symmetric cloud , and combine models for the most important processes , including radiative transfer , photoionization , star formation , and feedback . the problem therefore becomes tractable , and the results which we obtain may be included in larger - scale simulations . we simulate the evolution of the gas within a fixed halo potential . similar to the approach adopted by kepner et al . ( 1997 ) , we choose a form for the potential found by burkert ( 1995 ) , which has been shown to be a good fit to many dwarf systems . the corresponding dark matter density distribution is given by @xmath18 where @xmath19 is the central density . the dark matter density tends to @xmath20 for @xmath21 , and @xmath22 for @xmath23 . the rotational velocity , and mass at @xmath24 are given by @xmath25 @xmath26 in the numerical simulations presented below , the above mass distribution is arbitrarily assumed to extend out to a halo radius @xmath27 and so the total mass of the dark matter halo is @xmath28 . the dark matter halo contains ordinary matter in the forms of stars and gas . the evolution of the gas density , @xmath29 is described by the continuity equation , @xmath30 the sink term @xmath31 , due to star formation , denotes the rate by which the gas becomes gravitationally unstable ( see equation [ eq : rhos ] ) . in the above expression , @xmath32 is the material derivative and @xmath33 is the gas velocity . the momentum equation for the gas is @xmath34 where the force @xmath35 is due to the gradient of the gravitational potential @xmath36 , @xmath37 is the gas pressure , and we introduce a kinetic pressure , @xmath38 to represent the momentum input due to stellar wind and supernovae of massive stars ( see equation [ eq : pk ] ) . massive stars emit intense uv radiation , providing another mechanism for feedback . in addition , observations of ly@xmath39 clouds along quasar lines - of - sight suggest that , at high redshifts , the universe is permeated by a metagalactic uv flux . at @xmath40 , this background is estimated to have a strength @xmath41 at the lyman limit ( bechtold et al 1987 ) . both of these energy sources are modelled in the energy equation of the gas , @xmath42 where @xmath43 is the specific internal energy , @xmath44 , and @xmath45 and @xmath46 represent radiative heating and cooling terms . in order to consider the effect of feedback , the formation rate of massive stars must be specified . here , we introduce a simple prescription for massive star formation . in general , gas is destabilized by its collective self - gravity and stablized by internal pressure and external tidal stress . when the residual gas within a galaxy is fully ionized , it attains hydrostatic equilibrium , with the jeans mass @xmath47 of the system comparable to or larger than the total mass of gas . but , if the gas is able to cool well below the ionization temperature , @xmath47 decreases to masses comparable to high mass stars . the density of the residual gas also increases as it contracts to a new hydrostatic equilibrium within the dark matter potential . in those regions of the dwarf galaxy where @xmath48 , with @xmath49 the tidal field of the potential is insufficient to prevent the collapse of gas within the jeans radius @xmath50 . in the absence of a strong magnetic field , the gas collapses to form stellar mass objects when these conditions are satisfied . ionizing photons and stellar winds from young , massive stars act to raise the jeans mass of surrounding gas , inhibiting further star formation . the collapse of the gas cloud does not , however , lead to immediate star formation and feedback . a minimum time delay between the instant when the gas becomes gravitationally unstable and the onset of main sequence evolution for the resulting generation of stars is the dynamical free - fall time scale @xmath51 . if the collapsing gas has angular momentum , protostellar disks would form , through which gas is accreted onto the growing protostellar objects on a viscous time scale . after the formation of the protostars , an additional time interval , @xmath52 , must elapse before the emission of intense uv radiation . for most of our models , we choose @xmath53 myr . but in some delayed star formation models , we choose @xmath54 either 5 or 50 myr , to represent delays due to disk accretion ( yorke & sonnhalter 2002 ) , merger processes ( murray & lin 1996 , bonnel _ et al . _ 1998 ) , or sequential star formation ( elmegreen & lada 1977 ) . thus , changes in the stellar and gas mass densities ( @xmath55 and @xmath29 respectively ) are related via a time - delay , such that @xmath56 and , @xmath57 where @xmath58 refers to the local free - fall timescale at the epoch when the collapse begins , i.e. @xmath59 . we adopt a form for the star formation efficiency , @xmath60 , of @xmath61 \left[1+\left(\frac{r_j}{r_r}\right)^{\lambda_r}\right]}\ ] ] where @xmath62 is the critical temperature for the onset of @xmath63 star formation , and @xmath64 is the roche radius of the gas in the potential of the dwarf galaxy . the three factors in @xmath60 act to decrease the star formation efficiency in situations where it is expected to be low . when the local free fall time is longer than the timescale for the onset of main sequence evolution , then it would be expected that , because of the spread in star formation times , the first stars to form would inhibit further star formation . the efficiency is reduced for high stellar masses , due to the fact that the dynamical timescale for the evolution of the gas exceeds the evolution time of massive stars , in which case the first stars to form would re - ionize the surrounding gas before it , too , could collapse to form stars . finally , star formation is also expected to be inefficient whenever the roche radius of a jeans unstable region exceeds its tidal radius within the dwarf galaxy . we set the star formation quenching factors @xmath65 , such that star formation is rapidly shut off whenever one or more of the above criteria is met . unless otherwise stated , we set the maximum efficiency @xmath66 . in some cases , we choose @xmath67 , representing a conversion of at most ten percent of the total residual gas into stars within a free fall time . accompanying each massive progenitor star , there are many coeval lower - mass stars . thus , the corresponding formation rate , per unit volume , of massive uv and wind producing stars is given by @xmath68 the total mass of stars formed per uv , wind , and supernova producing star , @xmath69 , can be estimated from a scalo initial mass function ( imf ) to be 150 @xmath70 ( scalo 1986 ; miller & scalo 1979 ) . for more general initial mass functions , we set @xmath71 where @xmath72 for a steeper initial mass function with a relatively small cut off in the stellar mass function at the high end . unless otherwise stated , @xmath73 is chosen for most of our models . in order to explore the effects of variations , we examine some models in which @xmath74 . the formation of massive stars strongly influences the environment . the massive stars emit uv radiation with a well - determined flux ( see e.g. osterbrock 1989 ) . when averaged over a miller - scalo initial mass function , the rate of ionizing photon emission per 10 @xmath70 star is @xmath75 s@xmath1 . photoionization by uv radiation from massive stars can be included in the energy equation of the gas , and in determining its ionization state ( see below ) . stars are assumed to emit uv photons at an average rate of @xmath76 for a time equal to the average lifetime of a massive star , @xmath77 myr . after that time , they are taken to be `` dead stars , '' contributing to the overall mass of the system , but providing no feedback to the gas . after a brief stage of adiabatic expansion away from the massive stars , the main effect of the stellar winds and supernovae is to increase the momentum of the surrounding gas . we approximate this feedback process as a a `` kinetic pressure '' term in equation ( [ eq : momentum ] ) , defined for each given volume ( zone ) of the model as @xmath78 where the surface area at the center of the zone is @xmath79 . the first term in equation [ eq : pk ] is the momentum flux due to supernovae , and the second is that due to stellar winds . in the above equation , @xmath80 is the momentum of a single supernova explosion ( we adopt @xmath81 g cm s@xmath1 ) , @xmath82 is the total supernova rate ( number of sn per unit time ) within the volume of the zone . the value of @xmath82 is obtained from the the supernova rate per unit volume , which itself is determined by the star formation rate at the earlier time @xmath83 , i.e. @xmath84 the total number of massive stars on the main sequence within the volume of the zone is @xmath85 , while the wind mass loss rate of a typical massive star and the typical wind speed are @xmath86 y@xmath1 ) and @xmath87 km s@xmath1 ) , respectively . with these prescriptions for the feedback process , we use a one - dimensional lagrangian hydrodynamic model for the evolution of the baryonic gas component within the fixed external gravitational potential set by the dark matter ( richtmyer & morton 1995 ) . as indicated above , uv radiation from both massive stars and the background can lead to photoionization and heating of the residual gas . a comprehensive treatment of the full 3d radiative transfer requires the determination of the contribution to the flux at every point from every other point along all paths for each wavelength , which is at the minimum a 6d problem . however , in most instances , symmetries can be introduced , resulting in a more tractable situation . the greatest simplification occurs when the gas can be assumed to be optically thin throughout . this approximation is sufficient in the majority of cosmological situations ( katz , weinberg & hernquist 1996 ; navarro & steinmetz 1996 ; anninos , norman & clarke 1994 ) and only breaks down in the cores of halos that have undergone sufficient cooling , a situation that is usually made intractable by the complexities of star formation . the next simplest geometry is that of a slab ( or a sphere under the assumption of a radially perpendicular radiation field ) , which reduces the flow to an intrinsically 2d problem . this approach is the most common in radiative transfer , and has been used to address similar situations to those considered here ( hainman , rees & loeb 1996 ; haiman , thoul , & loeb 1996 ) . in this paper , we adopt a similar approach for spherical geometry . we relax the assumption of a radially perpendicular radiation field by treating radiative transfer using the strmgren shell model . similar to the original strmgrem model , the ionized `` h ii '' shell is assumed to be surrounded isotropically by a neutral @xmath88 region ( figure [ fig : strom ] ) . the rate of entry of ionizing photons into the ionized volume balances the total number of recombinations to excited levels within the volume . in our implementation of the strmgrem shell model , we use a discretized 1d lagrangian hydrodynamic scheme . we define @xmath89 to be the number of photons required to ionize zone @xmath90 , with @xmath91 being the volume of the zone , @xmath92 being the number density of hydrogen , and @xmath93 being the case b recombination coefficient ( osterbrock 1989 ) . the total number of uv photons emitted by the massive stars in the zone is @xmath94 , while @xmath95 is the rate of uv photons entering into the zone from larger radii , and @xmath96 is the rate of uv photons entering into the zone from smaller radii ( figure [ fig : geom ] ) . in the innermost zone , there is only @xmath97 and @xmath98 . the total rate of ionizing photons either produced within or entering into the zone is @xmath99 . if @xmath100 , then the zone is only partially ionized , and the ionization fraction of the gas is given by @xmath101 if @xmath102 then @xmath103 , and any photons in excess of @xmath104 give rise to @xmath105 and @xmath106 of adjacent zones , as follows ; @xmath107 and @xmath108 the term @xmath109 with @xmath110 and @xmath111 being , respectively , the inner and outer radii of any zone , is a factor that takes into consideration the spherical geometry of the model . in the limit of isotropic radiation , that geometry causes slightly more than half of the photons within a zone to exit in the outward direction . the case of either slab geometry or of a radially - directed radiation field can be represented by @xmath112 . the above form of @xmath113 can be arrived at by considering the optically thin limit with only inflowing uv photons . in that case , the uv intensity @xmath114 , and the flux of photons @xmath115 , thus @xmath116 . care is also taken to deal with the special cases when either @xmath106 or @xmath105 is negative . iterations by sweeping inwards and then outwards several times are taken to ensure convergence . for the innermost zone , @xmath117 , while for the outermost zone @xmath98 is given by the external radiation field . the external uv flux @xmath118 ergs s@xmath1 @xmath119 ster@xmath1 hz@xmath1 at the lyman limit at low red shift , and @xmath120 times more intense at redshift @xmath121 ( cf . haardt & madau 1996 ) , is sufficient to ionize most dwarf spheroidal galaxies throughout . for a less intense external flux , neutral residual gas is surrounded by ionized gas . the presence of metals would increase the critical @xmath122 needed for fully ionizing the dwarf galaxy , but , because we are considering the early evolution of low - metallicity systems , the modification is expected to be limited . the temperature of the photoionized gas is taken to be 15,000 k , appropriate for low - metallicities . we calculate the temperature for the @xmath88 region according to equation [ eq : energy ] . we adopt equilibrium cooling rates , including contributions from hydrogen , helium , and metals ( dalgarno & mccray 1972 ; fall & rees 1985 ) . because the dwarf galaxies being modeled have low metallicities , and we are modeling them at early epochs , the metallicity is taken to be low , [ fe / h]=-2 . in the absence of photoionization , such metal - poor gas gas may cool to a minimum temperature of @xmath123 k ( kang et al . 1990 ; lin & murray 1992 ) . using the above treatments for heating , cooling , and radiative transfer , we can compute the ionization fraction and the temperature of the gas as well as the attenuation of the ionizing photons . we note that the model described above follows the assumptions used in calculating static , `` initial '' strmgren spheres . these are formed extremely rapidly , on timescales of the order of the recombination time of the gas ( @xmath124 s ) . subsequently , the gas expands in response to the photoionization heating . to compute , on galactic scales , the evolution of individual strmgren spheres , as well as that of the ionization front propagating inwards from the external uv radiation , is a multidimensional problem , beyond current computational capabilities . our models do capture the expansion of the ionized gas within each zone in response to heating , and therefore account for its global effects upon feedback and star formation . the essential physics of the problem is therefore included in the models . our models also assume the gas to be uniformly distributed within each zone . clumpiness in the ism would increase the value of @xmath104 from that given above . clumpiness would result from either temperature variations ( a multi - phase medium ) or self - gravity . in the former case , the presence of a cold medium implies the lack of sufficient uv photons to completely ionize the galaxy , and that subsequent star formation shall rapidly lead to re - heating and ionization of the gas , ie . in our models , this corresponds to a dwarf galaxy undergoing star formation . if the gas is completely heated , the jeans mass is sufficiently large that self - gravity shall be unable to cause significant clumping . finally , our models do not include metallicity evolution of the gas , due to star formation . metallicity does not strongly affect the size of strmgren spheres , and so this shall not significantly affect our assumptions regarding ionization by external and internal uv sources . the primary affect of increasing metallicity is to enhance cooling in gas that has lost its heat source , potentially enhancing the star formation efficiency . this , too , is not likely to be a strong effect , because the timescales for star formation are dominated by the initial free - fall time of the gas , and the pre - main sequence evolution time of the stars . the main affect would , therefore , likely be in the stellar mass function . higher metallicity gas can cool to lower temperatures , possibly enhancing the formation of low mass stars , and reducing the effectiveness of internal feedback . such effects must be included in the uncertainties of our models , and we explore a wide range of parameters in order to examine their potential significance . the currently preferred ( e.g. de bernardis et al 2002 ) cosmological parameters are as follows : the hubble constant is @xmath125 the age of the universe is @xmath126 gyr ; the baryonic density parameter is @xmath127 ; and the dark matter density parameter is @xmath128 . these parameters are relevant to the issues of delayed star formation in dwarf galaxies . for instance , kepner et al . ( 1997 ) simulated halos in the range of @xmath129 , which , using their model , translate to : @xmath130 . in their models , @xmath131 , such that the baryonic fraction is : @xmath132 with the very small value of @xmath133 , dwarf galaxies at the low end of the mass range could not form stars even today because the recombination efficiency is insufficient to offset photoionization due to external background uv radiation . kepner et al . ( 1997 ) deduced such a small value for @xmath134 under the assumption that the ordinary matter is accreted onto the dark matter halo from an ionized intergalactic medium with temperature @xmath135 k. however , if the dwarf galaxies formed prior to the era of reionization or if the first generation stars are primarily formed in the progenitors of dwarf galaxies , then cool , neutral gas may be retained by their dark matter haloes with the cosmological value of @xmath136 . observation of dwarf galaxies in the local group find high baryonic fractions ( mateo 1998 ) . indeed , in some systems , such as ngc 147 , the observationally inferred estimate of @xmath134 approaches unity . such values indicate that these systems may have undergone complicated and highly varied evolutionary histories . it is possible that dwarf galaxies had larger values of @xmath134 in the past , and have since preferentially lost a fraction of their baryonic matter through star formation feedback ( dekel & silk 1986 ) and tidal stripping ( murray , dong , & lin 2003 ) . it is also possible that the observed dwarf galaxies are completely different from the dwarf galaxy building blocks in the cdm scenario . in view of these uncertainties , it is useful to consider a wider range of parameters than adopted by kepner et al . we find below that the magnitude of @xmath134 is a crucial factor in regulating the star formation efficiency in dwarf galaxies . in the analysis of kepner et al . ( 1997 ) , gas is assumed to be in quasi - hydrostatic equilibrium at all times . yet , both pre - star formation evolution and post - star formation feedback processes lead to flow and evolution of the residual gas mass that should be treated with a time - dependent dynamical calculation . in addition , external perturbations , such as tidal disturbances by nearby large galaxies or collisions with other gas - rich dwarf galaxies are also important in modifying the density and velocity of the gas . for some models , we approximate the possibility of a strong tidal compression ( or a mild collision ) with an initial velocity perturbation in the form of a hubble inflow @xmath137 when scaled with the halo radius and the characteristic velocity dispersion , the above expression is equivalent to @xmath138 . the main purpose of our calculation is to study the conditions under which old stars formed within dwarf galaxies . the age of these stars , 10 - 12 gyr , corresponds to high - red - shift epochs , when the external uv flux is intense . we approximate the frequency dependence of the external uv energy flux by @xmath139 for the isotropic external flux , the rate at which photons enter the dwarf galaxy ( @xmath140 ) , whose outer surface area is @xmath79 , is @xmath141 the flux of photons across the surface would involve integrating over @xmath142 where @xmath143 , and @xmath144 is the angle from the normal . the resulting value of @xmath145 would be smaller by a factor of two . this introduces some uncertainty into our results , but is smaller than the uncertainty in the magnitude of the uv background , and much smaller than the range in @xmath145 experienced by dwarf galaxies over their lifetimes . we adopt @xmath146 in all our calculations . ionizing photons are also generated by the massive stars formed within these galaxies . the uv production rate by massive stars in each zone ( @xmath140 ) is @xmath147 where @xmath148 is the number of massive stars of zone , and @xmath76 is the value of @xmath149 averaged over massive stars according to the miller - scalo imf ( @xmath150 s@xmath1 ) . the parameters of the models that we have examined are shown in table [ tab : galaxymodels1 ] . in the table are listed , for each model , the scale radius ( @xmath24 ) , the initial baryonic - to - dark matter ratio ( @xmath134 ) , the ratio of the background flux to its value at the current epoch , the upper limit upon the star formation efficiency ( @xmath151 ) , the falue of @xmath152 , the value of @xmath52 , and the fraction of gas converted into stars , @xmath153 . in the first series of three models , a1 , a2 , and a3 , we simulate the evolution of an unperturbed dwarf galaxy ( @xmath154 ) at the present epoch ( @xmath155 ) . the scale length of the potential @xmath156 kpc , such that @xmath157 @xmath70 , @xmath158 , and @xmath159 . these model parameters for the halo strcuture are comparable to those of leo i and other dsph ( mateo _ et al . _ 1998 ) . in order to demonstrate the dependence of the star formation efficiency on the ordinary matter content , we consider @xmath160 and @xmath161 . in these and all other models , the ordinary matter has the same initial spatial density distribution as the dark matter ( @xmath162 ) , and an initial temperature of @xmath163 k everywhere . in figure [ fig : r006 ] , we plot the fraction , @xmath153 , of ordinary mass that is converted into stars as a function of time . for the tenuous ordinary matter distribution in model a1 , the residual h is fully ionized , and no stars are formed . despite the low initial temperature , the residual gas is ionized and heated to @xmath164 k almost instantly . as the gas evolves toward a new equilibrium , most of it leaves the galaxy . for the residual gas , the recombination process is inefficient compared with the photoionization even for the present low flux of background ionizing photons . for models a2 and a3 , the outer regions of the galaxy ( beyond 0.6 and 1 kpc respectively ) , quickly become photoionized while the gas that is initially located near the central regions of the galaxy collapses to form stars with @xmath165 and @xmath166 . the total stellar content is @xmath167 so that the total mass of the stars formed in these two models are @xmath168 and @xmath169 , comparable to stellar content of the ursa minor and leo i dsph respectively . the results of these models indicate that the hydrogen gas has sufficiently large column density to become self - shielded against the external uv flux if the ionization fraction is low . but , in both models , the onset of star formation quickly leads to photoionization from stellar uv emission , and an outward expansion of the gas . in the shallow potential of low - velocity dwarf galaxies , a large fraction of the ionized gas escapes from the system eventually , the residual gas evolves toward new equilibria , with a density that is too low for effective recombination . consequently , an inflated , hot , and highly ionized state is attained , in which star formation is quenched ( figure [ fig : a2xd ] ) . the comparison between models a1 , a2 , and a3 indicates that unperturbed dwarf galaxies with very low ordinary matter contents are likely to remain fully ionized . nevertheless , a small population of stars may be formed within these galaxies if a critical amount of gas can become neutral . the conversion efficiency is a function of the baryonic mass . hydrogen may become neutral either due to initial conditions , by accretion of additional gas from the intergalactic medium , or it may be induced by cloud - cloud collisions ( see below ) . although mass loss can reduce @xmath170 to well below both its initial value and the cosmological @xmath171 , the amount of stars formed in these systems provide a lower limit upon its initial value , @xmath134 . although the potential in models a1 , a2 , and a3 is relatively small , most dwarf spheroidal galaxies have even shallower potentials and smaller velocity dispersions . in order to assess the efficiency of initial star formation in such systems , we consider models b1 , b2 , and b3 , in which @xmath24 is set to be 0.3 , such that @xmath172 km s@xmath1 , @xmath173 , and @xmath174 . the value of @xmath134 for these three models are 0.1 , 0.2 , and 0.5 , so that the total baryonic mass in model b3 equals that in model a2 . the external uv flux all three models is @xmath175 . in comparison with model a2 , model b3 is more compact with a higher @xmath176 and more rapid decline in the density distribution . although the actual baryonic mass contained in the initial strmgren radius in model b3 is smaller than that in model a2 , the gravitational potential for @xmath177 is shallower than that for @xmath178 . in both models , the gas was not in a hydrostatic equilibrium initially . because the gas collapses faster in the more massive model a2 than in model b3 , a slightly larger fraction of the gas was turned into stars during the initial formation episode . the subsequent evolution of the gas and stars in models b1 , b2 , and b3 are similar to those in models a2 and a3 . after an initial episode of star formation , the residual gas becomes fully ionized , a substantial amount of mass is lost , and star formation is quenched . the asymptotic values of @xmath153 for these models are plotted in figure [ fig : r003]a . the trend of @xmath153 increasing with @xmath134 generally holds . the magnitude of @xmath179 is @xmath180 and @xmath181 for models b2 and b3 respectively . thus , even with @xmath134 substantially larger than the cosmological @xmath171 , the mass of stars formed is substantially below the stellar contents in dsphs . in the above models , we adopt the background uv flux for low redshift . the old stellar population in dsphs was , however , formed at high redshifts , when the background uv flux was considerably more intense . in models c1 , c2 , and c3 ( figure [ fig : r003]b ) , we adopt @xmath182 , with @xmath183 , and , respectively , @xmath184 and @xmath185 . all other model parameters are similar to those in the b series of models . although @xmath186 in models c1 , c2 and c3 is @xmath187 larger than that in the previous models , the resulting @xmath153 and hence @xmath179 for models b3 and c3 are comparable to each other . the density at the strmgren surface is proportional to @xmath188 , and so the c - series of models are photoionized closer to the galaxy center . but the initial episode of star formation in both model b3 and c3 is quenched primarily by the feedback , rather than by photoionization from external uv radiation . the difference in @xmath153 is substantial , however , between the most tenuous models , b1 and c1 . the value of @xmath186 is sufficient to completely ionize the gas in model c1 prior to the onset of collapse . consequently , the star formation efficiency is essentially zero for model c1 . the models considered above have relatively shallow gravitational potentials , such that the residual gas can escape after it is ionized . we now consider a series of models with deeper gravitational potentials . in models d1 , d2 , d3 , d4 , and d5 , we adopt , respectively , @xmath177 , @xmath189 , @xmath190 , @xmath191 , and @xmath192 . we set @xmath193 for all five models . in each , we also use @xmath194 , such that models d1 and d2 are identical , respectively , to models b1 , and a2 . for models d3 , d4 , and d5 , respectively , @xmath195 , @xmath196 , and @xmath197 ; @xmath198 , @xmath199 , and @xmath200 ; and @xmath201 , @xmath202 and @xmath203 . the potentials of these models are deeper than those of the dsphs but comparable to those of typical dwarf elliptical and dwarf irregular galaxies . in figure [ fig : b01]a , we plot the evolution of @xmath153 for various models . these results clearly indicate that the residual gas is retained by the gravitational potential of the dwarf galaxies in models d4 and d5 . star formation in each of these cases is self - regulated . in model d4 , the mass of the stellar component reaches @xmath204 with a comparable amount of residual gas bound to the galaxy after 1 gy . thereafter , star formation continues at a self - regulated rate of @xmath205 yr@xmath1 . for the more massive model d5 , more than half of its original gas is converted into stars relatively early . after @xmath206 myr , the mass of the stars becomes @xmath207 with @xmath208 of residual gas . the residual gas is much more extended than originally , which slows down the star formation rate . the external uv flux in models d1 - 5 corresponds to the present - day value . in order to consider the conditions under which stars as old as 10 - 12 gyr formed , we consider models e1 - 5 ( figure [ fig : b01]b ) . these models are identical to models d1 - 5 , except that @xmath186 is increased by two orders of magnitude ( _ i.e. _ @xmath209 ) . with this enhanced uv flux , the gas in model e2 is ionized immediately , and no star formation occurs . by contrast , the lower uv flux present in model d1 allows for the initial formation of a few stars . the intense uv radiation increases the density needed for ionization equilibrium , and so the radius of the strmgren sphere is reduced . this effect is most evident in the marginal models d3 and e3 . photoionization due to the background is weak in the low @xmath186 model d3 . in that model , the residual gas undergoes rapid collapse , leading to high early levels of star formation and feedback , resulting in two early , strong ` bursts ' of star formation . the momentum injected to the gas by the star formation is sufficient to drive the gas outwards , such that it becomes too tenuous to recombine , and subsequent star formation is quenched . in contrast , the outer regions of model e3 are ionized . but the gravitational potential is sufficiently deep to retain the ionized gas . star formation is therefore able to proceed , and does so at a nearly constant rate , regulated by the supernova injection of momentum , and reduced relative the initial rate of model d3 due to the higher level of ionization . somewhat counterintuitively , therefore , model e3 by late times is able to convert a higher fraction of its gas into stars than is model d3 . in figure [ fig : snap ] , we show the density profile snapshots of model d3 ( solid lines ) and e3 ( dotted lines ) at @xmath210 and @xmath206 myr . evolution of the star formation rate for the more massive models e4 and e5 are nearly indistinguishable from those of d4 and d5 . this implies that the rich supply of gas in both cases limits the effects of the external uv radiation to the gas in the tenuous outer halo , well outside of the central , star - forming regions , which have much shorter recombination and dynamical timescales . in order to verify this tendency , we examine models f1-f4 , in all four of which @xmath211 . in models f1 and f2 , @xmath212 kpc , whereas it is 6 kpc for models f3 and f4 . the external @xmath213 for models f1 and f3 , whereas @xmath214 for models f2 and f4 . the evolution of @xmath153 for all four models is shown in figure [ fig : b001 ] . the comparison between models f1 and f2 ( as well as that between f3 and d4 ) indicates that the external flux does not significantly modify the star formation rate . the nearly steady star formation rate is self - regulated . the above results indicate that , in systems with @xmath215 kpc , the residual gas expands towards a new equilibrium after an initial epoch of star formation . consequently , mass is lost from the galaxy and the residual gas density becomes sufficiently depleted that the external background uv flux can severely suppress the star formation rate . the magnitude of @xmath153 may be enhanced if the feedback effect is weaker than in the standard models . such might be the case if @xmath151 is relatively small ( corresponding to an inefficient conversion of gas into stars ) , or if @xmath152 is relatively large ( corresponding to a suppressed production rate of the massive supernovae progenitor stars ) . in models g1-g4 , we consider some variations of model a3 ( which is also referred to as g0 for comparison purposes ) . in each , @xmath216 , @xmath217 , and @xmath218 kpc . in model g1 , we set @xmath67 and @xmath73 . in models g2 and g3 , we set @xmath219 and @xmath220 , while in model g4 , @xmath221 and @xmath222 . in models g0 - 2 , @xmath223 myr , but in models g3 and g4 , @xmath224 myr ( see fig . [ fig : param ] ) . the magnitude of @xmath153 for model g3 is almost an order of magnitude larger than that of model g0 ( a3 ) . in comparing between models g1 and g2 , we see that @xmath153 depends more sensitively on the value of @xmath151 than upon @xmath152 and @xmath52 . star formation in the more massive galaxies , with larger @xmath24 , discussed above is self - regulated by the feedback process . in the models presented so far , self - regulated star formation leads to quasi - equilibria , such that the star formation rate is essentially constant . for models d4 , d5 , e4 , and e5 there are small variations in the star formation rate during the early evolution of the galaxy . in these models , the star formation rate is relatively steady , because the local values of @xmath225 are much shorter than the average dynamical timescale of the star forming region , which itself is regulated by the feedback process . in contrast , for model d3 , with a more compact structure , the time delay can lead to episodic variations in the star formation rate . however , the supply of baryonic mass is limited , and after one or two episodes of star formation the density is reduced sufficiently that the residual gas remains ionized and star formation is quenched . some dwarf galaxies show evidence for episodic star formation histories . in order for episodic star formation to take place , three conditions must be satisfied : 1 ) the galaxy must have sufficient baryonic mass to become self - shielded against the background uv radiation , 2 ) the collapse and feedback switch - on delay times must be longer than the global dynamical timescale , in order that a substantial amount of baryonic matter may accumulate prior to the onset of each episode of star formation , and 3 ) the star formation efficiency must be modest , so that an adequate amount of residual gas may be preserved for the subsequent episodes of star formation . models h1-h5 , have the same values of @xmath226 kpc@xmath227 , and @xmath186 as model d4 ( which we aso refer to as the fiducial model h0 ) . in models h1 and h2 , we set @xmath224 my and 50 my respectively . in both cases , the delay in the switch on of the feedback effects allows more material to accumulate into the central regions of the galaxy . the delay enhances the eventual rate of star formation leading to ferocious feedback , driving all the residual gas out of the galaxies . in models h3 and h4 , we set , respectively , @xmath67 and @xmath228 , motivated to satisfy requirement 3 ) above for episodic star formation . in figure [ fig : r4par ] , we see that these models indeed lead to episodic variations in the star formation rate . finally , in model h5 , we also include the effect of a bottom - heavy initial mass function by setting @xmath222 . in this case , a large fraction of the ordinary matter is retained by the galaxies , and converted into stars . one difficulty associated with the results presented above is the low gas - to - star conversion efficiency for the lowest mass dwarf galaxies . in addition to reducing the severity of the feedback effect , we now explore the possibility of compressions , such as would result from either tidal perturbations or collisions with other galaxies . the models which we consider are shown in table [ tab : galaxymodels2 ] . in that table are listed , for each model , the values of @xmath24 , @xmath134 , @xmath229 , @xmath230 , @xmath152 , @xmath52 , and @xmath153 . the value of @xmath231 indicates the strength of a linear compression imposed upon the models , as described in equation [ eqn : compeqn ] , and is normalized in the table to @xmath232 km s@xmath1 kpc@xmath1 . models i1-i4 have values of @xmath186 , @xmath24 , and @xmath134 identical to those of model a3 . in models i1-i4 , respectively , @xmath233 and 150 km s@xmath1 kpc@xmath1 ( @xmath234 , and @xmath235 ) . the compressional timescale across the galaxies , @xmath236 1 , 0.2 , 0.1 , and 0.02 gyr , respectively , for models i1 - 4 . for the region which contains most of the gas in models i3 and i4 , the compressional speed is supersonic and the density jump at the shock fronts induces large increases in the local star formation rate and efficiency . consequently , @xmath153 is substantially enhanced . the results are shown in figure [ fig : hubble ] , where it can clearly be seen that @xmath153 is an increasing function of @xmath231 . the enhancement is associated with the increased densities , and resulting enhanced recombination rates , which provide a better shield for the residual gas in the galaxy . we note that the compression is excessive for model i4 . such a large impact is more appropriate for cloud - cloud collisions , rather than tidal compression . these results indicate a significant fraction of the baryons in dwarf galaxies can be converted into stars if the galaxies are subjected to supersonic compressions . finally , we consider the effect of impact on relatively massive dwarf galaxies similar to models d3 and e3 ( which is refereed to as models k0 and k2 here ) . we compare the results of these models without external perturbation to those of models k1 and k3 with a modest external perturbation ( @xmath237 km s@xmath1 kpc@xmath1 ) . both k0 and k1 have an external uv flux of @xmath216 ; while both k2 and k3 @xmath238 . the results are shown in figure [ fig : hbblr2 ] . for models k1 and k3 , the compression speed is slightly slower than the sound speed of the ionized gas in the outer regions of the galaxies . nevertheless , it can confine the thermally expanding ionized shell . thus , the compressional speed not only leads to a density enhancement , but also retention of the gas , enabling the gas in model k1 to undergo protracted and self - regulated star formation . we also note that even with such a modest compressional speed , gas in these dwarf galaxies can recombine efficiently even in early cosmological epoches when the intensity of the uv flux is relatively high . the comparison between models k0 and k1 indicates that this compression can double the star formation efficiency . in this paper , we consider a series of one - dimensional numerical models to examine the efficiency of star formation in dwarf galaxies under the influence of local cooling , heating by uv photons from both the background radiation and hot stars , and the winds and supernovae associated with massive stars . radiative transfer is handled using a simple `` strmgren shell '' approximation . completely ionized gas is assumed to have temperature of 15,000 k , while heating and cooling are followed dynamically in gas that is not completely ionized . the effects of supernovae and winds from massive stars are included in the momentum equation of the gas . stars can form in gas that has cooled sufficiently for the jeans mass to be within the range of stellar masses , which is not tidally limited , and for which the free - fall timescale is short compared to the pre - main sequence evolution time of massive stars , @xmath52 . the rate of conversion of gas into stars is a fraction of the gas free - fall timescale . feedback from the massive stars which form is delayed by the free - fall time , plus @xmath52 . the most important parameters explored in the models are : 1 ) the depth of the galactic potential , 2 ) the baryonic fraction , 3 ) the background uv flux , 4 ) @xmath52 , 5 ) the imf , and 6 ) compressional perturbation ( see tables 1 and 2 ) . we initially confirm the results of kepner et al ( 1997 ) , that for a low mass system ( @xmath239 ) with baryonic fraction @xmath240 , even today s background uv flux is adequate to totally ionize the gas such that no stars can form . with increasing baryonic fraction , however , these systems eventually become self - shielded from the background uv , and the gas is able to cool and collapse , leading to star formation . for @xmath217 ( comparable to the ratio of @xmath171 today ) , a low mass system with @xmath241 kpc is able to convert @xmath242 gas ( @xmath243 ) into stars in about 2 billion years . feedback from star formation leads to rapid expansion of the gas within the shallow galactic potential . the lower recombination efficiency of the expanded gas allows it to be completely ionized by the external uv radiation , and star formation is terminated after the initial , weak burst . modest retention of gas is possible , if the imf of the stars formed in these system is biased towards the low - mass stars . in the presence of much stronger background uv radiation , appropriate to early epochs , star formation may still occur in small galaxies having @xmath244 . in such systems , self - shielding is sufficient to prevent ionization initially within the cores , even with the higher uv background . star formation then proceeds , governed solely by internal time scales and feedback , while the external radiation acts only at late times , to prevent further star formation . the requirement for a relatively large concentration of ordinary matter in these shallow potential is in contrast to the low-@xmath134 estimate made by kepner _ et al _ ( 1997 ) based on the assumption of bondi accretion into dark matter potentials . although the accretion process may be more effective when the pristine gas is cold before the epoch of reionization , all present - day dwarf galaxies must have acquired their ordinary matter content prior to the formation of massive stars in other nearby dwarf galaxies . the relatively short dynamical time scales within the dwarf galaxies suggests that such a precise timing may be difficult to achieve . alternatively , prior to the epoch of dwarf galaxy formation , density fluctuation of the baryonic matter may be well correlated and coupled to that of the dark matter . star formation is much more efficient in higher mass systems , due both to increased self - shielding and to the enhanced ability of the galactic potential to retain ionized gas . a ten - fold increase in the dark matter mass , from @xmath245 typical of dsphs , to @xmath246 typical of dwarf ellipticals ( des ) or dwarf irregulars ( dirrs ) , may lead to an order of magnitude increase in the gas - to - star conversion efficiency . in the most massive systems , the star formation rate is self - regulated by internal feedback , independent of the external uv flux . for marginal systems , with low external flux , star formation occurs in a rapid initial burst , followed by expansion and ionization by the external flux , similar to the low - mass systems . the presence of a higher external uv flux , however , lessens the initial star formation rate , reducing the subsequent feedback , and allowing star formation to proceed at a slow pace for an extended period . observational data suggest that in the satellite dsphs around the galaxy , the baryonic matter content varies widely , while the dark matter not only dominates but also has widely varying masses . but , in dwarf galaxies with mass more than @xmath247 , the total mass - to - light ratio has a nearly constant value @xmath248 2.0 ( see mateo _ the results of our calculation are consistent with this apparent dichotomy . in order to make a direct comparison with fig . 7 in mateo _ ( 1998 ) , we plot the total mass to light ratio @xmath249 within @xmath24 as a function of the absolute visual magnitude @xmath250 of the different dwarf galaxies in models d1 - 5 , a1 - 3 , and g0 - 5 . for very low - mass ( @xmath251 ) dark matter halo , star formation may be completely suppressed even if @xmath134 is 0.5 . in these low - mass systems , the pressure scale height of the photoionized gas is comparable to the galaxy s radius and in the vicinity of the galactic disks , and the tidal perturbation of nearby host galaxy would induce efficient mass loss ( murray _ et al . it is entirely possible that that the halo of the milky way may be populated by many dark matter halos of the ` missing dwarf galaxies ' . for the intermediate - mass ( @xmath252 ) dwarf galaxies , the gas retention efficiency depends on many factors such as the background uv flux , the imf , the baryonic to dark matter ratio etc . in figure [ fig : mlmv ] , the @xmath253-@xmath250 correlation for this mass range merely indicate that a transition mass for the gas retention efficiency . for the high - mass dwarf galaxies , the baryonic matter represents a significant fraction of the total mass and the constant @xmath249 is a reflection of the self - regulated star formation process . by contrast , in low mass systems , @xmath254 exceeds the dynamical time on which the gas evolves within the galactic potential . if sufficient gas is present to shield the core from external uv , then gas in the core may accumulate significantly before the onset of feedback , leading to strong bursts of star formation . the same is observed in larger systems , for which we have considered larger values of @xmath52 . if the initial burst of star formation causes the gas to expand , and its density to decrease to the point at which the external uv is able to ionize the galaxy completely , then star formation is terminated after a single burst . in somewhat more massive systems , gas in the core is still self - shielded after the massive stars formed in the initial burst fade away . the gas can then cool , contract , and trigger additional bursts of star formation , as is observed in some of our models . in all systems which undergo either one or multiple episodes of star formation , the feedback is eventually sufficiently strong as to drive the gas out of the system . episodic star formation may also result from reducing the efficiency of feedback , either by reducing the conversion efficiency of gas into stars , or the fraction of massive stars which form . in high mass models with such variations , we observe episodic star formation . within the environment of a galaxy cluster , or near to a massive galaxy , a dwarf system shall experience tidal forces , that shall vary over time , resulting in occasional , or even periodic compressions . if dwarf systems form in large numbers around more massive overdense regions , they shall also experience collisions . the merger process which builds larger galactic entities can also lead to collisions and disturbances . both mechanisms lead to enhancements in the density , and therefore the recombination efficiency of the gas within dwarf galaxies . in small systems at early epochs , this may allow the galaxy to form some stars , when it might otherwise be completely ionized by the external uv radiation . in more massive systems , the star formation rates are enhanced by as much as an order of magnitude relative to unperturbed systems for realistic compression speeds . perturbations such as considered here would be expected to be stronger in dense cluster environments , or near to massive galaxies . in the former , collisions would be more frequent , while in the latter , tidal variations due to elliptical orbits around the parent system are strongest . these are also environments where tidal stripping , ram pressure , and external uv radiation are all enhanced , which would be expected to act to reduce star formation efficiencies , and so it is unclear whether the effect of perturbations alone would lead to environmental dependences for stellar populations within dwarf galaxies . as indicated in earlier work , star formation in the lowest mass dwarf galaxies is difficult at early epochs , and the observed population of dsphs appear to be at the very edge of the ability of small galaxies to form stars . the difficulty is due to the strong uv background , which easily ionizes the galaxies throughout , preventing the gas from cooling and contracting within the weak gravitational potential of the dwarf system . star formation may still occur , even in the smallest systems , however , if the gas within them can become self - shielded from the external uv . such might be the case if they form with a relatively large baryonic fraction , if they accrete additional gas , or if they undergo compressions due to either tidal forces or collisions . such processes may help to explain the existence of old stellar populations in even the smallest dsphs . feedback from star formation also forms a natural means for avoiding the `` over - cooling '' problem among the dwarf progenitors of massive galaxies . in small systems , star formation tends to occur in bursts , the feedback from which either expels the gas completely from the galaxy , or drives it to large radii , making it vulnerable to either tidal or ram - pressure stripping . subsequent generations of star formation would , therefore , appear to be unlikely . the gas lost to a dwarf system , however , leaves at small speeds ( @xmath255 km s@xmath1 ) . it shall not , therefore , move to large distances from the dwarf as it orbits within the potential of either a massive galaxy or within a cluster , and may be re - accreted by the dwarf at apogalacticon , possibly leading to subsequent bursts of star formation . some small dwarf systems , therefore , might be expected to show periodic bursts of star formation , with a burst period determined by the orbital period of the dwarf . by contrast , in larger systems , where the feedback time is short compared to the dynamical time of the system , and where the potential is capable of retaining ionized gas , feedback tends to lead to steady , self - regulated star formation . the gas within the system does , however , expand under the influence of stellar feedback , and so may also be subject to stripping within galactic cluster environments ( murray , dong , & lin 2003 ) . the possibility of severe mass loss from dwarf galaxies following star formation indicates that the currently observed baryonic - to - dark matter ratios can provide only lower limits , possibly very weak limits , to the values present when the galaxies formed . our models indicate that the star formation efficiencies , and the ability of galaxies to retain gas following star formation depend very sensitively dependences upon the evolutionary histories of the galaxies , and upon the galactic potentials , which may account for the large dispersion in the stellar to dark - matter fraction observed today among the dwarf galaxies . we thank drs . a. burkert , and k - s . oh for useful conversations . we also thank the anonymous referee for valuable comments . this work was performed under the auspices of the u.s . department of energy by university of california , lawrence livermore national laboratory under contract w-7405-eng-48 . this work is partially supported by nasa through an astrophysical theory grant nag5 - 12151 . cclrcrcr a1 & 0.6 & 0.001 & 1 & 1.0 & 1 & 0.5 & 0.0 + a2 & 0.6 & 0.1 & 1 & 1.0 & 1 & 0.5 & 2.6 + a3 & 0.6 & 0.2 & 1 & 1.0 & 1 & 0.5 & 3.7 + b1 & 0.3 & 0.1 & 1 & 1.0 & 1 & 0.5 & 1.3 + b2 & 0.3 & 0.2 & 1 & 1.0 & 1 & 0.5 & 1.9 + b3 & 0.3 & 0.5 & 1 & 1.0 & 1 & 0.5 & 2.5 + c1 & 0.3 & 0.1 & 100 & 1.0 & 1 & 0.5 & 0.0 + c2 & 0.3 & 0.2 & 100 & 1.0 & 1 & 0.5 & 0.7 + c3 & 0.3 & 0.5 & 100 & 1.0 & 1 & 0.5 & 1.5 + d1 & 0.3 & 0.1 & 1 & 1.0 & 1 & 0.5 & 1.3 + d2 & 0.6 & 0.1 & 1 & 1.0 & 1 & 0.5 & 2.6 + d3 & 2.0 & 0.1 & 1 & 1.0 & 1 & 0.5 & 14.2 + d4 & 4.0 & 0.1 & 1 & 1.0 & 1 & 0.5 & 68.4 + d5 & 6.0 & 0.1 & 1 & 1.0 & 1 & 0.5 & 83.2 + e1 & 0.3 & 0.1 & 100 & 1.0 & 1 & 0.5 & 0.0 + e2 & 0.6 & 0.1 & 100 & 1.0 & 1 & 0.5 & 0.0 + e3 & 2.0 & 0.1 & 100 & 1.0 & 1 & 0.5 & 19.7 + e4 & 4.0 & 0.1 & 100 & 1.0 & 1 & 0.5 & 66.3 + e5 & 6.0 & 0.1 & 100 & 1.0 & 1 & 0.5 & 67.6 + f1 & 4.0 & 0.01 & 0 & 1.0 & 1 & 0.5 & 24.5 + f2 & 4.0 & 0.01 & 1 & 1.0 & 1 & 0.5 & 24.1 + f3 & 6.0 & 0.01 & 0 & 1.0 & 1 & 0.5 & 47.7 + f4 & 6.0 & 0.01 & 1 & 1.0 & 1 & 0.5 & 45.4 + g0 & 0.6 & 0.2 & 1 & 1.0 & 1 & 0.5 & 3.7 + g1 & 0.6 & 0.2 & 1 & 0.1 & 1 & 0.5 & 3.5 + g2 & 0.6 & 0.2 & 1 & 1.0 & 10 & 0.5 & 11.7 + g3 & 0.6 & 0.2 & 1 & 1.0 & 1 & 5.0 & 6.3 + g4 & 0.6 & 0.2 & 1 & 0.1 & 10 & 5.0 & 22.4 + h0 & 4.0 & 0.1 & 1 & 1.0 & 1 & 0.5 & 68.4 + h1 & 4.0 & 0.1 & 1 & 1.0 & 1 & 5.0 & 71.7 + h2 & 4.0 & 0.1 & 1 & 1.0 & 1 & 50 . & 35.0 + h3 & 4.0 & 0.1 & 1 & 0.1 & 1 & 5.0 & 69.7 + h4 & 4.0 & 0.1 & 1 & 0.01 & 1 & 5.0 & 49.4 + h5 & 4.0 & 0.1 & 1 & 0.1 & 10 & 5.0 & 85.7 + cclrcrcr i0 & 0.6 & 0.1 & 1 & 0 & 1 & 0.5 & 3.7 + i1 & 0.6 & 0.1 & 1 & 3 & 1 & 0.5 & 4.2 + i2 & 0.6 & 0.1 & 1 & 15 & 1 & 0.5 & 6.3 + i3 & 0.6 & 0.1 & 1 & 30 & 1 & 0.5 & 27.3 + i4 & 0.6 & 0.1 & 1 & 150 & 1 & 0.5 & 89.3 + k0 & 2.0 & 0.1 & 1 & 0 & 1 & 0.5 & 14.2 + k1 & 2.0 & 0.1 & 1 & 3 & 1 & 0.5 & 32.1 + k2 & 2.0 & 0.1 & 100 & 0 & 1 & 0.5 & 19.7 + k3 & 2.0 & 0.1 & 100 & 3 & 1 & 0.5 & 23.7 +

the chemical compositions of the c - rich pagb stars iras 05113 + 1347 and iras 22272 + 5424 are determined from high - resolution optical spectra using standard lte model atmosphere - based techniques . the stars are c , n , and-process enriched suggesting efficient operation of the third - dredge up in the agb star following a first dredge - up that increased the n abundance . lithium is present with an abundance requiring li manufacture . with this pair , abundance analyses are now available for 11 c - rich pagbs . a common history is indicated and , in particular , the-abundances , especially the relative abundances of light to heavy-process elements , follow recent predictions for the third dredge - up in agb stars .

astro-ph0109068

the post - asymptotic giant branch ( pagb ) stars , as the immediate successors of agb stars , can be used to probe nucleosynthesis and mixing processes in agb stars . the spectra of pagb stars are simple relative to spectra of agb and planetary nebulae ( pne ) . in the case of agb stars , the presence of numerous molecular lines in their spectra makes it difficult to derive accurate abundances . the central stars of pne to which pagbs evolve are so hot that many elements do not provide lines in optical spectra . general properties of pagbs are well reviewed by kwok ( 1993 ) and recently by hrivnak ( 1997 ) . in the most advanced agb stars , the predicted changes of surface composition arise from the ( third ) dredge - up of material from the he - shell below the h - rich envelope . principal products added to the envelope and atmosphere are @xmath1c , and heavy elements synthesised by the @xmath0-process involving neutron capture on a long timescale . cool carbon stars , i.e. , stars with a carbon - rich ( c / o @xmath2 1 ) atmosphere , are nature s counterparts to the theoreticians advanced agb stars . carbon - rich pagb stars have been identified , and this paper presents abundance analyses of two such stars . many pagb stars are _ not _ c - rich , and many , if not all , of the o - rich pagb stars are also not enhanced in @xmath0-process products . examples include the rv tauri variables ( giridhar , lambert , & gonzalez 2000 ) at the cool end of the pagb range , and high galactic latitude b - type supergiants , e.g. , lsiv -12@xmath3111 ( conlon et al . 1993 ) towards the hot end of the pagb range . evidently , many stars depart the agb before the third dredge - up has transformed the o - rich giant to a c - rich star . the number of c - rich pagb stars for which detailed abundance analyses are available is small : 9 stars in total by reddy et al . ( 1997 , 1999 ) , and van winckel & reyniers ( 2000 ) . their compositions do resemble the predictions for advanced agb stars , but a large sample of stars is needed to test thoroughly the theoretical predictions . the two pagb stars - iras 05113 + 1347 and iras 22272 + 5435 - analysed here are classified as c - rich because their near- and far - infrared spectra show emission features at 3.3 @xmath4 m , 3.4 @xmath4 m , and 21 @xmath4 m attributed to carbon - bearing pah features ( kwok , volk , @xmath5 hrivnak 1989 ; hrivnak , kwok , @xmath5 geballe 1994 ) . basic data for the two pagb stars are given in table 1 . in this paper , we present abundance results of these two stars . our analysis of iras 05113 + 1347 is the first to be published . iras 22272 + 5435 was analyzed previously by zas et al . our results differ significantly from theirs , as we discuss below . spectra of iras 22272 + 5435 and iras 05113 + 1347 were obtained with the 2.7 m mcdonald observatory and the coud cross - dispersed echelle spectrograph ( tull et al . 1995 ) on june 14 , 2000 and march 26 , 2001 , respectively . spectra have a resolving power of r@xmath655,000 as measured from fwhm of th emission lines , and s / n ratios of 80 - 400 at 6500 . the spectra run from 4000 in the blue to 8800 in the red with gaps in the orders beyond 5600 . a high - resolution ( r@xmath645,000 ) spectrum for iras 05113 + 1347 was also obtained on december 31 , 1999 with 3.5 m apache point observatory telescope and an echelle spectrograph . the spectrum runs from 4000 in the blue to 10000 in the red with overlapping echelle orders . spectra have been reduced by standard procedures using the spectral reduction package called iraf . telluric lines in the spectrum of iras 05113 + 1347 have been removed by dividing with a spectrum of a rapidly rotating hot star . selected spectral regions at 6785 and 7965 of both pagb stars are displayed in figures 1 and 2 . atomic spectra are strikingly similar in both stars ( fig 1 ) . both stars show strong @xmath0-process lines . the metallic lines are weak suggesting metal deficiency . iras 05113 + 1347 shows photospheric cn lines ( fig 2 ) . absence of photospheric cn lines in the spectrum of iras 22272 + 5435 may be due to its higher @xmath7 . selection of clean lines with reliable @xmath8-values is very important in reducing uncertainties in the abundances . in selecting lines we made use of the very high resolution spectrum of the cool giant arcturus ( hinkle et al . 2000 ) , as well as the solar spectrum ( wallace , hinkle @xmath5 livingston 1993 ) . to avoid overcrowding of lines and uncertainties in continuum placement , we examined the green and red parts but not the blue part of the spectra . we limited the analysis to lines of moderate strength , equivalent width ( w@xmath9 ) @xmath10 200 m ) . our line list comprised about 150 atomic lines belonging to 25 elements . choosing accurate @xmath8-values is important in the reduction of errors . we concentrated on lines which have well determined @xmath8-values experimentally or theoretically . the @xmath8-values for most of the fei and feii lines come from laboratory measurements of either the oxford or kiel - hanover groups ; lambert et al . ( 1996 ) discussed the @xmath8-value measurements . for other elements , we chose accurate @xmath8-values from a variety of sources : cr ( blackwell et al . 1986a ) , si ( garz 1973 ) , tii ( blackwell et al . 1986b ) , v ( whaling et al . 1985 ) , na , al ( lambert @xmath5 warner 1968 ) , ca ( smith @xmath5 raggett 1981 ) , yii ( hannaford et al . 1982 ) and for c , n , and o ( wiese et al . 1996 ) . li , sc , and mn exhibit hyperfine structure ( hfs ) . for sc and mn , we adopted hfs data of kurucz ( 2000 ) , as discussed by prochaska & mcwilliam ( 2000 ) . for li we adopted the hfs and isotopic structure given by andersen et al . for the rest of the lines , the @xmath8-values come either from r.e . luck s compilation ( private communication ) or kurucz s ( 2000 ) atomic line list . to check the reliability of the @xmath8-values we derived abundances for the solar spectrum using solar equivalent widths measured either by meylan et al . ( 1993 ) or from our solar spectrum at r@xmath660,000 obtained from the mcdonald observatory 2.7 m telescope . we used the kurucz ( 1995 ) convective solar model with a microturbulence of 1.15 km s@xmath11 . we accepted all the individual @xmath8-values which yield abundances within @xmath120.2 dex of mean solar abundances ( grevesse @xmath5 sauval 1998 ) determined using the lines having reliable laboratory @xmath8-values . in this study , we adopted plane parallel , and line blanketed lte model atmospheres computed by kurucz ( 1995 ) using the atlas9 code and extracted from http://cfaku5.harvard.edu . final models were found by interpolation . we assume that the standard helium abundance is appropriate to these pagb stars . a good discussion of the kurucz model atmospheres is given by castelli et al . ( 1997 ) . initial values of model parameters @xmath13 and log @xmath14 were estimated from their spectral classification and photometry . hrivnak ( 1995 ) classified iras 05113 + 1347 as g8ia and iras 22272 + 5435 as g5ia using low - resolution spectra . the spectral types of g8 and g5 translate into @xmath13 of 4260 k and 4460 k , respectively ( bhm - vitense 1972 ) . we estimate log @xmath14@xmath15 1.0 from the luminosity class ia and log @xmath14 tables . the values of @xmath13 can also be estimated from intrinsic b - v colors . the interstellar extinction values in the direction of the stars were taken from hrivnak ( 1995 ) . the intrinsic b - v colors for supergiants ( bhm - vitense 1972 ) yield 5100 k for both stars . using these initial estimates , we derived more accurate model parameters : @xmath13 , log @xmath14 , microturbulent velocity ( @xmath16 ) , and metallicity for the program stars from the high - resolution spectra . the values of @xmath17 are derived from the excitation of fei lines . we chose 23 fei lines with lower excitation potentials ranging from 0.85 ev to 4.6 ev . for a given log @xmath14 = 1.0 dex and an assumed @xmath16 = 5.0 km s@xmath11 , the values of @xmath17 are determined from the requirement that the individual fei abundances be independent of excitation potentials ( fig 3a , 3b ) . the values of log @xmath14 are determined from the requirement that neutral and ionized iron lines yield similar abundance for the chosen @xmath13 . the values of @xmath16 are derived by forcing a zero slope on the abundances of lines fei , feii , and nii as a function of w@xmath9 . this method was iteratively repeated until we obtained a self - consistent set of parameters . the adopted model parameters are given in table 2 . radial velocity measurements not only provide information on the systemic velocity , but also yield information on the atmospheric structure : velocity gradients in the atmosphere , dust shells around the stars etc . radial velocities of various atomic and molecular absorption lines are given in table 3 . with the exception of resonance lines , the atomic lines are of photospheric origin and provide the systemic velocity : the mean heliocentric velocities are @xmath18 = 7.6 @xmath12 1.5 km s@xmath11 for iras 05113 + 1347 and @xmath1942.4 @xmath121.0 km s@xmath11 for iras 22272 + 5435 . the velocity of iras 22272 + 5435 is consistent given the probability of a small amplitude variation with the value @xmath18 = @xmath1939.7 @xmath12 0.5 km s@xmath11 also measured using optical lines ( zas et al . observations of co millimeter emission lines also provide an estimate of the systemic velocity provided that the emitting gas is uniformly distributed around a star . these estimates match our velocities : @xmath18 = 7.2 km s@xmath11 ( hrivnak @xmath5 kwok 1999 ) , and @xmath1941.0 @xmath12 1.0 km s@xmath11 ( loup et al . 1993 ) for iras 05113 + 1347 and iras 22272 + 5435 , respectively . circumstellar c@xmath20 and cn lines are seen in the spectra ( fig 4 ) . in the case of iras 05113 + 1347 , photospheric cn but not c@xmath20 lines are also seen but at the systemic velocity , v@xmath21 . for iras 05113 + 1347 , bakker et al . ( 1997 ) reported the detection of c@xmath20 swan and phillips , and cn red system lines from several vibrational bands . many rotational lines of the c@xmath20 bands , as expected for a homonuclear molecule , are seen for each band . the circumstellar lines of the cn bands are restricted to low-@xmath22 lines . our measurements of c@xmath20 swan 0 - 2 and 0 - 3 and cn red system 1 - 0 lines shows that they are shifted by @xmath1910.2 @xmath12 1.0 km s@xmath11 relative to the systemic velocity , a velocity difference identified as the expansion velocity ( v@xmath23 ) of that region of the circumstellar shell containing the c@xmath20 and cn molecules . this expansion velocity is similar to the estimate of 13.1 km s@xmath11 from the width of the co millimeter emission lines ( hrivnak @xmath5 kwok 1999 ) . our detections of optical circumstellar molecular blue - shifted lines for iras 22272 + 5435 are the first for this star which is too warm to show photospheric molecular lines . swan system lines give an expansion velocity of 6.3 @xmath12 0.3 km s@xmath11 which compares with 8.0 to 12 km s@xmath11 from the widths of co millimeter lines ( loup et al . 1993 ) . an exciting observation is the apparent presence of the lii 6707 resonance doublet weakly in absorption in both spectra . but excitement is tempered by the additional observation that the line is displaced redward by about 0.2 from the wavelength expected of a photospheric line . understanding this shift and the line profile is a prerequisite to determining the stellar lithium abundance . a clue to the correct interpretation of the lii line is , in the case of iras 05113 + 1347 provided by inspection of the resonance line profiles of the more abundant alkali atoms na and k. figure 5a shows the profiles of the nai d@xmath24 , the ki 7699 , and the lii doublet as observed on 1999 december 31 . emission in the core of the nai d lines exceeds the local continuum . this emission , which is from the star and not the nightsky , divides the absorption profile into two components which we label a and b. similar emission may be imagined in the ki line but it is evidently weaker with the line profile more strongly suggesting the presence of the two absorption components . the peak nai emission is at a velocity of 7 km s@xmath11 , which is the photospheric / systemic velocity ( table 3 ) . this suggests that emission may arise from a large shell , presumably a part of the expanding circumstellar envelope responsible for the co millimeter emission . the spectrum from march 26 , 2001 does not show na d emission but the components a and b remain . profiles of the li i and ki resonance lines are almost unchanged between the two observations . emission in the na d lines , if present in iras 22272 + 5435 , is very weak . the absorption for this star consists of at least four components : a and b with the same presumed origins as the a and b components in iras 05113 + 1347 , and two additional components assigned an interstellar origin and labelled is in figure 5b . for both stars , absorption component a is redshifted relative to the systemic velocity by about 10 km s@xmath11 . similar velocity shifts were found for the two pagb stars studied previously ( reddy et al . 1999 ) . that the component is similar for four stars suggests that it has a common origin , and is not , for example , simply an intruding interstellar line two possibilities may be considered : the component is primarily of photospheric origin but redshifted because the line is preferentially formed in cool downdrafts ; the line is formed in infalling circumstellar gas . present evidence suggests the latter is more probable because it appears that only resonance lines of low ionization potential atoms are affected . if the former explanation were valid , one would expect to see a velocity shift between low excitation lines of nai , and cai ( for example ) and high excitation lines of ci and oi but table 3 shows no differences . of course , the presence of infalling ( component a ) with outflowing ( component b ) implies a complicated circumstellar shell but such a juxtaposition is not unknown for giant stars . if component a is taken to be the photospheric line shifted by convective downflow , central emission is seen in all observations of the na d and ki profiles . the predicted photospheric ki profile ( fig . 5 ) is based on the inferred normal abundances for [ fe / h ] @xmath25 @xmath190.8 , and the kurucz model . the true profiles in the actual convective photospheres are surely somewhat different . the emission with component b constitutes a p cygni profile provided by circumstellar alkali atoms . lithium may be similarly affected . elemental abundances were determined using the code moog ( sneden 1973 ) in combination with the adopted model atmosphere . most of the selected lines are unblended and computing abundances from equivalent width measurements works well . for a few elements which suffer from blends and/or hyperfine structure , abundances were determined using spectrum synthesis . a summary of abundances for iras 05113 + 1347 and iras 22272 + 5435 is given in table 4 where @xmath26 is the number of lines , and @xmath27 is the standard deviation representing scatter among lines . the quantity log @xmath28(x ) = log ( @xmath29)+12 is the abundance of an element x relative to a hydrogen abundance of 12 on a logarithmic scale . the ratio [ x / h ] = log @xmath30)@xmath19 log @xmath31 refers to the stellar abundance relative to the solar photospheric abundances of grevesse @xmath5 sauval ( 1998 ) . the abundances relative to iron [ x / fe ] = [ x / h]@xmath19[fe / h ] are computed using our derived stellar iron abundances of [ fe / h ] . errors in the abundance analysis come from different sources . errors in the adopted @xmath8-values and in the measured @xmath32 introduce uncertainties . these errors can be represented by the standard deviation , @xmath27 ( table 4 ) that measures the scatter among the individual lines . the true error , i.e. , standard deviation of the mean @xmath33 , is less for elements represented by many lines . for abundances which are derived by fewer than 3 lines the real error may be large , and we adopted @xmath27 = 0.2 dex for these sources of error . the abundances are also model dependent . the four parameters : @xmath17 , log @xmath14 , @xmath34 , and [ m / h ] that are required to select the atmospheric model are somewhat coupled . the uncertainty in the @xmath7 can be estimated by examining the relation between the lower excitation potential and abundance . the slope of the fit to the observations is quite sensitive to @xmath7 . the error ( standard deviation ) in the slope @xmath35 0.03 translates into an uncertainty in the @xmath7 of 150 k. similarly , we estimated uncertainty in @xmath34 . in the case of log @xmath14 , the error is estimated by introducing 0.1 dex difference in abundance between neutral and ionized lines . thus , the estimated uncertainties in the input model parameters are as follows : @xmath36@xmath37 = 150 k , @xmath36log @xmath14 = 0.50 dex , @xmath36@xmath34 = 0.50 km s@xmath11 , and @xmath36[m / h ] = 0.25 dex . the uncertainties in the abundance ratios as a result of uncertainties in the model parameters are given in the table 5 for iras 05113 + 1345 . the values in the table 5 show that the abundance ratios are less affected by uncertainties in the [ m / h ] and @xmath34 . the uncertainty in the gravity of 0.5 dex changes the ratios up to 0.15 dex . the abundance ratios determined from the lines of high excitation are found to be sensitive to changes in @xmath17 ( table 5 ) . for each element the quadratic sum of the uncertainties in the four model parameters is given as @xmath38 . our results for iras 22272 + 5435 differ quite considerably in several key respects from those published by zas et al . the differences are not related to slight differences in the adopted fundamental parameters . there is a small difference in the derived iron abundances : we obtain [ fe / h ] = @xmath190.8 but zas et al . got [ fe / h ] = @xmath190.5 . large differences are found for many other elements . for example , they reported several cases of extraordinary enhancements relative to iron : e.g. , ratios for elements around iron include [ cr / fe ] = @xmath392.0 , [ ni / fe ] = @xmath391.5 , and [ zn / fe ] = @xmath392.2 . in all cases , our ratios are close to the expected values for a moderately metal - poor star , that is [ cr / fe ] = [ ni / fe ] = [ zn / fe ] @xmath25 0.0 . although our analyses agree that the @xmath0-process elements are severely overabundant , abundances in the two analyses differ by up to 1.0 dex . these differences may perhaps be traceable to the lower resolution of the spectra used by zas et al . , and their tendency to include strong lines in the abundance analysis . scrutiny of their list of measured lines showed several cases where we identified much weaker lines with the transition that they used . the abundance of carbon is determined using three different indicators : permitted ci lines , the forbidden ci line at 8727.1 ( this line is not measured in the spectrum of iras 22272 + 5435 ) , and lines of the cn red system at 8035 . the forbidden ci line is a superior c abundance indicator as it is unaffected by nlte ( gustafsson et al . however , this line is blended with fei and cn lines . fortunately , the fei and cn contributions to the blend at this metallicity are small . all known blends are taken into account in the spectrum synthesis . for iras 05113 + 1347 , the carbon abundance from this forbidden line is 0.2 dex less than that derived using the permitted ci lines , but the error estimate @xmath36@xmath40 = 150 k causes a difference of 0.19 dex in abundances derived from permitted and forbidden ci lines . non - lte effects may also be a factor . the cn red system lines yield a higher abundance by 0.3 dex than the forbidden line . this may be due to the uncertainty in the abundance of n which is an input in the synthesis of cn lines and is based on two weak ni lines in the red . given the uncertainties in the model parameters the derived c and n abundances from different abundance indicators are in good agreement . the abundance of oxygen is determined using both forbidden and permitted lines . the oxygen abundance ( log @xmath28([oi])=8.37 ) determined from [ o i ] lines at 6300.3 and 6363.8 for iras 05113 + 1347 is found to be in good agreement with the abundance derived from the permitted oi line at 9265.9 . for iras 22272 + 5435 , the oxygen abundance comes from the [ o i ] line at 6300.3 . final results of c , n , and o abundances are given in table 6 . abundances of c and o are the simple mean of abundances derived using different indicators . total sum of c , n , and o abundances ( @xmath41 ) relative to fe and ratios c / o and n / o are also given in table 6 . a majority of the s - process lines used here come from kurucz s ( 2000 ) line list . many of these lines do not exist in the solar spectrum . given the large number of lines in the kurucz list it is difficult to know that the @xmath0-process line is the dominant contributor . to make sure , we computed equivalent widths for all the nearby lines . molecular cn and c@xmath20 contributions to the lines were also taken into account . this procedure allowed us to identify quite a number of reasonably clean @xmath0-process lines . the @xmath0-process lines , their atomic data , and abundances are given in the table 8 . extraction of the li abundance is important because lithium plays an important role in understanding the evolutionary state of pagb stars . here , the abundance analysis is complicated by the fact that the dominant absorption attributable to the lii 6707 line is at the velocity of the infalling material labelled as component a. component b , definitively circumstellar , may also contribute absorption to confuse further a contribution from a lii photospheric line . finally , emission may fill - in the photospheric lii absorption in the case of iras 05113 + 1347 . to within the errors of measurement , identical velocity shifts of a line likely to be the photospheric lii line are found for this pair of pagbs as well as and the pagbs iras 02229 + 6208 and iras 07430 + 1115 ( reddy et al . 1999 ) . our syntheses of the li region used atomic lines listed by kurucz ( 2000 ) , and cn red system lines ( cunha , smith , & lambert 1995 ) . the arcturus spectrum was synthesized and , as a result , minor modifications were made to the adopted @xmath8-values . we also included the @xmath0-process ionized line required by lambert , smith , & heath ( 1993 ) to fit spectra of baii giants . selected syntheses are shown in figure 6 where it is obvious that the observed spectrum can not be matched satisfactorily . these syntheses leave unmatched the absorption to the red that is attributed to a lii line from the circumstellar shell ( component a ) . the presence of two lii components is not at all unprecedented having been reported for a li - rich ( o - rich ) giant in the globular cluster ngc362 ( smith , shetrone , & keane 1999 ) , and for the li - rich semiregular variable , also o - rich , w lmi ( giridhar , lambert , & gonzalez 2000 ) . alternative interpretations seem implausible . if photospheric lithium is assumed to be rich in @xmath42li , a fair fit to the spectrum is obtained , as noted previously by reddy et al . ( 1999 ) ; the @xmath43li - @xmath42li wavelength shift is about 7 km s@xmath11 or only slightly less than the redshift of component a , but the presence of @xmath42li is considered most improbable . a vi line at 6708.07 might be deemed responsible for component a but the @xmath8-value needed to fit the line in arcturus must be increased by nearly 3 dex to match component a in the pagbs . finally , it is unlikely that the component is an unrecognized @xmath0-process line ; no such line was required by lambert et al . ( 1993 ) in their syntheses of spectra of baii giants including stars for which the ceii lines are of comparable strength to the lines in the pagb stars . ( technetium appears not to have a line at 6708.0 . ) if component a is assigned to circumstellar lithium atoms and circumstellar emission is neglected , a fair assumption given that the spectrum of iras 05113 + 1347 at 6707 is not significantly different when the na d emission was very prominent , the syntheses of the photospheric component give log @xmath28(li ) @xmath25 1.8 for iras 05113 + 1347 , and @xmath25 2.1 for iras 22272 + 5435 . to derive the carbon isotopic ratio , we selected lines of the cn red system in the spectral region 8000 - 8010 . this is the region where a strong blend of @xmath45c features at 8000.4 and clean @xmath1cn lines are located . there are other @xmath45cn features but most of them are buried in stronger @xmath1cn lines . the molecular data : line positions , excitation values , and oscillator strengths are taken from the investigation of de laverny @xmath5 gustafsson ( 1998 ; private communication from de laverny ) . the list includes all the @xmath1cn lines given by davis @xmath5 philips ( 1963 ) and @xmath45cn lines given by wyller ( 1966 ) . additional input data required to produce synthetic spectrum is the dissociation energy that we adopted 7.75 ev ( see lambert 1993 ) for both the molecules . the region 8000 - 8010 was computed using the adopted line list for the arcturus model atmospheres and the abundances . the resultant spectrum was compared with the observed high - resolution spectrum of arcturus to test the accuracy of the molecular and atomic line data . the predicted spectrum matches nicely the observed arcturus spectrum and this also yields @xmath1c/@xmath45c = 6.0 @xmath12 1.0 which is in very good agreement with the isotopic ratio measured for arcturus ( griffin 1974 ) . we computed the spectrum for iras 05113 + 1347 . ( photospheric cn lines are not seen in the spectrum of iras 22272 + 5435 ) . we fit the @xmath1cn for the derived n abundance by changing the c abundance . for a c abundance of [email protected] dex the @xmath1cn lines are well fitted ( fig 7 ) . the @xmath45cn feature blended with @xmath1cn at 8002.0 , the blends of pure @xmath45cn line at 8004.5 , and at 8006.0 are sensitive to the changes in the isotopic ratio @xmath1c/@xmath45c . the spectrum was predicted for three different carbon isotopic ratios and compared with the observed spectrum of iras 05113 + 1347 . the features at 8002.0 and 8004.5 indicate a lower limit of @xmath1c/@xmath45c@xmath2 25 . a slightly smaller value is suggested by the @xmath45cn feature at 8006.0 . we give greater weight to the feature at 8004.5 as this is stronger and pure @xmath45cn . the limit @xmath1c/@xmath45c @xmath2 25 is in good agreement with the typical lower limit of @xmath1c/@xmath45c @xmath46 20 derived for pagbs , including iras 05113 + 1347 , by bakker et al . ( 1997 ) using circumstellar molecular cn red lines . inspection of the compositions of the two pagb stars ( table 4 & 6 ) leads to four broad conclusions , which we amplify below : ( i ) their composition is normal except for those elements expected to be affected during evolution prior to the pagb phase ; ( ii ) lithium is present at an abundance that probably implies lithium production in a post main sequence phase ; ( iii ) adjustments to initial c and n abundances occured as a result of mixing with h- and he - burnt material ; ( iv ) substantial enrichment with @xmath0-processed material is traceable to mixing from the he - shell of their agb progenitor . collectively , these conclusions suggest the stars have evolved from thermally - pulsing agb stars . the pair are compared with published results for 9 other c - rich pagb stars ( table 7 ) , which support the four broad conclusions . even in the case of thermally - pulsing agb stars , many elements are expected to retain their initial abundances ( relative to iron ) . this is the case here : na , al , si , s , ca , ti , and the iron - group from v to ni , and zn show the ratio [ x / fe ] reported for normal dwarfs and red giants with [ fe / h ] @xmath25 @xmath190.8 . ( possibly , [ ca / fe ] @xmath25 0.0 is less than the expected value of 0.2 dex as a result of galactic chemical evolution . ) this demonstration of normal abundances serves to exclude the possibility that the photospheres are affected by processes of dust - gas selective fractionation that are seen to have affected some warmer pagb stars ( waelkens et al . 1992 ) , and the warmest rv tauri variables ( giridhar , lambert , & gonzalez 1999 ) . in particular , our pair of pagb stars show normal [ s / fe ] and [ zn / fe ] ratios which are potent indicators of dust - gas fractionation . carbon , nitrogen , and oxygen abundances are key tracers of the evolutionary and nucleosynthetic history of red giants and their descendants . low mass red giants are predicted to experience three episodes in which the deep convective envelope brings nuclear - processed material from the interior to the atmosphere . the first episode known as the first dredge - up occurs when the star ascends the red giant branch , after exhaustion of hydrogen in the core of the main sequence star . this brings mildly cn - cycled material to the surface and , therefore , reduces the c abundance and enhances the n abundance with the constraint that the sum of the c and n abundances is preserved ( iben 1964 ) . oxygen is predicted and observed to be unaffected by the first dredge - up . additional changes of surface abundances occur through what is known as the third dredge - up phase known to affect agb stars . in this phase , material from the he - shell around the c - o electron degenerate core is mixed into the base of the deep convective envelope . the added material has been exposed to he - burning via the 3@xmath47-process and to @xmath0-processing . the principal product of the former is @xmath1c with an admixture of @xmath48o . an agb star is expected to experience the third dredge - up repeatedly until it evolves to the pagb phase and beyond . ( even then a final episode is possible . ) through the third dredge - up , the o - rich giant may be converted to a c - rich agb star . in its simplest theoretical implementation , the series of third dredge - up events do not involve exposure of the h - rich envelope to the h - burning cn - cycle , that is the carbon added from the he - shell is not converted to nitrogen . very luminous agb stars can experience h - burning at the base of the h - rich envelope in the long intervals between times when the he - shell is ignited . such stars said to experience hot bottom burning ( hbb ) , convert carbon to nitrogen , reduce the @xmath1c/@xmath45c ratio , and may be reconverted from a c - rich to an o - rich star . in order to compare the observed c , n , and o abundances with theoretical expectations , we begin with establishing the initial abundances for these slightly metal - poor stars , and , then , we comment on the changes expected from the first dredge - up which establish the expected surface abundances for the agb star prior to the onset of the third dredge - up . observed abundances for our two stars and eight others analysed previously are summarized in table 6 . initial abundances for c and o are taken from gustafsson et al . ( 1999 ) who analysed the [ ci ] 8727 line in a sample of main sequence stars , and collated results on the o abundance for a similar and overlapping sample . fits to the derived abundances gave @xmath49 = ( -0.17\pm0.03)\times [ fe / h ] + ( 0.065\pm0.008)$ ] and @xmath50 = ( -0.36\pm0.02)\times [ fe / h ] - ( 0.044\pm0.010)$ ] measurements of the nitrogen abundance are less extensive . we adopt [ n / fe ] = 0 ( clegg , lambert , and tomkin 1981 ) . the inferred initial abundances are given in table 6 along with the observed values for the pagb stars . relative to initial values , the stars are clearly enriched in carbon and nitrogen , but oxygen has retained about its initial abundance . these are in line with our sketch of the expected changes : the carbon enrichment is attributed to the third dredge - up , and the nitrogen enrichment to the first dredge - up , and , as expected , oxygen is almost unaffected by the dredge - ups . from table 6 , the mean difference between the o abundance of the pagb stars and their inferred initial o abundance is 0.2 dex . given the possibility of systematic errors in comparing abundances derived from main sequence and pagb stars , this difference is compatible with the expectation that oxygen is unaffected by evolution from the main sequence to the pagb stage . it may be noted that a study of post first dredge - up red giants with metallicities similar to those of stars in table 6 concluded that oxygen was undepleted ( cottrell & sneden 1986 ) . there is evidence , primarily from the low observed and higher predicted @xmath1c/@xmath45c ratios , that predicted changes for carbon and nitrogen brought by the first dredge - up are less severe than observed changes . it is , however , expected that a red giant s nitrogen abundance will approach the sum of the star s initial c and n abundances . considering this , we list in table 7 this sum alongside the observed n abundances for the pagb stars . on average , the latter are equal to the sum of the initial c and n abundances : the mean difference is less than 0.1 dex . this is certainly compatible with the idea that the nitrogen abundance of agb and pagb stars was unaltered by the third dredge - up and subsequent events . cottrell & sneden s analyses , however , do not fully support this conclusion . their carbon abundances show the expected underabundance , for the first dredge - up , especially if the initial abundances are taken from gustafsson et al . low @xmath1c/@xmath45c ratios reported by the authors clearly show that a first dredge - up has occurred . oddly , nitrogen was reported to be underabundant : [ n / fe ] = @xmath190.05 was found in the mean . this is surprising . we suppose that this anomaly is either an artifact of the analysis and/or the assumption that [ n / fe ] = 0 is incorrect for the initial abundances . if , of course , [ n / fe ] @xmath51 initially , our conclusion that the first dredge - up accounts for the observed nitrogen abundances of pagb stars is invalid . the first dredge - up reduces the surface @xmath1c/@xmath45c ratio . cottrell & sneden s measurements give a mean ratio of 16 for their sample of mildly metal - poor giants . available determinations for the pagbs including our lower limit for iras 05113 + 1347 give higher values . these are consistent with addition of @xmath1c to the envelope of a red giant without exposure to the cn - cycle , which would convert some @xmath1c to @xmath45c and @xmath52n . thus , the @xmath1c/@xmath45c ratios and the nitrogen abundances support the idea that little of the @xmath1c from the third dredge - up has been processed by protons . carbon in pagbs is considerably enhanced relative to its initial abundance and certainly to its post first dredge - up abundance . the stars are clearly c - rich , i.e. , c / o @xmath2 1 . table 6 shows that c / o @xmath25 3 for iras 05113 + 1347 , and 2 for iras 22272 + 5435 . the sensitivities of the abundances to the atmospheric parameters are so similar for the carbon and oxygen lines that the derived c / o ratio can not be plausibly decreased to less than unity . we assume that fresh carbon was added through the third dredge - up events . in summary , the c , n , and o abundances are compatible with the view that these c - rich pagb stars evolved directly from agb stars that were converted to carbon stars as a result of the third dredge - up . it remains to examine if the @xmath0-process and lithium abundances may be integrated into this picture . both stars are severely enhanced in @xmath0-process elements ( table 7 ) which we identify as a signature that the stars experienced an efficient third dredge - up in their agb phase . in this respect , our pair of pagbs are similar to the other analyzed c - rich pagbs . models indicate that low mass agb stars run their @xmath0-processing in the he - shell with the neutron source @xmath45c(@xmath47 , n)@xmath48o . to provide sufficient neutrons , @xmath45c must be generated in the h - he intershell region by a slow mixing of protons into the he - rich layers . presently , the mass of @xmath45c in the shell at the onset of @xmath0-processing is not calculable from the first principles but must be specified as a free parameter . this free parameter is a controlling influence on the neutron flux . the higher the neutron flux the greater is the abundance of @xmath53 relative to @xmath54 @xmath0-process elements . this is customarily characterized by the ratio [ hs / ls ] ( luck & bond 1991 ) where we take la and nd as heavy-@xmath0 ( hs ) and y and zr as light-@xmath0 ( ls ) elements . the abundances of these four @xmath0-process elements are represented by several lines and available for all the c - rich pagbs . busso et al . ( 2001 ) also used the same @xmath0-process elements in their theoretical computations . the results would not be significantly changed if we adopted any other combinations of heavy @xmath0-process elements . the ratios [ hs / ls ] for 10 pagbs including the two pagbs studied here are given in table 7 . the ratio [ hs / ls ] varies with metallicity across the pagb sample . this is clearly shown in figure 8 . for comparison we show the model predictions from busso et al . ( 2001 ) for the case in which the mass of @xmath45c in the he - shell is assigned a constant and plausible value independent of metallicity . the chosen value designated as st/1.5 for a model of 1.5 m@xmath55 by busso et al . offers a good fit to [ hs / ls ] measurements for @xmath0-process enriched stars over a wide range of metallicity . the model predictions offer a good fit to the observed trend set by the pagbs . two pagb stars ( iras 07134 + 1005 and iras 02229 + 6208 ) deviate from this trend and they can be matched well with st/3 model predictions . the models also account for the observed enhancements [ ls / fe ] and [ hs / fe ] . at [ fe / h ] @xmath35 @xmath190.8 , the maximum lithium abundance seen in main sequence stars is @xmath56(li ) @xmath25 2.3 ( chen et al . at the first dredge - up , the surface lithium abundance is reduced by a factor of about 50 , i.e. , @xmath56(li ) @xmath25 0.6 is predicted for red giants . shetrone , sneden , & pilachowski ( 1993 ) found lithium abundances from 0.3 to @xmath57 @xmath190.5 in old disk giants with [ fe / h ] similar to that of our pagbs and a low @xmath1c/@xmath45c ratio . hence , the pagb stars with @xmath56(li ) @xmath25 2.0 are very li - rich relative to their putative progenitors . therefore , the li - rich stars have either experienced lithium production , or external replenishment of lithium ( e.g. , engulfed a planet ) . production is possible by what is known familiarly as the cameron - fowler ( 1971 ) @xmath58be transport mechanism in which @xmath59he by @xmath47-capture is converted to @xmath58be and subsequently to @xmath58li by electron capture . the @xmath59he supply is partly primordial @xmath59he and partly the product of incomplete running of the @xmath60-chain in the outer parts of the low mass main sequence star . a high temperature is needed for @xmath47-capture on @xmath59he but a low temperature is needed for @xmath58li ( and @xmath58be ) to avoid destruction by protons . thus , lithium production is efficient only in a convective region / envelope with a hot base and cool outer boundary . in intermediate mass ( 4 to 7m@xmath61 ) agb stars , predictions and observations show that lithium production is possible . this has been most clearly demonstrated for the most luminous agb stars in the magellanic clouds ( smith & lambert 1989 , 1990 ; plez , smith , & lambert 1993 ; sackmann & boothroyd 1992 ) . the great majority of these li - rich stars are o - rich , a not unexpected result because the temperatures required for @xmath47-capture by @xmath59he also ignite the cn - cycle which will convert carbon to nitrogen and convert c - rich material to n - rich material with a low @xmath1c/@xmath45c ( @xmath35 3 ) ratio . attribution of lithium production to intermediate - mass luminous agb stars implies that the li - rich pagbs evolved from stars with masses of about 4 @xmath62 , which is unlikely , although perhaps not impossible , for stars as metal - poor as [ fe / h ] @xmath35 @xmath190.8 . examination of the reaction rates shows that the time scales ( @xmath63 ) for destruction of @xmath59he and @xmath1c are similar over the likely range of temperature . the rate constant for @xmath59he(@xmath64)@xmath58be is approximately an order of magnitude greater than for @xmath1c(p , @xmath65 ) @xmath45n ( angulo et al . 1999 ) but this difference is offset by the ratio @xmath66 . then , @xmath59he and @xmath1c are destroyed at about the same rate . unfortunately , the amount of @xmath59he that was destroyed to produce the observed lithium abundance is unknown because the efficiency of the conversion of @xmath58be to @xmath58li is not obtainable from the observations . certainly , the observed lithium abundance is a small fraction ( perhaps , a 10@xmath67th ) of the @xmath59he reservoir predicted by standard models . even at low efficiency , a small fractional change in the @xmath59he abundance results in the observed lithium abundance with a similarly small fractional change in the @xmath1c and , hence , the @xmath52n abundance . provided that fresh @xmath1c is added from thermal pulses occuring during and after lithium production , stars can remain c - rich with a high @xmath1c/@xmath45c ratio yet present a lithium enrichment . lithium - rich giants are known at much lower luminosities ( cf . charbonnel & balachandran 2000 ) . several proposals exist in which the @xmath68he reservoir is tapped to provide the @xmath43li . it is unlikely that these stars are progenitors of the li - rich pagbs . the lithium in the giants will most probably be destroyed as the stars evolve to the and up the agb . certainly , conversion of these o - rich giants to c - rich agb ( and pagb ) stars will remove the lithium unless it is replenished from the @xmath68he reservoir . replenishment of lithium is in principle possible when the agb star engulfs a planet or a brown dwarf . siess & livio ( 1999 ) discuss accretion by a 3 m@xmath61 agb star . these calculations suggest that surface lithium is enhanced by a factor of a few , a factor barely sufficient to account for the observed lithium abundances . although the parameter space of the scenario was not fully explored and other choices of agb star ( e.g. , lower mass and lower metallicity ) and accreted object may result in a higher lithium abundance , this scheme will have difficulty in accounting for the fact that a lithium enrichment is very common among the c - rich pagb stars . five of the 10 stars in table 7 have been reported to show lithium enrichment . spectra of the other five stars have been examined for lithium but the 6707 line was apparently absent . assuming a detection limit of 5 m for the line we obtain the abundance limits shown in table 7 . all are consistent with either lithium enrichment or the low lithium abundance predicted for an agb star that does not manufacture lithium . our abundance analysis based on high - resolution optical spectra of iras 05113 + 1347 and iras 22272 + 5435 shows that these stars classified as pagb stars have the surface composition expected of a highly - evolved agb star : carbon and nitrogen enrichments ( relative to their presumed initial abundances ) are identified as products of the third and first dredge - ups , respectively , and a strong @xmath0-process enrichment is also attributable to the third dredge - up . an intriguing lithium enrichment may reflect lithium production from @xmath68he in the agb or pagb star . the two stars analysed in this paper seem typical of the sample ( now , 11 stars in total analysed ) of c - rich pagb stars . if the transition from agb to pagb star was made without drastic alterations of surface composition , the immediate progenitor of a c - rich pagb was a cool carbon star of the n - type . bright cool carbon stars analysed by lambert et al . ( 1986 ) and ohnaka , tsuji , & aoki ( 2000 ) have c , n , and o abundances quite similar to the pagbs , and are @xmath0-process enriched to similar levels too . however , the carbon stars are of approximately solar metallicity ( lambert et al . 1986 ; abia et al . 2001 ) not metal - poor like the pagbs . lithium provides a discordant note . the n - type stars are not lithium rich . the lithium - rich carbon stars are the j - type stars rich in @xmath45c and with no @xmath0-process enhancement . there are metal - poor carbon stars in the halo that are as metal - poor as the pagbs : kipper et al . ( 1996 ) analysed 5 stars with [ fe / h ] from @xmath190.7 to @xmath191.2 , all with considerable @xmath0-process enrichments . with one exception , the carbon abundances are similar to those of the pagbs . but with the same exception , the nitrogen abundances are much less than those of the pagbs . in two of the stars li was found to be enhanced ( log @xmath28(li ) @xmath46 1.7 ) . the noticeable difference is that the @xmath1c/@xmath45c ratio of the carbon stars are much lower ( 5 to 10 ) than the limits set for the pagbs . in summary , the immediate progenitors of the pagbs have yet to be identified . this is odd given that the pagbs evolve rapidly and probably more rapidly than their agb progenitor . it is quite possible that the progenitor is shrouded by dust . we acknowledge the support of national science foundation ( grant ast-9618414 ) and the robert a. welch foundation of houston , texas . we thank dr . pandey and dr . allende prieto for many helpful comments . this research has made use of the simbad database , operated at cds , strasbourg , francs , and the nasa ads service , usa . abia , c. , busso , m. , gallino , r. , domnguez , i. , straniero , o. , & isern , j. 2001 , apj , ( in press ) andersen , j. , gustafsson , b. , & lambert , d.l . 1984 , a&a , 136 , 75 angulo , c. , arnould , m. , rayet , m. , ( nacre collaboration ) , 1999 , nuclear physics a , 656 , 1 bakker , e.j . , van dishoek , e. f. , waters , l.b.f.m . , & schmaker , t. 1997 , a@xmath5a , 323 , 469 blackwell , d.e . , booth , a.j . , menon , s.l.r . , & petford , a.d . 1986a , mnras , 220 , 303 blackwell , d.e . , booth , a.j . , menon , s.l.r . , & petford , a.d . 1986b , mnras , 220 , 289 bhm - vitense , e. 1972 , a&a , 17,335 busso , m. , gallino , r. , lambert , d.l . , travaglio , c. , & smith , v.v . 2001 , apj ( in press ) cameron , a.g.w . , & fowler , w.a . 1971 , apj , 164 , 111 castelli , f. , gratton , r.g . , & kurucz , r.l . 1997 , a&a , 318 , 841 charbonnel , c. , & balachandran , s.c . 2000 , a&a , 359 , 563 chen , y.q . , nissen , p.e . , benoni , t. , & zhao , g. 2001 , a&a , 371 , 943 clegg , r.e.s . , tomkin , j. , & lambert , d.l . 1981 , apj , 250 , 262 conlon , e.s . , dufton , p.l . , mccausland , r.j.h . , & keenan , f.p . 1993 , apj , 408 , 593 cottrell , p.l . , & sneden , c. 1986 , a&a , 161 , 314 cunha , k. , smith , v.v . , & lambert , d.l . 1995 , apj , 452 , 634 davis , s.p . , & phillips , j.g . 1963 , the red system of the cn molecule , university of california press , berkeley . de laverny , p. , & gustafsson , b. 1998 , a&a , 332 , 661 garz , t. 1973 , a&a , 26 , 471 giridhar , s. , lambert , d.l . , & gonzalez , g. 1999 , pasp , 111 , 1269 giridhar , s. , lambert , d.l . , & gonzalez , g. 2000 , pasp , 112 , 1559 grevesse , n. , & sauval , a.j . 1998 , space . 85 , 161 griffin , r. 1974 , mnras , 167 , 645 gustafsson , b. , karlsson , t. , olsson , e. , edvardsson , b. , & ryde , n. 1999 , a&a , 342 , 426 hannaford , p. , lowe , r.m . , grevesse , n. , bimont , e. , & whaling , w. 1982 , apj , 261 , 736 hinkle , k. , wallace , l. , valenti , j. , & harmer , d. 2000 , visible and near infrared atlas of the arcturus spectrum 3727 - 9300 hrivnak , b.j . 1995 , apj , 438 , 341 hrivnak , b.j . 1997 , in iau symp . 180 , planetary nebulae , ed . habing and h.j.g.l.m . lamers , dordrecht : kluwer , 303 hrivnak , b.j . , & kwok , s. 1991 , apj , 371 , 631 hrivnak , b.j . , & kwok , s. 1999 , apj , 513 , 869 hrivnak , b.j . , kwok , s . , & geballe , t.r . 1994 , apj , 420 , 783 iben , i.jr . 1964 , apj , 140,1631 kipper , t. , jorgensen , u.g . , klochkova , v.g . , & panchuk , v.e . 1996 , a&a , 306 , 489 kurucz , r.l . 2000 , http://cfaku5.harvard.edu kurucz , r.l . 1995 , http://cfaku5.harvard.edu kwok , s. 1993 , ara&a , 31 , 63 kwok , s. , volk , k.m . , & hrivnak , b.j . 1989 , apj , 345 , l51 lambert , d.l . , 1993 , in molecules in the stellar environment , ed . jorgensen , springer - verlag ( berlin ) , 1 lambert , d.l . , gustafsson , b. , eriksson , k. , & hinkle , k.h . 1986 , apjs , 62 , 373 lambert , d.l . , heath , j.e . , lemke , m. , & drake , j.1996 , apjs , 103 , 183 lambert , d.l . , smith , v.v . , & heath , j. 1993 , pasp , 185 , 568 lambert , d.l . , & warner , b. , 1968 , mnras , 138 , 181 loup , c. , forveille , t. , omont , a. , & paul , j.f . 1993 , a&as , 99 , 291 luck , r.e . , & bond , h.e . 1991 , apjs , 77 , 515 meylan , t. , furenlid , i. , wiggs , m.s . , & kurucz , r.l . 1993 , apjs , 85 , 163 ohnaka , k. , tsuji , t. , & aoki , w. 2000 , a&a , 353 , 528 plez , b. , smith , v.v . , & lambert , d.l . 1993 , apj , 418 , 812 prochaska , j.x . , & mcwilliam , a. 2000 , apj , 537l , 57 reddy , b.e . , bakker , e.j . , & hrivnak , b.j . 1999 , apj , 524 , 831 reddy , b.e , parthasarathy , m. , gonzalez , g. , & bakker , e.j . 1997 , a&a , 328,331 sackmann , i - j . , & boothroyd , a.i . 1992 , apj , 392 , l71 shetrone , m.d . , & sneden , c. , & pilachowski , c.a . 1993 , pasp , 105 , 337 siess , l. , & livio , m. 1999 , mnras , 304 , 925 smith , v.v . , & lambert , d.l . 1989 , apj , 345 , l75 smith , v.v . , & lambert , d.l . 1990 , apj , 361 , l69 smith , g. , & raggett , d.st.j . 1981 , j.phys.b 14 , 4015 smith , v.v . , shetrone , m.d . , & keane , m.j . 1999 , apj , 516 , 73 sneden , c. 1973 , ph.d . thesis , univ . of texas at austin tull , r.g . , macqueen , p.j . , sneden , c. , & lambert , d.l . 1995 , pasp , 107 , 251 van winckel , h & reyniers , m. 2000 , a&a , 354 , 135 waelkens , c. , van winckel , h. , bogaert , e. , & trams , n.r . 1992 , a&a , 264 , 159 whaling , w. , hannaford , p. , lowe , r.m . , bimont , e. , & grevesse , n. 1985 , a&a , 153 , 109 wallace , l. , hinkle , k. , & livingston , w. 1993 , an atlas of the spectrum of the solar photosphere 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 nist , 453 wyller , a.a . 1966 , apj , 143 , 828 zas , l. , klochkova , v.g . , & panchuk , v.e . 1995 , mnras , 275,764 ccccccc + & & & & + @xmath69 & lep & & @xmath32 & & @xmath32 + ( ) & ( ev ) & log @xmath8 & ( m ) & log @xmath28(x ) & ( ) & log @xmath28(x ) + & & & & & & + 6363.63 & 1.12 & -1.419 & 35.0 & 1.64 & 28 & 1.92 + 6413.68 & 1.13 & -0.853 & ... & ... & 36 & 1.49 + 6281.28 & 0.96 & -0.570 & 117.0 & 1.36 & ... & ... + 6165.89 & 0.92 & -0.205 & 199.0 & 1.59 & 166 & 1.74 + 6093.06 & 1.46 & -0.648 & 39.0 & 1.32 & ... & ... + 6087.53 & 1.12 & -0.618 & 124.0 & 1.66 & ... & ... + 6086.18 & 1.30 & -0.801 & 92.0 & 1.81 & ... & ... + 6016.51 & 1.00 & -0.798 & 152.0 & 1.92 & ... & ... + 6182.34 & 1.44 & -0.387 & ... & ... & 33.0 & 1.31 + 6046.65 & 1.12 & -0.686 & ... & ... & 58.0 & 1.60 + & & & & & & + 6248.27 & 1.22 & -1.600 & 93.0 & 2.56 & 60 & 2.67 + 6250.44 & 1.16 & -1.620 & 112.0 & 2.66 & 79 & 2.80 + 6591.43 & 0.20 & -2.510 & 126.0 & 2.54 & 76 & 2.66 + 6637.19 & 1.45 & -1.080 & 143.0 & 2.63 & 125 & 2.88 + 6650.52 & 1.95 & -0.170 & 184.0 & 2.60 & 147 & 2.64 + 6669.63 & 1.04 & -1.720 & 115.0 & 2.61 & 83 & 2.78 + 6680.14 & 1.69 & -0.810 & 148.0 & 2.66 & ... & ... + 6698.64 & 1.64 & -1.360 & 80.0 & 2.65 & 63 & 2.86 + & & & & & & + 7085.49 & 1.07 & -1.963 & 100.0 & 2.52 & ... & ... + 6844.70 & 1.36 & -1.529 & 95.0 & 2.38 & ... & ... + 6734.05 & 1.37 & -1.400 & 106.0 & 2.37 & 83 & 2.53 + 6731.81 & 1.17 & -1.372 & 145.0 & 2.38 & 117 & 2.55 + 6694.72 & 1.37 & -2.209 & 30.0 & 2.43 & ... & ... + 6693.56 & 1.69 & -1.089 & 104.0 & 2.39 & ... & ... + 6679.22 & 1.07 & -1.691 & 108.0 & 2.34 & ... & ... + 6632.27 & 1.67 & -1.344 & 53.0 & 2.19 & 38 & 2.33 + 6616.61 & 1.35 & -2.220 & 30.0 & 2.42 & ... & ... + 6544.61 & 1.17 & -2.180 & 30.0 & 2.18 & 13 & 2.14 + 6542.76 & 1.17 & -1.737 & 77.0 & 2.26 & 45 & 2.31 + 6484.55 & 1.26 & -1.792 & 81.0 & 2.46 & ... & ... +

a simple physical model for long - duration gamma ray bursts ( grbs ) is used to fit the redshift ( ) and the jet opening - angle distributions measured with earlier grb missions and with . the effect of different sensitivities for grb triggering is sufficient to explain the difference in the distributions of the pre- and samples , with mean redshifts of and , respectively . assuming that the emission properties of grbs do not change with time , we find that the data can only be fitted if the comoving rate - density of grb sources exhibits positive evolution to 5 . the mean intrinsic beaming factor of grbs is found to range from 42 , with the average opening half - angle , compared to the pre- average of . within the uniform jet model , the grb luminosity function is , as inferred from our best fit to the opening angle distribution . because of the unlikely detection of several grbs with , our analysis indicates that low redshift grbs represent a different population of grbs than those detected at higher redshifts . neglecting possible metallicity effects on grb host galaxies , we find that grb occurs every 600,000 yrs in a local spiral galaxy like the milky way . the fraction of high - redshift grbs is estimated at 8 12% and 2.5 6% at and , respectively , assuming continued positive evolution of the grb rate density to high redshifts .

astro-ph0610043

gamma - ray bursts ( grbs ) are brief flashes of @xmath13 rays occurring at an average rate of a few per day throughout the universe . the ultimate energy source of a grb is believed to be associated with a catastrophic event associated with black - hole formation @xcite . this event takes place through the collapse of the core of a massive star in the case of long - duration grbs , and due to merger or accretion - induced collapse events for the short - hard class of grbs . for the class of long - duration grbs , which is the only type considered in this paper , liberation of energy from the grb source is thought to result in a very high temperature , baryon - dilute fireball that expands to reach highly relativistic speeds and form a collimated outflow . dissipation of the directed kinetic energy of the relativistic blast wave through internal and external shocks is the standard model for producing the prompt radiation and afterglow ( see * ? ? ? * for a recent review ) . evidence of jetted grbs is obtained from radio @xcite and optical @xcite observations of achromatic breaks in the afterglow light curves . the structure of these jets is , however , still an open question . the two leading models are ( 1 ) the uniform jet model , in which the energy per solid angle @xmath14 is roughly constant within a well - defined jet opening angle @xmath15 , but drops sharply outside of @xmath15 , with @xmath15 differing between grbs ( see * ? ? ? * ; * ? ? ? * ) ; and ( 2 ) the universal structured jet model @xcite , in which all grb jets are intrinsically identical with the directional energy release @xmath16 dropping approximately as the inverse square of the opening angle from the jet axis . @xcite and @xcite found that the absolute @xmath13-ray energy releases , after correcting for the jet opening angles inferred from the afterglow light curves , are clustered near @xmath17 ergs . @xcite , treating a larger sample , confirmed this clustering around @xmath18 ( the total energy release @xmath19 depends on the assumed efficiency for @xmath13-ray energy conversion ) . in a later analysis by @xcite , still in the pre- era , a broader spread by about one order of magnitude is found in the beaming corrected @xmath13-ray energy release around a mean value of @xmath20 ergs ( see fig since we use jet opening angles derived from these analyses , which are based on the uniform jet model , we also adopt this model in our treatment . the sample of @xcite is taken to provide a good representation of the pre- observations of grbs . + 0.1 in -0.46 in -ray energy release @xmath21 for pre- grbs @xcite.,title="fig:",width=364 ] -2.28 in grbs can in principle exist ( e.g. , * ? ? ? * ; * ? ? ? * ) and be detected to very high redshift , @xmath22 @xcite , which holds promise that grbs can be used to probe the early universe . their utility for this purpose depends on the frequency and brightness of high - redshift grbs , which in turn depends on the rate - density evolution of grb sources and the evolution of grb source properties with cosmic time . given the assumption that the grb rate density is proportional to the measured star formation rate ( sfr ) history of the universe @xcite , the redshift distribution of grbs can be predicted @xcite . correlations between spectral properties and energy releases may also permit grbs to be used to determine cosmological parameters ( e.g. , * ? ? ? * ; * ? ? ? + 0.1 in -0.46 in and @xmath23 , respectively.,title="fig:",width=364 ] -2.21 in with the launch of the satellite @xcite , rapid follow - up studies of grbs triggered by the burst alert telescope ( bat ) on became possible . a fainter and more distant population of grbs than found with the pre- satellites cgro - batse , bepposax , integral , and hete-2 is detected @xcite . the mean redshift of 41 pre- grbs that also have measured beaming breaks @xcite is @xmath24 , while 16 grbs discovered by have @xmath25 @xcite ( see fig . [ fig2 ] ) . @xcite performed five different statistical tests to compare the redshift distributions of and pre- grbs assuming , as the null hypothesis , that they are identical distributions . they conclude that the redshift distributions are different , that is , all five tests reject the null hypothesis at the @xmath26 significance level . thus , the pre- and data sample distinct subsets of the overall grb population due to different detector characteristics and followup capabilities . here , we develop a physical model to understand the differences between the redshift distributions of and pre- grbs , taking into account the different detector triggering properties . the grb model is based on the uniform jet model for grbs , though the typical properties are assigned in accordance with grb observations . the grb rate density and distribution of jet opening angles are the principle unknowns , with parameter values for the mean intrinsic duration and absolute @xmath13-ray energy release of grbs determined from observations and best fits to the data . a grb is detected if its flux exceeds the flux threshold within the energy window of a detector ( the actual detection criteria are more complicated ; see * ? ? ? * ; * ? ? ? * and discussion below ) . for a given rate density of grbs , the redshift and opening angle distributions are calculated and fit to the data . we consider only data for the long , soft population of grbs . the separation of the long duration grbs from the short , hard grbs , which clearly comprise a separate source population is complicated both by the extended emission observed with swift from grbs that would normally be classified as short hard grbs ( e.g. , * ? ? ? * ) , and the appearance of the unusual grb 060614 that defies simple classification @xcite . we therefore additionally assume that the data set used to test our model represents only , or mostly , classical long duration grbs . we return to this point in the discussion . in 2 , we develop the formulation for bursting sources , and we fit the data in 3 . inferences for the differences between the and pre- redshift distributions are also made in 3 , including implications for beaming factors and low-@xmath0 grbs . we conclude in 4 by discussing the derived grb source rate density , the nature of the low - redshift grbs with weak energies , and the predicted fraction of high - redhift grbs . derivation of the bursting rate of sources in a flat @xmath27 cold dark matter cosmology is given in the appendix . we approximate the spectral and temporal profiles of a grb occurring at redshift @xmath0 by an emission spectrum that is constant for observing angles @xmath28 to the jet axis during the period @xmath29 ; thus the duration of the grb in the observer s frame is simply @xmath30 ( stars refer to the stationary frame , and terms without stars refer to observer quantities ) . the @xmath31 spectrum is therefore written as @xmath32 where the spectral function @xmath33 at @xmath34 . at observer time @xmath35 , @xmath36 is the dimensionless energy of a photon in units of the electron rest - mass energy , and @xmath37 is the photon energy at which the energy flux @xmath38 takes its maximum value @xmath39 . the quantity @xmath40 is the heaviside function such that @xmath41 when @xmath42 , that is , the angle @xmath43 of the observer with respect to the jet axis is within the opening angle of the jet , and @xmath44 otherwise . one possible approximation to the grb spectrum is a broken power law , so that @xmath45 \ , \\ \label{lambda}\ ] ] where the heaviside function @xmath46 of a single index is defined such that @xmath47 when @xmath48 and @xmath49 otherwise . the @xmath31 spectral indices are denoted by @xmath50 and @xmath51 . note that @xmath52 can easily be generalized to accommodate spectral energy distributions ( seds ) with multiple components . the bolometric fluence of the model grb for observers with @xmath53 is given by @xmath54 where @xmath55 is a bolometric correction to the peak measured @xmath56 flux . if the sed is described by eq . ( [ lambda ] ) , then @xmath57 , and is independent of @xmath58 . the beaming - corrected @xmath13-ray energy release @xmath21 for a two - sided jet is given by @xmath59 where the luminosity distance @xmath60 for a @xmath27cdm universe . substituting eq . ( [ eq2 ] ) for @xmath61 into eq . ( [ eq3 ] ) gives the peak flux @xmath62 finally , by substituting eq . ( [ eq5 ] ) into eq . ( [ eq1 ] ) , the energy flux becomes @xmath63 because of the reduced spectral sensitivity of swift to high - energy ( @xmath64 kev ) emission and the measurement of @xmath58 in many grbs , and to avoid complications from the potentially large number of parameters in our grb model , we consider in this paper the simplest grb sed with @xmath65 , so that @xmath66 ( see eqs . [ [ eq2 ] ] , [ [ eq5 ] ] and [ [ eq6 ] ] ) . the flat part of the spectrum takes the value of peak flux @xmath39 given by eq . ( [ eq5 ] ) , and is assumed to intersect the sensitive range of the grb detector . by making the above approximation , we modify the fractional number of bursts that occurs within a given redshift @xmath0 and within a jet opening angle @xmath15 ( see eqs . [ [ eq18 ] ] , [ [ eq19 ] ] , and [ [ eq20 ] ] ) ; however , the physical interpretation of the problem is expected not to change drastically when @xmath67 , which we plan to address in a subsequent paper . in the flat spectrum approximation , a @xmath31 spectrum ( @xmath65 ) has to be cut off at low ( @xmath68 ) and high ( @xmath69 ) energies in order to avoid divergent flux . the model spectrum is not likely to extend over more than two orders of magnitude , so that @xmath70 . thus we take @xmath55 , the bolometric correction factor , equal to @xmath71 in our calculations . to calculate the redshift and the jet opening angle distributions , we need to determine the rate at which grbs occur per steradian . for a @xmath72 cosmology with @xmath73 , @xmath74 , and @xmath75 @xcite , the event rate per unit time per sr for bursting sources is @xmath76 ( derivation of this result is given in appendix a ) . here , @xmath77 @xmath78 is the distribution of grb sources differential in @xmath79 @xmath80 @xmath81 @xmath82 and @xmath0 , and separability of the comoving rate density @xmath83 of grbs from their properties is assumed , where @xmath84 is grb star formation rate function . this is the crucial assumption of this treatment , and so needs restated . there is assumed to be no evolution of the emission properties of grb with cosmic time . thus , for example , the mean jet opening angles , the mean energies , and the mean values of @xmath85 ( which could be correlated with @xmath21 to satisfy the @xcite and @xcite relations when @xmath67 ) do not change through cosmic time . the phenomenological validity of the the amati and ghirlanda relations for grbs at different values of @xmath0 suggests the plausibility of this assumption . these relations can be introduced in more detailed treatments by considering source spectral properties where @xmath85 is correlated with @xmath21 . employing an integral formulation for beamed sources @xcite , the observed directional event rate for bursting sources - ray telescope by a @xmath31 flux sensitivity @xmath86 by the mean photon energy @xmath87 of the sensitive waveband of the detector . a further complication , not considered here , is that @xmath86 may depend on the duration @xmath88 of the grb . ] with @xmath31 spectral flux greater than @xmath86 at observed photon energy @xmath89 is given by @xmath90 @xmath91 @xmath92 . \label{eq9}\ ] ] this expression can also be written in terms of a heaviside function involving a minimum flux threshold condition @xmath93 . for discrete values of @xmath21 , @xmath85 , and @xmath29 , @xmath94 where @xmath95 is the jet opening angle distribution , and @xmath86 is the instrument s detector sensitivity . the sensitivity of a detector to a grb depends on many factors , namely , the detector area , the detector efficiency , the fraction of the detector that is active , the fraction of the coded mask that is open , the average solid angle , and the internal background @xcite . @xcite give the bat s instrument detector sensitivity to be about @xmath96 , which is about an order of magnitude more sensitive than batse and beppo - sax , particularly in the case of those grbs for which redshifts were obtained , as discussed below . we can make a rough estimate of the detector threshold for both and pre- instruments , noting that the mean flare size @xmath97 for a power law size distribution with index @xmath98 for a euclidean distribution ) . for swift grbs , we took the ratio of the measured grb fluences to their @xmath99 durations to find that the mean flux of a detected swift grb is @xmath100 . for an index @xmath101 1.2 , as appropriate to the swift range , we find that the swift is sensitive to grbs as weak as 10@xmath102 . we take this energy flux as the effective swift flux threshold , recognizing that the actual detection of grbs with swift is more complicated , involving both a rate and image trigger @xcite . the assignment of an effective flux threshold for the pre- detectors is complicated by the fact that they involve a variety of detectors batse , beppo - sax , hete - ii , and integral each with its own specific triggering criteria . the mean flux of the pre- sample can be obtained from the values of fluence and @xmath103 compiled by @xcite . we find that the mean flux of the pre- grbs is @xmath104 , meaning that the flux threshold sensitivity is a factor @xmath105 less than this value , because @xmath106 at fluxes @xmath107 @xcite . for simplicity , we use the value of 10@xmath108 for the mean sensitivities of the grb detectors prior to swift . hence , in this work we assume the detector threshold for and pre- grbs to be @xmath109 and @xmath110 , respectively . integrating over @xmath111 in eq . ( [ eq9 ] ) gives the factor @xmath112 describing the rate reduction due to the finite jet opening angle . hence , eq . ( [ eq9 ] ) becomes @xmath113 @xmath114 where @xmath115 is given by eq . ( [ eq6 ] ) . in order to close the system of equations and solve for the redshift distribution , we must also specify the functional form of @xmath95 as well as the comoving grbs rate density @xmath116 . grbs with small opening angles , that is , with @xmath117 , will radiate their available energy into a small cone , so that such grbs are potentially detectable from larger distances ( though not always , as a different portion of the spectrum is observed ) , with their rate reduced by the factor @xmath118 . by contrast , grb jets with large opening angles are more frequent but only detectable from comparatively small distances . the intrinsic opening angle distribution can be much different than the measured opening - angle distribution because of these effects @xcite . since the form for the jet opening angle @xmath95 is unknown , we consider the function @xmath119 where @xmath120 is the jet opening angle power - law index ; for a two - sided jet , @xmath121 . because the distributions are normalized to unity , @xmath122 + 0.1 in -0.46 in angle for @xmath123 , @xmath124 , @xmath125 , @xmath126 , @xmath127 , @xmath128 , @xmath129 in eq ( [ eq12 ] ) between @xmath130 and @xmath131 radians . , title="fig:",width=364 ] -2.2 in + 0.1 in equation ( [ eq12 ] ) hardly exhausts the possible expressions for @xmath95 , but contains interesting asymptotic behaviors . in figure [ fig3 ] , we illustrate the nature of this function by fixing the range of the jet opening angle for different value of @xmath120 . here we take @xmath15 lying between 0.05 and 0.7 radians ( @xmath132 @xmath133 ) . this figure shows that the fraction of bursts with small jet opening angles @xmath134 increases with decreasing values of @xmath135 . therefore most bursts have large jet opening angle when @xmath136 . since beaming effects are only important for negative @xmath120 , and because beaming effects are certainly important for grbs , we only consider negative value of @xmath120 in our analysis . furthermore , from equation ( [ eq13 ] ) it is clear that if @xmath137 , it is necessary to have a minimum opening angle @xmath138 in order to avoid a divergent fraction of grbs with small jet opening angles . note also that @xmath95 , eq . ( [ eq12 ] ) , approaches a @xmath139-function when @xmath140 . it has been suggested ( e.g. , * ? ? ? * ; * ? ? ? * ) that the grb formation history is expected to follow the cosmic sfr ( @xmath141 ) derived from the blue and uv luminosity density of distant galaxies , though with differences related to the metallicity dependence and fraction of the subset of high - mass stars that are grb progenitors . however , as we show in 3 , in order to obtain the best fit to the redshift and the pre- redshift and jet opening angle samples , we have to employ a grb formation history that displays a monotonic increase in the sfr at high redshifts ( sfr5 and sfr6 ; see fig . [ fig4 ] ) . to investigate the nature of the grb formation history , in this work we also assume that @xmath142 , and consider specific functional forms for @xmath143 . due to the large range of uncertainties in the sfr at @xmath144 , we first consider three different functional forms for the sfr . for the lower sfr ( lsfr ) and upper sfr ( usfr ) that should bound the actual sfr , we consider the analytic function @xmath145 @xcite , with @xmath146 , @xmath147 , and @xmath148 for the lsfr ( usfr ) . the lsfr ( sfr2 ) and usfr ( sfr4 ) describe extreme ranges of optical / uv measurements without and with dust extinction @xcite corrections , respectively . we also consider the sfr history ( sfr3 ) of @xcite , which is intermediate between the lsfr and usfr , using their analytic fitting profile given by @xmath149 where @xmath150 , @xmath151 , @xmath152 , and @xmath153 are their best fit parameters . figure [ fig4 ] shows these sfr functions , with the function normalized to unity at the present epoch @xmath154 . the constant comoving density , sfr1 , is also plotted . + 0.1 in -0.46 in ) ; sfr3 is the @xcite sfr history given by eq . ( [ eq14 ] ) , with ( @xmath155 , @xmath156 , @xmath157 , @xmath158 ) ; and sfr5 ( @xmath155 , @xmath159 , @xmath160 , @xmath161 ) and sfr6 ( @xmath162 , @xmath159 , @xmath160 , @xmath163 ) are the sfrs that give a good fit to the and pre- redshift distribution.,title="fig:",width=364 ] -2.2 in + 0.07 in note that sfr2 , sfr3 , and sfr4 are similar when @xmath164 , but differ greatly at higher redshifts . we also consider two alternative functions shown in fig . [ fig4 ] , sfr 5 and sfr6 , that display a monotonic increase in the sfr at high redshifts representing positive evolution of the sfr history of the universe to @xmath165 . later in the analysis we will show that for self - consistency we have to utilize sfr5 and sfr6 instead of sfr2 , sfr3 , and sfr4 , to obtain the best fit to the redshift , and the pre- redshift and jet opening angle samples . the directional grb rate per unit redshift with energy flux @xmath166 is thus given by @xmath167 @xmath168^{2+s } - ( 1-\mu_{\rm jmax})^{2+s}\}\ ; , \label{eq16}\ ] ] after substituting eq . ( [ eq12 ] ) into eq . ( [ eq11 ] ) and solving ; here @xmath169 when @xmath170 , where @xmath86 is the @xmath31 flux threshold sensitivity of a grb telescope that observes in a waveband centered at @xmath171 , eq . ( [ eq16 ] ) gives the redshift distribution . the size distribution of grbs in terms of their @xmath56 flux @xmath172 is then simply @xmath173 @xmath174^{2+s } - ( 1-\mu_{\rm jmax})^{2+s}\}\;. \label{eq18}\ ] ] the maximum redshift @xmath175 can be obtained by solving the luminosity distance @xmath176 using equation ( [ eq17 ] ) to obtain @xmath177 finally , the jet opening angle distribution ( see eq . [ [ eq11 ] ] ) is given by @xmath178 where @xmath95 is given in equations ( [ eq12 ] ) and ( [ eq13 ] ) , and the upper limit of the redshift integration is determined by equation ( [ eq19 ] ) . in our grb model , there are seven adjustable parameters : the @xmath56 spectral power - law indices @xmath179 and @xmath180 , the power - law index @xmath120 of the jet opening - angle distribution , the range of the jet opening angles @xmath181 and @xmath182 , the absolute emitted @xmath13-ray energy @xmath21 , and the detector threshold @xmath86 . as already mentioned , we consider , for simplicity , a flat @xmath31 grb sed , so that @xmath65 , leaving only the bolometric correction factor @xmath55 , which is set equal to 5 . the remaining parameters @xmath181 , @xmath182 , @xmath120 , and @xmath21 are constrained by observations , and the detector properties are used to estimate @xmath86 . besides these parameters , we must also assume a grb comoving rate density @xmath183 , which is also constrained by fitting our model to the observed data . adopting the uniform jet model , we assume that the energy per solid angle is roughly constant within the well - defined jet opening angle ( e.g. , * ? ? ? * ; * ? ? ? * ) . with these assumptions , we investigate the behavior of the redshift and opening angle distributions . before we proceed to compare our model of redshift and jet opening angle distributions with the samples obtained by the and pre- instruments , we need to explore some of the above parameter space . -0.46 in per day per @xmath184 per @xmath0 ] for an assumed @xmath139-function distribution of the jet opening angles , so that @xmath186 , with @xmath187 ( solid - curve ) , using sfr3 and @xmath188 . also shown are the directional event rates with @xmath189 radian , and @xmath190 , @xmath191 , and @xmath192 radians , with @xmath193 . , title="fig:",width=364 ] -2.2 in for illustrative purposes , we utilize sfr3 @xcite for the grb comoving rate density . furthermore , we also assume that @xmath194 , @xmath195 radian ( this assumption implies @xmath95 is a @xmath139-function , see eqs . [ [ eq11 ] ] , [ [ eq12 ] ] , and fig . [ fig5 ] ) , @xmath196 , @xmath197 ergs , and @xmath198 and @xmath199 for the flux sensitivity of detectors for and before , respectively . we choose @xmath29 = 10 s , and @xmath200 ergs because these are the mean values of the grb duration measured with batse assuming that batse grbs are typically at @xmath201 , and the mean beaming corrected @xmath13-ray energy release ( see fig . 1 ) , respectively . we shall use @xmath202 throughout this work . the calculation is normalized to a local ( @xmath203 ) event rate of one grb per day per @xmath204 , so that the normalization of the local rate density is @xmath205 . with the jet opening angle @xmath130 and @xmath206 radians , and the detector threshold @xmath207 , the maximum redshifts from which model grbs could be detected are @xmath208 and @xmath209 , respectivley ( see eq . [ [ eq19 ] ] ) . the directional event rate per unit redshift is shown in figure [ fig5 ] . setting @xmath210 , and @xmath211 , @xmath191 , @xmath192 radians ( the three dashed curves in figure [ fig5 ] ) , we plot the directional event rate per unit redshift for different values of @xmath181 ( see eq . [ [ eq18 ] ] ) . as @xmath212 , the directional event rate per unit redshift of the respective @xmath213-curves approaches the directional event rate per redshift of the @xmath139-function @xmath95 with @xmath206 radian . + 0.05 in -0.46 in , normalized to one event per @xmath214 per day , with @xmath215 , @xmath216 , and @xmath217 , using sfr3 . the thin and thick curves represent the instrument s energy flux sensitivity of @xmath218 ( ) and @xmath219 ( pre- ) , respectively.,title="fig:",width=364 ] -2.2 in in figure [ fig6 ] we examine three different values of the jet opening angle , @xmath220 , @xmath216 , and @xmath127 radians , with two values of flux sensitivity , namely @xmath221 and @xmath222 . as can be seen , the smaller the jet opening angle the larger the maximum redshift from which the grb is detectable . the maximum observable redshift @xmath0 associated with a particular opening angle is indicated by the vertical line in each curve . moreover , figure [ fig6 ] also shows that whatever the lower sensitivity telescope detects will also be detected by a telescope with better sensitivity , because a flat @xmath31 spectrum is considered . thus , being more sensitive than earlier telescopes , would in this model detect a sample of grbs encompassing those detected with instruments before . spectral behavior can be considered in more detailed treatments and will change the relations for maximum distances , though generally with better sensitivity the telescope can view to larger distances , especially when the more sensitive telescope can observe lower energy photons . + 0.1 in -0.46 in per day per @xmath184 per @xmath0 ] with @xmath223 and @xmath224 , @xmath225 , @xmath226 , @xmath227 , @xmath228 , @xmath229 , and @xmath230 , with sfr3 and @xmath231.,title="fig:",width=364 ] -2.2 in by varying the opening angle power - law index parameter @xmath120 over values @xmath225 , @xmath226,@xmath227 , @xmath228 , @xmath229 , and @xmath230 , with @xmath211 , @xmath232 and @xmath231 , while keeping all other assumed parameters as before , figure [ fig7 ] shows that the burst rate per unit redshift is progressively distributed to higher @xmath0 , as expected , since the more negative @xmath120 is , the larger is the fractional number of bursts that occurs at small jet - opening angles @xmath15 ( see also figure [ fig3 ] ) . more importantly , as @xmath233 , the directional event rate per unit redshift reaches an asymptotic limit ( see plots for @xmath234 and @xmath230 in figure [ fig7 ] ) . + .10 in -0.46 in per day per @xmath184 per @xmath0 ] with @xmath223 and @xmath224 for @xmath235 , @xmath236 , and @xmath237 , using sfr3.,title="fig:",width=364 ] -2.2 in the final parameter that we need to explore is the corrected @xmath13-ray energy @xmath21 . by setting @xmath238 , @xmath239 , @xmath240 , and @xmath241 with @xmath211 and @xmath242 , @xmath243 , and @xmath244 , figure [ fig8 ] shows that as the emitted @xmath13-ray energy increases , the bursting rate is shifted to higher @xmath0 , as expected . from @xcite , the pre- jet opening angle distribution extends from @xmath130 to @xmath245 radians . however , other workers have reported that the observed jet - opening angles are as large as @xmath246 ( e.g. , * ? ? ? * ) . hence , in this work we explore different ranges of @xmath15 , ranging from @xmath211 radians to @xmath247 , @xmath131 , and @xmath248 . we also use @xmath202 as the average grb duration in the stationary frame . we first consider sfr3 @xcite to fit the redshift distribution from the and pre- samples , and the opening angle distribution from the pre- sample . a large number of jet opening angles associated with the redshift sample is lacking , so our calculated jet opening angle distribution will be used as a prediction for . as indicated in figure [ fig1 ] , the corrected @xmath13-ray energy release is broadly distributed around the mean value of @xmath249 ; hence we can explore different values of @xmath13-ray energy that best fit the observed redshift and opening angle samples . it is important , however , that the best - fit value is not too far off from the mean value in order for our model to be in accord with observations . later in the analysis we will show that our best fit corrected @xmath13-ray energy @xmath21 is indeed only a factor of 2 larger than the mean . in figures [ fig9 ] , [ fig10 ] and [ fig11 ] , we plot the calculated redshift distribution at three different corrected total @xmath13-ray energies , @xmath197 , @xmath250 , and @xmath251 , respectively . in each panel , each curve represents the calculated cumulative redshift distribution for the power - law index @xmath120 of the jet opening angle distribution . the values @xmath252 , @xmath253 , @xmath254 , and @xmath255 are plotted , from far left to far right , respectively . in each figure , from top to bottom , we have @xmath256 and @xmath257 radians , @xmath256 and @xmath258 radians , and @xmath256 and @xmath259 radians , respectively , as the jet opening angles . the cumulative redshift distribution ( bold solid curves ) of the and pre- samples are plotted for comparison . we do not plot @xmath260 , since there is little variation in the redshift distribution ( see fig . [ fig7 ] ) . the results in figures [ fig9 ] , [ fig10 ] , and [ fig11 ] clearly show that the redshift distributions for different values of @xmath120 from the model do not fit the redshift sample . furthermore , the fits to both the and pre- samples do not correlate together with the associated range of the jet opening angle . in fact it indicates that we need larger burst rates at higher redshifts ( @xmath261 ) . these results rule out the lower sfr , sfr2 , as can be seen in figure [ fig12 ] . it is important to note at this point that the fit to the pre- and redshift distribution should rely on a single comoving rate density of grb with the same physical parameters . if one particular grb rate model can not fit the observed pre- and redshift distribution at the same time , then that grb model has to be eliminated , unless additional complications are introduced into the model , such as the change of grb model parameters with time the comparisons also rule out the possibility that the upper sfr explains the grb data , even though sfr4 has a higher sfr than sfr2 at large redshifts . in figure [ fig13 ] we fit the pre- and redshift samples using three different @xmath13-ray energies , namely @xmath235 , @xmath262 , and @xmath251 , with @xmath252 , @xmath253 , @xmath254 , and @xmath255 using sfr4 . the results for the range of the jet opening angle @xmath256 , and @xmath263 radians show that the calculated redshift distribution between and pre- are anticorrelated , that is , smaller values of @xmath120 , which would improve the fit for grbs , makes the fit for pre- grbs worse . however , with @xmath252 and @xmath264 ergs ( two bottom panels ) , the redshift distributions are marginally acceptable by the model , noting that the kolmogorov - smirnov one - sided test statistic for a sample of 41 and 31 grbs at the 10% level are 0.167 and 0.187 , respectively ( that is , only 10% of randomly chosen samples should exhibit maximum vertical deviations from the model cumulative distribution ; if it is greater than these values , then the model is an inadequate representation of the data at the 90% confidence level ) . only 2.5% of randomly chosen samples should have vertical excursions greater than 0.212 and 0.238 for a sample of 41 and 31 grbs , respectively . when we calculate the opening angle distribution to fit the pre- opening angle sample in figure [ fig14 ] , however , the fit is adequate for @xmath264 ergs and @xmath265 but not for @xmath266 . hence , the upper sfr ( sfr4 ) is statistically eliminated . nevertheless , the results in figure [ fig13 ] suggest that a greater grb burst rate at larger redshifts is required to fit the redshift distributions . the cumulative jet opening angle distribution ( bold solid curves ) of the pre- sample is plotted for comparison , but not for , since the available opening angles for the associated redshift sample are lacking . as a result , we consider a grb comoving source rate density that exhibits positive evolution above redshift @xmath267 . we therefore modify the hopkins & beacom sfr ( sfr3 ) to obtain the comoving rate - density that attains ( 1 ) a constant comoving rate - density at high redshift @xmath0 ( sfr5 ) , and ( 2 ) a monotonically increasing comoving rate - density with increasing redshift ( sfr6 ) in order to provide a high grb burst rate at large @xmath0 ( see caption to figure [ fig4 ] for coefficients of the sfr models ) . in figures [ fig15 ] and [ fig16 ] we begin to see some improvement in fitting the pre- and redshift samples when using sfr5 and sfr6 with @xmath268 . nevertheless , the fitting is still somewhat inconsistent between the pre- redshift distribution and the jet opening angle . for example , in figure [ fig15 ] using sfr5 , the fitting to the pre- and redshift distribution shows a good fit with @xmath269 for a range of jet opening angles between @xmath256 and @xmath259 radians ( two bottom panels ) . however , the result from the fitting between the pre- redshift distribution and the opening angle distribution with @xmath269 is inconsistent ( see figures [ fig15 ] and [ fig16 ] ) . above redshift @xmath270 , the calculated redshift distribution underestimates the fractional number of bursts that should occur at high redshifts @xmath0 . on the other hand , the calculated opening angle distribution overestimates the fractional number of grbs that should occur at small opening angles . recall that if a burst occurs at a small opening angle , then this translates into the possibility of observing this particular burst at a high redshift @xmath0 . hence , the fit with the range of the jet opening angle @xmath256 , and @xmath271 radians is not yet acceptable . in figures [ fig17 ] and [ fig18 ] we see similar results when using sfr6 . however , it is interesting to note that we can resolve this problem if we could increase the fractional number of grbs at both high and low redshifts @xmath0 . as a result , this will provide a self - consistent fitting between the pre- redshift and the jet opening angle distributions , as well as the redshift distribution . this problem can be solved by adjusting the @xmath13-ray energy @xmath21 . using @xmath272 , figures [ fig19 ] and [ fig20 ] , and figures [ fig21 ] and [ fig22 ] show a statistically acceptable fit to the pre- and redshift distribution and the pre- opening angle distribution with @xmath269 , and the range of the jet opening angle @xmath256 and @xmath258 radians for sfr5 and sfr6 , respectively . for sfr5 , the fitting can be improved with @xmath273 , as indicated by the solid curves ( see figures [ fig19 ] and [ fig20 ] ) . the above analyses show that @xmath274 gives the best fit to the pre- redshift and the jet opening angle samples , and the redshift sample . this best - fitted @xmath13-ray energy is interesting since @xcite and @xcite show that the standard - energy reservoir is @xmath275 . however , with a larger sample set , the data from @xcite shows that @xmath21 is broadly distributed between @xmath276 , as shown in figure [ fig1 ] . because we have been able to obtain an acceptable fit to the data with discrete values of @xmath21 and @xmath29 , a more complicated model where a range of values of @xmath21 and @xmath29 were allowed would certainly permit excellent fits to the data . we can utilize our best - fitted parameters to estimate the average beaming factor @xmath277 . the beaming factor average over the opening angle distribution is given by @xmath278 \ , \nonumber \label{eq21}\end{aligned}\ ] ] where @xmath95 is given in equation ( [ eq12 ] ) . using the best - fitted parameters @xmath269 and @xmath273 and the range of the jet opening angles between @xmath256 and @xmath258 , we obtain @xmath279 as the beaming factor for @xmath269 and @xmath273 , respectively . our value of @xmath280 is about a factor of 2 smaller than the value obtained by @xcite . we can also infer the luminosity function from the jet opening angle distribution . from equations ( [ eq8 ] ) and ( [ eq9 ] ) , the jet opening angle distribution is @xmath281 , which is related to luminosity by the expression @xmath282 where @xmath10 is the grb apparent isotropic luminosity , and @xmath95 is the jet opening angle distribution given in equation ( [ eq12 ] ) . since the energy flux is defined as @xmath283 or @xmath284 , then the isotropic luminosity in the bursting frame is @xmath285 , so that @xmath286 . utilizing equation ( [ eq17 ] ) , the jet opening angle in terms of isotropic luminosity is given by @xmath287 substituting equation ( [ eq24 ] ) into equation ( [ eq23 ] ) , we obtain @xmath288 after utilizing the jet opening angle distribution in equation ( [ eq12 ] ) . since our best fit to the pre- redshift and opening angle samples and the redshift sample constrains the jet opening angle distribution power - law index @xmath120 between @xmath289 and @xmath253 for using sfr5 and sfr6 , respectively , our apparent isotropic luminosity function is @xmath7 within the uniform jet model . in the universal structured jet model , the luminosity function is suggested to be @xmath290 ( e.g. , * ? ? ? * ; * ? ? ? * ) ; others have suggested that the luminosity function is @xmath291 ( e.g. , * ? ? ? * ) . for our best fitting model with jet opening angles ranging from @xmath130 to @xmath292 rad , this implies bolometric @xmath13-ray luminosities in the range @xmath293 to @xmath294 ergs s@xmath295 . the results for our best fitting model are shown in figure [ fig23 ] . the fraction of high - redshift grbs in the high - redshift universe implied by our fits using sfr5 and sfr6 is estimated at 8 12% and 2.5 6% at @xmath11 and @xmath12 , respectively . if this fraction is not observed with swift , then it suggests that sfr5 and sfr6 do not continue with positive evolution @xmath296 . indeed , it is likely that the formation rate of grbs at high redshift declines at sufficiently high redshift . we can also derive the fractional number of grbs with @xmath297 from figure [ fig23 ] . we find that @xmath298% of long duration grbs should occur in the low - redshift ( @xmath299 ) universe . in this work , we considered whether the differences between the pre- and redshift distributions can be explained with a physical model for grbs that takes into account the different flux thresholds of grb detectors . the model presented here parameterizes the jet opening angle distribution for an assumed flat @xmath31 spectrum , and finds best fit values for the @xmath13-ray energy release for different functional forms of the comoving rate density of grbs , assuming that the properties of grbs do not change with time . adopting the uniform jet model , we assumed that the energy per solid angle is roughly constant within a well - defined jet opening angle . the pre- redshift sample suggests that the range of the jet opening angle is between @xmath300 and @xmath131 radians . we explored three different ranges of jet opening angle with @xmath301 , and @xmath257 , @xmath131 , and @xmath302 radians . the results in 3 show that an intrinsic distribution in the jet opening angles and single values for all other parameters yields a good fit to the pre- and redshift samples , and furthermore provides an acceptable fit to the distribution of opening angles measured with pre- grb detectors . a good fit was only possible , however , by modifying the hopkins & beacom sfr to provide positive evolution of the sfr history of grbs to high redshifts ( sfr5 and sfr6 ; see fig . [ fig4 ] ) . the best fit values were obtained with @xmath303 ergs , @xmath301 , @xmath258 radians , and @xmath304 . the best - fit value for the absolute @xmath13-ray energy released by a grb is about a factor of 2 larger than the mean value of the @xcite sample of pre- grbs . the results of our fitting therefore indicate that grb activity was greater in the past and is not simply proportional to the bulk of the star formation as traced by the blue and uv luminosity density of the universe . this is contrary to the conclusion of @xcite based upon a study of batse size distribution and a small number of grbs with known redshifts in the pre - swift era . they obtain worse fits for sfrs that increase at high redshift . the batse analysis of @xcite gives results that are consistent with the required monotonically increasing grb sfr that we find . @xcite also conclude that a rising grb sfr is implied , and also note that the star formation rate can not increase in proportion to greater past grb activity , as this would result in overproduction of metals . changes in the progenitor population or grb properties , for example , due to changing metallicities with time , might account for the large number of high - redshift grbs observed with swift , but this would introduce a large number of unconstrained parameters . moreover , one would then have to explain why the amati and ghirlanda correlations relating spectral properties and energy releases are insensitive to redshift . assuming that the grb properties do not change with time , we therefore find that the grb source rate density must display positive evolution to at least @xmath165 . why would grb activity be greater in the past ? one possibility suggested by recent studies @xcite on the metallicities of host galaxies of grbs is that grbs derive from progenitor stars with a lower metallicity than the progenitors of type ii sne . in this case , one might expect that the peak of grb activity would occur earlier in the universe than during the peak of star formation activity at @xmath305 3 . nevertheless , the high - redshift fraction of grbs detected with swift will test the validity of sfr5 and sfr6 used to model the grb data . the models that fit the data imply that 8 12% of swift grbs occur at @xmath165 , in accord with the data as shown in figure [ fig23 ] . our model predicts that 2.5 6% of grbs should be detected from @xmath12 . if no grb with @xmath306 is detected by the time @xmath307 swift grbs with measured redshifts are found , then we may conclude that the grb formation rate has begun to decline above @xmath308 . by contrast , if a few @xmath309 grbs have been detected within the next year or two with swift , then this would be in accord with our model and would support the conjecture of @xcite that a second episode of population i and ii star formation took place at @xmath165 , possibly even including some grbs that originate from population iii stars . our model parameters suggest that grbs can be detected with swift to a maximum redshift @xmath310 , even given the limits placed on the range of jet angles , in particular , @xmath311 . the relationship between the galaxies that host grbs and their metallicities remains controversial , with evidence for high - redshift , low - metallicity grb host galaxies @xcite countered by examples of grbs found in galaxies with moderate metallicity @xcite . @xcite find that 5 grbs , or several per cent of the grb population with known redshifts , are found at low ( @xmath312 ) redshift in galaxies that have very low metallicity compared with solar metallicity and with the distribution of metallicities of low - redshift galaxies in the sloan digital sky survey . in 4 out of the 5 low-@xmath0 grbs , the apparent isotropic energy release of these grbs is much smaller than @xmath313 ergs , and so these grbs are unrepresentative of typical long duration grbs . moreover , the large number of these low redshift grbs already means that they are not typical of the sample of grbs considered in this paper , as the model prediction is that @xmath314 0.2% of the classical " grbs should be detected at low redshifts.% can be estimated by considering the fractional volume within @xmath315 compared with @xmath316 , and that the sfr activity at @xmath317 was a factor @xmath318 greater than at @xmath319 . ] these low - redshift grbs may also belong to a completely distinct population of grbs compared to their high - redshift counterparts , as also suggested by the differing luminosity functions of the two populations @xcite . alternatively , the low - redshift grbs could be normal grbs that appear weak because , for example , their jetted emission is observed off - axis ( e.g. * ? ? ? although we have restricted our treatment to the long - duration , soft - spectrum class of grbs , the separation between the short hard and long soft class of grbs is not distinct . moreover , recent observations of nearby grbs show that the nearby ( @xmath320 ) grb 060614 , a long duration grb in terms of its light curve @xcite , lacks a supernova excess @xcite , does not fit neatly into either the short hard or long soft classes @xcite , and could represent a separate population of grbs @xcite . phenomenological studies have also indicated that the long grb population is bimodally distributed in terms of the autocorrelation function of the light curves ( e.g. , * ? ? ? @xcite argues for multiple populations by examining the size distributions of different groups of batse grbs separated according to spectral properties . from an analysis of durations of batse , @xcite identifies an intermediate duration population that displays the softest spectra and comprises @xmath321% of the total batse grbs . that we were able to obtain good fits to the different data sets using a single grb model may indicate that the contribution of sub - populations to the total number of grbs is small . alternately , the model parameters might have been adjusted to take into account the aggregate long duration grbs . this can be tested in future work using a larger swift data set that allows subgroups to be separately studied based on spectra or durations . figure [ fig24]a shows the model integral size distribution ( see eq . [ [ eq18 ] ] ) of grbs predicted by our best fit model . the plots are normalized to the current total number of observed grbs per year from batse , which is 550 bursts per @xmath322 sr exceeding a peak flux of 0.3 photons @xmath323 in the @xmath324 kev band for the @xmath325 s trigger time @xcite . from the model size distribution , we find that @xmath326 to 360 grbs per year should be detected with a batse - type detector over the full sky above an energy flux threshold of @xmath327 , or photon number threshold of @xmath328 . this range of values is determined by the two model fits with sfr6 and sfr5 , respectively . we estimate from our fits that @xmath329 1370 grbs take place per year per @xmath330 sr with a flux @xmath331 , or @xmath332 . the field of view of the bat instrument on swift is 1.4 sr @xcite , implying that swift should detect @xmath333 150 grbs per year ; currently , observes about 100 grbs per year . this minor discrepancy may be a consequence of using a flat @xmath31 grb spectrum in our model . we also show the differential size distribution of batse grbs from the fourth batse catalog @xcite in comparison with our model prediction in figure [ fig24]b . as can be seen , our model gives a good representation of the size distribution of the batse grb distribution within the statistical error bars , except for a slight overprediction of the number of the brightest grbs ( see also * ? ? ? these brightest grbs typically originate from @xmath164 , possibly suggesting a slight reduction in the number of grbs with @xmath203 compared to the grb rate densities considered . below a photon number threshold of @xmath334 in the 50 300 kev band , the observed number of grbs falls rapidly due to the sharp decline in the batse trigger efficiency at these photon fluxes ( see * ? ? ? * ) . the size distribution of the grbs will extend to much lower values , @xmath335 , and we can use our model to predict the peak flux size distribution to fit the data , noting however that the triggering criteria are more complicated than a simple rate trigger , particularly near threshold @xcite . we can derive the burst rate @xmath336 in our local universe by normalizing our best fit models to the batse results of 340 360 grbs per year full sky with peak fluxes exceeding @xmath337 ergs @xmath338 s@xmath295 . comparing this value with the rate calculated from eq . ( [ eq18 ] ) for our best fit models , we obtain local grb rate densities of @xmath339 gpc@xmath340 yr@xmath295 for sfr5 and @xmath341 gpc@xmath340 yr@xmath295 for sfr6 . since the mean volume occupied by an @xmath342 galaxy in the local universe is @xmath343 mpc@xmath340 @xcite , we obtain the local event rate in our galaxy , assumed to be a typical @xmath342 galaxy , to be @xmath344 and @xmath345 events per year per @xmath10 galaxy , for sfr5 and sfr6 , respectively . this implies that about 1 event occurs every 0.6 myrs in our galaxy . comparing this to the rate of type ib / c supernovae of @xmath9 every 360 years in our galaxy @xcite , we estimate that the rate of grbs is only @xmath346% of the rate of sn ib / c supernovae in the galaxy . @xcite find that @xmath347% of grbs could originate from sn ib / c by comparing the radio brightnesses of sn ib / c with that of sn 1998bw associated with grb 980425 . our results fall comfortably within the observational limits . we also find that about 1 out of every 20,000 sne make a grb , compared to the value of 1 out of every @xmath348 @xmath349 found by @xcite . if grbs do occur at the rate of more than once per myr in the milky way , depending on uncertain metallicity effects @xcite , then the earth will intercept the beam of a grb about once every 25 myrs , given the beaming factor @xmath350 obtained in our analysis . if a milky way grb is located at an average distance of 10 kpc , then the average fluence deposited by the grb is on the order of @xmath351 ergs @xmath338 . strong astrobiological effects begin to be important when the grb fluence is @xmath352 @xmath353 ergs @xmath338 , or when the grb occurs within @xmath9 2 kpc with its beam pointing in our direction @xcite . this will happen perhaps a few times per gyr , so that grbs could have significant effects on terrestrial evolution @xcite . the number of jet opening angles measured with follow - up observations of grbs is not yet sufficient to produce a statistically reliable sample . thus our jet opening angle distribution shown in figures [ fig20 ] and [ fig22 ] with @xmath273 and @xmath253 , for sfr5 and sfr6 , respectively , provide a prediction for . with and pre- thresholds of @xmath354 and @xmath355 , our model predicts the mean measured jet opening angle @xmath356 , and @xmath357 rad , respectively , which can be compared with the measured average value of @xmath358 for the pre- sample . thus we expect to detect more faint low - redshift , large opening angle grbs that pre- satellites could not detect ( see also * ? ? ? it is interesting to note that with a better detector sensitivity , the mean jet angle is shifted to a higher angle and not vice versa , as might be expected . this suggests that on average we expect to see more long - duration bursts with larger opening angle ( @xmath359 radian ) , meaning that the mean times for the achromatic breaks in the light curves could be longer than for the pre- grbs when one also takes into account the larger mean redshift of the sample . in conclusion , we have developed a physical model for grbs and obtained parameters to the model by fitting the grb redshift and opening angle distributions . we could only obtain statistically acceptable joint fits if the comoving rate density of grbs increases monotonically to @xmath165 . our fitting results give a rate of @xmath9 grb every 600,000 yrs in our galaxy if metallicity effects do not play a large role in the galaxies in which grbs are formed . in this case , the increase in the grb rate density to high redshifts would have a different cause , for example , an episode of enhanced star formation at @xmath360 @xcite . the opening angle distribution of grbs measured with , and the fraction of low-@xmath0 and high-@xmath0 standard long - duration grbs will test this model . we thank d. band , j. beacom , e. berger , j. bloom , and a. friedman for discussions and correspondence , and the referee for comments , corrections , and useful suggestions . t. l. is funded through nasa _ glast _ science investigation no . dpr - s-1563-y . the work of c. d. d. is supported by the office of naval research . the differential number of bursts per unit proper time per unit volume with beaming - corrected absolute @xmath13-ray energy release between @xmath361 and @xmath361 + @xmath362 , jet opening cosine angle @xmath363 between @xmath364 and @xmath365 , peak energy between @xmath366 and @xmath367 , burst duration between @xmath368 and @xmath369 , that occur in volume element between @xmath370 and @xmath371 within observer angle @xmath372 between @xmath111 and @xmath373 during time interval between @xmath374 and @xmath375 is given by @xmath376 here the parameter @xmath377 represents the various parameters of the problem , namely @xmath361 , @xmath364 , @xmath366 , and @xmath378 . hence , the directional burst rate per unit observer time @xmath35 is given by @xmath379 in a flat @xmath27cdm universe , the robertson - walker metric describes the line element @xmath380 for a homogeneous , isotropic expanding universe , and is given by @xmath381 \ , \label{a3}\ ] ] where @xmath382 is the dimensionless scale factor of the universe and ( @xmath383 ) are the _ comoving coordinates _ of a point at rest in the hubble flow . the curvature index @xmath384 , @xmath126 , and @xmath385 represents an open , flat , and closed universe model , respectively , where @xmath386 is the radius of curvature at the present epoch . at the present epoch , we let @xmath387 and @xmath388 correspond to the physical distance between two comoving points . the ratio of the scale factor , @xmath389 , at the emitted time , @xmath374 , to the scale factor @xmath390 , at the present time , @xmath391 , is given by , @xmath392 the following relations hold from the above definition , namely @xmath393 where the last expression applies to photons and relativistic particles in an expanding universe with @xmath394 , and @xmath395 . to calculate the differential volume element @xmath396 in equation ( [ a2 ] ) , we utilize the fact that the product of differential length elements corresponding to infinitesimal changes in the coordinates @xmath388 , @xmath397 , and @xmath398 equals the differential volume element @xmath399 . hence , the proper volume element of a slice of the universe at time @xmath374 is , from eq . ( [ a3 ] ) , @xmath400 specializing to a flat universe , where @xmath401 represents the solid angle at the emitted time . substituting equation ( [ a5 ] ) into equation ( [ a2 ] ) for @xmath396 , we obtain @xmath402 where in a homogeneous isotropic expanding universe with no intervening gravitating matter to distort the paths of the light rays , @xmath403 . light travels along a null geodesic , @xmath404 , which is a path along which @xmath397 and @xmath398 are constants . thus in the robertson - walker metric the proper time element elapsed for a photon to propagate a differential comoving distance at earlier time @xmath374 is just @xmath405 or @xmath406 . upon substituting @xmath407 into equation ( [ a6 ] ) we obtain , @xmath408 where we have also expressed everything in terms of redshift z by transforming from the variable @xmath374 to @xmath0 . an expression for @xmath409 is derived by recalling the relationship between energy flux @xmath410 and luminosity distance @xmath411 , namely , @xmath412 or @xmath413 where the area element @xmath414 of the robertson - walker metric is @xmath415 hence @xmath416 where we have utilized equation ( [ a4 ] ) and use @xmath417 . substituting equation ( [ a9 ] ) into equation ( [ a8 ] ) for @xmath414 , we obtain @xmath418 for a flat @xmath27cdm universe @xcite , @xmath419 where @xmath420 and @xmath421 are the ratios of the energy densities of total mass , dominated by dark matter , and dark energy , respectively , compared to the critical density . substituting equations ( [ a10 ] ) and ( [ a11 ] ) into equation ( [ a7 ] ) gives @xmath422 for randomly oriented grb sources . after using the relation that the proper density @xmath423 increases with redshift @xmath424 in the comoving density @xmath425 , that is @xmath426 , we obtain the result given in equation ( [ eq8 ] ) . + .20 in -0.85 in , @xmath427 , @xmath131 , and @xmath428 , respectively . the right column has results as in the left column but for data . in each panel , each curve represents the cumulative redshift distribution for different jet opening angle power - law indices @xmath120 , where @xmath252 , @xmath253 , @xmath254 , and @xmath255 , from far left to far right , respectively . the assumed gamma - ray energy is @xmath429.,title="fig:",width=768 ] -1.8 in + .2 in -0.75 in and @xmath432 . from top to bottom , the assumed gamma - ray energies are @xmath433 , @xmath434 , and @xmath251 , respectively . each curve represents the cumulative redshift distribution for different power - law indices @xmath120 , where @xmath252 , @xmath253 , @xmath254 , and @xmath255 , from far left to far right , respectively.,title="fig:",width=576 ] -0.6 in + .20 in -0.85 in , @xmath239 , and @xmath251 , and an assumed range of the jet opening angles between @xmath256 and @xmath435 . the left and right columns show the model distributions for pre- and data , respectively . in each panel , each curve represents the cumulative redshift distribution for different indices @xmath120 , where @xmath252 , @xmath253 , @xmath254 , and @xmath255 , from far left to far right , respectively.,title="fig:",width=768 ] -1.8 in + 0.2 in , and an assumed range of the jet opening angles between @xmath256 and @xmath436 . each curve represents the cumulative opening - angle distribution for different indices @xmath120 , where @xmath252 , @xmath253 , @xmath254 , and @xmath255 , from far right to far left , respectively.,title="fig:",width=576 ] -3.45 in + 0.20 in -0.85 in to @xmath437 , @xmath131 , and @xmath428 , respectively . in the right column we have the same thing as in the left column but for data . in each panel , each curve represents the cumulative redshift distribution for indices @xmath120 , where @xmath252 , @xmath253 , @xmath254 , and @xmath255 , from far left to far right , respectively . the assumed gamma - ray energy is @xmath438.,title="fig:",width=768 ] -1.8 in + 0.20 in -0.85 in to @xmath257 , @xmath131 , and @xmath428 , respectively . in the right column we have the same thing as in the left column but for the data . in each panel , each curve represents the cumulative redshift distribution for indices @xmath252 , @xmath253 , @xmath254 , and @xmath255 , from far right to far left , respectively . the assumed gamma - ray energy is @xmath439.,title="fig:",width=768 ] -1.8 in

we study finite - frequency quantum noise and photon - assisted electron transport through a wide and ballistic graphene sheet sandwiched between two metallic leads . the elementary excitations allow as to examine the differences between effects related to fabry - prot like interferences and signatures caused by correlations of coherently scattered particles in electron- and hole - like parts of the dirac spectrum . we identify different features in the current - current auto- and cross - correlation spectra and trace them back to the interference patterns of the product of transmission- and reflection amplitudes which define the integrands of the involved correlators . at positive frequencies the correlator of the auto - terminal noise spectrum with final- and initial state associated to the measurement terminal is dominant . phase jumps occur within the interference patterns of corresponding integrands , which also reveal the intrinsic energy scale of the two - terminal graphene setup . the excess noise spectra , as well as the cross - correlation ones , show large fluctuations between positive and negative values . oscillatory signatures of the cross - correlation noise are due to an alternating behavior of the integrands .

1203.2010

ballistic electron transport @xcite in two - terminal graphene systems is in the focus of intensive studies ever since the pioneering experiments on single - layer carbon . @xcite the dirac hamiltonian @xcite describes charge transport close to the charge - neutrality point and leads to a linear dispersion relation @xmath0 . this allows to observe several relativistic phenomena in solid - state system , such as klein tunneling @xcite or the zitterbewegung . @xcite in the very early works on graphene the minimal conductivity @xcite @xmath1 per valley and pseudo - spin at the charge - neutrality point has been found and stimulated the research on current and noise properties . the current - current correlations around the minimal conductivity lead to a zero frequency sub - poissonion fano factor with a maximal value of @xmath2 , @xcite remarkably similar to diffusive systems as disordered metals . @xcite the suppression of the fano factor below the poissonian value originates from noiseless , open quantum channels that are found at all conductance minima in graphene - based two - terminal structures @xcite , and can be explained as an interplay between klein tunneling , resonant tunneling and pseudo - spin matching . this pseudo - diffusive behavior @xcite is due to the special band - structure of graphene . without impurity scattering , coherent transport through such a graphene sheet @xcite gives rise to the same shot noise as in classical diffusive systems . the opening of a gap @xcite in the quasiparticle spectrum leads to an enhanced fano factor . @xcite such a gap can be opened for example by photon - assisted tunneling , as shown recently for the case of a graphene p - n junction @xcite with a linear potential drop across the interface . @xcite there , landau - zener like transitions stimulated by photon emission or absorption via resonant interaction of propagating quasiparticles in graphene with an irradiating electric field lead to hopping between different trajectories . + ) sandwiched between two heavily - doped , metallic graphene leads . the fermi - level of the sheet can be tuned by a center gate voltage @xmath3 . electron- and hole states are injected via the time - dependent bias - voltages in left- and right leads @xmath4.,title="fig : " ] + the scattering approach as put forward by landauer and bttiker @xcite has been applied to ac - driven charge transport @xcite through a metal - graphene interface with an abrupt potential change . @xcite the metal can be formed by a graphene lead strongly electrostatically doped by a gate potential , thus shifting the dirac point far away from the fermi energy . in this work we adopt the formalism and parameterization introduced in refs . @xcite and calculate the finite - frequency current - current correlations at finite dc- and ac - bias voltages in the system depicted in fig . we complement recent results on ac - transport in fabry - prot graphene devices of ref . @xcite , in which the influence of different boundary conditions , i.e. zigzag or armchair configurations , on the fabry - prot patterns in a combined tien - gordon / tight - binding approach has been investigated . the influence on transmission properties of a time - dependent potential barrier in a graphene monolayer has been investigated . @xcite in our work the transverse boundary effects are central and we assume so - called infinite mass boundaries @xcite describing a short but wide ( @xmath5 ) graphene strip . we focus on the interplay between the dirac - spectrum with the fabry - perot interferences . interestingly , the well - known oscillations as function of gate voltage on a scale of the return frequency @xmath6 , related to the length @xmath7 of the graphene sheet , can be seen as a reminiscence of zitterbewegung . @xcite the role of the complex reflection amplitude and the onset of contributions of scattering states coming from terminal @xmath8 and being scattered into terminal @xmath9 will be the key characteristics in our discussion of the results for the noise as function of bias voltage and frequency . as a consequence of these onsets the oscillations add up de- or constructively depending on the precise values of voltage and frequency . in our setup , the separation of oscillations caused by the fabry - prot reflections and effects caused by the band - structure of the dirac hamiltonian is a priori not obvious . in both cases phase - coherent transport is essential . however , for charge injection either into the conduction or the valence band only , effects like zitterbewegung should not be present and all oscillating features of the noise spectra have to be of fabry - prot nature . the ballistic graphene @xcite sheet considered in the following can be described by the two - dimensional dirac equation for the two - component spinor @xmath10 with indices referring to the two pseudo - spins of the carbon sub - lattices . throughout this work we will neglect inter - valley scattering and coulomb interactions . we only consider the interaction of the electrons with the radiation field in the form of photon - assisted transitions . with fermi velocity @xmath11 the dirac equation can be cast into the form @xmath12{\hat{\psi}}({\mathbf{x}},t ) \nonumber \\ & = i \hbar \partial_t { \hat{\psi}}({\mathbf{x}},t ) \ , . \label{diraceq}\end{aligned}\ ] ] the electrochemical potential @xmath13 includes static and harmonically driven potentials in the leads plus a static gate voltage in the graphene sheet . @xmath14 making use of the tien - gordon ansatz , we write the solution to the time - dependent dirac equation as a sum over pat modes : @xmath15 the advantage of this ansatz is that the scattering problem has to be solved for the time - independent case only . therefor , in terminals @xmath16 we define stationary solutions @xmath17 by the equation @xmath18{\hat{\psi}}_{\mathrm{0}}({\mathbf{x } } ) \\ & = \epsilon { \hat{\psi}}_{\mathrm{0}}({\mathbf{x } } ) \ , . \label{diraceq2}\end{aligned}\ ] ] as a function of energy and transverse momentum q. ] the basis states in graphene can be constructed as a superposition of left- and right movers , @xmath19 \ , .\ ] ] @xmath20 describes the angle between the momentum of a quasiparticle and it s y - component @xmath21 in region @xmath22 of the graphene sheet . then the pseudo - spinors can be parametrized as @xmath23 here the dispersion is given by @xmath24 . the wave vector @xmath25 and the angle @xmath20 are defined as @xmath26 therewith , and neglecting transverse momentum due to high doping , we have the basis states @xmath27 in the leads . additionally shifting the fermi surface of the graphene sheet away from the dirac point , and thus changing the concentration of carriers , is incorporated into the gate voltage @xmath28 . for @xmath29 we have evanescent modes , @xcite with imaginary @xmath20 and @xmath25 . otherwise we have propagating modes and scattering is only at @xmath30 . + irradiating the two - terminal structure with a laser @xcite can be described by a harmonic ac - bias voltage with driving strength @xmath31 as discovered in the pioneering paper by tien and gordon @xcite . their theory can be incorporated into the scattering formalism @xcite and we are applying it here to the two - terminal graphene structure . we take the two valleys and two pseudo - spin states of the carbon lattice into account in the pre - factor of the current operator of reservoir @xmath32 , which reads @xmath33 indices @xmath34 run over reservoirs @xmath35 . summation over all modes of y - momentum is replaced by an integral since @xmath36 . scattering is contained within the current matrix @xmath37 of a current between leads @xmath38 and @xmath39 measured in lead @xmath32 via the energy - dependent scattering - matrix as a function of dc - voltage applied across the two - terminal setup . we show curves for various ac - driving strengths @xmath8 applied to a ) both reservoirs ( @xmath40 ) and b ) to the left reservoir only ( @xmath41 ) with @xmath42 . , title="fig : " ] + as a function of dc - voltage applied across the two - terminal setup . we show curves for various ac - driving strengths @xmath8 applied to a ) both reservoirs ( @xmath40 ) and b ) to the left reservoir only ( @xmath41 ) with @xmath42 . , title="fig : " ] @xmath43 the scattering matrix connects in- and outgoing scattering states at the two barriers and is calculated in appendix [ app : bounds ] by matching the wave functions at @xmath30 . here we write the results for reflection and transmission amplitudes in an alternative version : @xmath44 we assume identical scattering for quasi - particles incident from left and right , so @xmath45 and @xmath46 . @xmath47 vanishes if @xmath48 , with integer @xmath49 . the corresponding modes in @xmath50-direction are determined by @xmath51^{1/2}\ , , \label{qjumps}\ ] ] giving rise to special features of the current fluctuations , going along with the phase jumps of @xmath52 in @xmath47 we discuss later on . at the dirac point transmitted quasi - particles at perpendicular incidence perform klein tunneling via evanescent modes , leading to finite transmission probability @xmath53 at small transverse momentum , see fig . + since the average current has only a zero - frequency component , pat events in the conductance@xcite can only be studied by inducing photon - exchange via a time - dependent voltage as it is , for example , generated by irradiating the setup with a laser beam . different polarizations of the coupled light field lead to different ac - driving in left and right leads . such an asymmetry can be described by a parameter @xmath54 $ ] which varies the driving in the leads via @xmath55 . we call the driving symmetric ( in the amplitudes @xmath56 ) if @xmath40 and asymmetric if @xmath57 . for clearness we will only discuss @xmath58 since intermediate values are just a mixture of those limiting cases . for arbitrary @xmath59 the differential conductance can be derived from eqn.([currentop ] ) by taking the statistical- and time average and differentiating with respect to voltage . at @xmath60 it reads @xmath61 different orders @xmath62 of pat do not mix but have to be summed up resulting in independent contributions @xmath63 to differential conductance . since @xmath64 only depends on the besselfunctions squared , these pre - factors will always be positive . the influence of the driving strength @xmath8 on conductivity @xmath65 as a function of dc - bias is plotted in fig . pat events lead to a substantial enhancement of the conductivity around zero dc - bias , because more channels are available in comparison to the case without time - dependent voltages . at large dc - bias voltages this effect gets negligible since the transmission probability of the graphene sheet , see fig . [ fig2 ] , is not vanishing at large energies . thus , those contributions built a dominant background . conductance at arbitrary dc- and ac - bias is a sum of two integrated transmission probabilities , where the integrand exhibits crossings of the two independent interference patterns , as in region iii@xmath66 in fig . [ fig4 ] a ) . each @xmath67 shows a transition from a region with an oscillating , but in average not increasing contribution to conductance for dc - bias voltages @xmath68 , to a regime with a linear increasing background at larger dc - bias voltages . the photon - energy @xmath69 introduces a phase shift in the oscillations of @xmath63 as a function of dc - bias voltage , so for different @xmath62 we can have local minima or maxima at @xmath70 . after summation , conductivity can also show a local minimum or maximum at @xmath70 , as it can be observed for the various values of @xmath8 in fig . [ fig3 ] a ) . if @xmath71 tends to one this effect is hidden behind the contribution from the terminal where driving gets small , as in fig . [ fig3 ] b ) with @xmath41 . from the oscillations with period proportional to @xmath7 , we expect no measurable effect on conductivity or shot - noise @xcite , as in the case without ac - driving and for the zero - frequency fano factor . in the scattering approach they are simply because the transmission function oscillates as a function of energy . but imperfections of real samples , as impurities @xcite or lattice - mismatch , lead to scattering events . due to this randomizing effect on the path - lengths for propagating quasi - particles the calculated oscillations are averaged out in experiment @xcite . to get full informations on current - current correlations we study the non - symmetrized noise - spectrum as it can be detected by an appropriate measurement device in the quantum regime . @xcite + we allow harmonic ac - driving @xmath72 in the leads , so in fourier space the current - current correlations are defined as @xmath73 the non - symmetrized shot noise correlates currents at two times : @xmath74 with variance @xmath75 . of experimental interest are the fluctuations on timescales large compared to the one defined by the driving frequency @xmath76 . thus , as in @xcite , we introduce wigner coordinates @xmath77 and @xmath78 and average over a driving period @xmath79 . then , the noise spectrum is defined by the quantum statistical expectation value of the fourier - transformed current - operator @xmath80 via @xmath81 . @xmath82 is nothing but the fourier transform of @xmath83 . similarly , in the case without ac - driving the noise is only a function of relative times @xmath84 . in order to keep notation short , in the dc - limit we write @xmath85 . to get a deeper insight into the underlying processes of charge - transfer we split the noise into four possible correlators @xcite , defined by @xmath86 the correlators itself can be seen as the building - blocks of noise spectra where different combinations describe noise detected by corresponding measurement setups . @xcite first we discuss @xmath87 , the case when no ac - driving is present . we also skip @xmath76 in the arguments of the correlators . then evaluation of eqn . ( [ noisedef ] ) at @xmath60 leads to the expressions : [ correlators ] @xmath88 at finite dc - bias voltages correlations with initial and final state related to the measurement terminal @xmath7 are special in the sense that they can not be written in terms of probabilities at finite frequency . for symmetrized noise , bttiker @xcite discussed the essential role of the complex reflection amplitudes in elastic electron transport and how they determine the equilibrium current fluctuations . in the quantum regime at @xmath60 , the equilibrium fluctuations are given by @xmath89 for finite dc - bias the reflection amplitudes entering @xmath90 play the same essential role as in equilibrium , in the sense that finite - frequency current fluctuations are non - zero even for vanishing transmission . the combination of scattering - matrices of the correlators integrands which enter in the current - current cross - correlation spectrum @xmath91 are substantially different than in the ones for the auto - terminal noise . most of all , at finite frequency none of the complex correlators can be written as an integral over transmission- or reflection probabilities : [ correlatorscross ] @xmath92\label{correlatorscross : ltol } \\ c^{\mathrm{c}}_{r \rightarrow r}(\omega)= & \frac{e^2 \theta(\hbar \omega ) } { 2 \pi \hbar } \underset{\mu_r-\hbar \omega}{\overset{\mu_r}\int } d\epsilon \underset{-\infty}{\overset{\infty}\int } dq \ , \nonumber \\ & t^*(\epsilon ) t(\epsilon+\hbar\omega ) \left[1- r^*(\epsilon+\hbar\omega)r(\epsilon ) \right]\label{correlatorscross : rtor}\\ c^{\mathrm{c}}_{l \rightarrow r}(\omega)= & \frac{-e^2 \theta(\hbar \omega - ev)}{2 \pi \hbar } \underset{\mu_r-\hbar \omega}{\overset{\mu_l}\int } d\epsilon \underset{-\infty}{\overset{\infty}\int } dq \,\nonumber \\ & r^*(\epsilon)t(\epsilon ) r^*(\epsilon+\hbar\omega)t(\epsilon+\hbar\omega)\label{correlatorscross : ltor}\\ c^{\mathrm{c}}_{r \rightarrow l}(\omega)= & \frac{-e^2 \theta(\hbar \omega+ev)}{2 \pi \hbar } \underset{\mu_l-\hbar \omega}{\overset{\mu_r}\int } d\epsilon \underset{-\infty}{\overset{\infty}\int } dq \ , \nonumber \\ & t^*(\epsilon)r(\epsilon)t^*(\epsilon+\hbar\omega)r(\epsilon+\hbar\omega)\label{correlatorscross : rtol}\end{aligned}\ ] ] unlike for symmetrized noise , quantum noise @xcite spectra discriminate between photon absorption ( @xmath93 ) and emission ( @xmath94 ) processes between quasi - particles in graphene and a coupled electric field @xcite . energy for photon emission has to be provided by the voltage source , so at @xmath60 the heaviside - theta functions ensure that only terms satisfying this condition contribute at negative frequencies . in the dc - limit , our choice of chemical potentials @xmath95 and the fact that the measurement is performed at reservoir @xmath7 , leaves only @xmath96 if @xmath97 . when additional ac - voltages are present none of the correlators of eqn.([thenoise ] ) is given in terms of probabilities and integration boundaries are changed by @xmath98 . then all correlators can contribute at frequencies @xmath99 . ) contributing to the shot - noise , namely a ) @xmath100 , b ) @xmath101 , c ) @xmath102 and d ) @xmath103 . here the energy is fixed @xmath104 corresponding to vanishing dc - bias . the correlator in a ) can not be written in terms of a probabilities , except in the zero frequency limit the integrand results in @xmath105 . correlator b ) contains one transmission probability at zero energy that is only non - zero at small @xmath21 . since for small transversal momentum @xmath106 decays as @xmath107 the correlator c ) tends to zero in this regime and otherwise mimics the behavior of @xmath108 . integrand d ) is also restricted to low transverse momentum because @xmath109 otherwise . + ] , see also fig . [ fig5 ] ) contributing to the shot - noise for fixed energy @xmath110 . at finite @xmath111 there is an additional interference pattern along @xmath21 if @xmath112 , leading to phase jumps in the integrand of correlator a ) , the one where initial and final state belong to the measurement terminal @xmath7 . when the integrand can be written as a product of probabilities , see b)-d ) , the phase jumps are absent but two independent interference patterns are found . ] ) , see also fig . [ fig5 ] ) contributing to the shot - noise with fixed frequency @xmath113 . analogous to fig.[fig6 ] but as a function of @xmath114 . phase jumps occur in the intervall @xmath115 in integrand a ) , region iii@xmath116 of fig.[fig4]a ) . the interplay of the two interference patterns can also be observed at larger energies and transverse momenta for @xmath117 , @xmath118 in all integrands a)-d ) . ] ) contributing to the cross - correlation shot - noise for fixed energy @xmath110 ( top ) and fixed frequency @xmath113 ( bottom ) , namely a),c ) @xmath119 $ ] and b),d ) @xmath120 $ ] . due to symmetry reasons the integrands are identical when interchanging index labels @xmath35 . as a function of frequency integrand a ) leads to strongly oscillating contributions to the noise spectrum . these oscillations are reduced due to the alternating behavior along @xmath21 in cross - terminal contributions b ) . in c ) , d ) the integrands are plotted as a function of @xmath114 where they reveal a similar structural difference . ] ) contributing to the shot - noise for driving frequency @xmath121 and fixed frequency @xmath113 . the integrands are a ) @xmath122 , b ) @xmath123 , c ) @xmath124 and d ) @xmath125 . when two frequencies are present none of the correlators can be written in terms of probabilities and additional phase jumps come into play . ] and fixed energy @xmath110 . analogous to fig . ( [ fig9 ] ) , but as a function of @xmath126 ) . as a consequence of pat horizontal interference lines occur for transverse momenta @xmath127 as in figures a ) , c ) . ] a good starting point to interpret results for conductivity and shot - noise spectra is to examine the involved integrands in eqs . ( [ correlators ] ) and ( [ correlatorscross ] ) . figure [ fig4 ] provides a schematic overview of the different regions occurring in the 2d - plots of figs . [ fig5]-[fig10 ] . we show the real parts of integrands either as a function of @xmath114 as in scheme [ fig4 ] a ) or of @xmath128 as in scheme [ fig4 ] b ) . the former is divided by the four envelopes @xmath129 and @xmath130 into six areas : i , where the regimes ii@xmath131 and ii@xmath116 of evanescent modes are merging and the areas iii@xmath131,iii@xmath131 , iii@xmath131 of propagating modes . area iii@xmath116 is defined by the two lines with origins @xmath132 ) , ( @xmath133 and intersection @xmath134 . areas in scheme [ fig4]b ) are separated by @xmath135 and the dashed horizontal line @xmath136 . the transmission probability fits into this scheme when the horizontal separation is absent so we are left with areas 1@xmath131 and 2@xmath137 . then area 1@xmath131 includes the black region of fig . [ fig2 ] where no transmission is possible , and the regime of evanescent modes with finite transmission probability for small @xmath138 around @xmath139 due to klein tunneling . in regimes 2@xmath137 a hyperbolic shaped interference pattern with oscillations along @xmath111 is prominent , where the period of oscillations is on the order of @xmath140 for small @xmath141 . figure [ fig5 ] shows the relevant integrands of the four correlators @xmath142 contributing to the finite frequency quantum noise , plotted as a function of @xmath128 when @xmath139 . then the imaginary part of @xmath143 leads to finite contributions in the region i@xmath131 and i@xmath116 in figure [ fig5]a ) . @xmath144 is only non - zero for small @xmath21 , so integrands b ) and d ) vanish for large @xmath21 . since @xmath145 , integrand c ) vanishes when @xmath146 and otherwise resembles the shape of @xmath108 . + finite @xmath111 , as in fig . [ fig6 ] , introduces another interference pattern for propagating modes . in region 1@xmath131 non - zero values are possible and in 2@xmath131 and 2@xmath116 the usual interferences occur . for @xmath21-values below @xmath147 this additional pattern can be seen in region 1@xmath116 . the interplay of both patterns leads to phase jumps of @xmath52 in regions 3@xmath131 and 3@xmath116 . these phase jumps can be determined by requiring @xmath148 in eq.([correlators : ltol ] ) , fig . [ fig6]a ) . therefor @xmath149 has to vanish , what is fulfilled by the transversal momenta of eq . ( [ qjumps ] ) . the condition @xmath150 for a maximum in the integrand leads to modes which experience klein tunneling . actually , this correlator can be written as integral over @xmath151^{1/2 } \cos(\phi(\epsilon,\omega))$ ] including a scattering - phase @xmath152 $ ] . thus it describes events containing the scattering - phase between time - reversed paths of electron - hole pairs separated by the photon energy @xmath153 reflected back into the measurement terminal . the effect of the phase shifts on the integrands interference patterns is also obvious in the @xmath114-plot of fig . [ fig7 ] a ) , region iii@xmath116 . figs . [ fig7 ] b)-d ) show a similar interference pattern although the corresponding correlators are defined in terms of probabilities . + concerning cross - correlation noise , the integrands occurring in eqs.([correlatorscross ] ) show alternating patterns of positive and negative values . the ones which describe auto - terminal contributions to @xmath154 ( eqs.([correlatorscross : ltol ] ) and ( [ correlatorscross : rtor ] ) ) , as in fig . [ fig8]a ) , have an alternating sign along @xmath155 . in the cross - terminal ones ( eqs.([correlatorscross : ltor ] ) and ( [ correlatorscross : rtol ] ) ) , as in fig . [ fig8]b ) , the additional interference pattern along @xmath21 introduces another change of sign . plots [ fig8 ] c ) and d ) show a similar behavior as functions of @xmath114 . when ac - bias voltages introduce the driving frequency @xmath76 , the integrands structures become even richer but also less clear , as in fig . [ fig9 ] and fig . [ fig10 ] . then alternating signs in all contributions to auto - correlation noise are observed , except for the correlator with initial and final sates in the measurement terminal . this results in peculiar oscillatory features in the interference patterns at combinations of all involved energies @xmath156 . predicting the effect of such features on the noise spectra from the plotted integrands is then almost impossible because one still has to average over all possible energies and @xmath21-values by integration . in contrast to conductivity , the shot - noise spectrum in general couples different orders of pat events , expressed by the product of four besselfunctions of arbitrary order . but since the driving is fixed , non - vanishing contributions exist only up to a certain order depending on the precise value of @xmath8 . when time - dependent voltages are present , current fluctuations of eq . ( [ thenoise ] ) contain products of four scattering matrices , each with a different energy argument . after performing the dc - bias limit only transitions between @xmath111 and @xmath157 are left . . we compare a setup where dc - bias voltages are fixed symmetrically around the dirac point ( top , ) with the case when @xmath158 ( bottom ) . thick lines : shot - noise and correlators . thin lines : derivatives with respect to frequency . contributions from @xmath90 are dominant at positive frequencies . top : special features in the derivatives are seen for frequencies @xmath159 in the @xmath160 contribution , when the lower bound of the energy - integration interval approaches the dirac point ( compare to figs . [ fig6 ] , [ fig7 ] ) . bottom : the distance to the dirac point is increased by the offset voltage . therefore oscillatory features appear in a larger frequency interval and in all four correlators , since integration boundaries in all contributions are crossing the dirac - point with increasing @xmath155.,title="fig : " ] + . we compare a setup where dc - bias voltages are fixed symmetrically around the dirac point ( top , ) with the case when @xmath158 ( bottom ) . thick lines : shot - noise and correlators . thin lines : derivatives with respect to frequency . contributions from @xmath90 are dominant at positive frequencies . top : special features in the derivatives are seen for frequencies @xmath159 in the @xmath160 contribution , when the lower bound of the energy - integration interval approaches the dirac point ( compare to figs . [ fig6 ] , [ fig7 ] ) . bottom : the distance to the dirac point is increased by the offset voltage . therefore oscillatory features appear in a larger frequency interval and in all four correlators , since integration boundaries in all contributions are crossing the dirac - point with increasing @xmath155.,title="fig : " ] ( thick lines , upper panels ) and derivatives ( thin lines , lower panels ) with dc - bias symmetrically applied around the dirac point ( top ) and for finite @xmath161 ( bottom ) . by subtracting the noise at zero dc - bias the divergent background is removed . the structure and especially the oscillatory behavior are coined by auto - terminal contributions of eq.([correlatorscross : ltol ] ) related to the measurement terminal @xmath7 . the jump in the derivative of @xmath162 is present because this correlator does not contribute for frequencies @xmath163 . by applying an offset voltage @xmath164 a complicated structure emerges , best visible in the derivatives.,title="fig : " ] + ( thick lines , upper panels ) and derivatives ( thin lines , lower panels ) with dc - bias symmetrically applied around the dirac point ( top ) and for finite @xmath161 ( bottom ) . by subtracting the noise at zero dc - bias the divergent background is removed . the structure and especially the oscillatory behavior are coined by auto - terminal contributions of eq.([correlatorscross : ltol ] ) related to the measurement terminal @xmath7 . the jump in the derivative of @xmath162 is present because this correlator does not contribute for frequencies @xmath163 . by applying an offset voltage @xmath164 a complicated structure emerges , best visible in the derivatives.,title="fig : " ] in the regime @xmath165 , the scattering matrix can be treated as energy - independent . then , as for a single level quantum dot in the broad - band limit , asymmetric quantum noise as function of frequency is the sum of four straight lines , with kinks at @xmath166 . @xcite for vanishing dc - bias we have @xmath167 and @xmath168 , as long as @xmath169 . the richer regime , when @xmath170 , additionally exhibits strongly oscillating integrands . those oscillations are purely due to propagating modes as it is also clear from interference patterns of the integrands in figs . [ fig5]-[fig9 ] , regions ii@xmath171 and iii@xmath172 . in the special case of perpendicular incidence ( @xmath173 ) we have klein tunneling , thus the frequency - dependence of the correlators is linear for this mode . then @xmath174 if @xmath175 since @xmath176 otherwise the @xmath142 mirror the interference patterns of the integrands . so the noise spectrum ( fig . [ fig11 ] solid , thick curve ) shows oscillations on the scale of @xmath177 in the regime @xmath178 , similar to the shot - noise at zero - frequency as a function of gate voltage @xcite . although present in all four correlators , the oscillations show up in the noise spectrum mainly via @xmath179 of the terminal where the fluctuating currents are probed . that is because the correlator itself as well as the amplitude of the oscillations are significantly larger than for other contributions . therefore , in comparison to the absorption - branch ( positive frequencies ) the emission - branch of the spectrum ( negative frequencies ) shows only small shot - noise . indeed all correlators except @xmath180 vanish when @xmath181 since the energy for the emission of a photon has to be provided by the voltage source . especially the contribution dominant at positive frequencies vanishes : @xmath182 if @xmath183 . + we are considering the limit @xmath60 where the correlators integration windows are exactly determined by the chemical potentials . at finite temperature this so - defined onsets of the four contributions as a function of frequency are smeared out by the broadening of the fermi - functions . clearly a gate voltage does not affect these onsets since it does not enter in the fermi functions of the leads , but it still changes the transmission function resulting in a modified spectrum . as a function of dc - bias for fixed frequency @xmath153 ( thick lines , upper panels ) . we compare the symmetric setup without dc - bias offset ( top ) and when @xmath164 ( bottom ) . thin lines ( lower panels ) are used for the derivatives with respect to voltage . top : due to symmetrically applied bias voltage the noise and the auto - terminal contributions are symmetric in the voltage dependence and @xmath184 if @xmath185 . bottom : by applying an offset voltage we are breaking the setups symmetry . auto - terminal terms are then symmetric with respect to @xmath186 while the summed up noise is asymmetric . , title="fig : " ] + as a function of dc - bias for fixed frequency @xmath153 ( thick lines , upper panels ) . we compare the symmetric setup without dc - bias offset ( top ) and when @xmath164 ( bottom ) . thin lines ( lower panels ) are used for the derivatives with respect to voltage . top : due to symmetrically applied bias voltage the noise and the auto - terminal contributions are symmetric in the voltage dependence and @xmath184 if @xmath185 . bottom : by applying an offset voltage we are breaking the setups symmetry . auto - terminal terms are then symmetric with respect to @xmath186 while the summed up noise is asymmetric . , title="fig : " ] those limits of energy - integration , as well as their position relative to region iii@xmath116 , result in features in the noise spectra besides the discussed oscillations . in order to clarify the role of the dirac hamiltonian in comparison to the role of pure fabry - prot interferences , we compare results when the charge injection is only in the conduction or valence band by shifting the dc - bias voltages above the fermi energy of the graphene sheet via the offset voltage @xmath187 in @xmath188 . @xmath189 can never see the regime @xmath115 when @xmath190 , as in the upper plot of fig . [ fig11 ] . thus the oscillations visible in the derivate have a well defined period over the whole spectrum on top of a linearly increasing background . when an offset voltage @xmath191 is applied , as done when calculating the spectra for the lower plot of fig . [ fig11 ] , @xmath192 shows a complicated frequency dependence for small @xmath155 . contribution @xmath90 describes correlations of scattering states emanating from the left reservoir reflected back into the same reservoir . we will discuss this contribution now in detail : special features for small frequency are due to the interplay of the integration boundaries with the various regions in fig . [ fig4]a ) occurring in the integrands @xmath114-dependence of fig . [ fig7]a ) . integration is over all @xmath21-modes and from @xmath193 to @xmath194 . when @xmath195 this corresponds to @xmath196 , regions iii@xmath116 and partly ii@xmath171 of fig . [ fig4]a ) . now at finite @xmath197 as in fig . [ fig11 ] , the integration window can include region iii@xmath116 completely , partly , or not at all , resulting in variations of the spectrum . at small @xmath153 , features in the integrands interference patterns have stronger impact . this can be seen from strongly non - harmonic features of the noise spectrum , e.g. in @xmath90 and @xmath180 for @xmath161 . for large frequencies averaging leads to nearly harmonic oscillations on top of the increasing background . with the chosen parameters the distance of the chemical potential @xmath198 to the charge - neutrality point is given by @xmath199 . around the corresponding frequency the oscillatory behavior of the spectrum is modified and flattened due to a reduced fraction of propagating modes . raising the frequency further increases this fraction again and oscillations are roughly harmonic with period @xmath52 , best visible in the derivatives @xmath200 of fig . [ fig11 ] . that is also the point where the lower bound of energy integration starts to include the special interference pattern of the integrands around the energy interval @xmath201 , region iii@xmath116 . @xmath192 is not influenced by the measurement terminal itself , but probes transmission probabilities via scattering events which are related to the right terminal only . an analogous behavior of the spectrum as before is found , this time with a distance @xmath202 of the lower integration boundary to the charge neutrality point when @xmath203 . now increasing frequency is going along with a decreasing slope of the derivative with respect to frequency until the dirac point is reached . there the slope increases again since more open channels become available . the same interpretation also explains features in the interval @xmath159 of the auto - terminal correlators shown in fig . [ fig11 ] , when @xmath204 . e.g. the spectrum of the correlator eq.([correlators : rtor ] ) , with initial and final state in the right lead , exhibits a reducing slope until @xmath205 from where on the oscillations have a well defined period . the @xmath206 curve has a maximal slope at @xmath207 when positive and negative energies with same magnitude are present . for higher frequencies oscillations have again a well - defined phase . + we also study the excess noise at finite frequencies : @xmath208 . subtracting the noise at zero bias - voltage removes the divergent contributions from the noise spectrum . then oscillating features due to bias - voltages are more obvious since they are now also prominent in the noise spectra of fig . [ fig12 ] , not only in derivatives . when @xmath190 the excess noise ( thick , black , solid curve ) is purely positive for @xmath209 while for @xmath210 it is oscillating around zero , because then cross - terminal contributions @xmath142 cancel each other up to a constant offset acquired at small @xmath155 . this offset is compensated by the @xmath211 contribution . oscillations of this contribution have again a considerable impact on the excess noise spectrum . in the lower plot of fig . [ fig12 ] the offset voltage is fixed to @xmath161 . for low frequencies @xmath159 , complicated oscillations occur in all contributions to excess noise and are accompanied by a strongly increasing slope up to frequencies @xmath212 . as for the noise itself , the frequency of the oscillations is determined by @xmath213 and equals the frequency expected from the zitterbewegung of relativistic dirac fermions @xcite . this frequency corresponds to a period of @xmath214 in our plots . it would be interesting to test experimentally if those much more pronounced oscillation , compared to the overall shot - noise , can be detected in spite of randomization effects of imperfections on the quasi - particles path lengths . in summary , i ) the impact of the dirac hamiltonian on the frequency - dependence of auto - terminal current fluctuations leads to peculiar oscillation for energies in the vicinity of the dirac point as an interplay of klein tunneling , phase - jumps in the correlators and their energy - integration limits . and ii ) oscillations due to the fp setup have a constant phase for high energies when propagating modes are dominant . then @xmath215 oscillates between positive and negative values with a period as it is expected from the effect of zitterbewegung . when @xmath190 ( top ) and @xmath216 ( bottom ) . without an offset voltage ( @xmath190 ) auto - terminal contributions are identical , as well as cross - terminal ones at large frequencies @xmath217 . at finite @xmath218 the asymmetric bias voltage is reflected in the frequency - dependence of the auto - terminal correlators by their different heights and the shift of the oscillations maxima . , title="fig : " ] + when @xmath190 ( top ) and @xmath216 ( bottom ) . without an offset voltage ( @xmath190 ) auto - terminal contributions are identical , as well as cross - terminal ones at large frequencies @xmath217 . at finite @xmath218 the asymmetric bias voltage is reflected in the frequency - dependence of the auto - terminal correlators by their different heights and the shift of the oscillations maxima . , title="fig : " ] analogous to the spectrum , the dc - bias dependence for fixed frequency is featureless in the regime @xmath165 , except the pronounced onsets of the four correlators . this is not surprising when looking at the derivatives with respect to voltage : [ derivativesv ] @xmath219 \label{derivativesv : ltol } \\ & \frac{dc_{r \rightarrow r}}{dv}= \frac{e^2 \theta(\omega ) } { 4\pi \hbar}\underset{-\infty}{\overset{\infty}\int } dq\nonumber\\ & \left[t(ev/2)t(ev/2 + \hbar \omega ) -t(ev/2-\hbar\omega)t(ev/2)\right]\label{derivativesv : rtor}\\ & \frac{dc_{l \rightarrow r}}{dv}=\frac{e^2 \theta(\omega - ev ) } { 4\pi \hbar}\underset{-\infty}{\overset{\infty}\int } dq\nonumber\\ & \left [ t(\hbar\omega - ev/2)r(-ev/2)-t(ev/2)r(ev/2-\hbar\omega ) \right ] \label{derivativesv : ltor}\\ & \frac{dc_{r \rightarrow l}}{dv}=\frac{e^2 \theta(\omega+ev ) } { 2\pi \hbar } \underset{-\infty}{\overset{\infty}\int } dq\nonumber\\ & \left [ t(ev/2)r(ev/2+\hbar\omega ) -t(-ev/2-\hbar\omega ) r(-ev/2)\right ] \label{derivativesv : rtol}\end{aligned}\ ] ] scattering amplitudes are roughly constant for a given q - mode in this regime , then correlators are straight lines as a function of dc - bias voltage . e.g. a special situation that could exhibit interesting physics is when some derivatives are zero . but this is , due to symmetry arguments , only possible at @xmath220 , proofing a zero slope of the correlators at their onsets but revealing no additional effect . by this means , as in the shot - noise spectrum , the dependence on the bias voltage reveals again the onsets of the four correlators . since we have chosen positive @xmath153 , the auto - terminal contributions are non - zero over the whole bias range . as before , cross - terminal ones vanish if no energy is provided by the voltage source : @xmath221 if @xmath222 and @xmath223 if @xmath224 . as it is clear from the bottom plot of fig . [ fig13 ] , the oscillations of the components are not in phase , thus adding up to complicated oscillations in @xmath225 . but , as mentioned in the beginning , we doubt this could be a measurable effect . the shot - noise and the auto - terminal correlators are symmetric in the voltage dependence if @xmath204 , whereas the cross terminal ones obey @xmath226 . here the charge - neutrality point and the width of the region iii@xmath116 are revealed as a minima in the slope of the correlator @xmath227 at @xmath228 and in the change of sign in @xmath229 at @xmath70 . the explicit expressions of eq.([correlatorscross ] ) for the cross - correlation current noise spectrum of fig . [ fig14 ] can be extracted from the general expression eq.([noisedef ] ) in the same way as we did when deriving eq.([correlators ] ) . from figs . [ fig8 ] a ) and b ) it is also clear that the spectrum of auto - terminal correlators are oscillating as a function of @xmath155 with larger amplitude than cross - terminal ones , since they show an alternating behavior between positive and negative integrands . dependence on @xmath21 in the relevant frequency range is weak , as shown in fig . [ fig8 ] a ) . contrary , cross - terminal contributions as in fig . [ fig8 ] b ) show features with an alternating sign along both variables , @xmath155 and @xmath21 . thus , integration along y - momentum leads to averaging and therefore significantly smaller oscillation amplitudes occur . as discussed for the excess noise of the auto - correlation noise spectral function , we find the oscillations have a frequency @xmath230 what is tantamount to a period @xmath231 in the plots . complex conjugation corresponds to time - reversed states . again , as the product of scattering matrices of the integrands in eq.([noisedef ] ) suggests , it is probing transmission- and reflection amplitudes of electron - hole pairs separated by an energy quanta @xmath153 . so , for cross - terminal noise not only the reflection but also the complex transmission amplitude is essential even without ac - bias voltages . again it would be interesting to test if the resulting oscillations could be detected in the challenging task of a finite - frequency cross - correlations experiment . analyzing the integrands reveals the symmetry @xmath232 if @xmath233 as we show in fig . [ fig15 ] . this symmetry is distorted by applying an offset voltage @xmath187 . the spectrum of the correlator @xmath234 shows a shift of the maxima and minima of the oscillations with respect to @xmath235 for finite @xmath187 . this shift is due to the fact that the distance between neighboring maxima of the integrand is not constant when varying @xmath153 at given q - mode ( see the bending of the maxima towards higher frequencies for larger @xmath21 in the integrands , e.g. fig . [ fig6 ] ) . derivatives of the correlators @xmath236 with respect to voltage show a sequence of pairs of different maxima . this observation is traced down to the same origin as above , and so the appearance of peculiar oscillations in the summed up cross - correlation shot - noise @xmath154 is explained . at @xmath203 current conservation and the unitarity of the s - matrix require @xmath237 . therefore the correlator described by eq . [ correlatorscross : rtol ] is negative . , as function of symmetrically applied dc - bias voltage and for fixed frequency . jumps in the derivatives at @xmath238 are due to the onsets of the cross - terminal contributions . ] by applying an ac - bias voltage at the leads one can inject charge - carriers at positive and negative energies of the dirac cone without applying a dc - voltage . analogous to the minimal conductivity , in the non - driven case going along with a maximal fano factor , the shot - noise at zero frequency but finite ac - bias @xmath239 mirrors the behavior of the conductivity in fig . [ fig3 ] . with respect to frequency as function of frequency . we have chosen a symmetrically applied dc - bias voltage with additional harmonic ac - driving ( @xmath240 , @xmath241 and @xmath41 ) in lead @xmath7 . dashed vertical lines mark the step - positions , coloring specifies the correlator which shows the step at corresponding @xmath242 . ] the noise spectrum fig . [ fig16 ] for the driven setup ( @xmath41 ) is similar to the one without driving but with additional steps in the derivatives . for arbitrary ac - bias these steps can appear at frequencies @xmath243 due to the onset of higher - order pat events . since we set @xmath41 in fig . [ fig16 ] , the correlator with states @xmath244 shows no ac - induced steps in the derivative . but when @xmath245 all integrands ( fig . [ fig9 ] ) are not given in terms of probabilities and can take negative values as mentioned in section [ sec : qualitative ] . for the shot - noise spectrum there are then two possible sources of contributions that could reduce noise : either a correlators integrand or the the product of besselfunctions is negative . when the driving voltage is applied symmetrically ( @xmath40 ) more pat - induced steps in the derivatives of the noise spectrum are visible and finite contributions at negative @xmath155 are possible for all correlators . as proposed by trauzettel et . @xcite a time - dependent voltage could be used to induce interference between states in particle- and hole - like parts of the dirac spectrum . this should correspond to zitterbewegung like in relativistic quantum mechanics , but we are not aware of any unique feature caused by zitterbewegung that can be distinguished from other oscillations , especially of fabry - prot nature . we have analyzed conductivity and non - symmetrized finite - frequency current - current correlations for a fabry - prot graphene structure . oscillations on the intrinsic energy scale @xmath177 are still present in the finite frequency noise . emission spectra are diverging for large frequencies , whereas the absorption branch of the spectrum has to vanish at @xmath246 . as expected from the integrands , the current - noise also diverges for voltages @xmath247 . since the onset of the different noise contributions is defined by the four possible combinations of the chemical potentials , the noise built by all correlators consists of contributions oscillating with the same period but different phases . although dominated by @xmath90 when correlating the currents in terminal @xmath7 at large frequencies , this interplay is revealed in the spectra and voltage dependence of all correlators . each contribution can show peculiar oscillations at low enough frequencies or voltages . in this regime features in the integrands @xmath114-dependence can have a prominent impact whereas they tend to be averaged out at large frequencies . another aspect is the appearance of a special region showing phase jumps in the energy dependence of the integrands when @xmath248 . this interplay of the dirac spectrum and the fabry - prot physics @xcite can be probed purely by applying an appropriate combination of dc - bias and offset voltage @xmath249 , thus connecting electron- and hole part of the dirac spectrum symmetrically when @xmath250 . the way the scattering amplitudes are combined in this approach spoils the clear picture in terms of transmission- and reflection probabilities . instead , in the dc - limit it gives rise to the interpretation of the @xmath251 contribution in terms of jumps in the scattering - phase between time - reversed electron - hole states separated by the photon energy @xmath153 . in the same way the complex correlators for cross - correlation noise or for the driven setup exhibit phase jumps and can not be written in terms of probabilities . complex contributions of the scattering matrices lead to large oscillations between positive and negative values of cross - correlation noise or in the derivatives with respect to frequency of the auto - terminal noise spectral function . these oscillations have a frequency of @xmath213 , what corresponds to a period of @xmath231 in our plots . this frequency corresponds to the zitterbewegung frequency as it is known for relativistic dirac fermions . again , strongly non - harmonic features can occur when the transition between different regimes is probed , especially when region iii@xmath116 around the dirac point comes into play . additional ac - bias complicates the picture because combinations of @xmath252 define additional phase jumps , onsets of the correlators and therefore steps in the noise when higher - order pat events occur . then the special role of the complex reflection and transmission amplitudes is essential for all possible correlators . we would like to acknowledge the financial support by the dfg ( grant no . sfb 767 ) and thank b. trauzettel for validating and sharing corrections to reference @xcite . the non - symmetrized noise spectrum under harmonic ac - driving without interactions is determined by @xmath253 f_{{\gamma}}(\epsilon - l \hbar \omega ) \left(1-f_{{\delta}}(\epsilon + \hbar \omega -k \hbar \omega ) \right ) \ , . \label{thenoise}\end{aligned}\ ] ] for the fermi distribution function in lead @xmath38 we use the shorthand @xmath254 + 1)$ ] . in the limit of @xmath60 the distribution functions in the leads are given by heaviside - theta functions @xmath255 that define the integration intervals . explicitly writing down the expression above for chosen @xmath256 then leads to the four possible contributions to auto - correlation noise of eq . ( [ correlators ] ) and to cross - correlation noise of eq . ( [ correlatorscross ] ) via summation over reservoir indices @xmath257 . as in @xcite we confine the charge - carriers along the @xmath50-directions by infinite mass boundaries that diverges at the edges @xmath258 . this corresponds to the boundary conditions @xcite @xmath259 unlike the procedure for the schrdinger equation , in graphene one only has to match the wave function itself and no constraint is given for the derivatives . now exploiting the boundary conditions along the @xmath260-direction , transmission- and reflection amplitudes are fully determined by these constraints of the field operators at the fermi levels . therefor it is sufficient to match the wave functions at @xmath30 without ac - driving . a plain wave ansatz to solve the dirac equation ( [ diraceq ] ) for an electron incident from the left ( @xmath261 ) with energy @xmath111 is given by @xmath262 where @xmath263 are the complex wave vectors in the reservoirs and and @xmath25 the one in the sandwiched graphene strip . matching conditions at boundaries ( continuity at @xmath30 ) combined with high doping in the reservoirs lead to the following set of coupled , complex equations : @xmath264 the solution is straightforward and determined by the four complex coefficients @xmath265 . the transmission @xmath266 and reflection @xmath47 amplitudes define the s - matrix and thus the current and noise of the scattering device at all voltages . @xmath267 d. a. wharam , t. j. thornton , r. newbury , m. pepper , h. ahmed , j. e. f. frost , d. g. hasko , d. c. peacock , d. a. ritchie and g. a. c. jones , j. phys . c * 21 * , l209 ( 1988 ) b. j. van wees , h. vanhouten , c. w. j. beenakker , j. g. williamson , l. p. kouwenhoven , d. vandermarel , and c. t. foxon , phys . rev . lett . * 60 * , 848 ( 1988 ) a. h. castro neto , f. guinea , n - m . r. peres , k. s. novoselov , and a. k. geim , rev . phys . * 81 * , 109 ( 2009 ) a. k. geim and k. s. novoselov , nat . mat . * 6 * , 183 ( 2007 ) p. a. m. dirac , proc . a * 117 * , 610 ( 1928 ) o. klein , z. phys a * 53 * , 157 ( 1928 ) n. dombey , and a. calogeracos , phys 315 * , 41 ( 1999 ) m. i. katsnelson , k. s. novoselov , and a. k. geim , nat . * 2 * , 620 ( 2007 ) e. b. sonin , phys . b * 79 * , 195438 ( 2009 ) a. f. young and philip kim , nat . phys . * 5 * , 222 ( 2009 ) a. v. shytov , m. s. rudner , and l. s. levitov , phys . lett . * 101 * , 156804 ( 2008 ) m. i. katsnelson , eur . j. b * 51 * , 157 ( 2006 ) m. i. katsnelson , eur . j. b * 52 * , 151 ( 2006 ) k. ziegler , phys . b * 75 * , 233407 ( 2007 ) i. snyman , j. tworzydo , and c. w. j. beenakker , phys . rev . b * 78 * , 045118 ( 2008 ) k. s. novoselov , a. k. geim , s. v. morozov , d. jiang , m. i. katsnelson , i. v. grigorieva , s. v. dubonos , a. a. firsov , nature * 438 * , 197 ( 2005 ) y. zhang , y .- w . tan , h. l. stormer , p. kim , nature * 438 * , 201 ( 2005 ) j. tworzydo , b trauzettel , m titov , a rycerz , and c. w. j. beenakker , phys . lett . * 96 * , 246802 ( 2006 ) r. zhu and y. guo , appl . * 91 * , 252113 ( 2007 ) l. dicarlo , j. r. williams , y. zhang , d . t. mcclure , and c. m. marcus , phys . lett . * 100 * , 156801 ( 2008 ) c. w. groth , j. tworzydo , and c. w. j. beenakker , phys . . lett . * 100 * , 176804 ( 2008 ) r. danneau , f. wu , m. f. craciun , s. russo , m. y. tomi , j. salmilehto , a. f. morpurgo , and p. j. hakonen , phys . lett . * 100 * , 196802 ( 2008 ) c. w. j. beenakker and m. bttiker , phys . b * 46 * , 1889 ( 1992 ) k. e. nagaev , phys . lett . a * 169 * , 103 ( 1992 ) c. h. lewenkopf , e. r. mucciolo , and a. h. castro neto , phys . b * * 7**7 , 081410 ( 2008 ) a. rycerz , p. recher , and m. wimmer , phys . rev . b * 80 * , 125417 ( 2009 ) e. b. sonin , phys . rev . b * 77 * , 233408 ( 2008 ) v. p. gusynin , s. g. sharapov , and j. p. carbotte , phys . * 96 * , 256802 ( 2006 ) f. j. lopez - rodriguez and g. g. naumis , phys . b * 78 * , 201406 ( 2008 ) j. r. williams , l. dicarlo , and c. m. marcus , science * 317 * , 638 ( 2007 ) d. a. abanin and l. s. levitov , science * 317 * , 5838 ( 2007 ) m. v. fistul and k. b. efetov , phys . lett . * 98 * , 256803 ( 2007 ) s. v. syzranov , m. v. fistul and k. b. efetov , phys . b * 78 * , 045407 ( 2008 ) ya . blanter and m. bttiker , phys . rep . * 336 * , 1 ( 2000 ) p. k. tien and j. p. gordon , phys . rev . * 129 * , 647 ( 1962 ) m. grifoni and p. hnggi , phys . rep . * 304 * , 229 ( 1998 ) g. platero and r. aguado , phys . rep . * 395 * , 1 ( 2004 ) s. kohler , j. lehmann , and p. hnggi , phys . rep . * 406 * , 379 ( 2005 ) . b. trauzettel , ya . m. blanter , and a. f. morpurgo , phys . b * 75 * , 035305 ( 2007 ) c. w. j. beenakker , phys . rev . lett . * 97 * , 067007 ( 2006 ) c. g. rocha , l. e. f. foa torres , and g. cuniberti , phys . b * 81 * , 1185435 ( 2010 ) m. ahsan zeb , k. sabeeh , and m. tahir , phys . b * 78 * , 165420 ( 2008 ) m. v. berry and r. j. mondragon , proc . r. soc . a * 412 * , 53 ( 1987 ) f. miao , s. wijeratne , y. zhang , u. c. coskun , w. bao , c. n. lau , science * 317 * , 1530 ( 2007 ) m. a. laakso and t. t. heikkila , phys . b * 78 * , 205420 ( 2008 ) a. schuessler , p. m. ostrovsky , i. v. gornyi , and a. d. mirlin , phys . b * 79 * , 075405 ( 2009 ) m. i. katsnelson and f. guinea , phys . rev . b * 78 * , 075417 ( 2008 ) d. c. guhr , d. rettinger , j. boneberg , a. erbe , p. leiderer , and e. scheer , phys . rev . lett . * 99 * , 086801 ( 2007 ) m. h. pedersen and m. bttiker , phys . rev . b * 58 * , 12993 ( 1998 ) m. bttiker , a. prtre and h. thomas , phys . rev . lett . * 70 * , 4114 ( 1993 ) n. m. r. peres , a. h. castro neto , and f. guinea , phys . b * 73 * , 195411 ( 2006 ) m. titov , eur . phys . lett . * 79 * , 17004 ( 2007 ) r. j. schoelkopf , p. j. burke , a. a. kozhevnikov , d. e. prober , and m. j. rooks , phys . lett . * 78 * , 3370 ( 1997 ) g. lesovik and r. loosen , jetp letters * 65 * , 295 ( 1997 ) u. gavish , y. levinson , and y. imry , phys . rev . b * 62 * , 10637 ( 2000 ) u. gavish , y. levinson , and y. imry , cond - mat 0211681 ( 2002 ) c.w.j . beenakker and h. schomerus , phys . lett . * 86 * , 700 ( 2001 ) l. reydellet , p. roche , d. glattli , b. etienne , and y. jin , phys . rev 90 * , 176803 ( 2003 ) b. reulet , j. senzier , and d. e. prober , phys . 91 * , 196601 ( 2003 ) quantum noise in mesoscopic physics , edited by y.v . nazarov ( kluwer , dordrecht , 2008 ) . j. gabelli and b. reulet , phys . * 100 * , 026601 ( 2008 ) j. gabelli and b. reulet , j. stat . mech . , p01049 ( 2009 ) e. zakka - bajjani , j. sgala , f. portier , p. roche , d. c. glattli , a. cavanna and y. jin , phys . . lett . * 99 * , 236803 ( 2007 ) h .- p . engel and d. loss , phys . lett . * 04 * , 136602 o. entin - wohlman , y. imry , s. a. gurvitz , and a. aharony , phys . rev . b * 75 * , 193308 ( 2007 ) e. a. rothstein , o. entin - wohlman , and a. aharony , phys . rev . b * 79 * , 075307 ( 2008 ) c. schnenberger , and c. beenakker , physics today * 56 * , 37 ( 2003 ) ramn aguado and l. p. kouwenhoven , phys . rev . lett . * 84 * , 1986 ( 2000 ) ramn aguado and tobias brandes , phys . lett . * 92 * , 206601 ( 2004 ) d. marcos , c. emary , t. brandes , and r. aguado , phys . b * 83 * , 125426 ( 2011 ) m. bttiker , phys . rev . b * 45 * , 3807 ( 1992 ) m. bttiker , h. thomas , and a. prtre , z.phys . b * 94 * , 133 ( 1994 ) m. bttiker and t. christen in quantum transport in semiconductor submicron structures , edited by b. kramer ( kluwer , dordrecht ) , 263 ( 1996 ) w. liang , m. bockrath , d. bozovic , j. h. hafner , m. tinkham & h. park , nature * 411 * , 665 ( 2001 ) l. g. herrmann , t. delattre , p. morfin , j .- m . berroir , b. plaais , d. c. glattli , and t. kontos , phys . lett . * 99 * , 156804 ( 2007 ) m. creux , a. crepieux , t. martin , phys . rev . b * 74 * , 115323 ( 2006 ) a. v. lebedev , a. crpieux , and t. martin , phys . rev . b * 71 * , 075416 ( 2005 ) s. oberholzer , e. v. sukhorukov , and c. schnenberger , nature * 415 * , 765 ( 2002 )

we show that a dynamical supersymmetry can appear in a purely fermionic system . this supersymmetry without bosons " is constructed by application of a recently introduced boson - fermion dyson mapping from a fermion space to a space comprised of collective bosons and ideal fermions . in some algebraic fermion models of nuclear structure , particular hamiltonians may lead to collective spectra of even and odd nuclei that can be unified using the dynamical supersymmetry concept with pauli correlations exactly taken into account .

nucl-th9606043

in its original inception supersymmetry pertains to a system of fermions and bosons exhibiting an invariance with respect to exchange between these two classes of particles . it may therefore be somewhat surprising to discover that a fermion system on its own can also exhibit _ dynamical supersymmetry_. states with even and odd fermion numbers are then unified in a single representation of a supergroup . this is discussed in general below and demonstrated for a specific nuclear model . states of even and odd nuclei are in principle eigenstates of the same hamiltonian obtained for different particle numbers . although there is no fundamental difference between even and odd nuclei from this point of view , their properties are , however , quite different . a unification of spectra of even and odd nuclei into a single framework is therefore a challenging possibility , with the prospect of unveiling a basic underlying symmetry . the notion of supersymmetry has in fact proved to be fruitful in nuclear structure physics@xcite . properties of some neighboring even and odd nuclei can namely be classified and understood in terms of an assumed supersymmetry within the framework of the interacting boson - fermion model ( ibfm)@xcite . although this phenomenological supersymmetry does not necessarily imply dynamical supersymmetry on the microscopic level , the ibfm@xcite does achieve a unification of even and odd states on the phenomenological level . starting from a common boson - fermion hamiltonian one finds that in some instances states of an even nucleus , described by many - boson wave functions , are linked by supersymmetry to states in a neighboring odd nucleus in which the odd fermion is treated explicitly . it is important to realize , however , that pauli correlations between the odd particle and even core are not fully taken into account in the ibfm . in this sense the link between observed supersymmetry and an underlying microscopy is tenuous and no detailed microscopic understanding of the ibfm hypothesis in fact exists so far . in this paper we investigate the link between dynamical supersymmetry and some fermion algebraic models . for this purpose we apply the recently developed dyson boson - fermion mapping @xcite which has the key property of mapping an even fermion system ( comprised of collective pairs ) to a purely boson system , and at the same time an odd fermion system ( comprised of collective pairs and an odd fermion ) to a system of collective bosons and a single ideal fermion . the construction is valid for any number of collective pairs and non - collective fermions treated individually , and preserves the pauli principle exactly . the paper is organized as follows . in sections [ sec6 ] and [ sec2 ] we discuss respectively the general properties of dynamical supersymmetry as it is manifested in nuclear physics , and the boson - fermion mapping which we use to construct it . [ sec3 ] presents a simple example of the appearance of a dynamical supersymmetry in the su(2 ) seniority model . then , in sec.[sec4 ] , we describe a more realistic system with the algebraic structure of so(8)@xmath0so(5 ) which also exhibits a dynamical supersymmetry similar to that appearing in the ibfm . conclusions are drawn in sec . the present paper is an expanded version of a short unpublished communication @xcite . the ibfm@xcite generalizes the phenomenological interacting boson model ( ibm)@xcite to odd systems by coupling single fermion degrees of freedom to the ibm bosons , _ assuming _ these fermions to be kinematically independent from the bosons . the most general one- plus two - body ibfm hamiltonian contains pure boson and pure fermion parts , as well as an interacting part , the only requirement being that boson and fermion numbers are conserved separately . even for the most restricted fermion subspaces this hamiltonian still contains far too many parameters to serve as the basis for a purely phenomenological analysis . microscopically inspired considerations can , on the one hand , serve as a guideline to effect a reduction to a few manageable terms@xcite , while dynamical bose - fermi symmetry and , eventually , dynamical supersymmetry can also serve to restrict the dynamics . these dynamical symmetries , their consequences and eventual comparison with experimental data are discussed in detail in refs.@xcite . ( see also ref . @xcite for a recent analysis where ( extended ) supersymmetry is successfully used to account for properties of negative parity states in @xmath1ir . ) here we only briefly recount the most salient aspects of , respectively , dynamical bose - fermi and supersymmetry on the phenomenological level in order to make it clear what we accomplish in our construction of a dynamical supersymmetry on the microscopic level . recall that dynamical symmetry refers to a hamiltonian which can be written in terms of casimir invariants of the algebras appearing in a _ single _ chain of associated subgroups only , whence an analytical expression can be found for the energy spectrum by exploiting a basis labelled according to irreps of the highest group in the chain@xcite . this situation implies various relationships between strength parameters of the general hamiltonian and thus represents a considerable restriction . from the phenomenological point of view one studies whether dynamical symmetries indeed appear among nuclear spectra ; from the microscopic point of view one would endeavour to motivate a set of strength parameters conforming to a dynamical symmetry . an extensive literature exists for both types of study see e.g. ref . @xcite for a presentation of phenomenological studies and ref . @xcite for a discussion of microscopic investigations . the general form of the highest or first group in the subgroup chain pertaining to a bose - fermi symmetry of the ibfm type can be written as g@xmath2g@xmath3 and in the simplest version of the ibm with one @xmath4- and five @xmath5-bosons this takes the form u(6)@xmath0u(2@xmath6 ) with @xmath7 the dimension of the fermion space . dynamical symmetries for situations where the boson and fermion subchains are essentially independent are the least restrictive and not very interesting . however , when some of the algebras appearing in the two chains are the same or isomorphic , one has the possibility to couple them to a spinor algebra . when this happens beyond the level of the rotation group , the dynamical structure seems to be sufficiently restrictive to lead to interesting and manageable classification schemes which are also realised in some nuclei@xcite . it is important to realize that a dynamical bose - fermi symmetry with a given set of strength parameters pertains to one specific odd nucleus , the number of bosons @xmath8 being fixed as @xmath9 with @xmath10 the number of valence nucleons . a more ambitious attempt to implement dynamical symmetry is to consider u(@xmath11)@xmath0 u(@xmath7 ) , with @xmath11 the number of single boson states , as being embedded in the supergroup u(@xmath12 ) . this implies that one is considering the classification of nuclear states in a set of neighbouring even and odd nuclei within a single supersymmetric representation @xmath13 , where @xmath14 , with @xmath15 the number of unpaired nucleons , and where the hamiltonian parameters and all other parameters appearing in physical operators are the same for all the nuclei under consideration clearly a severe dynamical restriction . note that the supersymmetric representation is specified not by the effective number of fermions , @xmath16 , but by the number of bosons plus fermions , @xmath17 , which of course underlies the possibility to use the single supersymmetric representation for a set of adjacent nuclei . although this type of dynamical supersymmetry is invariably found to be broken , there seems to be enough evidence @xcite that it is a useful scheme for classification and calculation of nuclear properties for a number of nuclei . what concerns us mostly in the remainder of this paper is the question whether dynamical supersymmetry , as summarized above , can be compatible with the pauli principle or , alternatively , if dynamical supersymmetry can be an exact property of a fermion system . from the point of view that there are important pauli corrections to the lowest order association between collective fermion pairs and ibm bosons@xcite , one might anticipate a negative answer to this question . nevertheless , we show that the implementation of appropriate boson - fermion mappings indeed reveals instances where this compatibility holds . these mappings , introduced and refined in refs.@xcite , and discussed and exploited below , introduce an equivalence between a system of interacting fermions on the one hand , and on the other a system of interacting bosons and fermions which are _ by construction kinematically independent_. although no guarantee in itself , this clearly fulfills the minimum requirement for a dynamical supersymmetry to exist , namely to have the appropriate degrees of freedom . moreover , as elaborated below , one can indeed find instances where dynamical supersymmetry emerges as an exact classification scheme of fermion states . this is simply an alternative , but equivalent , classification to whatever classification scheme may have been adopted on the fermion level . what the boson - fermion mapping accomplishes is to make the inherent supersymmetric nature transparent in these instances , although they should also be identifiable on a purely algebraic level in terms of relationships between standard and supersymmetric representations . apart from providing a concrete link between fermion dynamics and dynamical supersymmetry , the use of boson - fermion mappings also allows one to _ construct _ various transition operators appropriate to the boson - fermion description . this is in contrast to the phenomenological situation where one is obliged to truncate an infinite series of combinations of boson and fermion operators with phenomenological parameters and terms only restricted by their tensor and particle number changing properties@xcite . it should be emphasized that the choice of these transition operators in phenomenological models such as the ibm or ibfm is _ not _ dictated by the hamiltonian parameters in general , specifically also in the case of dynamical symmetry or supersymmetry . ( see also ref.@xcite for a discussion of this point . ) in the present context we show e.g. how the single fermion transfer operator is uniquely specified in the boson - fermion description by the appropriate mapping and how it compares with phenomenological results . finally , our work provides the possibility of a direct link between phenomenological models which introduce dynamical supersymmetry in an inherently fermion problem , such as is in the ibfm , and fermion models which exploit a similar structure by considering both vector and spinor representations of the the overall symmetry group , such as in the very recent fermion dynamical symmetry model ( fdsm ) analysis@xcite . our formalism is thus tailor made to investigate the similarities and differences between the two approaches . in order to keep the paper as self - contained as possible , we briefly retrace the basic steps in the derivation of the dyson boson - fermion mapping @xcite used in subsequent sections . we start from a fermion collective pair algebra comprised of collective fermion pair creation operators @xmath18=@xmath19 , pair annihilation operators @xmath20 , and their commutators @xmath21 $ ] . the notation exploits a summation convention in which upper ( lower ) indices denote creation ( annihilation ) operators , as in @xmath22=@xmath23 and @xmath18=@xmath24 . we also assume@xcite that the collective fermion operators obey the algebraic closure relations @xmath25,a_k\right ] = c^{jl}_{ik}a_l , $ ] i.e. , operators @xmath18 , @xmath20 , and @xmath21 $ ] are assumed to form a collective spectrum generating algebra@xcite . these relations guarantee an exact decoupling of the even collective space @xmath26 from all other even fermion states . similarly , by adding an odd fermion , one obtains an exactly decoupled odd collective space @xmath27 . in a boson - fermion description an even state @xmath28 should be represented by an ideal space state @xmath29 ) , say , which contains _ bosons only _ , with the odd ideal space states @xmath30 containing an additional fermion @xmath31 . following traditional terminology this is also referred to as an ideal fermion@xcite . by definition it commutes with all boson operators , @xmath32$]=@xmath33$]=0 . the required representation in the ideal space can be achieved by constructing an appropriate boson - fermion mapping ( see the appendix ) , [ e228]@xmath34 - \textstyle{\frac{1}{2}}c^{jl}_{ik}b^i b^k b_l , \label{e228c } \\ \big[a_i,\!a^j\big ] & \longleftrightarrow & \big[{\cal{a}}_i,{\cal{a}}^j\big ] - c^{jl}_{ik}b^kb_l , \label{e228b } \\ a_j & \longleftrightarrow & b_j , \label{e228a } \\ a^{\nu } & \longleftrightarrow & x^{-1}\left(\alpha^{\nu } + b^i\big[{\cal{a}}_i,\alpha^{\nu}\big]\right)x , \label{e228e } \\ a_{\nu } & \longleftrightarrow & x^{-1}\alpha_{\nu}x , \label{e228d } \end{aligned}\ ] ] where @xmath35= @xmath36 are collective pairs of _ ideal fermions_. the similarity transformation @xmath37 has the explicit form @xmath38 where @xmath39=@xmath40 is an operator which leaves invariant the ideal - fermion core subalgebra composed of operators @xmath41 $ ] , i.e. , @xmath42\right]$]=0 . here we use the notation@xcite of a deferred - action operator @xmath43 which should be evaluated at the position indicated by `` @xmath44@xmath44 '' . equivalently , one can write eq . ( [ eq3 ] ) by using multiple sums over eigenstates of @xmath39 , in which case @xmath45 can be replaced by typical energy denominators . we can now discuss the structure of the images of single fermion operators , eqs . ( [ e228e ] ) and ( [ e228d ] ) . an explicit general evaluation of these images is difficult because the operators @xmath46 and @xmath47 do not commute with the ideal - fermion invariant operator @xmath39 and one has to consider branching rules of the collective algebra . before presenting solutions in particular cases we can , however , analyze some general properties of these images . we note that the similarity transformation ( [ eq3 ] ) does not change any state in which there are no bosons @xmath48 , and therefore the single - fermion states are mapped onto single ideal fermion states , @xmath49 . since the images of pair creation operators ( [ e228c ] ) do not change the ideal fermion number , collective odd states @xmath50 are mapped onto ideal states with one ideal fermion only . the two - fermion states are mapped as @xmath51 and in general contain the non - collective pair of ideal fermions @xmath52 . however , when the collective pair @xmath18 is formed by summing the pairs @xmath53 with collective amplitudes @xmath54 , the ideal non - collective pairs above recombine ( note that @xmath55=@xmath56 ) and only the boson state @xmath57 remains . again , since the images ( [ e228c ] ) conserve the ideal fermion number , the same recombination mechanism is also valid for any even state . similarly as in the even case , spurious states may appear in spectra of mapped operators when diagonalized in the complete ideal boson - fermion space . however , they do not contaminate physical states and can be identified after the diagonalization . the methods which can be used to this effect are analogous to those discussed in ref . @xcite , and will not be repeated here . the textbook su(2 ) seniority model @xcite is very often used in group theory models to illustrate the key aspects of the methods used . although nothing new can be said about the model itself , its simplicity and intuitiveness allows for a very suitable test ground where advanced approaches can be explained in simple terms . in this section we apply to this model our boson - fermion mapping and discuss the resulting dynamical supersymmetry . the su(2 ) model is defined by considering in a single-@xmath58 shell the monopole pair creation operator @xmath59 with @xmath60 . it fulfills the commutation relation @xmath61 = \omega - n , \ ] ] where @xmath10 is the fermion number operator in the single-@xmath58 shell . the su(2 ) algebra can be generalized to describe odd systems by constructing the superalgebra generated by the operators @xmath62 , @xmath63 , @xmath6@xmath64@xmath10 , @xmath65 , and @xmath66 . the relevant commutation relation is @xmath67 = -\tilde{a}_{jm } \ ; , \ ] ] with @xmath68 , while the single - fermion operators obey the standard anticommutation relations . clearly this is a rather trivial superalgebra as the elements of the odd sector ( single fermion operators ) anti - commute only to the identity . alternatively , by considering the _ commutator _ of single - fermion operators , the set of bi- and single - fermion operators may of course also be viewed as generators of a standard ( orthogonal ) algebra . in the single-@xmath58 shell we consider the pairing hamiltonian @xmath69 which has the energy spectrum @xcite @xmath70 with the seniority quantum number @xmath71 denoting the number of fermions not coupled to angular momentum zero . this hamiltonian describes both the even and odd systems , and the spectra in both cases are given by the same expression ( [ senen ] ) with @xmath71 even or odd , respectively . we can apply to this model the dyson boson - fermion mapping described in sec . [ sec2 ] to find an equivalent description in the boson - fermion space . for the su(2 ) algebra , the operator @xmath39 depends on the ideal fermion number operator @xmath72 only , and one may derive an explicit form of the similarity transformation@xcite . as we have @xmath73 the transformation ( [ eq3 ] ) is @xmath74 \ : { \textstyle \raisebox{-1ex}{$\widehat{}$}}\qquad.\ ] ] specializing from the general case to su(2 ) , we have introduced here the ideal fermion operators @xmath75 and @xmath76 , which commute with the ideal boson operators @xmath77 and @xmath78 , and the ideal fermion pair operators @xmath79 and @xmath80 obtained from @xmath62 and @xmath63 by replacing @xmath81 by @xmath82 . the general dyson boson - fermion mapping ( [ e228 ] ) is then obtained for the su(2 ) case in the form [ sulmap ] @xmath83 where @xmath14 . we see that the single fermion images ( [ su141e ] ) and ( [ su141d ] ) are finite and contain terms changing the ideal fermion number by one only . furthermore , they preserve _ exactly _ the anti - commutation relations on the full ideal space , i.e. , as operator identities . this is guaranteed by our construction method , and can be verified by explicit calculation . the preservation of the commutation and anticommutation relations ensures the exact preservation of the pauli exclusion principle , once the original fermion problem is mapped into the boson - fermion space . clearly , the mapping ( [ sulmap ] ) transforms the 2-body hamiltonian ( [ pairh ] ) into a 1-plus-2-body boson - fermion hamiltonian of the form @xmath84 hamiltonians ( [ pairh ] ) and ( [ pairhbf ] ) have exactly the same spectrum ( [ senen ] ) . hamiltonian ( [ pairhbf ] ) can also be expressed in a form which stresses its dependence on the total number of bosons and fermions @xmath85 , i.e. , @xmath86 note also that the boson - fermion interaction term in ( [ pairhbf ] ) , which reads @xmath87 , can be expressed in terms of the odd generators , @xmath88 and @xmath89 , of the u(1/2@xmath6 ) superalgebra . since the boson and ideal fermion number operators can be linked to even generators ( see also expression ( [ bosfch ] ) and its discussion below ) , this identification makes it possible to write the hamiltonian ( [ pairhbf ] ) in yet another form in terms of both even generators and _ supergenerators _ of u(1/2@xmath6 ) : @xmath90 after the mapping states with a given fermion number @xmath10 are classified according to the number of bosons @xmath8 and the number of ideal fermions @xmath91 , these numbers still being related to the fermion number @xmath10 through @xmath10=@xmath92+@xmath15 . the number of ideal fermions now corresponds to the seniority quantum number , because all fermion pairs coupled to zero angular momentum are mapped onto bosons . this latter identification follows by construction , as any attempt to create an ideal fermion pair of angular momentum zero by applying ( [ su141e ] ) successively ( with appropriate coupling ) , results in the creation of the boson @xmath78 . as long as we do not overfill the @xmath58 level by too many particles , i.e. , restrict the ideal space to physical states with @xmath93 , the fermion space results are exactly reproduced in the boson - fermion space . in general it is not possible , as in this simple case , to make , for a given fixed particle number , an _ a priori _ selection of simple ideal space states which will span the physical subspace . however , as elaborated in ref . @xcite , one can exploit the simplicity of the ideal space basis for the diagonalization of the mapped hamiltonian . the physical eigenstates do not mix with the unphysical states and can by identified after the diagonalization . ( see ref . @xcite for a recent application of this procedure . ) in the original fermion space of the su(2 ) seniority model , the hamiltonian eigenstates are classified according to the representations of the subgroup chain @xmath94 while in the boson - fermion space they are classified by the quantum numbers derived from the chain @xmath95 in the second chain the supergroup u(1/2@xmath6 ) is generated by @xmath96 , @xmath97 , @xmath98 , and @xmath99 , with the first two operators belonging to the even sector and the remaining two to the odd sector of the superalgebra , respectively . the appearance of u(1/@xmath7 ) in the chain is clearly suggested by the form ( [ pairhbfb ] ) and also dictates that the same hamiltonian parameters in ( [ pairhbf ] ) are used for both even and odd states , if the hamiltonian should be viewed as a phenomenological boson - fermion hamiltonian . from the mapping point of view there is of course no real choice in this matter , as the original fermion hamiltonian pertains to any number of particles . nevertheless , it is interesting to note already here how the phenomenological extension to superalgebras _ la _ ibfm may be suggested from a microscopic point of view . in the dynamical supersymmetry concept , those states of the system with the same number of _ ideal _ particles @xmath85 belong to a single supersymmetric representation @xmath13 of the supergroup , here u(1/@xmath7 ) . ( note that @xmath85=@xmath17 differs from the fermion number @xmath10 and its ideal space image ( [ su141b ] ) , @xmath100 . starting from , say , the state with no fermions @xmath101 , the other states of the multiplet are obtained by successive application of supergenerators @xmath98 . although one can embed all states of the su(2 ) model in a sequence of representations of the supergroup u(1/@xmath7 ) , this does not really provide other interesting physical consequences . this is so because in the chain ( [ bosfch ] ) the supergroup u(1/@xmath7 ) is immediately split into the boson and fermion sectors which remain separate down to the bottom of the chain . as discussed in sec . [ sec6 ] , potentially interesting situations which are restrictive from the supersymmetry point of view , occur if at a certain level of subgroups one may find the same subgroups in the boson and fermion sectors , and combine them together into a given boson - fermion subgroup . such a situation arises in a fermion model with a richer structure than the present su(2 ) case and is discussed in the next section . in terms of algebras appearing in the chain ( [ bosfch ] ) , we here only find a trivial example of the boson subalgebra u@xmath102(1 ) , generated by the boson number operator @xmath8 , and the fermion subalgebra u@xmath3(1 ) , generated by the fermion number operator @xmath15 , which can be combined into the boson - fermion subalgebra u@xmath103(1 ) , generated by @xmath85=@xmath8+@xmath15 . at the same time , even the simplest example of the su(2 ) seniority model discussed here provides interesting consequences for the structure of the single - fermion transfer operators and the spectroscopic factors . after the mapping , the single - fermion operators acquire terms which are responsible for the pauli correlations between the even core and the odd particle . in the phenomenological supersymmetric models these terms are postulated together with some arbitrary numerical constants , whereas in the supersymmetric picture derived from the boson - fermion mapping they are fixed by the mapping procedure itself . for example , the image of the fermion annihilation operator ( [ su141d ] ) is a combination of the ideal fermion annihilation operator with two corrective terms . the first one corresponds to an annihilation of the same ideal fermion accompanied by an attempt to replace a boson by a collective ideal fermion pair . this term ensures that after the annihilation the remaining fermions still obey the correct statistics . the second term corresponds to the standard annihilation of the boson replaced by an appropriate fermion which has been the partner of the annihilated fermion in a collective pair . of course , the relative weights of these corrective terms depend on how many ideal fermions have already been present before the annihilation . to evaluate and understand different aspects of the original fermion and the mapped boson - fermion description , it is instructive to calculate the spectroscopic factors for the simplest states in the su(2 ) model . in the original fermion space this can be done by using the commutation relations ( [ coms ] ) and ( [ comas ] ) , and we arrive at the result @xmath104 to get this relation one has to calculate the normalization factors of the fermion states . evaluating matrix elements ( [ spect ] ) in the mapped boson - fermion space is much simpler . in particular , one can work with the ideal boson - fermion eigenstates of the hamiltonian ( [ pairhbf ] ) corresponding to the states with @xmath15=0 and 1 , e.g. , @xmath105 , and @xmath106 . using eqs . ( [ su141e ] ) and ( [ su141d ] ) we directly read off the results [ bfspec ] @xmath107 where @xmath108 and @xmath109 are the boson - fermion images of @xmath110 and @xmath111 , respectively . in this calculation , only simple boson normalization factors enter . after the hermitization ( see ref . @xcite and references therein ) and identification of 2@xmath112 , matrix elements ( [ bfspec ] ) reproduce the correct value of @xmath113 . it is probably worthwhile to conclude this section by carefully distinguishing two superalgebraic structures we identified in two different contexts . firstly , the algebraic structure obtained by extending a collective bifermion algebra with single fermion operators leads to a trivial superalgebra structure . nevertheless , this identification plays an important role in the construction of boson - fermion images developed in refs . @xcite . secondly , after mapping of the fermion hamiltonian , it becomes possible to analyse the resulting equivalent boson - fermion hamiltonian in terms of a subgroup chain with a supergroup as first member , and classify states of neighbouring even and odd systems according to a single representation of that supergroup . this analysis relies on the identification of the mapped hamiltonian as being constructed from the generators of subalgebras of a _ non - trivial _ superalgebra , generally different from the one appearing on the original fermion level . in this section we discuss the appearance of dynamical supersymmetry in a more complicated model , namely , in the ginocchio so(8 ) model @xcite . this model is defined by collective pairs [ so8gen ] @xmath114 with @xmath115 and @xmath116 integer . in ( [ so81 ] ) only @xmath62 ( @xmath117=0 ) and @xmath118 ( @xmath117=2 ) pairs are allowed , while in ( [ so82 ] ) @xmath117 takes values 0 , 1 , 2 , and 3 . in order to generalize the so(8 ) algebra to a superalgebra , we include the @xmath7=(2@xmath119 + 1)(2@xmath116 + 1 ) creation and annihilation operators @xmath120 and @xmath65 , where @xmath58=@xmath121@xmath64@xmath122+@xmath119 . a boson - fermion mapping of this algebra was derived in ref.@xcite . in this case the series defining the similarity transformation @xmath123 can not be explicitly summed up because the denominator @xmath124,\ ] ] contains the quadratic casimir operator of the spin(6 ) group which is not expressible in terms of number operators . here @xmath15 is the ideal fermion number operator @xmath125 and @xmath126 . the similarity transformation @xmath37 can nevertheless be applied with the result that the transformed mapping then reads [ so8images ] @xmath127 for the collective pair operators while for the single - fermion operators we obtain [ so8sinim ] @xmath128 - \frac{1}{c_{\rm f } - \widehat{c}_{\rm f } } { \cal f}^\dagger_{j_1}\cdot \tilde{b}_{j_1 } \ : \:{\textstyle \raisebox{-1ex}{$\widehat{}$}}\:\ : b^\dagger_j \cdot \left[\tilde{{\cal f}}_j,\alpha^{\dagger}_{jm}\right ] \nonumber \\ & & + b^\dagger_j \cdot \left[\tilde{{\cal f}}_j,\alpha^{\dagger}_{jm}\right ] \frac{1}{c_{\rm f } - \widehat{c}_{\rm f } } { \cal f}^\dagger_{j_1}\cdot \tilde{b}_{j_1 } \ : \:{\textstyle \raisebox{-1ex}{$\widehat{}$}}\:\:\ldots\:\:\ : , \label{so812 } \\ a_{jm } & \longleftrightarrow & \alpha_{jm } - \frac{1}{c_{\rm f } - \widehat{c}_{\rm f } } { \cal f}^\dagger_{j_1}\cdot \tilde{b}_{j_1 } \ : { \textstyle \raisebox{-1ex}{$\widehat{}$}}\ : \ : \alpha_{jm } + \alpha_{jm } \frac{1}{c_{\rm f } - \widehat{c}_{\rm f } } { \cal f}^\dagger_{j_1}\cdot \tilde{b}_{j_1 } \ : { \textstyle \raisebox{-1ex}{$\widehat{}$}}\:\:\ldots\:\:\ : . \label{so826}\end{aligned}\ ] ] the ideal fermion pair operators @xmath129 and @xmath130 are given by eqs . ( [ so81 ] ) and ( [ so82 ] ) , respectively , with the fermion operators @xmath65 replaced by ideal fermion operators @xmath76 . the dots @xmath131 refer to a class of higher order terms which will not contribute to matrix elements between physical states . ( see ref . @xcite for further discussion . ) a condensate of collective fermion pairs have now been mapped onto a condensate of @xmath132 and @xmath133 bosons , and a condensate of collective fermion pairs with one additional odd fermion , onto a condensate of @xmath134 and @xmath135 bosons with one additional ideal fermion . this situation is quite reminiscent of the phenomenological ibfm . the six bosons above span the symmetric representation of u(6 ) . the size of the fermion sector depends on @xmath116 . let us first discuss the situation with @xmath116=0 ( @xmath58=3/2 ) . ( [ eqso87 ] ) we observe that the boson - fermion images of the multipole operators are [ u64 ] @xmath136 \ ; , \\ p_{2 m } & \longleftrightarrow & 2\left[(d^\dagger s+s^\dagger \tilde{d})^{(2)}_m - ( \alpha_j^\dagger \tilde{\alpha}_j)^{(2)}_m\right ] \ ; , \\ p_{3 m } & \longleftrightarrow & -2\sqrt{2}\left[(d^\dagger \tilde{d})^{(3)}_m + \frac{1}{\sqrt{2 } } ( \alpha_j^\dagger \tilde{\alpha}_j)^{(3)}_m\right ] \ ; .\end{aligned}\ ] ] in the boson - fermion multipole operators ( [ u64 ] ) we recognize immediately the generators of the spin@xmath103(6 ) group applied in the phenomenological ibfm boson - fermion dynamical symmetry u@xmath2u@xmath3(4 ) @xcite . similarly , from eqs . ( [ so8sinim ] ) we identify single nucleon transfer operators in the form used in the above phenomenological dynamical symmetry , provided that we limit ourselves to @xmath137 states only and neglect the @xmath96 type terms : @xmath138 \ ; .\ ] ] clearly , once we construct a fermion dynamical symmetry hamiltonian from the casimir operators of the subgroups appearing in the chain @xmath139 its boson - fermion image would be just the ibfm u@xmath2u@xmath3(4 ) dynamical symmetry hamiltonian . since the original so(8 ) hamiltonian does not distinguish even and odd systems , the mapped hamiltonian must also be the same for both even and odd system . therefore , as already argued in the su(2 ) case , the classification scheme in the boson - fermion space should be extended by embedding the boson - fermion dynamical symmetry into the u(6/4 ) dynamical supersymmetry . however , there is a difference between the mapped boson - fermion system and the phenomenological ibfm u(6/4 ) supersymmetry . in the so(8 ) model considered here all particles occupy the same @xmath58=3/2 level while in the ibfm it is assumed that the bosons occupy the whole valence shell with only the fermion restricted to @xmath58=3/2 . as a more realistic situation , consider @xmath116=2 , corresponding to @xmath58=1/2 , 3/2 , 5/2 , and 7/2 . in the nuclear shell model this corresponds to the 3s@xmath140 , 2d@xmath141 , 2d@xmath142 , and 1g@xmath143 orbitals located between the @xmath144=50 and 82 nuclear magic numbers . in the ibfm , a related supersymmetry with the same single - particle content , is u(6/20 ) , realized in the au - pt isotopes@xcite . the group reduction chain is @xmath145 . a hamiltonian chosen as a linear combination of quadratic casimir operators appearing in this chain yields the analytical u(6/20 ) ibfm energy formula @xcite @xmath146 \nonumber \\ & & + b[\tau_1(\tau_1 + 3)+\tau_2(\tau_2 + 1 ) ] + c j(j+1 ) . \label{eqsy7}\end{aligned}\ ] ] in the simple extension of the so(8 ) ginocchio model to odd systems the inactive angular momentum @xmath116 of the odd fermion gives rise to an unrealistic degeneracy . one can , however , lift this degeneracy by adding to the so(8 ) algebra multipole operators corresponding to an interchange of the active and inactive angular momenta @xmath116 and @xmath119 . suppose we add two such operators @xmath147 for @xmath117=1 and 3 , @xmath148 with @xmath149 and @xmath150 . these operators form an so(5 ) algebra and commute with the so(8 ) generators ( [ so8gen ] ) . the resulting algebraic structure is therefore so(8)@xmath0so(5 ) . hence , we can form a new dynamical symmetry subgroup chain by combining the so(5 ) generators with the spin(5 ) generators to obtain an additional group denoted by @xmath151(5 ) . the fermion group reduction chain is then @xmath152 with @xmath153 generated by [ til5gen ] @xmath154 and @xmath155 by @xmath156 , with @xmath157 the original so(8 ) multipole operators ( [ so82 ] ) . we are now in a position to write a hamiltonian corresponding to the fermion dynamical symmetry ( [ spin5ch ] ) which is composed of quadratic casimir operators of the groups appearing in the chain : @xmath158 with [ casim ] @xmath159 the corresponding spectrum is obtained in a basis classified by quantum numbers characterizing the representations of the subgroups in the dynamical symmetry chain , namely @xmath160 + b[\tau_1(\tau_1 + 3 ) \nonumber \\ & & + \tau_2(\tau_2 + 1 ) ] + \tilde{b}[\tilde{\tau}_1(\tilde{\tau}_1 + 3 ) + \tilde{\tau}_2(\tilde{\tau}_2 + 1 ) ] + c j(j+1 ) , \label{eqsy5}\end{aligned}\ ] ] where @xmath161 are the @xmath162 irreps and @xmath163 are the irreps of @xmath153 . for an even system comprised of @xmath144 fermion pairs , the quantum numbers are : @xmath164 , with @xmath165 or @xmath166 , @xmath167 , with @xmath168 . one also has @xmath169 . in the odd case , when the system is formed by @xmath144 fermion pairs and a single odd fermion , the branching rules are @xmath170 , with @xmath165 or @xmath166 , @xmath171 , with @xmath172 . eventually , the values of @xmath173 , @xmath174 are deduced from the reduction pattern @xmath175 in table [ tab1 ] we present the angular momentum @xmath117 content of the lowest spin(5 ) representations . to arrive at an equivalent boson - fermion description we apply the boson - fermion mapping ( [ so8images ] ) and ( [ so8sinim ] ) , whereas the new so(5 ) algebra is simply mapped from the original fermion space to the ideal fermion space by replacing the operators @xmath176 by operators @xmath177 . @xmath178 obviously , the boson - fermion images of the generators of @xmath179 are obtained by combining the images of spin(5 ) and so(5 ) , as in expressions ( [ til5gen ] ) . it can be verified that , up to a normalization factor , these are just the generators of the subgroup spin@xmath103(5 ) of the ibfm u(6/20 ) ( [ eqsy7 ] ) given explicitly in ref . consequently , at the boson - fermion level we have now the following group chain @xmath180 the groups spin@xmath103(@xmath181 ) correspond to spin(@xmath181 ) and are generated by the images ( [ eqso87 ] ) , while the groups @xmath182 correspond to @xmath183 and their generators are obtained as outlined above . it is interesting to note that the construction ( [ oddmult ] ) and its implementation in the group chain ( [ spin5ch ] ) lead to a hamiltonian ( [ hamsp5 ] ) with spectrum ( [ eqsy5 ] ) which is the same as that used by jolie _ et al _ @xcite , but motivated from a different point of view . the construction in ref . @xcite introduces an _ ad hoc _ coupling between a pseudo - orbital angular momentum 2 and pseudospin 3/2 to obtain the appropriate set of single - particle orbits for their analysis of the a=130 mass region which is furthermore based on the introduction of phenomenological s- and d - bosons to establish a u(6/20 ) supersymmetry framework . our construction therefore implies that the analysis of ref . @xcite can be linked to and equivalently carried out in an so(8)@xmath0so(5 ) framework , which becomes even more transparent after we introduce below the boson - fermion image of the hamiltonian ( [ hamsp5 ] ) . to obtain the boson - fermion image of the hamiltonian ( [ hamsp5 ] ) , all the generators are simply replaced by their boson - fermion images . the eigenenergy expression remains the same as in ( [ eqsy5 ] ) . also the representation reduction pattern remains the same as in the original fermion system , provided the boson number satisfies @xmath184 and the number of ideal fermions is either 0 , or 1 . consequently , for the present model application , there is a one - to - one correspondence between fermion states and states in the boson - plus - one - fermion space , so that spurious states do not appear . furthermore , following the same reasoning as in sec.[sec3 ] , since the fermion interaction strengths pertain to even and odd systems alike , boson - fermion eigenstates of the mapped hamiltonian are contained in a single supersymmetric representation of u(6/20 ) . as a simple application of the supersymmetric @xmath185 energy formula ( [ eqsy5 ] ) , we compare the corresponding even and odd spectra in fig . the hamiltonian parameters @xmath186 kev , @xmath187 kev , and @xmath188 kev , are chosen so that the even part coincides with the one of ref . . in the odd spectrum we then find more low - lying @xmath117=@xmath189@xmath190 states than one gets from the phenomenological ibfm expression ( [ eqsy7 ] ) . the dotted lines connect states belonging to the same @xmath153 representations . we observe that these representations are ordered with increasing energy as @xmath163= @xmath191 , @xmath192 , @xmath191 , @xmath192 , @xmath193 , @xmath194 . on the other hand , the phenomenological ibfm u(6/20 ) supersymmetry ( [ eqsy7 ] ) gives , approximately , the ordering of spin@xmath103(5 ) representations as follows : @xmath161= @xmath191 , @xmath192 , @xmath194 , @xmath192 , @xmath193 , @xmath191 . a different ordering of representations also changes the selection rules of the allowed transitions . for example , the observed @xmath195 transition in @xmath196au @xcite is forbidden in u(6/20 ) ( [ eqsy7 ] ) as it violates the @xmath197=0 , @xmath198 selection rule . in the scheme presented here the @xmath199 state belongs to the @xmath192 representation and the above transition would be allowed . however , it is at the moment too speculative to claim that the dynamical supersymmetry discussed here may be realized in the pt - au region and allow such selection rules to be tested . it is interesting , though , to note that the discussed @xmath200 scheme , which we introduced in ref.@xcite , was also used recently in ref . @xcite and appears to have experimental evidence in the xe - ba region . see e.g. fig . 3 of ref.@xcite with more experimental low - lying low spin states in the odd systems than seems typical for phenomenological supersymmetric spectra derived from ( [ eqsy7 ] ) . we have discussed the cases of one or zero unpaired fermions only . however , it may be shown that a one - to - one correspondence between the fermion states and the boson - fermion states holds when more unpaired particles are considered as well , provided that the ideal fermions do not form collective so(8 ) pairs and , moreover , provided that the condition @xmath201 is satisfied . in summary , we have applied the recently constructed generalized dyson boson - fermion mapping of collective algebras to the seniority su(2 ) and so(8)@xmath0so(5 ) models . the mapping gives finite non - hermitian boson - fermion images of collective pairs and single fermion operators expressed in terms of ideal boson and fermion annihilation and creation operators . in these models , and for the specific interactions chosen , we have revealed a dynamical supersymmetric structure analogous to what is found in the interacting boson - fermion model , but with full recognition of the pauli principle . the findings discussed in this paper are based on the construction of a hamiltonian from the generators of the fermion spectrum generating algebra which defines a class of dynamical symmetries@xcite . the hamiltonian itself is not invariant with respect to the symmetry group , but its eigenstates can be classified by the group and subgroup representations . at this level the supersymmetric structure is not readily visible , and even and odd states simply belong to different representations of the fermion group . however , we showed that an equivalent description which does reveal a dynamical supersymmetry can be obtained after a boson - fermion mapping . whereas the hamiltonian in the su(2 ) model was constructed from the pair operators which are mapped in a non - hermitian way ( but nevertheless yields a hermitian image for the pairing hamiltonian ) , in the so(8 ) model application the hamiltonian consisted of multipole operators with manifestly hermitian boson - fermion images . apparently , a dynamical supersymmetric construction like the one applied to the so(8 ) model could also be extended to any fermion algebraic model based on a pseudo - spin and pseudo - orbital recoupling , as e.g. the sd sp(6 ) @xcite and sdg so(12 ) @xcite models . in such cases the unification of the collective and non - collective algebras can always be achieved on the spin(3 ) level . in the so(8 ) example the corresponding group was denoted by @xmath202 . the interesting question , however , is if one can find a unification of some of the groups relevant to a given model at some higher level , as in the case of @xmath153 for the so(8 ) example . this would be analogous to a dynamical supersymmetry , or a boson - fermion dynamical symmetry in the usual ibfm sense , where the unification is usually considered to be formed at the highest level possible in the group chain . obviously , a fermion algebraic model puts stronger restrictions on the possible extensions to odd systems . in any case , should an algebraic model be applied to describe even - even nuclei , like e.g. the fermion dynamical symmetry model ( fdsm ) @xcite , its extension to the description of odd nuclei should also be investigated . as we have shown in this paper , the identification of dynamical symmetries ( the usual starting point of such analyses ) can sometimes be incorporated into a framework of dynamical supersymmetries after the boson - fermion mapping , thus unveiling a possibly richer dynamical symmetry structure than is immediately visible on the original fermion level . it is of interest to enquire which other hidden " non - trivial dynamical supersymmetries exist in algebraic fermion models and to understand the relationship between representations on the fermion level and those on the boson - fermion level from a purely algebraic point of view , without necessarily relying on the boson - fermion type mapping considered here . let us finally remark that the dynamical supersymmetry discussed in the present context presupposes that all ideal space states enter the analysis , as in the case of the so(8 ) and so(8)@xmath0so(5 ) models . if some of these states do turn out to be spurious in other cases , as they most likely will , one will have a situation which is the equivalent of what is termed the dynamical pauli effect " in the fdsm @xcite . an incomplete supersymmetric spectrum will result , but the classification and description of the remaining physical states in terms of ( ideal ) bosons and fermions will still be perfectly in order . in order to prove that the boson - fermion mapping ( [ e228 ] ) preserves the commutation and anticommutation relations from the original fermion space we proceed in two steps . first we use the boson - fermion mapping , [ e128]@xmath203 - \textstyle{\frac{1}{2}}c^{jl}_{ik}b^i b^k b_l , \label{e128c } \\ \big[a_i,\!a^j\big ] & \longleftrightarrow & \big[{\cal{a}}_i,{\cal{a}}^j\big ] - c^{jl}_{ik}b^kb_l , \label{e128b } \\ a_j & \longleftrightarrow & b_j , \label{e128a } \\ a^{\nu } & \longleftrightarrow & \alpha^{\nu } + b^i\big[{\cal{a}}_i,\alpha^{\nu}\big ] , \label{e128e } \\ a_{\nu } & \longleftrightarrow & \alpha_{\nu } , \label{e128d } \end{aligned}\ ] ] derived in ref . @xcite using supercoherent states . this mapping preserves _ exactly _ the commutation relations of the collective algebra , as well as the commutation relations between single - fermion and pair operators , and also the anticommutation relations between single - fermion operators . however , a repeated application of the image of the pair creation operator ( [ e128c ] ) fails to produce an ideal even space spanned by bosons only . ( see also ref . @xcite . ) second , to prove that the similarity transformation , transforming the mapping ( [ e128 ] ) into the desired mapping ( [ e228 ] ) , is given by eq . ( [ eq3 ] ) , we proceed as follows . we begin by noticing that @xmath204=@xmath205 , because @xmath37 explicitly commutes with the boson annihilation operator @xmath205 , accounting for ( [ e228a ] ) . next we show that @xmath206=@xmath207@xmath64@xmath35 by proving the identity @xmath208={\cal{a}}^jx \ ; .\ ] ] this is achieved by splitting @xmath37 into an infinite sum of terms which individually increase the number of ideal fermions by 0 , 2 , 4,@xmath209 , i.e. , @xmath37=@xmath210 . as @xmath207@xmath64@xmath35 must conserve the number of ideal fermions , from eq . ( [ e4 ] ) we find the recurrence relation @xmath211={\cal{a}}^jx_n $ ] which allows all @xmath212 to be determined , if @xmath213 is known . we are free to choose @xmath213 as an arbitrary operator which commutes with @xmath207@xmath64@xmath35 , and eq . ( [ eq3 ] ) reflects the simplest choice @xmath213=1 . then the term @xmath214 is a solution of the recurrence relation for @xmath10=0 . indeed we have @xmath215 & = & \frac{1}{c_{\rm f}-\widehat{c}_{\rm f } } \left[{\cal a}^i b_i , r^j-{\cal{a}}^j\right]\ : { \textstyle \raisebox{-2ex}{$\widehat{}$ } } \nonumber \\ & = & \frac{1}{c_{\rm f}-\widehat{c}_{\rm f } } ( c_{\rm f}{\cal{a}}^j-{\cal{a}}^jc_{\rm f})\ : { \textstyle \raisebox{-2ex}{$\widehat{}$ } } = { \cal{a}}^j , \end{aligned}\ ] ] where the first equality results from the fact that @xmath39 commutes with @xmath207@xmath64@xmath35 and the second one from the explicit calculation of the commutator . similarly , we prove by induction that the @xmath216 term of eq . ( [ eq3 ] ) is a solution of the recurrence relation provided the same is true for @xmath212 : @xmath217 & = & \frac{1}{c_{\rm f}-\widehat{c}_{\rm f } } \left[{\cal{a}}^i b_ix_n , r^j-{\cal{a}}^j\right]\:\ : { \textstyle \raisebox{-2ex}{$\widehat{}$}}\:\ : \nonumber \\ & = & \frac{1}{c_{\rm f}-\widehat{c}_{\rm f } } \left({\cal{a}}^i b_i { \cal{a}}^jx_{n-1 } + ( c_{\rm f}{\cal{a}}^j-{\cal{a}}^jc_{\rm f})x_n\right)\:\ : { \textstyle \raisebox{-2ex}{$\widehat{}$ } } \nonumber \\ % & = & \:{\textstyle % \raisebox{-2ex}{$\breve{}$}}\:\:{\cal{a}}^j & = & \:{\textstyle \raisebox{-1ex}{$\breve{}$}}\:\:{\cal{a}}^j \frac{1}{\breve{c_{\rm f}}-\widehat{c}_{\rm f}}\bigl(1 + \frac{\breve{c_{\rm f}}-c_{\rm f}}{c_{\rm f}-\widehat{c}_{\rm f}}\bigr ) { \cal{a}}^i b_ix_{n-1}\:\ : { \textstyle \raisebox{-2ex}{$\widehat{}$}}\:\ : = { \cal{a}}^jx_n \end{aligned}\ ] ] where the operator @xmath218 is in turn evaluated at `` @xmath44@xmath44 '' . finally , to see that eq . ( [ e228b ] ) is invariant under the transformation ( [ eq3 ] ) we use the facts that @xmath42\right]$]=0 and @xmath219 - c^{jl}_{ik}b^kb_l\right]=0 $ ] which follow from the algebraic closure and the jacobi identities . p. navrtil , h.b . geyer and j. dobaczewski , report no . nucl - th/9405021 ( unpublished ) . f. iachello and a. arima , _ the interacting boson model _ ( cambridge university press , cambridge , 1987 ) . balantekin , i. bars , and f. iachello , nucl . phys . * a370 * ( 1981 ) 284 . balantekin , _ a study of dynamical supersymmetries in nuclear structure _ , ph.d . thesis , yale university ( 1981 ) ( unpublished ) . j. vervier , riv . nuov . cim . * 10 * ( 1987 ) 1 . j.jolie and p.e . garrett , nucl . a596 * ( 1996 ) 234 . a. klein and e.r . marshalek , rev . * 63 * ( 1991 ) 375 . j. dobaczewski , f.g . scholtz , and h.b . geyer , phys . * c48 * ( 1993 ) 2313 . geyer , p. navrtil and j. dobaczewski , proc . int . conf . on perspectives for the interacting boson model , padova , italy , ( 1994 ) , eds . casten and a. vitturi ( world scientific , singapore , 1995 ) p.189 . pan , j .- l . ping , d.h . feng , j .- q . chen , c .- l . wu and m.w . guidry , phys . c * 53 * ( 1996 ) 715 . j. dobaczewski , h.b . geyer , and f.j.w . hahne , phys . * c44 * ( 1991 ) 1030 . a. bohm , y. neeman , a.o . barut , and others , _ dynamical groups and spectrum generating algebras _ , ( world scientific , singapore , 1988 ) . geyer , phys . * c34 * ( 1986 ) 2373 . kim and c.m . vincent , phys . * c35 * ( 1987 ) 1517 . p. ring and p. schuck , _ the nuclear many - body problem _ ( springer - verlag , berlin , 1980 ) . p. navrtil , h.b . geyer j. dobe and j. dobaczewski , ann . ( n.y . ) * 243 * ( 1995 ) 218 . scholtz , h.b . geyer and f.j.w . hahne , ann . ( n.y . ) * 213 * ( 1992 ) 74 . ginocchio , ann . ( n.y . ) * 126 * ( 1980 ) 234 . f. iachello and s. kuyucak , ann . * 136 * ( 1981 ) 19 . _ , phys . lett . * b148 * ( 1984 ) 13 . fischer , a. aprahamian , x. wu , j.x . saladin , and m.p . metlay , proc . int . conf . on perspectives for the interacting boson model , padova , italy , ( 1994 ) , eds . casten and a. vitturi ( world scientific , singapore , 1995 ) p.443 . .angular momentum content of the lowest spin(5 ) representations is presented . in the upper ( lower ) part the representations relevant to the even ( odd ) system are shown , respectively . [ cols="^,<",options="header " , ]

we present results of chandra observations of two flanking fields ( ff ) in orion , outside the orion nebula cluster ( onc ) . the observations were taken with the acis - i camera with an exposure time of about 48 ks each field . we present a catalog of 417 sources , which includes x - ray luminosity , optical and infrared photometry and x - ray variability information . we have found 91 variable sources , 33 of which have a flare - like light curve , and 11 of which have a pattern of a steady increase or decrease over a 10 hour period . the optical and infrared photometry for the stars identified as x - ray sources are consistent with most of these objects being pre - main sequence stars with ages younger than 10 myr . we present evidence for an age difference among the x - ray selected samples of ngc 2264 , orion ff , and onc , with ngc 2264 being the oldest , and onc being the youngest .

astro-ph0405003

the orion nebula is one of the best studied star formation regions in the galaxy . the orion nebula cluster ( onc ) designates the inner 3 pc ( @xmath0 20 arcmin ) from @xmath1c orionis , and contains more than three thousand stars . the core of the onc , also known as the trapezium cluster @xcite , comprises the stars located within 0.3 pc ( @xmath0 2 arcmin ) of @xmath1c orionis , it has an inferred stellar density of about 8000 stars pc@xmath2 and it is one of the densest known region of star formation @xcite . most of the onc members are visible , primarily because the massive stars in the trapezium cluster have swept the gas from the molecular cloud creating a cavity facing towards us . the structure , dynamics and stellar content of the onc have been extensively studied @xcite . the stars in the onc are less than 1 myr old and their masses range from @xmath30.05 m to @xmath050 m @xcite . studies of the outer regions of the orion nebula ( flanking fields ) have revealed the presence of young stars of ages between 1 and 3 myrs old @xcite , with an accretion disk fraction of about 40% @xcite . the orion nebula provides an excellent laboratory for studying pre - main sequence ( pms ) stars , given its proximity ( 470 pc ) , its age and its stellar diversity . pms stars are strong x - ray emitters ; typical pms stars have x - ray luminosities @xmath4 to @xmath5 above those observed in older main sequence stars @xcite . the x - ray emission in low mass stars comes from magnetic reconnection . in the dynamo model , the strength of the magnetic field , and hence the x - ray activity , depends on the rate of differential rotation and on the depth of the outer convective envelope @xcite . there is a clear relationship between rotation rate ( period ) and x - ray luminosity ( @xmath6 ) found in late - type stars in clusters as old as ngc 2547 ( 15 - 40 myrs , * ? ? ? * ) through the hyades ( @xmath0500 myrs , * ? ? ? the ratio between the x - ray and bolometric luminosity , @xmath6/@xmath7 , increases with increasing rotation rate , until the most rapidly rotating stars reach a maximum x - ray luminosity ( or saturation level ) such that @xmath6/@xmath7@xmath8 ( see * ? ? ? * and references within ) . it is much less clear that rotation and @xmath6/@xmath7 are related in younger clusters . @xcite studied a 4.5 square degree region centered in the orion nebula , using x - ray data from the einstein observatory . they found that at least 100 sources were associated with late - type pms stars , and their x - ray activity was not correlated with published rotational periods and spectroscopic rotational velocities . @xcite used rosat x - ray data in a similar study in the orion nebula , finding no dependence of x - ray activity on rotation . more recently , the chandra observatory has provided significantly improved spatial resolution and sensitivity for x - ray observations . @xcite reported x - ray observations of the onc using the chandra observatory acis - i detector . they found more than a thousand x - ray sources , 91% of them associated with known stellar members of the cluster . @xcite also see no obvious correlation between rotation and @xmath6/@xmath7 for their onc sample , and conclude that the x - ray generation mechanism for young pms stars must be different from that responsible in young main sequence stars . @xcite analyzed data for a number of young associations and clusters ( including orion ) , and agree that there is little correlation between @xmath6/@xmath7 and rotation at very young ages , but conclude that the data are consistent with a single physical mechanism , where the orion - age stars are simply all at or near the saturation level . we intend to use our orion flanking field data and other new chandra data for ngc 2264 @xcite to help determine at what age the relationship between rotation and @xmath6/@xmath7 emerges . in the present paper , we present results of chandra observations of two of the flanking fields ( ff ) in the orion nebula ( fields # 2 & # 4 from * ? ? ? * ) , just outside the onc . we discuss source detection , variability and @xmath6 determination , providing a catalog of 417 x - ray sources . we present evidence that the x - ray selected sample of pms stars in the flanking fields in orion are older than a similarly selected sample in the onc . in paper iii @xcite , we will discuss in more detail the relationships found here between rotation rate , mass accretion rate , disk indicators , and x - ray luminosity , in our field of ngc 2264 and the two orion flanking fields . two flanking fields in orion were observed with the advanced ccd imaging spectrometer ( acis ) detector on board the _ chandra x - ray observatory _ @xcite . the imaging array ( acis - i ) consists of four 1024@xmath91024 front - side illuminated ccds and covers an area on the sky of about 17@xmath917. the north orion flanking field ( noff ) is centered at @xmath10 , about 36 @xmath11 ( @xmath0 5 pc , at a distance of 470 pc ) north of the trapezium cluster , and it was observed on 2002 august 26 with a total exposure time of 48.8 ks . the south orion flanking field ( soff ) , centered at @xmath12 , about 17 @xmath11 ( @xmath0 2.5 pc , at a distance of 470 pc ) south of the trapezium cluster , was observed on 2002 september 6 with a total exposure time of 47.9 ks . figure [ dss ] shows a 1@xmath13 image of orion from the palomar digital sky survey @xcite , with both chandra acis - i fields of view marked as boxes . the fields were selected to maximize the number of stars in the field of view with known periods from @xcite and with minimal overlap with other contemporaneous chandra orion observation @xcite . data analysis was performed in the same manner as described in @xcite , where we discussed similar observations of a field in ngc2264 . in summary , we started with the level 1 processed event list , applied charge transfer inefficiency ( cti ) correction as described in @xcite , then filtered by asca grades ; by time intervals ; and by background flaring due to solar activity . removal of the background flares identified by the latter step reduced the exposure time to 46.9 ks for the noff and to 46.3 ks for the soff . finally , the energy range was restricted to 0.3 10 kev . the filtering process was done using the chandra interactive analysis of observations ( ciao ) package . figures [ orion_n ] and [ orion_s ] show the acis - i image of the filtered observations for the noff and the soff , respectively . x - ray sources were identified using the _ wavdetect _ tool within the ciao package . the _ wavdetect _ tool works well in detecting x - ray sources in crowded fields . this tool performs a mexican hat wavelet decomposition and reconstruction of the image as described in @xcite . we used wavelet scales ranging from 1 to 16 pixels in steps of @xmath14 , and the default source significance threshold of @xmath15 . the _ wavdetect _ tool was run separately in the four acis - i ccds images for both fields and it produced an original list of 256 sources for the noff and 280 sources for the soff ( see sec [ xray_phot ] ) . the positions of the sources obtained by _ wavdetect _ were correlated with 2mass positions . each x - ray source was manually checked to confirm that the 2mass counterpart lay within the radius for count extraction ( as defined in @xcite ) . we checked the astrometry of the chandra observations using all x - ray sources located within an off - axis angle ( @xmath16 ) less than 5 arcmin that have 2mass counterparts . a total of 85 noff sources and 70 soff sources meet this criteria . we determined a mean offset in r.a . of @xmath17 0.02 and a mean offset in dec of @xmath18 0.03 between the noff chandra and 2mass coordinates and a mean offset in r.a . of @xmath19 0.03 and a mean offset in dec of @xmath20 0.03 between the soff chandra and 2mass coordinates . the chandra positions were corrected by these mean offsets to put the x - ray sources in the same reference frame as their 2mass counterparts . x - ray aperture photometry was initially performed on the 536 sources detected by the ciao tool _ wavdetect_. the radius for count extraction , @xmath21 , and the annulus for background determination were defined in the same manner as in @xcite . the x - ray counts ( @xmath22 ) and background counts were extracted using the ciao tool _ dmextract_. we computed the background in counts@xmath9arcsec@xmath23 as a function of off - axis angle ( @xmath16 ) , performing a 3@xmath24 rejection fit to avoid background counts from annuli that have other sources within . the computed background was constant as a function of @xmath16 , and it had a value of @xmath25=([email protected])counts@xmath9arcsec@xmath23 for the noff and a value of @xmath25=([email protected])counts@xmath9arcsec@xmath23 for the soff . the net count , _ n. c. _ and the count rate , _ c. r. _ were computed for each source as in @xcite . we carefully inspected the light curves of all the sources ( see sec . [ variability ] ) and their appearance in the image of the chandra field of view . a total of 119 ( 22% ) sources were rejected from the original list : 107 sources had light curves consistent with cosmic ray afterglows , 5 sources were detected twice , since _ wavdetect _ was run separately in each ccd , and 7 sources were located at the edge of the field of view . of the 107 light curves having a cosmic ray shape , 106 ( 99% ) contain cosmic ray afterglows flagged by the pipeline . our final list of 417 x - ray sources ( 203 from the noff , 214 from the soff ) is provided in table [ tab_xray ] . we selected all sources with more than 500 net counts , extracted their spectra and fit them to measure their x - ray fluxes . a total of 24 sources from the noff and 20 sources from the soff meet this criteria and they are listed in table [ tab_spectra ] . the spectra were extracted within @xmath21 using the ciao tool _ dmextract_. the spectra of the noff and the soff sources are shown in figures [ spectra_n ] and [ spectra_s ] , respectively . we used the same spectral fitting process , as described by @xcite , with the redistribution matrix files ( rmf ) provided by the cti corrector @xcite , auxiliary response files ( arf ) created by the ciao tool _ mkarf _ and later corrected for the acis low energy quantum efficiency degradation . we adopted a photoelectric absorption model ( @xmath27 ) , which uses wisconsin cross sections from @xcite . this model has one parameter that is the equivalent hydrogen column density ( @xmath28 ) . the hydrogen column density was fixed to a value of 0.08@xmath29 to match an extinction value of @xmath30 = 0.41 , which is the most likely value of the observed extinction towards both orion flanking fields ( see * ? ? ? * for more discussion ) . as discussed in @xcite , the error in @xmath6 that comes from fixing @xmath28 is negligible . finally , we adopted a thermal emission model ( @xmath31 ) based on the model calculations of @xcite , and @xcite with fe l calculations by @xcite . this model includes line emissions from several elements . the remaining two parameters in the model are the plasma temperature ( kt ) and a normalization factor . as in the analysis of ngc 2264 chandra observations @xcite , we used a two temperature model to fix the x - ray spectra . the spectral parameters obtained from the two plasma temperature models are listed in table [ tab_spectra ] . the x - ray flux for the brightest sources is determined from the best spectral model derived from the mean model parameters . we computed mean plasma temperatures of ( [email protected] ) kev and ( [email protected] ) kev for the noff sources and mean plasma temperatures of ( [email protected] ) kev and ( [email protected] ) kev for the soff sources . the mean plasma temperatures are held constant and the integration of the best fit between 0.3 and 8.0 kev provides the x - ray flux . the resulting x - ray fluxes are listed in table [ tab_spectra ] . we use the x - ray fluxes of the bright sources to compute a x - ray flux weighted conversion factor between count rate and x - ray flux . we obtained a conversion factor of ( [email protected])@xmath32 ( erg/@xmath33/s)/(counts / ks ) for the noff sources and a conversion factor of ( [email protected])@xmath32 ( erg/@xmath33/s)/(counts / ks ) for the soff sources . the values of these conversion factors are in reasonably good agreement with the derived conversion factor for our similar x - ray observation in ngc 2264 ( [email protected]@xmath32 ( erg/@xmath33/s)/(counts / ks ) * ? ? ? * ) and other published values ( see * ? ? ? the x - ray flux for our catalog of x - ray sources in both orion fields is listed in column 10 of table [ tab_xray ] . the x - ray luminosity , @xmath6 , listed in column 11 of table [ tab_xray ] , is computed assuming a distance to orion of 470 pc . the limiting luminosity in our x - ray observations varies within the field of view , because of the variation of the psf across the field . the faintest source is located at @xmath34 and it has a count rate of 0.08 counts / ks , corresponding to a x - ray luminosity of 28.15 at the distance of orion . about 85 % of our sources are located within @xmath35 . at that off - axis angle the limiting count rate has increased to 0.20 counts / ks , corresponding to a x - ray luminosity of log(@xmath6)=28.5 at the distance of orion . at @xmath36 , we can not detected x - ray sources fainter than 0.45 counts / ks ( log(@xmath6 ) = 28.9 ) . therefore , we adopt a value of log(@xmath6)=28.5 dex as the limiting x - ray luminosity for our observations , keeping in mind that this value holds for sources located within @xmath37 . light curves were determined for all 417 sources detected by _ wavdetect _ using the ciao tool _ lightcurves _ with a bin time of 2500 s. the statistics of the light curves of the sources of our x - ray catalog were obtained using the xronos tool _ lcstats_. this provides , among other values , the probability of the light curve being constant , @xmath38 , as derived from the chi - square value . the @xmath38 values are listed in column 12 of table [ tab_xray ] . the light curves of the sources with @xmath39 90% were analyzed further , since they are the most likely to be variable . we defined a variable source as those having 2500 s bin light curves with @xmath39 90% , and 5000 s and 7500 s bin light curves with reduced @xmath40 2.5 , as in @xcite . there are 91 variable sources that meet this criteria . variable sources are marked with a v in column 13 of table [ tab_xray ] . there are 33 variable sources which show a flare shape , defined as a rapid increase and a slow decrease in the x - ray flux . variable sources with a flare - like light curve are marked with an additional f in column 13 of table [ tab_xray ] . there are sixteen sources that show a possible flare pattern , described as an increase in x - ray flux happening at the end of our observations or a decrease in x - ray flux occurring at the beginning of our observations . variable sources with a possible flare pattern are marked with an additional p in column 13 of table [ tab_xray ] . we have detected a fraction of about 11% flaring sources ( 8% excluding possible flares ) in both orion flanking fields . a comparable fraction ( about 8% ) was observed in a similar length observation in ngc 2264 @xcite . there are 11 sources that show a steady increase or decrease in their x - ray flux . those sources are marked with an additional s in column 13 of table [ tab_xray ] . a similar pattern is observed in two ngc 2264 sources @xcite and three x - ray ic 348 sources @xcite . these patterns might be understood by rotational modulation of x - ray flares @xcite . in figure [ light_curves ] , we show a selection of light curves of sources of comparable luminosity . in the top panels , we show two light curves with a flare shape ; in the middle - top panels , there are two sources that show a steady decrease in their x - ray flux ; in the middle - bottom panels , there are two variable light curves . finally in the bottom panels , two constant light curves are plotted . we correlated the x - ray sources with a catalog of optical and near - ir stars in the orion region ; see tables [ tab_optical ] and [ tab_xids ] . we constructed the catalog by merging @xmath030 published catalogs of optical and infrared photometry , spectral types , previous x - ray detections , proper motion surveys , periods , and spectroscopic projected rotational velocities ( @xmath41 ) . although our x - ray data presented are located in two flanking fields ( # 2 and # 4 from * ? ? ? * ) , we constructed a catalog over this whole region in an attempt to increase the number of stars with known @xmath6 and optical counterparts . the optical photometry were taken from @xcite , @xcite , @xcite , @xcite , and @xcite , in that order . for near - ir photometry , @xcite averaged over their light curves , presenting the most reliable average @xmath42 magnitude . magnitudes for the remaining stars then came from 2mass , @xcite , @xcite , and @xcite , in that order . spectral types were taken in this order : @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite . @xcite and @xcite collected much of the optical and spectroscopic literature for stars in this region , reporting on x - ray detections using einstein and rosat , respectively . @xcite analyzed einstein data for some stars in this region . the largest x - ray data base , and the one obviously most similar to our own data , is @xcite ; these chandra acis x - ray detections are in the onc region ; their fields have marginal spatial overlap with the observations reported here . finally , there are several surveys in this region devoted specifically to obtaining rotation data . periods were taken from @xcite , @xcite , @xcite , @xcite , @xcite , in that order . spectroscopic rotational velocities came from @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and finally @xcite . we found that 356 ( 85% ) of our 417 x - ray sources have optical and/or infrared counterparts ( 130 ( 31% ) have known periods ) . each x - ray source was manually checked to confirm that the optical and/or infrared counterparts lay within the radius for count extraction ( @xmath21 , as defined in @xcite ) . the 356 sources and their corresponding optical and/or infrared photometry are listed in table [ tab_optical ] . other names of the sources given by the different catalogs used in our compilation are listed in table [ tab_xids ] . there were cases in which two optical / near - ir counterparts were located within @xmath21 , which is a measure of the size of the psf in the chandra field . this was the case for 9 x - ray sources , shown in figure [ find_chart ] . these sources are marked with a c in table [ tab_xray ] and are matched to the closest counterpart in each case . source n134 seems to be have two components separated by @xmath0 1 . the optical ( par 1970 ) and 2mass counterparts coincide with the weaker component . par 1970 is a b2 star , which should be weak in x - rays . source n217 has two pairs of optical and 2mass counterparts separated by @xmath0 6.5 . source s025 has two pairs of optical/2mass counterparts within the extraction circle . one pair is at the edge of the extraction circle and the other within 1.3 of the x - ray position . source s091 is an extended source , possibly with two components . source s101 has two optical and one infrared source within the extraction circle . source s141 has two optical sources within the extraction circle . source s182 has a large aperture and its extraction circle contains emission from a nearby x - ray source . sources s187 and s191 are separated by 5.5 at an off axis angle of about 8.5. their circles of extraction ( @xmath43 10 ) overlap considerably . source s187 has an optical counterpart ( par 1897 ) , while source s191 has an infrared counterpart ( 2mass j05351696 - 0532464 ) . optical and/or infrared spectroscopy of these objects may help determine the true counterpart of these x - ray sources . there are 902 stars in our optical / infrared catalog with @xmath44 , @xmath45 , and @xmath46 photometry and with positions inside the field of view of our chandra observations . among those 902 stars , 316 ( 35% ) have x - ray chandra counterparts . figure [ j_histogram ] shows a @xmath44 magnitude histogram of all the 902 stars with @xmath44 , @xmath45 , and @xmath46 photometry and positions inside the chandra field ( solid line ) and the histogram of the 316 x - ray chandra counterparts ( dotted line ) . the completeness limit of the infrared sample is that of 2mass . we can see in figure [ j_histogram ] that our infrared sample goes deeper than the sample of stars with x - ray counterparts . a similar behavior is seen in the optical sample histogram . this means that all the x - ray sources should have been matched to sources in our optical / infrared catalog if they are associated with stars earlier than m4 at the distance of orion in the absence of extinction . among the orion x - ray sources in our catalog , 130 ( 31% of the 417 x - ray source ) have known periods from @xcite . among the orion x - ray sources with known periods , 46 ( 35% of the 130 x - ray sources with known periods ) are classified as variable x - ray sources , and 24 ( 18% of the 130 x - ray sources with known periods ) show a flare or possible flare pattern . nearly 52% of the 46 flaring x - ray sources are periodic in optical wavelengths , while only 31 % of all the x - ray sources show a periodic optical variation . there are twelve sources with short enough optical periods so that a significant fraction of the period is observed in the time length of the x - ray chandra observations . however , the low signal to noise of the light curves and the presence of x - ray flares prevent us from drawing any conclusions regarding the presence or absence of rotational modulation of the x - ray flux . there are 61 x - ray sources that do not have optical or infrared counterparts and they are marked with a x in column 13 of table [ tab_xray ] . based on the limiting magnitude of our optical catalog ( @xmath47 20 mag ) , the x - ray to optical flux ratio , @xmath48 , of the sources with only x - ray detection is @xmath49 . therefore , these sources could be either late m dwarfs or extragalactic objects @xcite . given the presence of a dark cloud behind orion , the detection of background m dwarfs in x - rays is unlikely , so , if these sources detected only in x - rays are m dwarfs , then they should be @xmath50470 pc . we have identified 17 of these sources detected only in x - rays with sources in a deep 2mass image ( kopan 2004 , priv comm , @xmath0 2 magnitudes deeper than the 2mass all sky point source catalog ) reducing the number of sources detected only in x - rays to 44 . based on the limiting flux of our x - ray sample , the maximun x - ray luminosity of m dwarfs @xcite , and the limiting magnitude of the deep 2mass image , we determined that all foreground m dwarfs present in our x - ray sample should have been detected in the deep 2mass image . source counts in the chandra south and north deep fields @xcite predict the presence of about 20 agn in each acis field of view at the flux limit of our observations and assuming a galactic extinction of @xmath51 11 magnitudes ( from ned extinction calculator ) . @xcite found 101 x - ray sources with no optical counterparts in their 82 ks chandra observation of the onc , all highly concentrated towards the center of the onc . they argued that most of those sources were indeed heavily embedded protostars or young stars seen behind the cloud . they estimated that the extragalactic contamination can not be more than 15 objects , due to the high extinction of the onc behind the stars . in our case , the extinction in front of the orion flanking field stars is significantly lower than on the onc @xcite and our only x - ray sources seem to be concentrated in areas of low extinction determined from a co map of the area @xcite and @xmath52 color of 2mass sources in those areas . thus , we believe that the most likely explanation is that most of the objects detected only in x - rays are active galaxies . deeper infrared photometry or deeper x - ray observations may help determine the true nature of these sources . there are 17 stars for which rotation and spectral type information exist in our optical / infrared catalog , and for which no x - ray counterpart was found . in order to allow us to use these stars in the next paper , we determined upper limits for their x - ray luminosity in the same manner as @xcite . the upper limit for the x - ray luminosity was computed assuming a distance to orion of 470 pc . the upper limit results are listed in table [ tab_upp_lim ] . in figure [ color_color ] , we have plotted the @xmath53 , @xmath52 color - color diagram of all the infrared sources in the field of view of our chandra observation . most of the x - ray sources with infrared colors are located near the locus of the classical t - tauri ( ctt ) stars @xcite . the remaining sources are located within the reddening vectors of late main sequence stars and ctt stars . there are 445 stars in our optical / infrared catalog with @xmath54 and @xmath55 photometry and with positions inside both fields of view of the chandra observations . among those 445 stars , 280 ( 63 % ) have x - ray chandra counterparts . in figure [ cmd1 ] , we have plotted ( @xmath56)@xmath57 color magnitude diagrams of the optical sources in the fields of view of our chandra observations . the dereddened @xmath56 color and the absolute magnitude @xmath57 were obtained assuming an average extinction of @xmath30 = 0.41 @xcite , and dereddening relationships from @xcite and @xcite . in both panels , we have also plotted isochrones from @xcite , as a reference . the dashed line corresponds to @xmath57 = 8.75 mag . , which is the lower limit for a low mass orion member rotating at the saturation level ( log(@xmath6/@xmath7)=3 ) with log(@xmath6)=28.5 ( limit luminosity of our x - ray sample ) . this means that the most slowly rotating stars earlier than m3 with optical counterparts should be detected by our x - ray sample . most of our x - ray sources are younger than @xmath58 years at the distance of orion , and furthermore the x - ray sources are heavily concentrated between the @xmath59 and the @xmath60 year old isochrones with respect to the general population . stars that are strong x - ray emitters are likely to be young stars and therefore true members of the orion cluster . in @xcite , as a result of the lack of membership information for stars this far from the trapezium , a provisional membership criterion was defined . a line was drawn through an empirically - discovered gap nearly coincident with the 3 myr isochrone to divide the stars clumping above it in a locus ( likely members ) from the ones below it ( likely field stars ) . this approximation was supported in @xcite by the locations of the periodic stars , also likely young and therefore members . the clustering of the x - ray sources above the 3 myr isochrone is additional confirmation that stars in the locus are more likely to be true members of the cluster than stars below this locus . there is one star ( s038 ) with @xmath57 brighter than 0 magnitude . its optical counterpart , par 1605 , has a spectral type of a0v . it has log(@xmath6)=30.1 and @xmath6/@xmath7=@xmath61 . x - ray emission from late b stars and early a stars is rare , since they are in the transition between wind driven x - ray emission from hot stars and magnetically driven coronae from late type stars . it is often speculated that the x - ray emission observed in late b and early a stars comes from unseen late type companions @xcite , that explanation could apply to par 1605 . there are 3 x - ray sources that appear to be much older than @xmath58 years . one of them ( 141s ) , has unreliable photometry because of background contamination from a nearby star . the other sources ( n068 and s115 ) lack both spectral type and membership information . if we deredden sources n068 and s115 ( see figure [ color_color ] ) along reddening vector towards the locus of ctt stars , we derive a @xmath30 value of 5.9 magnitudes for source n068 and @xmath30 value of 2.1 magnitudes for source s115 . the locations of sources n068 and s115 are indicated along with their dereddened positions in figure [ color_color ] . in figure [ cmd4 ] , we have plotted the @xmath62 color magnitude diagram of all the infrared sources with x - ray counterparts , with isochrones from @xcite . the locations of sources n068 and s115 are indicated along with their possible dereddened position . it is possible that these two objects are the youngest and most embedded of our x - ray sources , if they are indeed members of the orion complex . the unusual optical colors of these sources could be explained by scattered light from an edge on disk . figure [ xcmd ] shows the dereddened color magnitude diagram for three x - ray selected samples . on the left panel , we plot the ngc 2264 sample from @xcite , which includes 201 stars showing x - ray emission . the central panel shows orion ff data from the present work . finally , the right panel shows 584 x - ray onc sources @xcite that have optical counterparts from our database . the stars from the three samples have been dereddened using the following criteria . if the star has a known spectral type , we have used @xmath63 derived from the spectra . for the remaining stars , we have derived @xmath63 from @xmath42 photometry , assuming that the stars are pms stars . we use @xmath63 from infrared colors if this is greater than one magnitude . if not , we use the median @xmath63 for each field ( @xmath63=0.25 for ngc 2264 , @xmath63=0.25 for orion ff , and @xmath63=1.31 for the onc , herbig & terndrup 1986 ) . this criterion is reasonable for fields with variable reddening , as the onc , where @xmath63 can be as high as 4 magnitudes . there are several notable differences among the shown cmds . ngc 2264 stars appear to be more compactly distributed than the onc stars . there are fewer high mass stars in the orion ff sample than in the onc sample . the ngc 2264 sample appears to be the oldest of the three samples . one way to quantify these differences is through the use of statistical boxes . we have divided each x - ray selected sample into smaller samples containing stars of a determined color range of 0.5 magnitudes of width . for each color range we have determined a box , as defined by @xcite ( see also * ? ? ? the central horizontal line in each box is the median @xmath64 for the range in @xmath65 color , while the bottom and the top of the box shows its inter quartile range ( containing 50% of the sample ) , and the vertical lines coming out of the box mark the position of the adjacent points of the sample ( most extreme values in the sample that are not more than 1.5 times the inter quartile range ) . in figure [ xcmd_box ] we show the box plots for the three samples together with isochrones from @xcite . we have only plotted the boxes that overlap with all three samples , are within the range of isochrones ( @xmath66 ) , and are not subject to biases from the optical limit of the catalog or other intrinsic properties ( @xmath67 ; we have also plotted in the three panels horizontal dotted lines that correspond to the median value of the orion ff boxes . the median , upper and lower quartile @xmath64 values are listed in table [ tab_boxes ] for each of the @xmath65 color ranges and for three the samples of x - ray sources . all the ngc 2264 medians are systematically lower than the orion ff medians , with a mean difference of 0.3 magnitudes ( @xmath24=0.1 ) . all the onc medians are systematically higher than the orion ff medians , with a mean difference of 0.3 magnitudes ( @xmath24=0.1 ) . the effect of reddening is very small , since the reddening vectors go almost parallel to the isochrones . if we used the median reddening for each field , we find the same mean differences in the median @xmath64 for each color range . if we used a @xmath30 one magnitude higher or one magnitude lower , we get again the same differences in the median @xmath64 . the observed differences can be explained by a difference in age among the three samples . using the @xcite isochrones , we derive an age of @xmath02 myrs , @xmath01 myrs , and @xmath00.5 myrs for the x - ray selected samples of ngc 2264 , orion ff , and onc , respectively . alternatively , assuming vertical hayashi tracks , stellar luminosities scale as @xmath68 @xcite . thus , differences in luminosity of 0.3 magnitudes corresponds to ratios in age of a factor of @xmath01.5 . searching for young stars in x - rays is known to be a efficient way of finding young stars ( for example , the rass , e.g. , * ? ? ? does the age difference for the * x - ray selected * stars in the three regions apply as well for all low mass members in the three regions ? our own belief is that x - ray selection does not bias yso population statistics in terms of age and mass ( e.g. * ? ? ? * ) and therefore our age estimates should be valid for the entire set of stars in the three regions . we caution however that x - ray selection may result in bias for other physical parameters , such as rotation or disk properties ( e.g. * ? ? ? * ; * ? ? ? in order to provide some support to our assertion concerning x - ray selection bias and age , we have constructed plots like figure [ xcmd_box ] for low mass members of the three clusters located within the chandra field of view but not restricted to x - ray detection . we derived box - plots for the optical sample of likely members in all three regions using two different reddening prescriptions - one with a mean reddening for each field , and one where we derived reddening exactly as for figure [ xcmd_box ] . in both cases , we derived similar ages and age differences for the three fields . we note that this result differs from that reported in @xcite and @xcite , where no age difference was found between optical samples in the onc and the orion flanking fields . the earlier studies used larger fields and a different statistical method that may have smeared out any age differences that might be present . we note that none of the essential conclusions in earlier papers change in any way if the orion flanking fields are indeed older ( or contain a range of older ages ) than the onc . we present a catalog of orion x - ray sources from two flanking fields . the observations were taken with the acis - i on board the chandra x - ray observatory . the catalog , consisting of 417 sources , includes x - ray luminosity , optical and infrared photometry and x - ray variability information . we found 91 variable sources , 33 of which have a flare like light curve , and 11 of which have a pattern of a steady increase or decrease . from the optical and infrared counterparts of the x - ray sources , we have learned that most of the x - ray sources have colors consistent with ctts that are younger than @xmath58 years . we argue that the data are consistent with an age difference among the x - ray selected samples of ngc 2264 , orion ff , and onc , with ngc 2264 being the oldest , and onc being the youngest . this catalog of x - ray sources will be used to study the relationship between rotational properties and x - ray characteristics of orion and ngc 2264 stars in paper iii @xcite . we plan to discuss correlations of @xmath6/@xmath7 with rotation rate ( period and @xmath69sin@xmath70 ) , disk indicators ( @xmath71 , @xmath52 , @xmath72 , and @xmath73 ) , and mass accretion rate as derived from @xmath72 excess . we will also compare the @xmath6/@xmath7 values found here with those from other young clusters . we thank the anonymous referee for her / his careful review of the manuscript . financial support for this work was provided by nasa grant go2 - 3011x . this research has made extensive use of nasa s astrophysics data system abstract service , the simbad database , operated at cds , strasbourg , france , and the nasa / ipac infrared science archive , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . the research described in this paper was partially carried out at the jet propulsion laboratory , california institute of technology , under a contract with the national aeronautics and space administration . brandt , w. n. , alexander , d. m. , hornschemeier , a. e. , garmire , g. p. , schneider , d. p. , barger , a. j. , bauer , f. e. , broos , p. s. , cowie , l. l. , townsley , l. k. , burrows , d. n. , chartas , g. , feigelson , e. d. , griffiths , r. e. , nousek , j. a. , & sargent , w. l. w. , 2001 , , 122 , 2810 reid , i. n. , brewer , c. , brucato , r. j. , mckinley , w. r. , maury , a. , mendenhall , d. , mould , j. r. , mueller , j. , neugebauer , g. , phinney , j. , sargent , w. l. w. , schombert , j. , & thicksten , r. , 1991 , , 661 rosati , p. , tozzi , p. , giacconi , r. , gilli , r. , hasinger , g. , kewley , l. , mainieri , v. , nonino , m. , norman , c. , szokoly , g. , wang , j. x. , zirm , a. , bergeron , j. , borgani , s. , gilmozzi , r. , grogin , n. , koekemoer , a. , schreier , e. , zheng , w. , 2002 , , 566 , 667 rcllcccccccrc + n001 & cxorrs j053442.6 - 044215 & 5 34 42.70 & -4 42 15.294 & 10.93 & 17.50 & 1.00 & 35.9 & 17.25 & 0.114e-12 & 30.48 & 93 & + n002 & cxorrs j053444.5 - 044214 & 5 34 44.53 & -4 42 14.807 & 10.55 & 16.30 & 1.00 & 36.4 & 12.88 & 0.848e-13 & 30.35 & 0 & v , s + n004 & cxorrs j053448.2 - 044740 & 5 34 48.24 & -4 47 40.865 & 7.77 & 9.50 & 0.98 & 40.9 & 36.05 & 0.237e-12 & 30.80 & 0 & v , s + n005 & cxorrs j053448.5 - 044956 & 5 34 48.60 & -4 49 56.658 & 7.78 & 8.70 & 0.96 & 24.7 & 1.56 & 0.103e-13 & 29.43 & 99 & + n007 & cxorrs j053451.0 - 044341 & 5 34 51.05 & -4 43 41.614 & 8.39 & 10.10 & 0.98 & 38.7 & 10.48 & 0.690e-13 & 30.26 & 0 & v , f + n008 & cxorrs j053451.2 - 044757 & 5 34 51.28 & -4 47 57.405 & 7.00 & 9.40 & 0.99 & 41.6 & 55.40 & 0.365e-12 & 30.98 & 23 & v + n010 & cxorrs j053452.4 - 044941 & 5 34 52.46 & -4 49 41.023 & 6.85 & 6.70 & 0.97 & 36.5 & 1.04 & 0.684e-14 & 29.26 & 100 & + n011 & cxorrs j053453.0 - 044811 & 5 34 53.02 & -4 48 11.677 & 6.55 & 6.20 & 0.96 & 40.5 & 0.44 & 0.290e-14 & 28.88 & 100 & + n012 & cxorrs j053453.8 - 044340 & 5 34 53.82 & -4 43 40.195 & 7.83 & 8.80 & 0.98 & 36.7 & 0.44 & 0.290e-14 & 28.88 & 100 & + n015 & cxorrs j053454.3 - 045413 & 5 34 54.31 & -4 54 13.447 & 8.63 & 10.70 & 0.97 & 40.2 & 2.10 & 0.138e-13 & 29.56 & 97 & + n016 & cxorrs j053454.4 - 044540 & 5 34 54.44 & -4 45 40.342 & 6.73 & 6.50 & 0.97 & 41.8 & 0.26 & 0.171e-14 & 28.66 & 100 & + n017 & cxorrs j053454.5 - 045604 & 5 34 54.58 & -4 56 4.736 & 9.96 & 14.40 & 0.98 & 32.5 & 3.88 & 0.255e-13 & 29.83 & 99 & + n018 & cxorrs j053455.1 - 044827 & 5 34 55.12 & -4 48 27.968 & 6.05 & 5.30 & 0.96 & 40.6 & 24.67 & 0.162e-12 & 30.63 & 2 & v + n019 & cxorrs j053455.6 - 045611 & 5 34 55.67 & -4 56 11.579 & 9.89 & 14.20 & 0.98 & 39.3 & 3.78 & 0.249e-13 & 29.82 & 99 & + n020 & cxorrs j053456.3 - 044548 & 5 34 56.31 & -4 45 48.316 & 6.25 & 5.70 & 0.97 & 37.7 & 1.14 & 0.750e-14 & 29.30 & 100 & x + n021 & cxorrs j053456.3 - 044437 & 5 34 56.39 & -4 44 37.961 & 6.77 & 6.60 & 0.97 & 41.5 & 1.79 & 0.118e-13 & 29.49 & 100 & x + n022 & cxorrs j053456.8 - 044605 & 5 34 56.81 & -4 46 5.044 & 6.01 & 5.30 & 0.97 & 39.8 & 8.12 & 0.534e-13 & 30.15 & 95 & + n025 & cxorrs j053457.9 - 044913 & 5 34 57.90 & -4 49 13.139 & 5.43 & 4.40 & 0.95 & 19.6 & 14.61 & 0.961e-13 & 30.40 & 90 & + n026 & cxorrs j053458.2 - 045052 & 5 34 58.22 & -4 50 52.031 & 5.86 & 5.00 & 0.96 & 40.1 & 0.27 & 0.178e-14 & 28.67 & 100 & + n027 & cxorrs j053459.3 - 045011 & 5 34 59.35 & -4 50 11.837 & 5.36 & 4.30 & 0.95 & 42.7 & 2.40 & 0.158e-13 & 29.62 & 89 & + n029 & cxorrs j053459.6 - 044756 & 5 34 59.63 & -4 47 56.143 & 4.92 & 3.70 & 0.95 & 43.4 & 0.68 & 0.447e-14 & 29.07 & 100 & x + n030 & cxorrs j053459.7 - 045158 & 5 34 59.70 & -4 51 58.602 & 6.18 & 5.50 & 0.96 & 42.0 & 0.62 & 0.408e-14 & 29.03 & 98 & + n031 & cxorrs j053459.8 - 045526 & 5 34 59.87 & -4 55 26.028 & 8.66 & 10.80 & 0.98 & 40.3 & 1.18 & 0.776e-14 & 29.31 & 100 & + n034 & cxorrs j053500.7 - 044649 & 5 35 0.80 & -4 46 49.233 & 4.85 & 3.60 & 0.95 & 34.8 & 1.12 & 0.737e-14 & 29.29 & 99 & + n035 & cxorrs j053500.9 - 044819 & 5 35 0.92 & -4 48 19.319 & 4.60 & 3.30 & 0.95 & 43.7 & 0.93 & 0.612e-14 & 29.21 & 98 & + n037 & cxorrs j053501.9 - 044115 & 5 35 1.96 & -4 41 15.366 & 8.22 & 9.70 & 0.98 & 40.8 & 4.65 & 0.306e-13 & 29.91 & 13 & v + n038 & cxorrs j053502.0 - 044731 & 5 35 2.09 & -4 47 31.843 & 4.36 & 3.00 & 0.95 & 34.9 & 1.14 & 0.750e-14 & 29.30 & 100 & + n039 & cxorrs j053502.3 - 044755 & 5 35 2.33 & -4 47 55.508 & 4.21 & 2.90 & 0.95 & 32.3 & 2.64 & 0.174e-13 & 29.66 & 98 & x + cccccc + n093 & par 1817 & 209.4 & 0.77 & 3.40 & 14.20 + n256 & par 2257 & 65.6 & 0.70 & 2.87 & 4.03 + n218 & par 2140 & 60.6 & 0.56 & 1.96 & 3.61 + n087 & par 1798 & 56.0 & 1.05 & 3.62 & 3.78 + n008 & par 1621 & 55.4 & 0.40 & 2.63 & 3.60 + n097 & par 1834 & 52.5 & 0.00 & 3.58 & 3.35 + n126 & par 1950 & 43.5 & 0.65 & 2.22 & 2.65 + n044 & par 1701 & 35.4 & 0.00 & 2.74 & 2.54 + n004 & par 1598 & 36.0 & 0.00 & 2.11 & 2.21 + n137 & par 1967 & 30.3 & 0.31 & 1.39 & 1.88 + n165 & par 2043 & 41.4 & 0.30 & 1.31 & 2.54 + n124 & par 1935 & 27.8 & 0.00 & 1.23 & 1.65 + n190 & par 2081 & 29.5 & 0.85 & 4.40 & 1.77 + n222 & par 2145 & 27.0 & 0.61 & 2.40 & 1.54 + n018 & par 1651 & 24.7 & 0.84 & 3.88 & 1.60 + n195 & r01 2133 & 21.1 & 0.40 & 3.25 & 1.25 + n080 & par 1778 & 20.3 & 0.00 & 4.02 & 1.16 + n208 & par 2109 & 17.0 & 0.24 & 1.07 & 0.91 + n081 & smmv 1944 & 17.3 & 0.00 & 1.96 & 1.01 + n156 & par 2017 & 67.2 & 0.00 & 1.38 & 4.06 + n046 & par 1710 & 16.4 & 0.89 & 4.36 & 0.95 + n130 & par 1948 & 16.1 & 0.87 & 5.02 & 0.91 + n001 & r01 1413 & 17.2 & 0.00 & 6.76 & 0.82 + n175 & par 2064 & 12.9 & 0.00 & 1.01 & 0.75 + + s249 & par 2069 & 200.9 & 0.85 & 3.71 & 13.30 + s168 & par 1828 & 100.8 & 1.00 & 4.80 & 7.21 + s238 & par 2048 & 103.8 & 0.66 & 3.34 & 7.25 + s020 & par 1553 & 52.0 & 0.70 & 2.43 & 3.40 + s198 & par 1929 & 37.5 & 0.66 & 2.23 & 2.41 + s040 & par 1756 & 34.7 & 0.77 & 2.70 & 2.16 + s183 & par 1874 & 31.6 & 1.00 & 4.45 & 2.13 + s152 & par 1787 & 27.3 & 1.33 & 5.82 & 1.85 + s058 & par 1643 & 26.5 & 0.64 & 2.02 & 1.34 + s172 & par 1846 & 21.2 & 0.64 & 2.65 & 1.26 + s202 & par 1942 & 19.5 & 0.78 & 2.66 & 1.12 + s174 & par 1848 & 19.5 & 0.86 & 3.70 & 1.14 + s022 & par 1564 & 17.8 & 0.64 & 1.63 & 1.12 + s047 & par 1613 & 16.0 & 0.69 & 1.76 & 0.86 + s016 & par 1535 & 15.4 & 0.64 & 1.63 & 0.85 + s039 & par 1616 & 15.6 & 0.64 & 1.47 & 0.85 + s176 & par 1876 & 14.4 & 0.81 & 3.36 & 0.86 + s079 & r01 1603 & 13.1 & 0.32 & 3.58 & 0.64 + s025 & par 1571 & 13.4 & 1.04 & 1.88 & 0.84 + s104 & chs 7005 & 21.2 & 0.00 & 0.00 & 0.96 cllrrrrrrr r01 1413 & 5 34 42.616 & -4 42 14.86 & n001 & & 18.79 & 15.52 & 13.07 & 10.99 & 9.94 + par 1567 & 5 34 44.512 & -4 42 13.71 & n002 & 0.04 & 15.38 & 13.53 & 12.36 & 11.49 & 10.89 + par 1598 & 5 34 48.155 & -4 47 39.98 & n004 & 0.11 & 16.63 & 14.08 & 12.05 & 10.73 & 9.97 + chs 5807 & 5 34 48.548 & -4 49 56.72 & n005 & & & & & 13.97 & 12.07 + par 1620 & 5 34 50.988 & -4 43 41.35 & n007 & & 16.02 & 13.60 & 12.14 & 11.09 & 10.43 + par 1621 & 5 34 51.204 & -4 47 56.91 & n008 & & 15.85 & 13.45 & 11.88 & 10.66 & 9.96 + r01 1536 & 5 34 52.410 & -4 49 40.29 & n010 & & & 19.34 & 16.40 & 13.76 & 12.45 + par 1639 & 5 34 52.963 & -4 48 10.56 & n011 & & & 17.10 & 15.20 & & 12.23 + r01 1563 & 5 34 53.670 & -4 43 40.69 & n012 & & & 18.12 & 15.82 & 14.57 & 13.95 + 2mass j05345431 - 0454129 & 5 34 54.313 & -4 54 13.447 & n015 & & & & & 14.81 & 12.42 + r01 1570 & 5 34 54.391 & -4 45 39.22 & n016 & & & 19.70 & 16.11 & 13.83 & 13.11 + r01 1572 & 5 34 54.501 & -4 56 4.96 & n017 & & & 18.05 & 14.65 & 12.11 & 10.93 + par 1651 & 5 34 55.052 & -4 48 27.54 & n018 & 0.47 & 16.27 & 14.12 & 12.44 & 11.13 & 10.41 + r01 1586 & 5 34 55.574 & -4 56 11.23 & n019 & & & 19.11 & 15.25 & 12.47 & 11.00 + chs 6568 & 5 34 56.817 & -4 46 4.78 & n022 & & & & & & 11.46 + r01 1610 & 5 34 57.827 & -4 49 12.54 & n025 & & & 16.99 & 14.43 & 12.73 & 11.91 + r01 1620 & 5 34 58.184 & -4 50 51.45 & n026 & & & 18.61 & 15.72 & 13.30 & 12.38 + par 1672 & 5 34 59.269 & -4 50 11.41 & n027 & & & 15.85 & 13.38 & 11.77 & 10.93 + r01 1635 & 5 34 59.636 & -4 51 57.50 & n030 & & & 21.60 & 16.58 & 13.94 & 12.94 + 2mass j05345988 - 0455272 & 5 34 59.874 & -4 55 26.028 & n031 & & & & & 17.05 & 14.02 + r01 1647 & 5 35 0.740 & -4 46 49.07 & n034 & & & 16.52 & 13.99 & 12.44 & 11.69 + r01 1651 & 5 35 0.846 & -4 48 18.71 & n035 & & & 17.23 & 14.22 & 12.38 & 11.58 + r01 1659 & 5 35 1.890 & -4 41 14.54 & n037 & 1.35 & & 18.86 & 15.55 & 13.22 & 12.14 + r01 1660 & 5 35 2.000 & -4 47 30.73 & n038 & & & 21.16 & 15.93 & 13.49 & 12.86 + par 1692 & 5 35 2.303 & -4 49 15.92 & n040 & 0.24 & & 16.74 & 14.28 & 12.70 & 11.88 + r01 1663 & 5 35 2.574 & -4 49 29.12 & n042 & & 17.68 & 14.87 & 12.60 & 11.09 & 10.22 + par 1700 & 5 35 2.891 & -4 48 32.54 & n043 & & 16.57 & 16.52 & 14.61 & 13.21 & 12.51 + par 1701 & 5 35 3.229 & -4 49 20.29 & n044 & & 14.31 & 13.01 & 11.32 & 9.96 & 9.08 + r01 1677 & 5 35 3.252 & -4 56 42.63 & n045 & & 18.24 & 15.60 & 13.58 & 12.33 & 11.57 + par 1710 & 5 35 3.622 & -4 50 52.77 & n046 & & 16.47 & 15.11 & 13.19 & 11.47 & 10.36 + 2mass j05350376 - 0447516 & 5 35 3.748 & -4 47 51.576 & n047 & & & & & 14.24 & 13.47 + par 1709 & 5 35 4.292 & -4 46 42.20 & n048 & & & & & 9.79 & 9.26 + 2mass j05350469 - 0452418 & 5 35 4.668 & -4 52 41.560 & n049 & & & & & 14.35 & 12.67 + rlllll n001 & & r01 1413 & chs 5282 & 2mass j05344268 - 0442148 & smmv 1089 + n002 & par 1567 & r01 1429 & chs 5438 & 2mass j05344454 - 0442146 & tian 125,rosat 42,smmv 1126 + n004 & par 1598 & r01 1483 & & 2mass j05344823 - 0447401 & rosat 52 + n005 & & & chs 5807 & 2mass j05344854 - 0449568 & + n007 & par 1620 & r01 1521 & & 2mass j05345106 - 0443414 & + n008 & par 1621 & r01 1522 & chs 6064 & 2mass j05345128 - 0447570 & einstein 56,rosat 59 + n010 & & r01 1536 & & 2mass j05345249 - 0449404 & + n011 & par 1639 & r01 1542 & & 2mass j05345302 - 0448104 & + n012 & & r01 1563 & & 2mass j05345374 - 0443407 & + n015 & & & & 2mass j05345431 - 0454129 & + n016 & & r01 1570 & & 2mass j05345446 - 0445393 & + n017 & & r01 1572 & & 2mass j05345458 - 0456053 & + n018 & par 1651 & r01 1578 & chs 6412 & 2mass j05345513 - 0448277 & rosat 72 + n019 & & r01 1586 & chs 6469 & 2mass j05345566 - 0456117 & + n022 & & & chs 6568 & 2mass j05345682 - 0446047 & + n025 & & r01 1610 & & 2mass j05345791 - 0449127 & smmv 1452 + n026 & & r01 1620 & & 2mass j05345826 - 0450517 & + n027 & par 1672 & r01 1633 & & 2mass j05345935 - 0450117 & + n030 & & r01 1635 & & 2mass j05345971 - 0451578 & + n031 & & & & 2mass j05345988 - 0455272 & + n034 & & r01 1647 & & 2mass j05350082 - 0446491 & + n035 & & r01 1651 & & 2mass j05350093 - 0448188 & + n037 & & r01 1659 & chs 7084 & 2mass j05350196 - 0441145 & + n038 & & r01 1660 & & 2mass j05350208 - 0447308 & + n040 & par 1692 & r01 1662 & & 2mass j05350238 - 0449161 & + n042 & & r01 1663 & & 2mass j05350265 - 0449293 & + n043 & par 1700 & r01 1668 & chs 7175 & 2mass j05350298 - 0448326 & + n044 & par 1701 & r01 1676 & chs 7211 & 2mass j05350326 - 0449209 & tian 158 + n045 & & r01 1677 & & 2mass j05350333 - 0456430 & + n046 & par 1710 & r01 1678 & chs 7251 & 2mass j05350370 - 0450530 & + n047 & & & & 2mass j05350376 - 0447516 & + n048 & par 1709 & & & 2mass j05350421 - 0446435 & tian 161 + n049 & & & & 2mass j05350469 - 0452418 & + n050 & & & & 2mass j05350481 - 0447089 & + lllllccccc & & chs 10870 & & & @xmath3 6.30 & @xmath3 0.3447 & @xmath30.0227 & @xmath3 28.78 + par 2083 & & & & tian 250 & @xmath310.70 & @xmath3 0.4994 & @xmath30.0329 & @xmath3 28.94 + par 2131 & & & 2mass j05353948 - 0451216 & tian 258 & @xmath3 9.80 & @xmath3 0.2455 & @xmath30.0161 & @xmath3 28.63 + par 1654 & r01 1599 & & 2mass j05345622 - 0445574 & tian 150 & @xmath3 8.00 & @xmath3 0.2016 & @xmath30.0133 & @xmath3 28.54 + par 1708 & & & 2mass j05350478 - 0443546 & tian 162 & @xmath3 8.40 & @xmath3 0.2328 & @xmath30.0153 & @xmath3 28.61 + & & chs 8034 & 2mass j05351065 - 0442075 & & @xmath3 6.30 & @xmath3 0.1464 & @xmath30.0096 & @xmath3 28.41 + & & chs 8467 & 2mass j05351406 - 0453112 & & @xmath3 5.10 & @xmath3 0.1671 & @xmath30.0110 & @xmath3 28.46 + & r01 1937 & chs 9321 & 2mass j05351974 - 0448180 & & @xmath3 3.00 & @xmath3 0.1881 & @xmath30.0124 & @xmath3 28.51 + & r01 1720 & & 2mass j05350693 - 0449097 & & @xmath313.90 & @xmath3 0.8425 & @xmath30.0554 & @xmath3 29.17 + & r01 1806 & & 2mass j05351282 - 0539077 & jw 398 & @xmath3 3.00 & @xmath3 0.2237 & @xmath30.0150 & @xmath3 28.60 + & & chs 9056 & 2mass j05351795 - 0535157 & jw 579,smmv 2470 & @xmath3 6.70 & @xmath3 0.1530 & @xmath30.0103 & @xmath3 28.43 + & & & 2mass j05351898 - 0537234 & hbjm 10548 & @xmath3 3.80 & @xmath3 0.1427 & @xmath30.0096 & @xmath3 28.40 + & & & & jw 618 & @xmath3 5.60 & @xmath3 0.1889 & @xmath30.0127 & @xmath3 28.53 + par 1810 & r01 1795 & & 2mass j05351235 - 0536403 & jw 384 & @xmath310.10 & @xmath3 0.4175 & @xmath30.0280 & @xmath3 28.87 + par 1898 & & & 2mass j05351596 - 0539147 & jw 514,tian 196 & @xmath3 3.00 & @xmath3 0.1990 & @xmath30.0134 & @xmath3 28.55 + & r01 1949 & & 2mass j05352074 - 0537536 & jw 667 & @xmath3 8.50 & @xmath3 0.3344 & @xmath30.0225 & @xmath3 28.77 + & r01 1451 & & 2mass j05344609 - 0537312 & jw 79 & @xmath312.10 & @xmath3 0.3498 & @xmath30.0235 & @xmath3 28.79 + cccccccccc 1.01.5 & 3.81 & 4.21 & 4.44 & 3.18 & 3.71 & 4.22 & 2.66 & 3.43 & 3.90 + 1.52.0 & 4.30 & 4.88 & 5.25 & 4.15 & 4.53 & 4.88 & 3.63 & 4.30 & 4.69 + 2.02.5 & 5.10 & 5.58 & 5.98 & 4.81 & 5.30 & 5.72 & 4.57 & 4.94 & 5.45 +

we study a fokker - planck equation with double - well potential that is nonlocally driven by a dynamical constraint and involves two small parameters . relying on formal asymptotics we identify several parameter regimes and derive reduced dynamical models for different types of phase transitions . keywords : _ multi - scale dynamics , gradient flows with dynamical constraint , _ + _ phase transitions , hysteresis , fokker - planck equation , kramers formula _ msc ( 2010 ) : 35b40 , 35q84 , 82c26 , 82c31

1110.3518

in this paper we investigate the different dynamical regimes in a fokker - planck equation with multiple scales that was introduced in @xcite to describe the charging and discharging of lithium - ion batteries , a process that exhibits pronounced hysteretic effects @xcite . the model , to which we refer as ( fp ) , governs the evolution of a statistical ensemble of identical particles and is given by the nonlocal fokker - plank equation @xmath0 here @xmath1 is the free energy of a single particle with thermodynamic state @xmath2 , the probability density @xmath3 describes the state of the whole system at time @xmath4 , and @xmath5 reflects that the system is subjected to some external forcing . moreover , @xmath6 is the typical relaxation time of a single particle and @xmath7 accounts for entropic effects ( stochastic fluctuations ) . the model ( fp ) has two crucial features which cause highly nontrivial dynamics . first , the free energy @xmath1 is a double - well potential , hence there exist two different stable equilibria for each particle . second , the system is not driven directly but via a time - dependent control parameter . in our case this parameter is the first moment of @xmath8 , that means we impose the dynamical constraint @xmath9 where @xmath10 is some given function in time , and a direct calculation shows that is equivalent to @xmath11 provided that the initial data satisfy @xmath12 . the closure relation implies that is a nonlocal and nonlinear pde . well - posedness was proven in @xcite on bounded domains , but we are not aware of any result about the qualitative properties of solutions . an intriguing property of ( fp ) is that its dynamics involves three different time scales . on the one hand , there are the relaxation time @xmath13 and the time scale of the dynamical constraint . on the other hand there is the scale of probabilistic transitions between different local minima of the effective energy @xmath14 , that means particles can move between the different wells due to stochastic fluctuations ( large deviations ) . kramers studied such transitions in the context of chemical reactions @xcite and derived the characteristic time scale @xmath15 where @xmath16 is the minimal difference of energy between the local maximum any of the local minima of @xmath17 . in what follows we always assume that @xmath18 is of order @xmath19 , whereas both @xmath13 and @xmath20 are supposed to be small . our goal in this paper is to identify different parameter regimes and to describe the asymptotics of ( fp ) in the limit @xmath21 . to this end we focus on strictly increasing constraints and describe four different mechanisms of mass transfer between two stable regions . the corresponding four types of phase transitions are , roughly speaking , related to two main regimes , which we refer to as _ fast reaction regime _ and _ slow reaction region _ , respectively . the dominant effect in the fast reaction regime is mass exchange according to kramers formula . this appears for very small @xmath13 and covers , as limiting case , also the quasi - stationary regime @xmath22 . the slow reaction regime , however , corresponds to very small @xmath20 and kramers formula is not relevant anymore . instead , phase transitions are dominated by transport along characteristics and this causes rather complicated dynamics since localized peaks of mass can enter the spinodal region of @xmath1 . in both the slow reaction and the fast reaction regimes we are able to characterize the small parameter dynamics in terms of a few averaged quantities only . detailed descriptions of the corresponding limit models are given in the introductions to sections [ sec : fast ] and [ sec : slow ] , respectively . before we give a more detailed overview on the different dynamical regimes we specify our assumptions on @xmath1 and review some basic facts about fokker - planck equations . in this paper we assume that @xmath1 is an even double - well potential that satisfies the following conditions , see figure [ fig : potential ] , 1 . @xmath1 is even , sufficiently smooth ( at least @xmath23 ) , and @xmath24 grows at least linearly as @xmath25 . 2 . there exist constants @xmath26 and @xmath27 such that 1 . @xmath28 and @xmath29 , 2 . @xmath30 and @xmath31 for @xmath32 , 3 . @xmath33 and @xmath34 for @xmath35 . + in particular , the inverse of @xmath36 has the three strictly monotone branches @xmath37}\to { ( -\infty,\,-x_*]}\,,\quad x_0:{[-{{\sigma}}_*,\,{{\sigma}}_*]}\to{[-x_*,\,x_*]}\,,\quad x_-:{[-{{\sigma}}_*,\,\infty)}\to { [ x_*,\,\infty)}\,.\end{aligned}\ ] ] 3 . the functions @xmath38 and @xmath39 are concave on the spinodal interval @xmath40 . the assumptions in ( a1 ) and ( a2 ) are made for convenience and might be weakened for the price of further technical and notational efforts ; ( a3 ) is a geometric condition that becomes important in the slow reaction regime that is discussed in section [ sec : slow ] . notice that all assumptions are in particular satisfied for the standard double - well potential @xmath41 in what follows we refer to @xmath42 and @xmath43 as the stable intervals , whereas the spinodal region @xmath40 is called the unstable interval . this nomenclature is motivated by the different properties of the transport term in . in both stable intervals adjacent characteristics approach each other exponentially fast , hence there is a strong tendency to concentrate mass into narrow peaks . in the unstable region , however , the separation of adjacent characteristics delocalizes any peak with positive width . fokker - plank equations like ( fp ) are derived in @xcite from first principles and provide a thermodynamically consistent model for a many - particle system with dynamical constraint . in particular , the second law of thermodynamics can be stated as @xmath44 where @xmath45 is the free energy of the system and @xmath46 the dissipation . they are given by @xmath47 and @xmath48 , where @xmath49 denote the internal energy and entropy of the many - particle system , respectively . it is well known that the fokker - planck equation without constraint , that is with @xmath50 , admits several interpretations as a gradient flow . there is for instance a linear structure , which has been exploited in @xcite in order to derive the effective dynamics in the limit @xmath51 . of particular interest , however , is the nonlinear wasserstein gradient flow structure , see @xcite , as this structure is compatible with the constraint . more precisely , ( fp ) with @xmath52 is the wasserstein gradient flow for @xmath45 on the constraint manifold @xmath53 , and @xmath54 describes a drift transversal to this manifold . the entropic term @xmath20 is often supposed to be very small but it is important that @xmath20 is positive . more precisely , without the diffusive term the qualitative properties of solutions would strongly depend on microscopic details of the initial data , and hence it would be impossible to characterize the limit @xmath55 in terms of macroscopic , i.e. averaged , quantities only , see @xcite . a key observation is that the singular perturbation @xmath56 regularizes the macroscopic evolution , at least for some classes of initial data , in the following sense . microscopic small - scale effects are still relevant on the macroscopic scale , but they are independent of the initial details and affect the system in a well - defined manner . as a consequence we now obtain a well - posed limit model for macroscopic quantities . another approach to ensure well - defined macroscopic behavior is investigated in @xcite . the key idea there is to mimic entropic effects by assuming that each particle is affected by a slightly perturbed potential . the macroscopic evolution is then completely determined by the dynamical constraint , the macroscopic initial data , and the probability distribution of the perturbations . we next summarize some facts about the dynamics of with time - independent @xmath5 and small parameters @xmath57 . for @xmath58}$ ] , the effective potential @xmath14 possesses a single critical point @xmath59 that corresponds to a global minimum . the system then relaxes very fast ( on the time scale @xmath13 ) to its unique equilibrium state @xmath60 where @xmath61 is a normalization constant ensuring @xmath62 . this equilibrium density @xmath63 has a single peak of width @xmath20 located at @xmath59 and decays exponentially as @xmath64 . the situation is different for @xmath65 since @xmath17 now exhibits a double well - structure with two wells ( local minima ) at @xmath66 that are separated by a barrier ( local maximum ) at @xmath67 . initially the system relaxes very quickly , and approaches ( for smooth initial data ) a state composed of two narrow peaks located at the wells . both peaks have masses @xmath68 and @xmath69 with @xmath70 , but the precise values of @xmath71 depend strongly on the initial data . this fast transition reflects that each particle in the system is strongly attracted by the nearest well due to the gradient flow structure . the resulting state , however , is in general not an equilibrium but only a metastable state . the underlying physical argument is that particles can pass the energy barrier due to stochastic fluctuations . in the generic case , in which both wells have different energies , it is of course more likely for a particle to cross the barrier coming from the well with higher energy , and thus there is a net flux of mass towards the well with lower energy . this flux is , for small @xmath72 , given by kramers celebrated formula and guarantees that the system approaches its equilibrium on the slow time scale . the corresponding equilibrium solution is again given by and has now two peaks with a definite mass distribution between the wells . notice , however , that for small @xmath20 almost all the mass of an equilibrium solution is confined to the well with lower energy . due to the different time scales , the dynamics of ( fp ) can be very complicated , and we are far from being able to characterize the small parameter dynamics for all types of initial data and all reasonable dynamical constraints . we thus restrict most of our considerations to strictly increasing dynamical constraints and well - prepared initial ; only in section [ sect:332 ] we allow for non - monotone constraints . in what follows we consider functions @xmath10 with @xmath73 where @xmath74 , and @xmath75 are given constants . since @xmath13 is small , the system then relaxes very quickly to a local equilibrium state with @xmath76 . we can therefore assume that the initial mass is concentrated in a narrow peak , that means @xmath77 where the right hand side abbreviates the dirac distribution at @xmath78 . an even better approximation , that also accounts for the small entropic effects caused by @xmath79 , is @xmath80 the dynamical constraint implies @xmath81 , so that the peak starts moving to the right and the system quickly relaxes to a new local equilibrium state . for sufficiently small times , that means as long as @xmath82 , the system can therefore be described by the _ single peak model _ @xmath83 where we write @xmath84 and @xmath85 to indicate that all mass is confined in the left stable interval @xmath42 . moreover , at some time @xmath86 we have @xmath87 , and for @xmath88 the system again relaxes quickly to a local equilibrium state , that means we have @xmath89 where @xmath90 and @xmath91 now reflect that all mass has been transferred to the second stable region @xmath43 . the key question is what happens between @xmath92 and @xmath93 . it was already observed in @xcite that , depending on the relation between @xmath13 and @xmath20 , there are at least four types of phase transitions driven by rather different mechanisms of mass transfer from the left stable region into the right one . the main objective of this paper is to investigate the different regimes and to derive asymptotic formulas for the dynamics . we illustrate the different types of phase transitions for the driven initial value problem and by numerical simulations with @xmath94 and initial data @xmath95 as in . figure [ fig : ivp1 ] visualizes the numerical solutions by means of two curves @xmath96 these curves represent the _ macroscopic state _ of the system and the ( rescaled ) _ phase field _ , respectively , and are defined by @xmath97 the microscopic state of the system is illustrated in figure [ fig : ivp2 ] by snapshots of @xmath8 at different times . we emphasize that figures [ fig : ivp1 ] and [ fig : ivp2 ] also illustrate the different dynamical regimes for fixed @xmath79 , in the sense that the limit @xmath98 can be regarded as a passage from ( @xmath99 ) to ( @xmath100 ) . however , as our results will show , there is an exponential scale separation between the different regimes and thus it is very hard to capture all types of phase transitions in numerical simulations with the same value of @xmath20 . the numerical simulations illustrate that there exist the following types of phase transitions . @xmath101 at some time the narrow peak enters the unstable region due to the dynamical constraint and starts to widen due to the separation of characteristics . however , the transport is much faster than the widening , so that the peak can pass trough the unstable region . the peak still enters the unstable region but now the widening is much faster than before . in particular , the unstable peak delocalizes , and the system quickly forms new peaks in each of the stable regions . at a later time the left peak enters the unstable region , and the competition between transport and widening starts once more . the peak does not enter the unstable region anymore . instead , at some position in the left stable interval the peak stops moving and starts loosing mass to feed another peak in the right stable interval . this kramers - type process happens with @xmath102 and goes on until the left peak has disappeared . this is the quasi - stationary limit . the phase transition is similar to type iii but happens with @xmath103 . notice that the types @xmath104 and @xmath105 imply hysteresis . in fact , if we revert the situation by driving the system with @xmath106 and @xmath107 , the symmetry of the problem implies that the macroscopic state and the phase field are confined to the images of @xmath108 and @xmath109 under point reflections at @xmath110 . our main results are formulas for the macroscopic evolution in different parameter regimes . the corresponding scaling relations between @xmath13 and @xmath20 are summarized in table [ tbl : scalingregimes ] and the limit models are presented in the introductions to sections [ sec : fast ] and [ sec : slow ] . the rest of the paper is organized as follows . in section [ sec : fast ] we derive asymptotic formulas to describe type-@xmath105 transitions in the regime @xmath111 , where @xmath112 denotes the energy barrier of @xmath1 . we first recall in section [ sec : kramers.1 ] the formal derivation of kramers formula for the mass flux between the different wells of @xmath17 . in section [ sec : kramers.2 ] we then identify the critical value for @xmath113 for which such a mass flux can also take place in the constrained setting and show in section [ sec : kramers.3 ] that small variations in @xmath5 are sufficient to adjust the mass flux according to the dynamical constraint . finally , in section [ sec : kramers.4 ] we discuss type-@xmath114 transitions as these can be regarded as limits of kramers type transitions . in section [ sec : slow ] we consider the scaling regime @xmath115 with @xmath116 and discuss type-@xmath104 transitions which contain , as a special case , also type-@xmath100 transitions . in section [ sect:31 ] we first neglect all entropic effects and introduce a simplified two - peaks model that allows to understand how two dirac peaks interact due to the dynamical constraint . it turns out that the dynamical constraint can stabilize a peak in the unstable region , but also that at some point a bifurcation forces both peaks to merge instantaneously . in section [ sect:32 ] we introduce another simplified model that accounts for the stochastic fluctuations in the unstable region and allows to understand how an unstable peak delocalizes and splits into two stable peaks . in particular , we derive an asymptotic formula for the time at which such a splitting event takes place and introduce the _ mass splitting problem _ that determines the mass distribution between the emerging peaks . in section [ sect:33 ] we finally combine all result and characterize the limit dynamics as intervals of regular transport that are interrupted by several types of singular event . in this section we show that kramers type phase transitions are also relevant in presence of the dynamical constraint as long as @xmath117 where @xmath118 denotes the energy barrier of @xmath1 . the key idea is that a kramers type phase transition occurs during a time interval @xmath119 in which @xmath5 is positive and almost constant . during this time interval @xmath120 only changes to order @xmath121 but this is sufficient to accommodate the dynamical constraint . kramers formula therefore allows to understand phase transitions of type iii , and hence that a stable peak suddenly stops moving and starts loosing mass to feed a second stable peak . the situation is different for @xmath122 since then we expect to find phase transitions of type iv , that means the mass flows towards the second well as soon as it is energetically admissible . this regime is governed by the quasi - stationary approximation but can also be regarded as a limiting case of kramers regime . our main result concerning the fast reaction regime combines the formal asymptotics for the kramers regime and the quasi - stationary approximation and can be stated as follows . suppose that the dynamical constraint and the initial data satisfy and , and that @xmath13 and @xmath20 are coupled by @xmath123 for some constant @xmath124 . then there exists a constant @xmath125 such that 1 . the dynamical multiplier satisfies @xmath126 where @xmath127 and @xmath128 are uniquely determined by @xmath129 and @xmath130 , 2 . the state of the system satisfies @xmath131 where @xmath132 and @xmath133 moreover , the assertions remain true 1 . with @xmath134 if @xmath135 , 2 . with @xmath136 if @xmath137 but @xmath138 for all @xmath139 . to justify the limit dynamics we review kramers argument for constant @xmath120 in section [ sec : kramers.1 ] . in section [ sec : kramers.2 ] we then derive similar asymptotic formulas for the constrained case , which allow us to adjust the mass flux according to the dynamical constraints in section [ sec : kramers.3 ] . moreover , in section [ sec : kramers.4 ] we discuss the quasi - steady approximation , which governs the regime @xmath140 . we finally mention that the limit energy is given by @xmath141 and evolves according to @xmath142 where @xmath17 is defined by @xmath143 and @xmath144 denotes the usual characteristic function . to derive kramers formula for the unconstrained case we consider the fokker - planck equation with fixed @xmath145 and use the abbreviations @xmath146 at first we approximate @xmath8 outside the local maximum @xmath67 for small @xmath20 by the ansatz @xmath147 this is the outer expansion and reflects the assumption that the system has a peak in either of the stable regions , where the masses are given by @xmath148 for small @xmath20 we can simplify the integrals using laplace s method , that means we expand @xmath17 around @xmath149 to find @xmath150 the mass exchange between both peaks is then determined by @xmath151 where @xmath152 is the mass flux at @xmath153 . the key idea behind kramers formula is that @xmath152 can be computed from the quasi - stationary approximation of @xmath8 near @xmath153 . more precisely , with the change of variables @xmath154 we approximate @xmath155 and obtain the inner expansion @xmath156 where @xmath157 is a constant of integration . for small @xmath20 and @xmath158 we can simplify the integrals by laplace s method to obtain @xmath159 and using a similar formula for @xmath160 we find @xmath161 in order to match the outer and the inner expansions , we consider @xmath162 and compare the asymptotic formulas and . since both contain the factor @xmath163 , we equate the time dependent coefficients and arrive at the matching conditions @xmath164 we finally eliminate @xmath157 and find the desired expression for kramers mass flux , namely @xmath165 combining this with we easily verify that the characteristic time for kramers mass transfer is given by @xmath166 we also notice that in the generic case @xmath167 and for small @xmath20 the mass transfer is essentially unidirectional on the time scale , that means the mass flows from the well with higher energy to the well with smaller energy , see figure [ fig : kramers ] . we now derive a self - consistent description for kramers type phase transitions in the presence of the dynamical constraint . to this end it is convenient to replace @xmath13 by the parameter @xmath168 defined in , and to consider the functions @xmath169 these functions are well - defined for @xmath170 and satisfy @xmath171 with @xmath172 and @xmath173 . in terms of @xmath174 and @xmath175 , the rescaled flux @xmath176 from can be stated as @xmath177 where @xmath178 and @xmath179 as above . we next present some heuristic arguments for the dynamics of the rescaled flux terms @xmath180 and @xmath181 . to this end we assume @xmath182 and recall that , due to our assumptions on the initial data , the system evolves for small times according to the single peak evolution . in particular , the peak reaches the critical position @xmath183 at time @xmath184 with @xmath185 , and the dynamical constraint implies @xmath186 . assuming that @xmath5 changes regularly at @xmath184 , we then conclude that @xmath187 consequently , for small @xmath20 and sufficiently small times @xmath188 we expect to find @xmath189 , so crossing the energy barrier is very likely for a particle in the right well but very unlikely for a particle in the left well . however , the net transfer across the energy barrier is very small since there are essentially no particles in the right well . we thus expect that the partial masses @xmath71 stay constant in the limit @xmath190 , so the system can still be described by the single peak approximation . at some later time @xmath191 we have @xmath192 and the fluxes @xmath193 and @xmath194 have the same order of magnitude . however , both are very small due to @xmath195 , and so there is , for small @xmath20 , still no effective mass transfer between the two wells of @xmath17 . the situation changes completely at time @xmath127 defined by @xmath196 . at this time , @xmath197 becomes suddenly of order one and we can no longer neglect particles that move from the left well to the right one . the other flux @xmath198 , however , is now very small as it is very unlikely that a particle moves the other way around . as explained above , the main idea in the dynamical case is that there exist a constant @xmath199 and a time @xmath128 such that @xmath200 for all @xmath201 . kramers mass flux can hence stabilize to continuously transfer mass from the left well to the right one . at time @xmath128 , all mass has been transferred to the right well and the system again evolves according to the single peak evolution , now given by . before we describe the details of the mass transfer we proceed with two remarks . first , the above assumption @xmath202 is truly necessary : for @xmath203 there is still a time @xmath127 with @xmath204 , but then we have @xmath205 and hence @xmath206 , which shows that a net transfer from the left well to the right one is impossible . moreover , for @xmath207 both @xmath208 and @xmath198 are always very large . in both cases we expect that the phase transition occurs when @xmath103 and is not governed by kramers formula anymore but by the quasi - stationary approximation . second , the mass flux is already determined by the dynamical constraint and the assumption @xmath200 . in fact , the constraint implies that @xmath209 and thus , since @xmath5 does not change much , @xmath210 as a consequence we obtain @xmath211 for @xmath212 , where @xmath213 is defined by @xmath214 , and using @xmath215 it is easy to check that @xmath216 . to identify the formulas that relate the mass flux self - consistently to small temporal changes in @xmath5 we consider only times @xmath4 with @xmath201 and assume that both @xmath68 and @xmath69 are strictly positive . of course , in order to match the resulting approximations for @xmath8 to the single - peak evolution for @xmath217 and @xmath218 we must introduce transition layers at @xmath219 and @xmath220 corresponding to @xmath221 and @xmath222 , respectively , but since these transition layers do not contribute to the limit model , we do not investigate them in detail . . the critical value @xmath199 is defined by @xmath223 thanks to @xmath202 we find @xmath224 and @xmath225 , so yields @xmath226 since @xmath227 must be of order one , we introduce the rescaled multiplier @xmath228 and simplify by expanding @xmath229 around @xmath230 . using and we then conclude that kramers formula implies the mass transfer law @xmath231 comparing this with we finally conclude that kramers type phase transitions comply with the dynamical constraint if and only if @xmath232 this is the heart of our argument . if @xmath233 evolves according to , then @xmath234 is of order @xmath235 , and kramers formula provides a mass flux that satisfies the dynamical constraint . . in this limiting case , the phase transition happens when @xmath120 is close to @xmath236 , so both @xmath237 and @xmath153 are close to @xmath238 . moreover , the constants @xmath239 and @xmath240 approach zero , and hence we can no longer use the asymptotic expressions from the first case . however , if we expand all relevant quantities around @xmath236 and @xmath238 , it is still possible to derive an asymptotic formula for kramers flux that is consistent with the dynamical constraint . thanks to the identities @xmath241 , @xmath242 and @xmath243 , we deduce from the definition of @xmath244 and @xmath245 that @xmath246 to leading order in @xmath247 , we therefore find @xmath248 as well as @xmath249 so can be simplified to @xmath250 in kramers regime this flux @xmath152 should be of order @xmath13 . on the other hand , the asymptotic formula holds only if @xmath251 , and thus we shall guarantee that @xmath252 . both conditions can be satisfied if @xmath253 in fact , if we define for given @xmath72 and @xmath254 the large parameter @xmath255 by @xmath256 then @xmath152 becomes of order @xmath13 if @xmath247 is of order @xmath257 . finally , we proceed as in the first case . inserting into gives a formula for @xmath258 in terms of @xmath259 . on the other hand , the dynamical constraint implies @xmath260 via , and eliminating @xmath258 we obtain @xmath261 in terms of @xmath262 . to conclude this section we show that the quasi - stationary approximation of ( fp ) describes phase transitions with @xmath263 . notice that such type - iv transitions can be regarded as limits of type - iii transitions in the sense that @xmath264 as @xmath265 . in the quasi - stationary limit we approximate @xmath8 by the equilibrium solution that corresponds to the current value of @xmath5 . in other words , we set @xmath266 where the normalization factor @xmath267 is given by @xmath268 the dynamical multiplier @xmath269 is then determined by the dynamical constraint via @xmath270 we now derive asymptotic formulas that characterize the quasi - stationary dynamics for small @xmath20 . at first we notice that laplace s method applied to with @xmath271 yields @xmath272 and hence @xmath273 in particular , the system evolves according to the single - peak approximation as long as @xmath5 has a sign , and a phase transition can occur only for @xmath274 . to describe the details of such a transition it is convenient to rescale @xmath5 by @xmath275 . using the expansion @xmath276 and employing laplace s method once more , we find @xmath277 and hence @xmath278 on the other hand , the dynamical constraint , see , provides @xmath279 and we conclude that the rescaled multiplier evolves according to @xmath280 notice that this formula is well defined for all times @xmath4 with @xmath201 , where @xmath127 and @xmath128 are defined by @xmath281 and @xmath282 , and satisfy @xmath283 and @xmath284 . this section concerns the effective dynamics of ( fp ) in the slow reaction regime : both @xmath13 and @xmath20 are still supposed to be small , but @xmath20 is so small that ` reactions ' , that means continuous mass transfer between the stable regions as described by kramers formula , are not relevant anymore . instead , the dominant effect in ( fp ) is now transport along characteristics and this gives rise to new phenomena . in particular , localized peaks can enter the unstable region and peaks can split or merge rapidly . notice , however , that the small entropic effects caused by @xmath7 are still relevant and can not be neglected . they guarantee that each peak entering the unstable region is basically a rescaled gaussian , and hence that such a peak behaves in a well - defined manner . we now introduce an informal concepts that is motivated by numerical simulations and turns out to be useful for describing the asymptotic dynamics in the slow reaction regime . we say the system is in a _ two - peaks configuration _ if there exist two positions @xmath285 and two masses @xmath286 with @xmath287 such that the state can be approximated by @xmath288 a two - peaks configuration is called _ stable - stable _ if @xmath289 and @xmath290 , but _ unstable - stable _ if @xmath291 and @xmath290 . moreover , in case that one of the masses vanishes , we refer to a two - peaks configuration as a ( _ stable _ or _ unstable _ ) _ single - peak configuration_. our main result in this section is an asymptotic description of the dynamics in a certain parameter regime for @xmath13 and @xmath20 . the corresponding limit model is illustrated in figure [ fig : flow_chart ] and can be summarized as follows . suppose that the dynamical constraint and the initial data satisfy and , and that @xmath13 and @xmath20 are coupled by @xmath292 for some constant @xmath116 . then , in the limit @xmath98 the dynamics can be described in terms of single - peak and two - peaks configurations according to one of the following scenarios . 1 . _ transport with splitting ( type - ii phase transition ) _ : there are intervals of quasi - stationary transport which are interrupted by singular times . more precisely , depending on @xmath10 and the value of the parameter @xmath293 there exists an integer @xmath294 such that there are @xmath295 _ switching times _ , @xmath255 _ splitting times _ , and a final _ merging time_. 1 . during the intervals of quasi - stationary transport , the peaks do not exchange mass and are move according to the dynamical constraint . 2 . at each switching time a stable peak reaches the position @xmath238 to become unstable ; each switching time is followed by a splitting or the merging time . 3 . at each splitting time an unstable peaks splits and its mass is instantaneously transferred to the stable regions . in particular , the system jumps from an unstable - stable two - peaks configuration ( or , initially , from an unstable single - peak configuration ) to an emerging stable - stable two - peaks configuration . the precise values for the masses and the positions of the emerging peaks are determined by a _ mass splitting problem_. 4 . at the final merging time the two peaks of an unstable - stable configuration merge either continuously or discontinuously to form a single stable peak located in the region @xmath296 . 2 . _ pure transport ( type - i phase transition ) _ : the system is always in a single peak configuration and there exist only two singular times corresponding to switching ( the peaks enters the unstable interval ) and trivial continuous merging ( the peak leaves the unstable interval ) . notice that it is practically impossible to perform numerical simulations with @xmath20 exponentially small in @xmath13 and @xmath13 much smaller than @xmath19 . the phase transitions of type i and type ii presented in figures [ fig : ivp1 ] and [ fig : ivp2 ] are therefore not directly covered by our limit model , but a close look to the numerical data reveals that they are likewise dominated by the interplay between transport and widening of unstable peaks . it remains a challenging task to derive next order corrections to replace the singular times by transition layers with width depending on @xmath13 . another interesting but open question is whether there exist scaling laws different from that give rise to other reasonable slow reaction limits . to justify the limit dynamics we introduce several reduced models that allow us to study each of the different phenomena in a simplified setting . in section [ sect:31 ] we derive a _ two - peaks model _ that describes the transport of two peaks in terms of a simple ode system . moreover , this model also reveals that the separated peaks in an unstable - stable configuration can merge instantaneously due to the dynamical constraint . in section [ sect:32 ] we investigate the splitting of unstable peaks . to this end we propose a _ peak - widening model _ that accounts for the small entropic effects and allows us to derive a deterministic equation for the width of an unstable peak . in particular , we show that for small @xmath13 it can happen that this width blows up almost instantaneously , which in turn gives rise to rapid mass transfer from the unstable towards the stable regions . we then discuss the _ mass splitting problem _ , which consists of solving a nonlocal transport equation in order to determine how much mass is transferred to each of the stable regions . finally , in section [ sect:33 ] we combine all partial results and characterize the limit dynamics of the original model ( fp ) . in particular , we derive explicit formulas for the iterative computation of all switching , splitting , and merging times . moreover , at the end we sketch the slow - reaction limit for non - monotone dynamical constraints . a major tool in our analysis is a simple _ two - peaks model _ ( tp ) which describes the essential dynamics of ( fp ) as long as the system is a two - peaks configuration . the model governs the evolution of the peak positions @xmath297 and @xmath298 , and reads @xmath299 here @xmath300 and @xmath301 denote the constant masses of the peaks . notice that is just the characteristic ode for with @xmath302 , and this implies that ( tp ) is also a constrained gradient flow corresponding to the energy @xmath303 in particular , using the dissipation @xmath304 the energy balance is given by @xmath305 . since @xmath13 is small it seems natural to neglect the time derivatives in and . this gives rise to the _ quasi - stationary approximation _ to ( tp ) , which consists of the algebraic equations @xmath306 for our analysis it is important to understand in which sense approximates ( tp ) . this problem is not trivial because the non - invertibility of @xmath36 implies that has multiple solutions for @xmath307 , and thus we have to understand which solution branches are dynamically selected by solutions to ( tp ) . in the limit @xmath98 , solutions to ( tp ) exhibit two important dynamical phenomena which correspond to changing the solution branch of . both phenomena seem to be counter - intuitive at a first glance but are a consequence of the dynamical constraint . they can be described as follows . first , the dynamical constraint can drive the system from a stable - stable configuration to an unstable - stable configuration , that means the stable peak at @xmath298 can stabilize an unstable one at @xmath297 due to the dynamical constraint . the second effect is that this stabilization can break down eventually . when this happens , the separated peaks merge almost instantaneously to form a single stable peak . to describe both phenomena we consider times @xmath308 and suppose that the dynamical constraint @xmath262 is smooth and strictly increasing . we also suppose that the initial data for ( tp ) are well prepared via @xmath309 to elucidate the key ideas we proceed with discussing some numerical results and present some semi - rigorous analytical considerations afterwards . the left panel of figure [ fig : hyperbolicinstability ] depicts a typical numerical solution to ( tp ) with @xmath310 , potential , and initial data as in . the simulation reveals the existence of two critical times @xmath127 and @xmath128 that separate the three different regimes @xmath311 where @xmath312 denotes the final simulation time . for all times @xmath313 we have @xmath314 and thus we expect that in the limit @xmath98 the system can in fact be described by quasi - stationary peaks . the details , however , are different for @xmath315 , @xmath316 , and @xmath317 . more precisely , initially we have @xmath318 that means the solution resembles a stable - stable configuration , and @xmath319 implies @xmath320 , @xmath321 , and @xmath81 . at time @xmath127 , which is defined by @xmath322 , the peak at @xmath297 enters the unstable region and the configuration switches to unstable - stable . this means @xmath323 and hence @xmath320 , @xmath324 , and @xmath325 . at the second critical time @xmath128 , the two - peaks approximation breaks down , that means the system can no longer be described by two separated peaks . instead , both peaks merge discontinuously , in the sense that the system jumps from an unstable - stable two - peaks configuration to a stable single - peak configuration . the evolution for @xmath218 is still quasi - stationary , but involves only a single peak that evolves according to @xmath326 as illustrated in the right panel of figure [ fig : hyperbolicinstability ] , the existence of the critical time @xmath128 can be understood as follows . the quasi - stationary two - peaks approximation imposes the constraints @xmath327 which define a smooth curve in the @xmath328-plane . this curve has the two branches @xmath329 which meet smoothly at the point @xmath330 . due to the dynamical constraint the system is further confined to @xmath331 which is a straight line with slope @xmath332 . it can now easily be seen that in the quasi - stationary approximation for @xmath333 the state of the system corresponds to the unique intersection point of @xmath334 and @xmath335 , which moves towards @xmath330 since @xmath10 is increasing in time . at time @xmath127 , the system crosses this point @xmath330 , and for sufficiently small times @xmath336 the state of the system is given by the unique intersection point of @xmath337 and @xmath335 . at time @xmath128 however , the line @xmath335 becomes tangential to @xmath337 and the system can no longer follow the curve @xmath337 due to @xmath319 . instead both peaks merge rapidly and the system jumps almost instantaneously to @xmath338 , which is the only intersection point of @xmath335 and the diagonal @xmath339 . the tangency condition for the merging reads @xmath340 where the left hand side is positive for all times @xmath341 . of course , if @xmath342 is very small , then the slope of @xmath335 is very negative and the tangency condition can not be satisfied . in this case , there is a continuous transition from the unstable - stable two - peaks configuration to the stable single - peak configuration since both peaks merge via @xmath343 . it is then natural to define @xmath128 as the time at which this continuous merging takes place . the above considerations also apply to the limiting case @xmath344 and @xmath345 , provided that we accept that @xmath298 is undefined . in this case , the quasi - stationary approximation reads @xmath346 and the switching time @xmath127 is defined by @xmath347 . at time @xmath128 with @xmath348 the single peak leaves the unstable interval and in view of the above discussion it makes sense to interpret this event as ( trivial ) continuous merging . we emphasize that our characterization of the unstable - stable evolution relies on condition ( a3 ) , that means on the concavity of the function @xmath349}\mapsto x_2=x_+{{\left({h^\prime{{\left({x_1}\right)}}}\right)}}$ ] . otherwise it may happen at time @xmath128 that the system does not jump to the diagonal but instead to another unstable - stable configuration . we also mention that discontinuous merging of peaks implies that @xmath5 jumps down to @xmath350 , and that the emerging single peak is stable due to @xmath351 . moreover , the energy of the system also jumps down . this is in line with the gradient flow structure of ( tp ) , and holds because the energy decreases along the straight line that connects @xmath328 to @xmath338 . in fact , with @xmath352 we find @xmath353 thanks to @xmath354 , @xmath355 and @xmath356 . we finally show that the quasi - stationary approximation is linearly stable until it ceases to exist . to this end , we consider a given solution to ( tp ) with @xmath310 and initial data as in , and denote by @xmath357 and @xmath358 the quasi - stationary approximation to ( tp ) . this reads @xmath359 where the definition of @xmath357 involves @xmath244 for @xmath360 but @xmath245 for @xmath361 . here , @xmath362 is the time at which the quasi - stationary approximation ceases to exist , and @xmath363 is defined by @xmath364 and denotes the switching time at which the quasi - stationary approximation enters the branch @xmath337 . making the ansatz @xmath365 , we easily derive the linearized model @xmath366 with @xmath367 . by construction , we have @xmath368 and hence @xmath369 where @xmath370 . notice that @xmath371 , @xmath372 and @xmath373 depend continuously on @xmath4 since the function @xmath10 is smooth . the variations of constants formula now reveals that the quasi - stationary solution is dynamically stable as long as @xmath374 , that means as long as @xmath375 . moreover , a similar analysis reveals that the quasi - stationary single - peak solution @xmath376 is dynamically stable provided that @xmath377 , which is satisfied for @xmath378 due to @xmath379 . in the previous section we have seen that the two - peaks model ( tp ) allows a stable peak to enter the unstable region . our goal in this section is to show that under some conditions the same is true for the original model ( fp ) , but also that the entropic terms trigger new phenomena . to point out the main idea we start with some heuristic arguments . afterwards we derive and investigate a simplified model that allows to understand the key effects in a more formal way . suppose that we are given a solution to ( fp ) that at time @xmath184 consists of two narrow stable peaks located at positions @xmath380 and @xmath381 . suppose also that the width of these peaks is sufficiently narrow such that @xmath382 . for small @xmath13 we then expect that the system relaxes very fast to a meta - stable state of ( fp ) that meets the constraint @xmath383 . without loss of generality we may hence assume that the initial peaks have width of order @xmath20 and that the initial data are well prepared in the sense of for sufficiently small times @xmath188 we can expect that ( fp ) follows the two - peaks model ( tp ) , that means both peaks have width of order @xmath20 and are located at stable positions @xmath384 , @xmath385 with @xmath320 and @xmath321 due to @xmath319 . at some time @xmath127 , however , the peak located at @xmath297 reaches the critical position @xmath238 , and this time can be estimated by @xmath386 . when the first peak has crossed @xmath238 , its width widens very quickly because the characteristics of the transport term now separate exponentially with local rate @xmath387 . however , since the width of the first peak is initially exponentially small in @xmath13 , it remains small for some times although it is surely much larger than @xmath20 . moreover , the second peak located at @xmath298 still has width of order @xmath20 as it remains confined to the stable region @xmath35 . combining both arguments we conclude that the system can be approximated by ( tp ) even at times @xmath336 provided that @xmath388 is sufficiently small . the condition @xmath319 then implies @xmath320 , @xmath324 and @xmath389 . the key question now is how long the first peak remains localized . depending on the scaling parameter @xmath293 and the mass distribution between the peaks , it can happen that the first peak remains localized until both peaks merge continuously via @xmath390 . in this case the first peak can in fact pass through the whole unstable region . it can also happen that the peak remains localized till the quasi - stationary two - peaks approximation ceases to exist . in this case both peaks merge instantaneously and discontinuously to form a single stable peak , but ( fp ) still behaves like ( tp ) . there is , however , a third possible scenario , in which the width of the first peak becomes of order one before both peaks can merge continuously or discontinuously . below we will argue that if this happens at all , it happens at a precise time @xmath128 . more precisely , we show that there is a time @xmath128 at which the width of the first peak blows up instantaneously for small @xmath13 . some amount of the mass of this peak is then transported along characteristics to the left until it creates a new peak in the stable region @xmath391 . the remaining part , however , is transported towards the other stable region @xmath35 to feed the second peak . since the transport along characteristics if very fast , we expect that in the limit @xmath98 the first peak splits and disappears instantaneously and that the system jumps to another stable - stable configuration . after this jump the new peak in the stable region @xmath391 has mass smaller than @xmath300 , and this implies that the mass of the second peak is larger than @xmath342 . to analyze the relevant phenomena we study a simplified _ peak - widening model _ ( pw ) , which approximates the second peak at @xmath298 by a dirac mass but keeps the probabilistic description for the first peak located at @xmath297 . this gives rise to the equations @xmath392 here @xmath300 and @xmath301 are two constants that describe the mass distribution between the peaks , and as long as @xmath298 is confined to the stable region @xmath35 we can expect that each solution to ( pw ) defines via @xmath393 an ( approximate ) solution to the original model ( fp ) . in what follows we denote the width of the first peak by @xmath394 and aim to derive an asymptotic formula for the evolution of @xmath394 that involves only the dynamical constraint @xmath262 . for simplicity we assume again that the data at time @xmath184 are localized and well prepared in the sense of @xmath395 a key ingredient to any asymptotic analysis of the widening phenomenon is to give an appropriate description of the position of the first peak . our ansatz is to define @xmath396 as solution of the characteristic ode , that means we set @xmath397 this ansatz has the following advantages . as long as the first peak is narrow , we have @xmath398 and hence @xmath297 and @xmath298 evolve according to the two - peaks model ( tp ) . consequently , for @xmath310 we can describe @xmath297 and @xmath298 in terms of the quasi - stationary approximations of ( tpm ) , whose dynamics is completely determined by the dynamical constraint @xmath262 and the mass distribution between the peaks , see . a further advantage of is that it gives rise to a quite simple evolution law for @xmath394 . in order to analyze the growth of @xmath394 we introduce the rescaling @xmath399 where both the spatial scaling factor @xmath400 and the rescaled time @xmath401 will be identified below . we then have @xmath402 , where @xmath403 is the width of @xmath404 at @xmath405 . we readily verify that and imply that the evolution of @xmath404 is governed by @xmath406 for the subsequent considerations we now assume that @xmath396 is a given continuous function with @xmath322 and @xmath407 for some @xmath312 . heuristically , @xmath312 is the time at which both peaks merge ( continuously or discontinuously ) according to the two - peaks model ( tp ) . as long as @xmath394 is small but possibly much larger than @xmath20 we can expand the nonlinearity according to @xmath408 neglecting the higher order terms and defining @xmath409 and @xmath405 as solutions to @xmath410 the nonlinear pde transforms into the heat equation @xmath411 consequently , for large @xmath405 the rescaled profile @xmath404 evolves in an almost self - similar manner , that means we can approximate @xmath412 notice that this approximation implies @xmath413 for @xmath414 and holds as long as @xmath415 is of order @xmath19 . in order to characterize the width of the original peak it remains to understand how @xmath416 and @xmath405 depend on @xmath4 . a direct computation yields @xmath417 where @xmath418 abbreviates @xmath419 due to @xmath310 and @xmath420 we now infer that @xmath421 for all @xmath422 . this implies @xmath423 , and hence @xmath424 where @xmath293 is the scaling parameter from . we have now identified an explicit formula for @xmath394 , which involves only the function @xmath396 . notice that for small @xmath13 this function can be computed by the quasi - stationary two - peaks approximation , whose evolution is independent of @xmath13 and completely determined by @xmath300 and @xmath262 . formula can be restated as @xmath425 where the function @xmath426 is non - positive on the interval @xmath427 since @xmath127 is the only root of @xmath418 in this interval . we now aim to show that this formula implies that the critical time @xmath128 is determined by the conditions @xmath428 more precisely , if @xmath429 attains the value @xmath430 for some time @xmath128 with @xmath431 , then in the limit @xmath98 the width of the unstable peak explodes instantaneously at @xmath128 . if , however , @xmath429 is larger than @xmath430 for all times @xmath432 , then the width of the peak is exponentially small in @xmath13 even at time @xmath312 . in order to prove these assertions we now derive rough estimates for @xmath394 . at first we consider @xmath433 . in this case we have @xmath434 for all @xmath435 , so gives @xmath436 now let @xmath336 but suppose that @xmath4 is sufficiently small such that @xmath437 . using @xmath438 we then we estimate @xmath439 and find again that the width of the peak is exponentially small in @xmath13 . finally , we consider a time @xmath4 with @xmath432 and @xmath440 . to derive a lower bound for @xmath394 we now employ the continuity of @xmath418 at @xmath127 as follows . for @xmath441 and @xmath442 we estimate @xmath443 to find @xmath444 combining this with gives @xmath445 and we conclude that the width of the first peak is exponentially large in @xmath446 . we finally emphasize that the equation for @xmath394 can be simplified for @xmath220 as follows . exploiting the continuity of @xmath418 at @xmath128 we find @xmath447 . for each @xmath448 we therefore have @xmath449 where @xmath450 means arbitrary small for small @xmath13 . in particular , choosing @xmath362 with @xmath451 such that @xmath452 we find @xmath453 as explained above , at the critical time @xmath454 we expect that the system undergoes a rapid transition from the unstable - stable configuration to a new stable - stable configuration . in order to describe this transition , in particular , to predict the mass distribution between the emerging stable peaks , we propose to study a simplified _ mass - splitting model _ ( ms ) , which describes ( pw ) on the rescaled time scale @xmath455 in the limit @xmath190 . in other words , ( ms ) consists of the equations @xmath456 which have no diffusion and satisfy the constraint via @xmath457 . notice that this equation is the wasserstein - gradient flow for the energy @xmath458 in view of the above discussion we now impose asymptotic initial conditions at @xmath459 , which reflect that the mass splitting process starts in an unstable - stable configuration and that the unstable peak is a rescaled gaussian due to the entropic randomness . to this end we identify @xmath460 and denote by @xmath357 and @xmath358 the quasi - stationary two - peaks approximation at @xmath128 , i.e. we have @xmath461 in accordance with , and we now require that @xmath462 weakly@xmath463 in the sense of probability measures , and that @xmath464 the gradient flow structure of ( ms ) implies that each solution approaches a stable - stable configuration in the limit @xmath465 . this reads @xmath466 where @xmath467 and @xmath468 denote the positions of the emerging stable peaks and @xmath469 is precisely the amount of mass that is transferred during the splitting process from the unstable region into the stable region @xmath35 . of course , the asymptotic data at @xmath470 must comply with @xmath471 but these conditions do not determine the three quantities @xmath472 , @xmath473 , and @xmath474 completely . this is not surprising and reflects that the amounts of mass that are transferred towards the stable regions depends crucially on the asymptotic shape of the unstable peak . in our case , however , this shape is a rescaled gaussian and therefore we expect that the data at @xmath470 are uniquely determined by the data at @xmath459 . in particular , we conjecture that there is a _ unique _ and _ continuous _ mass transfer function @xmath475 such that @xmath476 both the existence and continuity of @xmath475 are not obvious because the mass splitting problem involves two subtle limits . at first one has to show that the asymptotic condition gives rise to a well - posed initial value problem at @xmath459 . second , one has to guarantee that solutions do not drift as @xmath465 along the connected one - parameter family of equilibrium solutions . a rigorous justification of the mass splitting function @xmath475 is beyond the scope of this paper and left for future research . it is , however , possible to compute numerical approximations of @xmath475 using the method of characteristics . more precisely , restricting to @xmath477 characteristics @xmath478 with @xmath479 , the mass splitting problem ( ms ) can be approximated by @xmath480 with @xmath481 moreover , to mimic the asymptotic initial conditions and , we choose a small parameter @xmath482 and set @xmath483 with @xmath484 being the inverse of @xmath485 . the resulting finite - dimensional initial value problem can be integrated numerically , for instance by means of the explicit euler scheme , which satisfies the constraint @xmath486 up to computational accuracy . in the limit @xmath487 the critical index @xmath488 finally determines @xmath472 via @xmath489 . the strategy for the numerical computation of @xmath475 is therefore as follows . we sample the two - dimensional parameter space of ( ms ) and solve for each choice of the parameters the ode with @xmath5 as in on a sufficiently large time interval . the results are illustrated in figure [ fig : msp_sim ] for the potential , where @xmath490 and @xmath491 are , as above , regarded as the two independent parameters . combining the arguments from the previous sections we are now able to characterize the slow reaction limit of ( fp ) in terms of single - peak and two - peaks configurations . to describe their evolution we consider piecewise constant mass functions @xmath492 and @xmath493 along with piecewise continuous position functions @xmath494 and @xmath495 , where we allow @xmath297 and @xmath298 to be undefined on intervals with @xmath496 and @xmath497 , respectively . these functions are coupled by the constraints @xmath498 with @xmath499 and define the limit energy via @xmath500 to formulate the limit model we first specialize to strictly increasing dynamical constraints and discuss possible generalizations afterwards . sections [ sect:31 ] and [ sect:32 ] reveal that the slow reaction limit for constraints with can be described as follows . the initial dynamics is governed by the quasi - stationary evolution of a single stable peak within the stable region @xmath391 and hence it seems natural to define @xmath501 , which implies @xmath346 and renders @xmath298 to be undefined . at the first switching time , which is defined by @xmath502 , the single peaks becomes unstable and its width starts to widen . it may happen that the motion of the unstable peak is faster than the widening , and then the peak remains localized until it leaves the unstable region via @xmath503 . in this case we find a type - i phase transition as there is no splitting of unstable peaks ; notice that this happens if @xmath504 is sufficiently large or if the scaling parameter @xmath293 is sufficiently large . for type - ii transitions , however , the width of the peak becomes eventually large within the unstable interval . this means there exists a first splitting time with @xmath505 at which the system jumps from the unstable single - peak configuration to a stable - stable two - peaks configuration , where the masses and the positions of the emerging stable peaks is determined by the mass splitting problem . in particular , after the first splitting event the positions @xmath297 , @xmath298 and the masses @xmath300 , @xmath342 are well defined , and both stable peaks move according to the dynamical constraint until the first peak located at @xmath297 becomes unstable at the next switching time defined by @xmath506 . afterwards we must carefully investigate the unstable - stable evolution in order to decide whether there is a further splitting of the unstable peak or whether both peaks finally merge continuously or discontinuously . in summary , the limit dynamics can be described as illustrated in figure [ fig : flow_chart ] , that means intervals of quasi - stationary transport are interrupted by singular times corresponding to the following types of events : @xmath101 the peak at @xmath297 enters the unstable region . the unstable peak at @xmath297 splits and the system jumps to a new stable - stable configuration with decreased mass @xmath300 and increased mass @xmath342 . the peaks in an unstable - stable configuration merge either continuously ( with @xmath507 ) or discontinuously ( with @xmath508 ) , or there is only a single peak with @xmath344 that leaves the unstable region ( with @xmath506 ) . more precisely , each limit trajectory comprises @xmath295 switching times , @xmath255 splitting times , and a final merging time @xmath509 where we have @xmath510 and @xmath294 for type - i and type - ii transitions , respectively . here @xmath511 is defined by @xmath512 , so proper two - peaks configurations can only exist for @xmath513 . notice that an infinite number of switching and splitting events is not possible because the splitting condition implies a lower bound for @xmath514 via @xmath515 with @xmath516 . notice also that we must truly book keep the switching events since doe not determine @xmath517 but only @xmath514 . we now describe the flowchart from figure [ fig : flow_chart ] in greater detail . . between consecutive singular times , and likewise initially for @xmath518 , the dynamics of @xmath297 and @xmath298 is governed by a rate - independent system of non - autonomous odes . more precisely , differentiating with respect to time yields @xmath519 where we recall that @xmath297 and @xmath298 remain undefined for @xmath496 and @xmath345 , respectively . these odes imply @xmath520 with @xmath521 since the dynamical constraint @xmath262 is given , the initial value problem to can at least in principle be integrated . . at switching times all functions are continuous , that means we have @xmath522\!|}}{{\left({t_{{\mathrm{sw}},k}}\right)}}={{|\![x_i]\!|}}{{\left({t_{{\mathrm{sw}},k}}\right)}}= { { |\![{{\sigma}}]\!|}}{{\left({t_{{\mathrm{sw}},k}}\right)}}={{|\![e]\!|}}{{\left({t_{{\mathrm{sw}},k}}\right)}}=0\,,\qquad i=1,2\,,\end{aligned}\ ] ] where @xmath523\!|}}{{\left({t}\right)}}:= f{{\left({t+0}\right)}}-f{{\left({t-0}\right)}}$ ] with @xmath524 denotes the jump of the function @xmath418 at time @xmath4 . at a splitting time all quantities do jump , where the jump heights are determined by the mass splitting problem ( ms ) . in particular , the gradient flow structure of ( ms ) ensures that @xmath525\!|}}{{\left({t_{{\mathrm{sp}},k}}\right)}}<0,\end{aligned}\ ] ] and using the mass transfer function @xmath475 from we find @xmath526\!|}}{{\left({t_{{\mathrm{sp}},k}}\right)}}=-{{|\![m_1]\!|}}{{\left({t_{{\mathrm{sp}},k}}\right)}}= m{{\big({m_1{{\left({t_{{\mathrm{sp}},k}-0}\right)}}},\,{{{\sigma}}{{\left({t_{{\mathrm{sp}},k}-0}\right)}}}\big)}}.\end{aligned}\ ] ] the jumps of @xmath297 , @xmath298 and @xmath5 are then determined by the algebraic constraints , and can be used to reinitialize the odes and . finally , at the merging time we have @xmath527 and the concavity condition ( a3 ) , see the discussion at the end of section [ sect:312 ] , provides @xmath525\!|}}{{\left({t_{{\mathrm{me}}}}\right)}}\leq0\end{aligned}\ ] ] with strict inequality for discontinuous merging . after the merging the precise values masses @xmath300 and @xmath342 are actually undefined , but it seems natural to set @xmath528 . after the @xmath529 splitting time ( we set @xmath530 to describe the initial evolution ) , the system is in a stable - stable configuration and the subsequent switching time @xmath531 is the smallest time larger than @xmath517 such that @xmath532 . this implies @xmath533 , @xmath534 , and @xmath535 which determines @xmath531 uniquely since @xmath300 and @xmath342 are known . after the switching event at @xmath536 we have to decide whether the next singular time corresponds to splitting or merging according to the conditions @xmath537 to discuss this decision in a simple case , let us suppose that @xmath18 is constant in time . due to @xmath325 we can replace @xmath4 by @xmath120 , and direct computations provide the following nonlinear equations @xmath538 in particular , both merging conditions are rate independent , whereas the switching condition depends on @xmath18 and can not be satisfied for large @xmath18 . this is not surprising because the merging conditions are completely determined by the quasi - stationary two - peak approximation , whereas splitting only happens if the peak widening due to the separation of characteristics is faster than the transport due to the dynamical constraint . instead of solving , we can now vary @xmath5 , starting from @xmath539 and moving towards @xmath540 , and check which conditions is satisfied at first . in any case , the time @xmath541 of the next singular event can be computed by @xmath542 we finally derive an alternative description of the slow reaction limit that covers arbitrary dynamical constraints @xmath10 . decreasing constraints can produce stable - unstable configurations with @xmath289 and @xmath543 , which in turn can split or merge . however , these effects can easily be described by adopting the formulas from sections [ sect:31 ] and [ sect:32 ] . the truly new effect is that non - monotone constraints can trigger _ inverse switching events _ , that means , for instance , an unstable - stable configuration can become stable - stable via @xmath544 with @xmath545 . for general constraints , it is convenient to describe the slow - reaction limit in terms of the following variables : @xmath546 the dynamical multiplier @xmath5 , @xmath547 an internal variable @xmath548}$ ] as introduced in section [ sect:32 ] to control the width of an unstable peak , and @xmath549 three nonnegative masses @xmath68 , @xmath550 , and @xmath69 with @xmath551 to describe the several types of single - peak and two - peaks configurations . using these parameters , the peak positions are dependent variables which satisfy @xmath552 for @xmath553 as long as @xmath554 is well - defined . at each non - singular time , the system is confined to one of the sets defined in table [ tbl : sets ] , where @xmath555 and @xmath556 are abbreviations for @xmath557 and the functions @xmath558 are defined by @xmath559within each of these sets , the peaks are transported according to the quasi - stationary two - peaks approximation and the widening of unstable peaks is governed by @xmath560 as described in and . this reads @xmath561 on the other hand , singular events happen when the system reaches the boundary of either @xmath562 or @xmath563 . more precisely , examing the conditions for switching , inverse switching , splitting , and merging we arrive at list from table [ tbl : events ] , where we now interpret trivial continuous merging as inverse switching of single - peak configurations . the corresponding jump and update rules for the variables @xmath564 , @xmath5 , and @xmath560 are summarized in table [ tbl : jumps ] . the slow reaction dynamics of two - peaks initial data can now be integrated iteratively by @xmath546 moving via along either one of the sets @xmath562 and @xmath563 , and @xmath547 jumping to another set when reaching the boundary . notice that at both splitting and discontinuous merging events the system jumps to inner points and that splitting events require to solve the mass splitting problem . the authors are grateful to wolfgang dreyer , clemens guhlke , and alexander mielke for stimulating discussions . this work was supported by the epsrc science and innovation award to the oxford centre for nonlinear pde ( ep / e035027/1 ) . [ 2 ] http://www.ams.org/mathscinet-getitem?mr=#1[#2 ] [ 2]#2 steffen arnrich , alexander mielke , mark a. peletier , giuseppe savar , and marco veneroni , _ passing to the limit in a wasserstein gradient flow : from diffusion to reaction _ , var . and pde ( 2011 ) , in press . richard jordan , david kinderlehrer , and felix otto , _ free energy and the fokker - planck equation _ d * 107 * ( 1997 ) , no . 2 - 4 , 265271 , landscape paradigms in physics and biology ( los alamos , nm , 1996 ) .

we investigate the dynamical breaking of local supersymmetry ( supergravity ) , including the deser - zumino super - higgs effect , via the corresponding one - loop effective potential for the simple but quite representative cases of , simple supergravity and a ( simplified ) conformal version of it . we find solutions to the effective equations which indicate dynamical generation of a gravitino mass , thus breaking supergravity . in the case of conformal supergravity models , the gravitino mass can be much lower than the planck scale , for global supersymmetry breaking scales below the grand unification scale . the absence of instabilities in the effective potential arising from the quantum fluctuations of the metric field is emphasised , contrary to previous claims in the literature . kcl - ph - th/2013-*21 * + lcts/2013 - 12

1310.4122

supersymmetry is an important theoretical and phenomenological subject that has dominated theoretical physics for decades , despite the lack ( so far ) of evidence that it actually exists in nature . it assists us in understanding several aspects of low energy physics , from the stability of the higgs vacuum to the unification of forces , and even the low scale of inflation ( relative to the planck mass ) , in a way that is much clearer than in non - supersymmetric frameworks . the embedding of supersymmetry into a gravitational framework by promoting global supersymmetry into a local ( gauged ) symmetry , termed supergravity @xcite , has initiated many interesting avenues of research toward understanding both the unification of gravity with the remaining three fundamental forces of nature , and to some extent quantisation of the gravitational field . in particular , it is well known that supergravity theories constitute consistent low - energy limits of superstring theories , which are thought to provide plausible paths to the quantisation of gravity in concert with other fundamental interactions . + however , supersymmetry is not phenomenologically observed in nature , and thus if it exists , it must be broken in the low energy world . it is therefore desirable to understand the spontaneous or dynamical breaking of supersymmetry under various possible circumstances , of which , thanks to the profusion of gauge and matter sectors that may be consistently coupled into supersymmetry , there are many @xcite . it would be preferable however , to seek a way of breaking local supersymmetry ( supergravity ) directly in a dynamical manner , without the need to couple it to a gauge sector . + one such approach would be to leverage the fermionic ( gravitino ) torsion terms that are generically present in supergravity theories , which consist of four - gravitino self - interaction terms . the latter could conceivably condense under certain circumstances , thereby producing condensates of the gravitino field , which in this way would dynamically acquire a non - zero mass whilst leaving the graviton massless . in this way supergravity would be broken _ dynamically _ , in the same spirit as the breaking of chiral symmetry in the nambu - jona - lasinio model @xcite . + it was conjectured in @xcite that such a dynamical breaking of supergravity could occur via the formation of condensates of the gravitino field , with analyses based on the one - loop effective potential of a simple @xmath2 supergravity model in a flat minkowski space - time ( necessary to allow an unambiguous definition of the gravitino mass via the condensate field ) . the effective potential of the gravitino condensate field , dependent on positive powers of the ultraviolet ( uv ) cutoff scale due to the well - known non - renormalisability of quantum gravity in four space - time dimensions , acquires a non - trivial minimum for some values of the cutoff relative to the gravitational constant ( reduced planck mass ) . in the analyses of @xcite the gravitino thereby acquires a dynamical mass of the order of the planck scale . + it was important for their analyses that the one - loop value of the effective potential _ vanishes _ at the non - trivial minimum , which implies the vanishing of the effective vacuum energy of the resulting low - energy theory . this arguably justifies _ a posteriori _ the minkowski space - time analysis of the effective potential . it was also important for both the entire approach and the double - well shape of the effective potential , appropriate for supergravity breaking , that the deser - zumino super - higgs effect @xcite was incorporated self consistently by coupling the supergravity action with the non - linear volkov - akulov action @xcite of the goldstone particle associated with the ( assumed f - type ) spontaneous global supersymmetry breaking . the latter is a majorana spin 1/2 fermion , the goldstino . + this formalism was essential for two reasons . firstly , the goldstino could be absorbed by the gravitino , under an appropriate field redefinition to make the latter massive , and thus disappear from the physical spectrum secondly , the only subsequent trace of the goldstino would be a ( negative ) cosmological constant @xmath3 in the effective action , which is associated with the scale of global supersymmetry breaking ; namely the vacuum expectation value of the appropriate f - term of the chiral scalar superfield responsible for spontaneous supersymmetry breaking . this is the super - higgs effect @xcite which is essential in keeping track of the right degrees of freedom in the problem of dynamical acquisition of a mass by the gravitino field , and the consequent breaking of local supersymmetry . + from the point of view of the effective potential , this would correspond to a _ positive _ term at the origin in field space of order of @xmath4 , which is responsible for the double - well shape of the potential at the broken symmetry phase . these considerations are consistent with the generic features of dynamical supersymmetry breaking outlined by witten @xcite , according to which the vacuum energy of broken global supersymmetry is _ necessarily positive _ , here @xmath5 , whilst a _ broken local supersymmetry _ ( supergravity ) can still be characterised by _ zero vacuum energy . _ + although physically appealing , the flat minkowski space - time approach of @xcite was criticised in @xcite due to the fact that it ignored the quantum fluctuations of the metric field . following the generic approach of fradkin and tseytlin @xcite of calculating the one - loop effective potential in four - dimensional supergravity theories by means of expanding about a generic ( anti)de sitter , rather than minkowski , background , the authors of @xcite have argued that integrating over metric fluctuations introduces _ imaginary _ parts in the effective potential , for any non - trivial value of the gravitino condensate field , and irrespective of the value of the background cosmological constant @xmath6 ( whose role is to effectively replace @xcite the flat - space uv cut - off of @xcite ) . + the presence of imaginary parts would of course be an indication that the non - trivial - gravitino - condensate ( broken supergravity ) vacuum is _ unstable _ , and thus there would be no possibility of breaking @xmath2 supergravity dynamically in this simple and direct way . of course , the traditional way of breaking supergravity via dynamical global supersymmetry breaking through , say , gaugino condensation @xcite , which would then be communicated to the ( super)gravity sector , evades the arguments of @xcite and thus has appeared to be the only consistent way of dynamically breaking supergravity , carrying the price of necessarily coupling the theory to matter fields . + in this work we shall revisit the arguments and the analysis of @xcite , with a view to incorporating the super - higgs effect that was not included in their analyses . we have been motivated to do so by the compelling nature and simplicity of the possibility of direct dynamical breaking of supergravity by means of its gravitino - torsion self - interactions . + as it turns out , and as is demonstrated below in great detail , the proper incorporation of the super - higgs effect in such a framework enables the dynamical breaking of supergravity ( prior to its coupling to matter or gauge fields ) , in the sense that a one - loop effective potential analysis considering fully metric fluctuations about ( anti)de sitter backgrounds , and thereby fully incorporating any ( weak ) quantum gravitational effects , reveals the existence of non - trivial vacua with _ no imaginary _ parts , contrary to the claims of @xcite . the double - well shape of the effective potential , and its vanishing at the non - trivial minima , fully justifies the flat - space approximate analysis of @xcite and demonstrates that the presence of the cosmological constant @xmath4 due to the super - higgs effect is responsible for the vanishing of the effective vacuum energy of the low - energy effective action of supergravity at the one - loop level . + the structure of the article is as follows . * in section [ sec : superhiggs ] we briefly review the super - higgs effect in the context of simple @xmath7 supergravity , in order to outline to the reader its important features , and the underlying physics that will be used in our analysis of the resultant one - loop effective potential . * in section [ sec : sugra ] we set up the basic formalism and notations underlying the model of @xmath2 supergravity that we shall consider for concreteness in this work . our formalism of course applies in general to more complicated theories of supergravity . * in section [ sec : quadr ] we discuss the quadratic action obtained from previous section by incorporating weak metric fluctuations about a given ( anti)de sitter background , in conjunction with the linearisation of four - gravitino terms by means of appropriate auxiliary fields . this is the first step towards the construction of a ` proper ' ( i.e. incorporating ( weak ) quantum gravitational effects ) one - loop effective potential . we discuss bosonic and fermionic sectors separately as well as their respective gauge fixing procedures . * in section [ sec : potential ] we construct the effective potential and discuss the shape that we require in order to obtain non - trivial local supersymmetry ( supergravity ) breaking , via dynamical condensation of the gravitino field . we discuss carefully the absence of imaginary parts in certain non - trivial vacua , contrary to the claims of @xcite , and explain carefully how this may be achieved . as we also show in subsection [ sec : confsugra ] , phenomenologically realistic values for the gravitino mass and the associated breaking scale of global supersymmetry can be achieved only in appropriate conformal extensions of the @xmath2 supergravity , which may also be relevant for inflation . * conclusions and outlook are then given in section [ sec : concl ] , and some technical aspects of our approach of constructing the one - loop effective potential are discussed in an appendix . the goldstino is a majorana fermion with a volkov - akulov lagrangian @xcite that arises from some spontaneous or dynamical breaking of global supersymmetry , as a result of the appropriate extension of goldstone s theorem to supersymmetry . of global supersymmetry transformations is a space - time spinor , and hence if @xmath8 denotes the goldstone field , then the commutator that is not annihilating the vacuum ( @xmath9 is an infinitesimal global - supersymmetry - transformation parameter ) @xmath10 \ , | \ , 0 \rangle - \overline{\varepsilon}^{\dot \alpha } \ , \langle 0 \ , | \ , [ ( { q}_\alpha)^\dagger \ , , \ , \omega \ , ] \ , | \ , 0 \rangle \ne 0 $ ] must be a scalar for reasons of lorentz invariance , and hence @xmath11 must be a fermionic field , to be identified with a spin-1/2 four - component majorana goldstone fermion ( goldstino ) , @xmath12 . ] here we consider the case where the breaking of global supersymmetry is of the so - called @xmath13-type , in which the @xmath13-term of some chiral superfield @xmath14 acquires a vacuum expectation value @xmath15 in the original works of @xcite , the goldstino lagrangian is written in terms of a four - component majorana spin-@xmath16 goldstino field ( denoted @xmath12 ) , with @xmath17 . the constant @xmath18 expresses the strength of global supersymmetry breaking , which as mentioned above occurs in our case at @xmath19 in planck units , and the lagrangian is characterised by a non - linear realisation of global supersymmetry with infinitesimal parameter @xmath20 : @xmath21 the coupling of the goldstino to supergravity may then generate a mass for the gravitino through the absorption of the goldstino , via the super - higgs effect envisaged in @xcite . + according to this model , the @xmath0 supergravity theory is coupled to the goldstino field @xmath22 . thus , we consider a spontaneously - broken supersymmetric theory with a majorana goldstino @xmath22 , whose action takes the non - linear form considered by volkov and akulov @xcite : @xmath23 where we used a weak field expansion expansion of the determinant , where @xmath24 is a two - component weyl spinor , employing fermionic spinor truncation , the volkov - akulov lagrangian acquires the exact form : @xmath25 for our purposes here , truncation to terms of first order in derivatives suffices in the weak field approximation we utilise herein . ] . here we keep the discussion general by allowing for an arbitrary value of the parameter @xmath18 . + as discussed in @xcite , one can promote the global supersymmetry to a local one , by allowing the parameter @xmath26 to depend on space - time coordinates , and coupling the action ( [ goldstino ] ) to that of @xmath0 supergravity in such a way that the combined action is invariant under the following supergravity transformations : @xmath27 where the @xmath28 in the @xmath22 transformation denote non - linear @xmath22-dependent terms ( _ cf . _ the variation ) . the action that changes by a divergence under these transformations is the standard @xmath0 supergravity action plus @xmath29 which contains the coupling of the goldstino to the gravitino . in the above notation , @xmath30 denotes the vierbein determinant . + the goldstino can then be gauged away @xcite by a suitable redefinition of the gravitino field and the tetrad . one may impose the gauge condition @xmath31 but this leaves behind a _ negative cosmological constant term _ , @xmath32 , so the total lagrangian after these redefinitions reads : @xmath33 where the lagrangian of @xmath0 supergravity is the subject of the next section . working in the 1.5 order formalism of the einstein - cartan formulation of supergravity , we have the lagrangian @xcite @xmath34}~,\label{rexpansion}\\ & r^{ab}{}_{\mu\nu}\left(e,\psi\right ) = \partial_\mu\omega^{ab}{}_\nu\left(e,\psi\right ) -\partial_\nu\omega^{ab}{}_\mu\left(e,\psi\right ) + \omega_\mu{}^{ac}\omega_{\nu c}{}^b\left(e,\psi\right ) -\omega_\nu{}^{ac}\omega_{\mu c}{}^b\left(e,\psi\right)~,\\ & \omega_\mu{}^{ab}\left(e,\psi\right ) = \omega_\mu{}^{ab}\left(e\right ) + \frac{\kappa^2}{4}\left(\overline\psi_\mu\gamma^a \psi^b -\overline\psi_\mu\gamma^b \psi^a + \overline\psi^a\gamma_\mu \psi^b\right)~,\quad d_\mu=\partial_\mu+\frac{1}{2}\omega_\mu{}^{ab}\left(e\right)\sigma_{ab}~ , \end{aligned}\ ] ] where @xmath35 $ ] is the scalar curvature in the absence of torsion , @xmath36 are the minimal set of auxiliary fields required for closure of the algebra and @xmath37 indicates interaction terms between the gravitino and graviton fields , and four - gravitino interactions involving @xmath38 . as we are in the 1.5 order formalism our spin connection @xmath39 is determined entirely by the associated field equation , so that we may neglect the variation @xmath40 . + the action is also invariant under the local supersymmetry transformations @xmath41 where @xmath42 is the ` supercovariantised ' gravitino field equation @xmath43 the gauge condition ( [ gravinogauge ] ) is understood from now as having been imposed in the fermionic sector of the supergravity action . as mentioned previously , it is in this gauge that the decoupling of the goldstino from the supergravity action , upon an appropriate redefinition of the gravitino field , occurs . + to investigate the possibility of dynamical mass generation for the gravitino we firstly introduce an auxiliary scalar field @xmath44 to linearise the four - gravitino interaction term in equation via the equivalence @xmath45 which follows as a simple consequence of the subsequent euler - lagrange equation for @xmath44 . taking into account the goldstino - induced negative cosmological constant term arising from equation , the relevant terms in equation are then @xmath46 following the normalisation for the gravitino mass of @xcite , @xmath47 we note that , if @xmath44 acquires a non - zero vacuum expectation value ( vev ) through the process of quantisation so that @xmath48 , then we have dynamically generated an effective mass of @xmath49 for the gravitino , thus breaking local supersymmetry . the other four - gravitino terms ( which involve @xmath38 ) can be neglected in this regard as they are not of canonical form for mass terms , as can the ( subleading at one - loop order ) gravitino - graviton interaction terms . + furthermore , we may then define @xmath50 so that we may identify @xmath51 with a tree - level cosmological constant . and @xmath52 to give a tree - level cosmological constant which we can tune in such a way that such contributions are absorbed in @xmath4 in ( [ bare ] ) , so that @xmath51 is understood to contain such contributions . ] + upon quantisation this bare cosmological constant is then dressed by quantum corrections to yield a renormalised cosmological constant @xmath6 . in this sense , although de sitter space - time may not be a solution of the bare equations of motion stemming from @xmath0 supergravity lagrangian in its broken symmetry phase , it may be the solution of a _ quantum effective action _ , after the fluctuations of the metric and other ( gravitino and ghost ) fields are taken into account . + for the purposes of our analysis here , and following @xcite , we assume that one may expand the one - loop effective action about a de sitter background with a positive renormalised cosmological constant , @xmath53 , whose value will be determined by minimisation of the one - loop effective action . it is however known that in supergravity an apparent ( but not actual ) gravitino mass term is required in the presence of a non - zero cosmological constant @xcite for the closure of the supersymmetry algebra , mandating that we consider the limit @xmath54 to ensure that any such gravitino mass generated is indeed physical . + it is important to stress at this point that since @xmath6 is equivalently the overall energy density at the level of the one - loop effective potential @xmath55 , we then furthermore require as a self - consistency condition that @xmath56 : i.e. that the energy density ( and thus @xmath6 ) is indeed zero at whatever non - trivial minima we find . this is achieved in practice by tuning @xmath4 , which is contained within the bare cosmological constant . + our algorithm is therefore to find the one - loop effective action for this theory in @xmath57 ( euclidean ) de sitter space ( @xmath58 ) , before solving the corresponding effective equations in the limit that the effective cosmological constant @xmath6 vanishes , enabling a straightforward interpretation of any resultant gravitino mass . + some technical but important remarks are in order at this stage . one may consider the excursion of our theory through de sitter space purely in the spirit of euclidean continuation , as appropriate for path integrals in a consistent quantisation of ( super)gravity , allowing us to arrive back at a physical theory only in the limit @xmath59 . we may also note a tension here in that whilst the gravitinos in are majorana , there are in fact no majorana representations on@xmath60 ( or equivalently , in @xmath61 ) . as however we are treating this continuation as a purely technical step , we will proceed with the understanding that we will arrive at a physical result only after completing these manipulations . to compute the one - loop effective action for a given theory there are a number of operations that we must firstly take account of . firstly , we must expand about a classical background to compute fluctuations of the action to quadratic order , via decompositions of the type @xmath62 , where for our purposes @xmath63 is the standard euclidean @xmath64 metric . working to one - loop order in this instance has the advantage of decoupling the gravitino and graviton sectors , as all gravitino terms in are already quadratic , and lorentz invariance forbids any fermionic background terms . + since we are interested in the one - loop effective potential for the auxiliary field @xmath44 , it is sufficient to identify the latter with its vev @xmath65 . indeed , the auxiliary field does not propagate at tree level , and its kinetic term ( obtained from the integration over other degrees of freedom , for a non - uniform configuration @xmath44 ) would therefore be purely one - loop . hence its influence on the effective potential , which is already of one - loop order , would be at least of two - loop order . we therefore replace from now on @xmath44 by @xmath66 . note that the effective potential obtained in this way would be exact if one integrated over @xmath67 fermions ( which is excluded here because of the matching of bosonic and fermionic degrees of freedom ) . + we must also decompose all fields present ( including ghosts ) into those that are ` natural ' to our background geometry ; more precisely , to those corresponding to irreducible representations of the underlying isometry group . it is the spectra of these operators that can be reliably found through knowledge of the underlying representation theory , which will then allow us to compute the effective action . starting with the gravitational lagrangian in , we firstly vary to quadratic order in @xmath68 , yielding ( in the metric formalism ) @xcite @xmath69~ , \end{aligned}\ ] ] where we have followed the convention that @xmath70 and @xmath71 are the spin and christoffel connections respectively , and have also made the standard decomposition @xmath72 . + it is important to also note that given the presence of fermions we must work in the vierbein formalism , which leads to extra terms ( which vanish on - shell ) in the quadratic action relative to the metric formalism @xcite . this can be understood as arising from the first variation of the gravitational action , which takes the form @xmath73 we can then see that the second variation in the vierbein formalism will coincide with the metric formalism , along with the additional term @xmath74 where we have used that @xmath75 and that @xmath76 . + we therefore re - express as @xmath77~,\\ & x_1=\frac{8}{3}\lambda-2\lambda_0+\frac{\lambda_0-\lambda}{2}=\frac{13}{6}\lambda-\frac{3}{2}\lambda_0 , \quad x_2=2\lambda_0+\frac{\lambda_0-\lambda}{2}=\frac{5}{2}\lambda_0-\frac{1}{2}\lambda~. \end{aligned}\ ] ] to further decompose into the ` irreducible action ' we will make use of the standard ` transverse traceless ' decomposition @xmath78 where @xmath79 is some vector field and we have also defined a class of bosonic operators for constant @xmath80 @xmath81 it is these operators whose spectra we shall ultimately compute . there is however an important caveat that we must bear in mind in that are extra zero - modes present in these decomposed operators , which must be correctly accounted for @xcite . we will take account of these later , where we will note that in the limit @xmath82 their contributions are of subleading order . + to streamline the process of computation we will make use of the following ( @xmath83 identities @xmath84 @xmath85 for some constants @xmath80 and @xmath86 . we have two symmetries present in this sector ; local lorentz and infinitesimal coordinate transformations . to fix the former we simply follow the convention of setting the antisymmetric part of the vierbein to zero @xcite , leading to non - propagating ghost fields which can then be disregarded here . to fix the coordinate gauge transformations we add a standard two parameter covariant gauge fixing term @xmath87 which necessitates the ghost action @xmath88 for some anticommuting complex vector field @xmath89 + this may be easily integrated after applying the first identity given previously to arrive at the ghost partition function @xmath90 where we have absorbed any prefactors into the normalisation of the functional measure . there is however a secondary line of approach by which we may instead address the issue of gauge fixing . this is via an appeal to so - called ` physical ' gauges ; gauges which represent an alternative path to quantisation than the conventional faddeev - popov method , inasmuch as they consist of isolating the gauge degrees of freedom present and essentially disregarding them as a ` physical ' gauge choice . + in practice this is achieved in the following manner . it is firstly well known that we must gauge fix in order to render path integrals well defined . without dividing out by the volume of the gauge group , we naturally overcount field configurations which are physically equivalent and related by gauge transformations . conventionally this is remedied via a gauge fixing condition specifying a phase - space hypersurface which intersects each orbit of the gauge group , along with the inclusion of a faddeev popov ghost determinant in the path integral measure to locally cancel the phase - space curvature of non - abelian gauge symmetries . this is however not the only path we may proceed by . + re - examining our gauge fixing condition we may consider the illuminative case @xmath91 , whereupon we may strongly impose ( i.e. use at the level of the action , rather than solely imposing via lagrange multiplication ) the condition @xmath92 substituting the decomposition and setting @xmath93 , this condition then becomes @xmath94 for @xmath95 . we may then straightforwardly note that @xmath96 must therefore be killing , and as such , it is @xmath96 that parametrises the underlying diffeomorphism symmetry present . our ` physical ' gauge condition is then to strongly impose that @xmath97 and therefore disregard the components of the graviton corresponding to general coordinate transformations ; i.e. the fields @xmath98 and @xmath99 in our notation . the functional integral over these gauge degrees of freedom then yields an infinite constant prefactor of the volume of the diffeomorphism group , which is unimportant for our purposes . + we should note at this point that there are some additional complications to this physical gauge procedure regarding the correct counting of the zero modes of @xmath96 ( which should not necessarily be disregarded even if @xmath97 ) , however as in the previous instance of the extra zero modes arising from our decompositions , these contributions are subleading as @xmath82 . + given the inevitable gauge dependence of effective potentials such as these ; a consequence of our artificial truncation to one - loop order , it is arguable that physical gauge techniques are more ` natural ' in this context , and certainly can offer significant computational simplifications . it is perhaps unsurprising then that they have been utilised effectively in a number of situations similar to these @xcite@xcite@xcite . + naturally of course we may apply these same considerations to the gauge fixing of the fermionic sector , however for our purposes it will suffice to utilise this procedure only in the bosonic case . as we will see , in this context it is ultimately the behaviour of the bosonic sector that dictates the stability of the effective potential . with the two extra contributions outlined above , along with our gauge fixing term , the quadratic gravitational action now becomes @xmath100~ , \end{aligned}\ ] ] performing the functional integral would be straightforward , were it not for the final term in . however , we may firstly note from that @xmath101 is conserved and so can not mix with @xmath102 . furthermore , we may also leverage the result that for einstein backgrounds there is no @xmath103 mixing @xcite . to eliminate the final @xmath104 mixing we could then also use a so - called ` diagonal gauge ' @xcite , however this would be incompatible with the landau - dewitt gauge choice , which we know to correspond to the unique gauge - invariant one - loop effective action for pure einstein gravity @xcite . + instead , we will proceed with our general - gauge calculation . schematically , the scalar part of the action is then of the form @xmath105 with matrix elements @xmath106 so that we may integrate this along with the other fields present to find the bosonic partition function @xmath107\otimes\delta_1\left(\alpha\left(\frac{2 } { 3}\lambda - x_1\right)-\lambda\right)\otimes \left(a_1a_2-b^2\right)}\right)^{1/2}\nonumber\\ & = \det\delta_1\left(-\lambda\right ) \otimes\delta_0\left(\frac{4 \lambda } { \beta -3}\right ) \otimes\left(\delta_2\left(x_1\right ) \otimes\delta_1\left(\alpha\left(\frac{2}{3}\lambda - x_1\right)-\lambda\right ) \otimes\delta_0\left(\frac{a_3\pm \sqrt{a4}}{6\left(\beta-3\right)^2}\right)\right)^{-1/2}~ , \end{aligned}\ ] ] @xmath108 where we have again disregarded an irrelevant multiplicative prefactor . + as a quick check we may verify that we can reproduce known results from the literature . if we consider the replacements @xmath109 , corresponding to einstein gravity in landau - dewitt gauge , we find @xmath110 arriving precisely at the partition function given in @xcite . equivalent results in other gauges follow similarly . on the fermionic side we follow largely the same approach as utilised in the previous section , with the exception that rather than starting from the euclidean ( @xmath60 ) action , we will utilise and perform the continuation at an opportune moment . we have the action @xmath111 which , given the absence of fermionic background terms , is already quadratic in quantum fields . from the standard decompositions @xmath112 where @xmath113 in our gauge choice . we then define a class of fermionic operators for constant @xmath80 @xmath114 where explicit @xmath6 terms are largely for future convenience and coherence with the literature . + we euclideanise via the transformations @xmath115 which , since the dirac operator ` squares ' to give the laplacian @xmath116 provides the useful transformation @xmath117 which we may then apply to simultaneously remove @xmath118 operators and euclideanise the theory . + as before , we may streamline computations via the @xmath119 identity given in @xcite ( using as appropriate ) @xmath120 where @xmath121 is a generic mass term . although we have already imposed the gauge condition via , to implement it consistently ( and with a view to preserving local supersymmetry ) we may firstly consider a more general gauge fixing strategy ( which will in fact supersede the condition ) , before specialising to the specific instance of @xmath122 . + in generality , to fix the local supersymmetry present in the gravitino sector we must supplement with some gauge - fixing term , which , to preserve on - shell supersymmetry , we may derive via considerations of the variation of with respect to the transformation @xmath123 ( where @xmath124 is assumed to be killing , thus obeying the @xmath60 relation @xmath125 ) . note that for an ds space in our conventions @xmath126 , and thus the latter relation can contribute imaginary terms , these vanish in the limit @xmath82 that we consider herein . as encountered in other circumstances , strict proportionality between the fermionic and bosonic gauge fixing terms is difficult due to the presence of @xmath127 terms in the variation of @xcite . as a compromise however , we may find a proportionality in the following manner . taking the variation and the subsequent @xmath128-trace ( and noting that we will apply the eventual constraint @xmath129 ) , we find @xmath130 which suggests the following gauge fixing term @xmath131 since @xmath132 and on - shell proportionality between and is ensured . + to now find the corresponding ghost action we vary @xmath13 about the classical background ( where @xmath133 ) by decomposing @xmath134 in to give @xmath135 which yields the ghost action @xmath136 for some commuting complex spin @xmath16 field @xmath137 . + to ensure on - shell gauge independence we must also take account of so - called third ( or nielsen - kallosh ) ghosts arising from the non - trivial @xmath138 operator in our gauge fixing condition . exponentiating , we find @xmath139 for some anticommuting majorana and commuting dirac spinor fields @xmath8 and @xmath140 , respectively . + integrating gives the fermionic ghost partition function @xmath141 where we have leveraged to equate @xmath142 which , as in the bosonic case , is true modulo the additional zero modes incurred by the decomposition we have used . combining these elements and noting that the gauge condition implies the vanishing of the spin 1/2 field @xmath143 , we find the quadratic gravitino terms @xmath144 which we can integrate as before to give the total fermionic partition function ( including jacobian factors ) @xmath145 and we have again leveraged to equate @xmath146 which , as in the bosonic case , is true modulo the additional zero modes incurred by the decomposition we have used . having derived the relevant partition functions , we are now in a position to compute the one - loop effective action via the relation @xmath147 in conjunction with the functional determinant techniques detailed in the appendix , i.e. @xmath148 where we have a cut - off @xmath149 , @xmath128 is the euler - mascheroni constant , and the operator @xmath150 is ` non - decomposed ' . the mass dimension of @xmath151 and @xmath152 is one , and it is important to note at this point that as @xmath124 arises from equation in the appendix , which can be thought of as a proper time integral , @xmath153 is a short time ( and thus high energy ) cutoff . flowing from the uv to ir therefore corresponds to the direction of increasing @xmath154 . + given that we are investigating a simple model of supergravity which we anticipate the embedding thereof in the context of a more uv - complete theory , we set aside the renormalisability of the action for now ( encoded in the divergent terms above ( as an aside , @xmath155 is always zero for supersymmetric theories ) ) . instead , we will focus on the finite parts of our ` decomposed ' effective potential @xmath156 and the resultant effective equations . we may represent the ( finite parts of the ) ` decomposed ' effective potential as @xmath157 where @xmath158 is the usual spacetime volume for an @xmath60 of radius @xmath159 , and @xmath160 @xmath161 @xmath162 where @xmath163 is the number of extra zero modes incurred by our decompositions ( as alluded to previously , and first elucidated in @xcite ) . + we may note at this point the importance of the asymptotic forms for @xmath164 and @xmath165 presented above ( and derived in the appendix ) ; it is these relations that allow us to express the effective potential in terms of elementary functions , rather than the awkward integrals of digamma functions from which they derive in this instance , and thus fully investigate the behaviour of @xmath156 . + we should also note that one may be alarmed by the presence of an @xmath166 term in , which would naturally dominate over any classical contributions as @xmath82 , suggesting a failure of our one - loop approach . we may however note that there is a precise cancellation of any such terms at the level of the effective potential . more specifically , for every @xmath167 in there is a corresponding @xmath168 carrying the same coefficient and opposite sign . as @xmath168 also gains a factor of @xmath169 in , this then cancels exactly with the corresponding term arising from . one may also naturally be alarmed by the presence of @xmath170 terms in the above , as @xmath80 has the potential to become negative . this would then yield imaginary terms in the effective potential , which , if not artefacts of our one - loop formalism , would indicate an instability and thus the impossibility of dynamical gravitino condensation in this context . we may address this issue in two distinct ways . + given that the parameter we have available to tune is @xmath18 , we will consider the case where @xmath171 : i.e. whereupon after varying @xmath66 we self - consistently find non - trivial minima @xmath172 . this is natural in the present context as we are considering a renormalised cosmological constant @xmath173 , and it is known generically that quantisation of metric fluctuations about @xmath64 leads to ( positive ) planckian values for @xmath6 @xcite . to thus arrive at the case @xmath174 we must tune @xmath51 ( [ bare ] ) to cancel out the positive energy density incurred via quantisation , mandating that @xmath171 . + given that for @xmath82 @xmath175 , it is firstly straightforward to note upon inspection of that in the limit @xmath82 the only @xmath165 functions from which may be problematic in this sense are @xmath176 corresponding to the fields @xmath98 , @xmath99 and @xmath102 . + working for convenience in the gauge @xmath91 , these become @xmath177 and it is simple to note that via we will then arrive at an imaginary term in @xmath178 which we may freely tune via the gauge parameter @xmath179 . this freedom to tune is of course suggestive of the conclusion that any such terms are non - physical . whilst the real - valued nature of @xmath179 prevents us from tuning any such terms exactly to zero , we may in practice tune such that they may no longer play any physical role ; i.e. that @xmath180 for some non - trivial minimum @xmath66 satisfying the self - consistency condition @xmath181 outlined in the next section . given the inelegance of this approach however , we will not make use of it in what follows . + we will instead address this issue via an appeal to so - called ` physical ' gauges as outlined in [ sec : physical gauge ] . in the current context , this would amount to disregarding the components of the graviton corresponding to general coordinate transformations ; i.e. the fields @xmath98 and @xmath99 in our notation . one may note that it is the @xmath182 term which yields the ( non - contributory ) last element of , and also by comparison with the pure gr case as in verify that it corresponds to the trace of the graviton @xmath102 . it is perhaps illustrative to then note that it is precisely @xmath98 and @xmath99 that correspond to the two remaining @xmath165 functions in , indicating that any imaginary terms in this context must arise purely from gauge , rather than physical , degrees of freedom . working in a physical gauge these @xmath165 functions would not be present , and reality of the action for negative @xmath51 would thus be assured . whilst , as discussed previously , we may tune @xmath179 to eliminate any imaginary terms , it is simpler in this context to utilise a physical gauge . having already derived the effective action @xmath183 in generality , it is straightforward to specialise to this gauge ; we set @xmath184 and disregard the fields @xmath98 and @xmath99 , along with the ghost fields that were originally introduced to cancel out their ( gauge ) degrees of freedom ( which at any rate do not contribute for @xmath59 ) . the functional integral over these gauge degrees of freedom then yields an infinite constant prefactor of the volume of the diffeomorphism group , which is unimportant for our purposes . as also noted in the previous section this process incurs extra zero modes , but whose contributions are subleading as @xmath82 and so can also be neglecting in the following . we find for @xmath184 @xmath185 where the limit @xmath54 is understood to have been taken . + we may firstly note the presence of the @xmath186 term in , which has the capability to destabilise the potential for @xmath187 . in some sense , given the general incompatibility of supersymmetry with de sitter space , this should perhaps be unsurprising : as our intention is to break local supersymmetry dynamically ( i.e. via loop corrections ) , breaking it first at tree level via a positive cosmological constant @xmath51 renders the subsequent breaking via a dynamically generated @xmath66 an impossibility . as such , we must tune @xmath18 for a given value of @xmath151 to find self consistent minima @xmath66 satisfying the condition @xmath188 to ensure @xmath171 and thus a real @xmath178 . if this condition is not met , @xmath178 will contain an imaginary contribution @xmath189 it is furthermore interesting to note that this problematic term arises precisely from the spin 2 part of the effective potential ; absent these contributions , we find the potential @xmath190 which is real for all @xmath18 , @xmath66 . we may compare this expression with the potential computed in @xcite , found in a similar context but via an expansion about a flat , rather than curved , spacetime ( thus neglecting fluctuations of the metric field ) . @xmath191 where @xmath192 is the uv cutoff implemented in @xcite . the shapes of the potentials and are qualitatively similar , however , we find generically that leads to a larger dynamically generated gravitino mass . + , @xmath193 , with non - trivial minima at @xmath194 . we may observe firstly that @xmath171 and @xmath195 at the minima , to respectively ensure reality and self - consistency with our previous limit @xmath82 , and secondly that the onset of imaginary terms ( represented by the transition from blue ( solid ) to green ( dashed ) in the curve ) occurs when @xmath51 changes sign . for higher values of the non - trivial minima , this transition occurs further and further away from the minimum , extending the range of the blue curve beyond the minimum at the cost of a higher dynamically generated mass.,scaledwidth=45.0% ] at this point we emphasise that the complexity of the potential beyond the self - consistent minima is unproblematic in the current context ; the minima @xmath66 is a constant which is self - consistently determined by the minimisation procedure , and in that sense is not a dynamical field . quantum fluctuations about the minima , are represented by a space - time dependent condensate field @xmath196 , which however is massive , with mass of order of the gravitino field , as can be deduced by the ( parabolic ) shape of the effective potential around the non - trivial minima . thus , such quantum fluctuations are suppressed and are not capable of destroying the stability of the broken phase minima . + a rigorous estimation of the mass of the gravitino condensate requires a computation of the wave - function renormalisation of the condensate field , following the steps outlined in @xcite . in the present context , such an analysis has to be performed in de sitter space prior to the limit @xmath54 . given that such estimates are of no interest to us at this stage , we will not perform these calculations here . the mass of the gravitino condensate , though , is essential when discussing the phenomenology of inflationary scenarios that such condensates may induce @xcite . this is postponed for a future publication . + it is straightforward to note ( see fig . [ fig : pot ] ) that the potential above has the correct shape to realise the super - higgs effect , yielding ( _ cf_. , in view of the normalisation of @xcite ) a minimum dynamically generated gravitino mass of order @xmath197 with a corresponding global supersymmetry breaking scale @xmath198 furthermore , we may isolate the fermionic and bosonic contributions at zero and one - loop order as @xmath199 so that we may identify @xmath200 and @xmath201 at the non - trivial minima given above we find @xmath202 , @xmath203 , while the tree - level cosmological constant is negative ( corresponding to anti de - sitter space time ) , @xmath204 . in this case , the one - loop cosmological constant ( vacuum energy ) vanishes , as a result of the stronger spin-2 contributions ( positive sign ) as compared with the ( opposite sign ) gravitino torsion terms . + the above - mentioned dynamical gravitino mass can be modified by tuning the free quantities @xmath18 and @xmath151 , with the above minimum value ( [ gravinomass ] ) , obtained when the non - trivial minima of the potential occur close to inconsistent regions of the parameter space , where @xmath51 changes from negative to positive , preventing the dynamical mass from being consistently decreased further . + and @xmath18 , whilst holding , respectively , @xmath18 and @xmath151 fixed . the arrows in the respective axes correspond to the direction of increasing @xmath151 and @xmath18 . we may firstly note that as we flow from uv to ir ( _ i.e_. in the direction of increasing @xmath151 , _ cf_. discussion following eq . ( [ mudef ] ) ) we obtain the correct double - well shape required for the super - higgs effect , and secondly that tuning @xmath18 allows us to shift @xmath178 and thus attain the correct vacuum structure ( _ i.e. _ non - trivial minima @xmath66 such that @xmath181).,title="fig:",scaledwidth=40.0% ] and @xmath18 , whilst holding , respectively , @xmath18 and @xmath151 fixed . the arrows in the respective axes correspond to the direction of increasing @xmath151 and @xmath18 . we may firstly note that as we flow from uv to ir ( _ i.e_. in the direction of increasing @xmath151 , _ cf_. discussion following eq . ( [ mudef ] ) ) we obtain the correct double - well shape required for the super - higgs effect , and secondly that tuning @xmath18 allows us to shift @xmath178 and thus attain the correct vacuum structure ( _ i.e. _ non - trivial minima @xmath66 such that @xmath181).,title="fig:",scaledwidth=40.0% ] moreover , as demonstrated in fig . [ fig : potmu ] , the shape of the effective potential changes , as one varies the ( renormalisation ) scale @xmath151 from ultra - violet ( uv ) to infra - red ( ir ) values ( _ i.e. _ flowing in the direction of increasing @xmath151 ) , in such a way that the symmetry broken phase ( double - well shaped potential ) is reached in the ir . for comparison , we also mention that this feature also characterises the flat - space time effective potential of @xcite , as shown in @xcite . this demonstrates that the dynamical generation of a gravitino mass is actually an ir phenomenon , in accordance with rather general features of dynamical mass in field theory . + given the phenomenological unsuitability of the above ( transplanckian ) value of the gravitino mass ( [ gravinomass ] ) and the global supersymmetry breaking scale ( [ fscale ] ) , of order of the reduced planck mass , it is natural to seek ways for generating much lower gravitino masses , for global supersymmetry scales below the grand unification scale , thereby making contact with realistic phenomenology of such theories . to this end , we shall consider an extension of the above analysis to a conformal @xmath2 supergravity scenario , e.g. of the type considered in @xcite , which was used in @xcite to discuss the role of gravitino condensates in providing inflationary scenarios compatible with the current astrophysical data @xcite . + although we shall postpone , for a forthcoming publication , a detailed analysis of dynamical breaking of local supersymmetry in such conformal supergravities , including their matter sectors ( that notably can accommodate the next - to - minimal supersymmetric standard model ) , we may nevertheless discuss some important consequences of such models in yielding phenomenologically realistic values for the global supersymmetry breaking scale @xmath18 and the dynamical gravitino mass . + in this respect , we first notice that , in conformal supergravities @xcite in the jordan frame , there is a non - trivial coupling of a dilaton superfield to the gravitational einstein term in the supersymmetric action . passing to the einstein - frame @xcite , where the ( bosonic ) einstein - hilbert scalar curvature term in the lagrangian assumes its canonically normalised form , the gravitational part of the action of @xmath2 conformal supergravity , of relevance to our purposes here , is given by : @xmath205 where the superscript @xmath206 denotes quantities in the einstein frame , @xmath207 is the ( dimensionless ) dilaton field , @xmath208 its potential , @xmath209 denotes the canonically - normalized gravitino with a standard kinetic term as in @xmath0 supergravity , and the @xmath37 in ( [ confsugra2 ] ) denote contributions from dilaton - derivative terms , dilatinos , as well as auxiliary , gauge - fixing ghost and matter fields , which will not play a role in our discussion here . + in general the form of the potential @xmath210 depends on the low - energy content of the action , and apart from the dilaton terms it may also contain terms that depend on matter multiplet fields that appear in next - to - minimal extensions of the standard model that can be accommodated in the low - energy limit of such frameworks . for our purposes in this work we simply assume @xcite that , upon appropriately minimising the potential @xmath208 , the dilaton field is stabilised to a ( space - time ) constant v.e.v . @xmath211 . in this way , we observe from ( [ confsugra2 ] ) that in conformal supergravities , the four - gravitino interaction terms carry an extra factor arising from the v.e.v . of the dilaton field , yielding a modified coupling in the gravitino self - interaction sector @xcite @xmath212 the presence of two couplings , one ( the standard gravitational one , @xmath213 ) for the metric tensor interactions , and the other ( @xmath214 , _ cf_. eq . ( [ kappatilde ] ) ) for the gravitino self - interaction terms , leads to the possibility of dynamical generation of much smaller gravitino masses in the instance of conformal supergravity models with @xcite @xmath215 than in the simple @xmath2 supergravity scenario discussed in previous sections . + it should be noted , however , that , as in the standard supergravity case , the tree - level cosmological constant at the broken symmetry minima ( [ bare ] ) must be negative ( anti - de - sitter space time ) in order to avoid imaginary parts in the potential , which still arise in the bosonic part of the potential when @xmath216 . this is a notable difference from the flat space - time case of @xcite , where @xmath217 at the non - trivial minima , and thus global and local supersymmetries are already broken at tree level . we next notice that , in the conformal case the gravitino torsion parts of the effective potential dominate over the contributions due to the spin-2 graviton quantum fluctuations , and thus it is such torsion condensates which drive the one - loop cosmological constant to zero in this case . since the situation is qualitatively similar to the flat space - time case of @xcite , we may say that in the case of conformal supergravity models with @xmath215 , the gravitino torsion condensates `` effectively '' _ parallelize _ the manifold , in the sense of the gravitino parts of the spin connection being responsible for `` flattening '' the space - time . , which is known to be a parallelisable manifold ) of higher - dimensional ( d=11 ) supergravities @xcite , which were argued to condense , cancelling any contributions from the metric in the corresponding components of the christoffel symbol . in this way , a vanishing cosmological constant arises @xcite in the four - dimensional space - time obtained after appropriate compactification . however , our result on `` torsion - induced parallelism '' in the 4d conformal supergravity case does not constitute an exact mathematical statement , since it is only demonstrated at one loop . unlike the higher - dimensional case of @xcite , it has not been demonstrated that the gravitino torsion parts in the connection cancel out exactly any bosonic metric contributions to the christoffel symbols , thereby leading to a vanishing riemann tensor for our four - dimensional space - time manifold . all we have shown here is that , at one loop order , the minimum of the effective potential vanishes , if there are lorentz - invariant bilinear gravitino condensates , corresponding to a non zero gravitino mass . the resulting effective einstein equations then , for vanishing cosmological constant , admit flat minkowski solutions under the one - loop approximation . hence there is no contradiction of the current results with any rigorous theorems on parallelism , known in the literature . ] + to see this , we first notice that in the conformal supergravity case , the contributions to the one - loop effective potential coming from the spin-2 ( graviton ) fluctuations are still given by ( [ boson ] ) , while the gravitino contributions are given by ( [ fermion ] ) upon the replacement of @xmath213 by @xmath214 : @xmath218 if we consider for concreteness the case @xmath219 , which is a value dictated by the inflationary phenomenology of the model @xcite , we may find solutions with a vanishing one - loop effective potential at the non - trivial minima corresponding to ( _ cf_. fig . [ fig : confpot ] ) : @xmath220 which lead to a global supersymmetry breaking scale @xmath221 and dynamical gravitino mass @xmath222 at the non - trivial minima we find @xmath223 , @xmath224 , with tree - level cosmological constant @xmath225 . these values are phenomenologically realistic , thereby pointing towards the viability ( from the point of view of producing realistic results of relevance to phenomenology ) of the scenarios of dynamical breaking of local supersymmetry in conformal supergravity models . + , @xmath226 , @xmath227 , @xmath228 . the flatness of the potential around the origin is pronounced compared to the simple @xmath2 supergravity case of fig . [ fig : pot ] , whilst the other features of the diagram remain unchanged . , scaledwidth=45.0% ] an additional important feature is the pronounced flatness ( near the origin , cf . [ fig : confpot ] ) of the one - loop effective potential for the gravitino condensate in such models , as compared to the pure supergravity case of the previous sections . this leads @xcite to inflationary scenarios , with the massive gravitino condensate playing the role of the inflaton field , that are in agreement with the planck satellite data on inflation @xcite . the ( slow - roll ) inflationary phase in such approaches is identified with the rolling of the inflaton / gravitino - condensate field towards its non - trivial minimum on either side of the potential , with a model - dependent duration . the initial value of the inflaton depends on the initial conditions for inflation , whilst the exit from inflation coincides with the phase in which the field oscillates coherently near its non - trivial minimum . in that regime , as becomes evident from the shape of the double well potential , the inflaton condensate is massive ( in the conformal model of @xcite the inflaton has mass of the order of the grand unification scale ) . reheating of the universe then could conceivably arise from the decays of the condensate field upon coupling supergravity to matter . it would be interesting to explore such scenarios within our framework , in which the full quantum gravity corrections have been accounted for . this will constitute the topic of a future publication . + finally , before closing this section we should also mention an additional possibility that might characterise the conformal supergravity models of inflation , as envisaged in @xcite . this regards viewing the broken - supergravity - phase vacuum as being a _ metastable _ one , with a lifetime longer than the age of the universe . in such a case , the restriction on the value of the condensate field at the non - trivial minima @xmath229 could be lifted ( thus implying local supersymmetry breaking already at tree - level ) , provided the conformally rescaled coupling @xmath214 is sufficiently larger than the gravitational coupling @xmath213 , in order to ensure metastability . detailed studies in this direction , making contact with realistic phenomenologies , will also be a topic for future work . in this work we have revisited the issue of dynamical breaking of local supersymmetry ( supergravity ) without coupling to either gauge or matter sectors . we have examined for simplicity the specific case of @xmath1 , @xmath2 simple supergravity , although our considerations can be extended to more complicated and extended supergravities . we have constructed one - loop effective potentials , which exhibit a double - well ( symmetric about the origin ) shape and which vanish at the non - trivial minima . in this way dynamical mass generation of the gravitino field can be understood most clearly . we have taken into account quantum fluctuations of the metric field by expanding about a euclidean de sitter background and taking the limit of a vanishing renormalised cosmological constant only at the end of the computations . + it is essential for our arguments that global supersymmetry is broken at a given scale . we have assumed f - type breaking for concreteness , which resulted in a positive value of the one - loop effective potential at the origin of the condensate field . this allows for the existence of a non - trivial minimum at which the one - loop effective potential vanishes , consistently with the assumed vanishing of the ( renormalised ) cosmological constant @xmath54 . + we have demonstrated the existence of vacua at which the imaginary parts of the effective action were absent , contrary to previous claims in the literature @xcite , thereby supporting the possibility of dynamical breaking of supergravity and the acquisition of a dynamical mass by the gravitino . the latter is found to be or order of the planck mass in this simple supergravity framework , although this conclusion changes in more complicated settings , such as conformal supergravities @xcite , where phenomenologically realistic values are obtained for the dynamical gravitino mass and the global supersymmetry breaking scale . our findings support therefore , at least qualitatively , the results of the flat - space analysis of @xcite on dynamical breaking of supergravity . + an interesting feature of the double - well potentials that we have calculated in the broken phase of conformal supergravity models is their flatness about the origin , prompting one to consider small - field inflationary scenarios with the gravitino condensate playing the role of the inflaton field . a preliminary study @xcite of such scenarios in the context of the flat - space potential of @xcite within a conformal supergravity model @xcite has shown agreement of the inflationary cosmology with the current cosmological data . a complete analysis in this direction , within the context of conformal supergravitry models , where one - loop quantum fluctuations of the metric field are fully taken into account , as in the current article , is still pending , and will be the subject of a future work . the work of n.h . is supported by a kcl gta studentship , while that of n.e.m . is supported in part by the london centre for terauniverse studies ( lcts ) , using funding from the european research council via the advanced investigator grant 267352 and by stfc ( uk ) under the research grant st / j002798/1 . in this appendix we give details of some mathematical aspects and notions of our approach towards the construction of the one - loop effective action of @xmath2 @xmath1 supergravity model . more precisely , we detail the use of the heat kernel in computing functional determinants , before specialising to the computation of the resultant zeta functions on @xmath60 . finally , we demonstrate an asymptotic expansion which allows these zeta functions to be explicitly evaluated in the limit @xmath59 . consider a second - order laplace - type differential operator of the form @xmath230 , for some constant @xmath80 , defined on a smooth vector bundle over a compact , smooth @xmath231-dimensional riemannian manifold without boundary . there exist a discrete number of eigenfunctions and corresponding eigenvalues of this operator , which may be decomposed spectrally into a complete orthonormal set of eigenfunctions @xmath232 with eigenvalues @xmath233 , of multiplicity @xmath234 . + the determinant of this operator may be expressed @xmath235 however as this obviously diverges , we shall instead define the zeta function @xmath236 convergent for @xmath237 , which can in practice be extended via analytic continuation to a meromorphic function of @xmath238 over the entire complex plane . it is important to note that @xmath239 is regular at @xmath240 , yielding the derivative @xmath241 so that we may define @xmath242 our task is then to compute the form of @xmath165 for a given operator . + a convenient way of encapsulating some of the behaviour of @xmath164 is via the ` trace over the heat kernel ' , defined thusly @xmath243 valid for @xmath244 , where @xmath245 satisfies the heat equation with boundary condition @xmath246 and the @xmath247 are the heat kernel coefficients , which integrate to give spectral invariants of @xmath248 . since the heat equation has the solution @xmath249 we can trace over and integrate to see that ( since the @xmath250 form an orthonormal basis ) @xmath251 finally , we note that @xmath239 is related to via a mellin transform @xmath252 so that for @xmath253 we may expand the sum ( thus preserving only the first few terms ) and extract information about @xmath164 as necessary . + taking the divergent parts given by the trace over the heat kernel , in conjunction with the known expression for the finite parts of zeta regularised determinants @xcite , we can then find for @xmath57 @xmath254 where we have a cut - off @xmath149 , and @xmath128 is the euler - mascheroni constant . the mass dimension of @xmath151 and @xmath152 is one , and it is important to note at this point that as can be though of as a proper time integral , @xmath153 is a short time ( and thus high energy ) cutoff . flowing from the uv to ir therefore corresponds to the direction of increasing @xmath154 . + computing the form of the @xmath255 is straightforward in practice as their general forms are known @xcite . in @xmath57 we have @xmath256 where the trace is performed in the space of fields . as the heat kernel coefficients are straightforward to find ( for low @xmath231 ) , our problem of evaluating functional determinants is now reduced to computing the form of @xmath257 and @xmath258 for a given background . as @xmath164 is defined by the eigenvalues and their degeneracies for a given operator , we must work in a framework where these quantities are known . this is achieved in practice by specialising to ` differentially constrained ' operators ; those corresponding to irreducible representations of the background isometry group . for @xmath259 these representations can be labelled by @xmath260 ( for a positive integer @xmath261 and a spin @xmath262 ) with corresponding eigenvalues @xmath263 and degeneracies then given by the dimension of the representation , @xmath264 for the various representations we therefore have @xcite ( where we have rescaled @xmath265 @xmath266 , for @xmath267 on @xmath60 ) + * spin 0 : @xmath268 * @xmath269 * spin 1 : @xmath270 * @xmath271 * spin 2 : @xmath272 * @xmath273 * spin 1/2 : @xmath274 * where both representations have the same spectra , for ( n-1/2,1/2 ) we have @xmath275 which is to be doubled . + * spin 3/2 : @xmath276 * where both representations again have the same spectra , for ( n-1/2,3/2 ) we have @xmath277 which is to be doubled . spinor representations may be incorporated into this framework via ` squaring ' the corresponding first - order operators to yield those of second - order . + with this in mind , our prior expression for @xmath164 can now be re - expressed more concretely as @xmath278 where ( following the conventions of @xcite ) we have defined @xmath279 @xmath280 where @xmath267 , and we note that our sum starts from the minimal @xmath281 such that @xmath282 , and therefore all possible negative and zero modes are included . + following the appendix of @xcite , it is firstly straightforward to show that @xmath283 in line with this derivation we can compute @xmath284 via a similar argument , making use of the identity @xcite @xmath285 where @xmath286 is the digamma function . we find @xmath287 we may then note that @xmath288 so that we may integrate both sides to arrive at @xmath289 where @xmath89 is a ( real ) constant of integration @xmath290 for large @xmath291 we can explicitly evaluate the integral above by shifting the measure via @xmath292 and inserting the asymptotic expansion @xmath293 where @xmath294 is the @xmath261th bernoulli number . integrating term by term , only the leading order remains relevant and we find @xmath295 combing these elements , we thus find that for small @xmath6 @xmath296 from this result , it is interesting to note that in the limit @xmath82 , the presence or absence of imaginary terms in the effective potential is a straightforward consequence of the sign of @xmath80 , the argument of the functional determinant being evaluated . this may be therefore leveraged to explicitly investigate the possibility of instabilities in the effective potential , along with the associated question of whether or not any such apparent instabilities are , in fact , one - loop artefacts which may then be disregarded . these issues are of course central to the topic of this paper . p. van nieuwenhuizen , phys . rept . * 68 * ( 1981 ) 189 . h. p. nilles , phys . * 110 * , 1 ( 1984 ) . y. nambu and g. jona - lasinio , phys . * 122 * , 345 ( 1961 ) ; phys . rev . * 124 * , 246 ( 1961 ) . r. s. jasinschi , a. w. smith , phys . * b173 * , 297 ( 1986 ) . d. v. volkov and v. p. akulov , phys . b * 46 * , 109 ( 1973 ) ; theor . phys . * 18 * , 28 ( 1974 ) [ teor . mat . fiz . * 18 * , 39 ( 1974 ) ] . e. witten , nucl . b * 188 * , 513 ( 1981 ) . i. l. buchbinder and s. d. odintsov , class . * 6 * ( 1989 ) 1955 . e. s. fradkin and a. a. tseytlin , nucl . b * 234 * ( 1984 ) 472 . p. k. townsend , phys . d * 15 * ( 1977 ) 2802 . r. percacci , m. j. perry , c. n. pope and e. sezgin , arxiv:1302.0868 [ hep - th ] . m. r. gaberdiel , d. grumiller and d. vassilevich , jhep * 1011 * ( 2010 ) 094 [ arxiv:1007.5189 [ hep - th ] ] . h. -b . zhang and x. zhang , class . * 29 * ( 2012 ) 145013 [ arxiv:1205.3681 [ hep - th ] ] . o. lauscher and m. reuter , phys . d * 65 * ( 2002 ) 025013 [ hep - th/0108040 ] . e. s. fradkin and a. a. tseytlin , nucl . b * 234 * ( 1984 ) 509 . s. -i . kojima , n. sakai and y. tanii , int . j. mod . a * 9 * ( 1994 ) 5415 [ hep - th/9311045 ] . g. a. vilkovisky , nucl . phys . b * 234 * ( 1984 ) 125 . s. w. hawking , commun . * 55 * ( 1977 ) 133 . s. m. christensen and m. j. duff , nucl . b * 170 * ( 1980 ) 480 . b. allen , nucl . b * 226 * ( 1983 ) 228 . j. ellis and n. e. mavromatos , arxiv:1308.1906 [ hep - th ] , phys . d in press . s. ferrara , r. kallosh , a. linde , a. marrani and a. van proeyen , phys . d * 82 * , 045003 ( 2010 ) [ arxiv:1004.0712 [ hep - th ] ] .

this paper is dedicated to the study of the topologies and nesting configurations of the components of the zero set of monochromatic random waves . we prove that the probability of observing any diffeomorphism type , and any nesting arrangement , among the zero set components is strictly positive for waves of large enough frequencies . our results are a consequence of building laplace eigenfunctions in euclidean space whose zero sets have a component with prescribed topological type , or an arrangement of components with prescribed nesting configuration .

1701.00034

for @xmath0 let @xmath1 denote the linear space of entire ( real valued ) eigenfunctions @xmath2 of the laplacian @xmath3 whose eigenvalue is @xmath4 @xmath5 the zero set of @xmath2 is the set @xmath6 the zero set decomposes into a collection of connected components which we denote by @xmath7 . our interest is in the topology of @xmath8 and of the members of @xmath7 . let @xmath9 denote the ( countable and discrete ) set of diffeomorphism classes of compact connected smooth @xmath10-dimensional manifolds that can be embedded in @xmath11 . the compact components @xmath12 in @xmath7 give rise to elements @xmath13 in @xmath9 ( here we are assuming that @xmath2 is generic with respect to a gaussian measure so that @xmath8 is smooth , see section [ monochromatic ] ) . the connected components of @xmath14 are the nodal domains of @xmath2 and our interest is in their nesting properties , again for generic @xmath2 . to each compact @xmath15 we associate a finite connected rooted tree as follows . by the jordan - brouwer separation theorem @xcite each component @xmath15 has an exterior and interior . we choose the interior to be the compact end . the nodal domains of @xmath2 , which are in the interior of @xmath12 , are taken to be the vertices of a graph . two vertices share an edge if the respective nodal domains have a common boundary component ( unique if there is one ) . this gives a finite connected rooted tree denoted @xmath16 ; the root being the domain adjacent to @xmath12 ( see figure 2 ) . let @xmath17 be the collection ( countable and discrete ) of finite connected rooted trees . our main results are that any topological type and any rooted tree can be realized by elements of @xmath1 . [ main theorem r a ] given @xmath18 there exists @xmath19 and @xmath20 for which @xmath21 . [ main theorem r b ] given @xmath22 there exists @xmath19 and @xmath15 for which @xmath23 . theorems [ main theorem r a ] and [ main theorem r b ] are of basic interest in the understanding of the possible shapes of nodal sets and domains of eigenfunctions in @xmath11 ( it applies equally well to any eigenfunction with eigenvalue @xmath24 instead of @xmath4 ) . our main purpose however is to apply it to derive a basic property of the universal monochromatic measures @xmath25 and @xmath26 whose existence was proved in @xcite . we proceed to introduce these measures . let @xmath27 be the @xmath28sphere endowed with a smooth , riemannian metric @xmath29 . our results apply equally well with @xmath30 replaced by any compact smooth manifold @xmath31 ; we restrict to @xmath30 as it allows for a very clean formulation . consider an orthonormal basis @xmath32 for @xmath33 consisting of real - valued eigenfunctions , @xmath34 . a monochromatic random wave on @xmath27 is the gaussian random field @xmath35 @xmath36 } a_j \phi_j,\label{e : rwdef}\ ] ] where the @xmath37 s are real valued i.i.d standard gaussians , @xmath38 , @xmath39 is a non - negative function satisfying @xmath40 as @xmath41 , and @xmath42\}$ ] . when choosing @xmath43 the @xmath44 s we consider in forming the @xmath45 s are the square roots of the laplace eigenvalues . to a monochromatic random wave we associate its ( compact ) nodal set @xmath8 and a corresponding finite set of nodal domains . the connected components of @xmath8 are denoted by @xmath7 and each @xmath15 yields a @xmath46 . each @xmath20 also gives a tree end @xmath16 in @xmath47 which is chosen to be the smaller of the two rooted trees determined by the inside and outside of @xmath48 . the topology of @xmath8 is described completely by the probability measure @xmath49 on @xmath9 given by @xmath50 where @xmath51 is a point mass at @xmath52 . similarly , the distribution of nested ends of nodal domains of @xmath2 is described by the measure @xmath53 on @xmath47 given by @xmath54 with @xmath55 is the point mass at @xmath56 . the main theorem in @xcite asserts that there exist probability measures @xmath57 and @xmath26 on @xmath9 and @xmath17 respectively to which @xmath49 and @xmath53 approach as @xmath41 , for almost all @xmath35 , provided one has that for every @xmath58 @xmath59\big . \right|=o(1),\ ] ] as @xmath60 . here , @xmath61 , @xmath62 is the localized wave on @xmath63 defined as @xmath64 , and @xmath65 is the gaussian random field on @xmath63 characterized by the covariance kernel @xmath66 ( see section [ monochromatic ] ) . the probability measures @xmath57 and @xmath26 are universal in that they only depend on the dimension @xmath67 of @xmath31 . monochromatic random waves on the @xmath67-sphere equipped with the round metric are known as random spherical harmonics whenever @xmath68 . it is a consequence of the mehler - heine @xcite asymptotics that they satisfy condition for all @xmath58 . also , on any @xmath27 the fields @xmath45 with @xmath69 satisfy condition for all @xmath58 . finally , monochromatic random waves @xmath45 on @xmath27 with @xmath70 , for some @xmath71 , satisfy condition for every @xmath58 satisfying that the set of geodesic loops that close at @xmath72 has measure @xmath73 ( see @xcite ) . on general manifolds one can define monochromatic random waves just as in @xmath27 . monochromatic random waves with @xmath68 on the @xmath67-torus are known as arithmetic random waves . they satisfy condition for all @xmath74 if @xmath75 , and on @xmath76 with @xmath77 provided we work with a density one subsequence of @xmath44 s @xcite . on general @xmath78 monochromatic random waves with @xmath70 , for some @xmath71 , satisfy condition for every @xmath79 satisfying that the set of geodesic loops that close at @xmath72 has measure @xmath73 ( see @xcite ) . examples of such manifolds are surfaces without conjugate points , or manifolds whose sectional curvature is negative everywhere . our main application of theorems [ main theorem r a ] and [ main theorem r b ] is the following result . [ main theorem]let @xmath27 be the @xmath67-sphere equipped with a smooth riemannian metric . let @xmath57 and @xmath26 be the limit measures ( introduced in @xcite ) arising from monochromatic random waves on @xmath27 for which condition is satisfied for every @xmath58 . 1 . the support of @xmath57 is @xmath9 . that is , every atom of @xmath9 is positively charged by @xmath57 . the support of @xmath26 is all of @xmath17 . that is , every atom of @xmath17 is positively charged by @xmath26 . theorem [ main theorem ] asserts that every topological type that can occur will do so with a positive probability for the universal distribution of topological types of random monochromatic waves in @xcite . the reduction from theorems [ main theorem r a ] and [ main theorem r b ] to theorem [ main theorem ] is abstract and is based on the ` soft ' techniques in @xcite ( see also section [ monochromatic ] ) . in particular , it offers us no lower bounds for these probabilities . developing such lower bounds is an interesting problem . the same applies to the tree ends . theorem [ main theorem ] holds for monochromatic random waves on general compact , smooth , riemannian manifolds @xmath78 without boundary . part ( i ) actually holds without modification . the reason why we state the result on the round sphere @xmath30 is that , by the jordan - brouwer separation theorem @xcite , on @xmath30 every component of the zero set separates @xmath30 into two distinct components . this gives that the nesting graph for the zero sets is a rooted tree . on general @xmath78 this is not necessarily true , so there is no global way to define a tree that describes the nesting configuration of the zero set in all of @xmath31 , for all @xmath80 . however , according to @xcite almost all @xmath12 s localize to small coordinate patches and hence our arguments apply . we end the introduction with an outline of the paper . theorem [ main theorem r a ] for @xmath81 ( which is the first interesting case ) is proved in @xcite by deformation of the eigenfunction @xmath82 the proof exploits that the space @xmath83 is simply the set of orientable compact surfaces which are determined by their genus . so in engineering a component of a deformation of @xmath2 to have a given genus it is clear what to aim for in terms of how the singularities ( all are conic ) of @xmath84 resolve . for @xmath85 , little is known about the space @xmath9 and we proceed in section [ topology ] quite differently . we apply whitney s approximation theorem to realize @xmath86 as an embedded real analytic submanifold of @xmath11 . then , following some techniques in @xcite we find suitable approximations of @xmath87 and whose zero set contains a diffeomorphic copy of @xmath86 . the construction of @xmath2 hinges on the lax - malgrange theorem and thom s isotopy theorem . as far as theorem [ main theorem r b ] , the case @xmath88 is resolved in @xcite using a deformation of @xmath89 and a combinatorial chess board type argument . in higher dimensions , for example @xmath81 we proceed in section [ nesting ] by deforming @xmath90 this @xmath2 has enough complexity ( as compared to the @xmath91 in ) to produce all elements in @xmath17 after deformation . however , it is much more difficult to study . unlike or @xmath89 , the zero set @xmath92 in has point and @xmath4-dimensional edge singularities . the analysis of its resolution under deformation requires a lot of care , especially as far as engineering elements of @xmath17 . the pay off as we noted is that it is rich enough to prove theorem [ main theorem r b ] . in section [ monochromatic ] we review some of the theory of monochromatic gaussian fields and their representations . section [ topology ] is devoted to the proof of theorem [ main theorem r a ] . section [ nesting ] is devoted to the proof of theorem [ main theorem r b ] . the latter begins with an interpolation theorem of mergelyan type , for elements in @xmath1 . we use that to engineer deformations of which achieve the desired tree end , this being the most delicate aspect of the paper . our interest is in the monochromatic gaussian field on @xmath11 which is a special case of the band limited gaussian fields considered in @xcite , and which is fundamental in the proof of ( * ? ? ? * theoem 1.1 ) . for @xmath93 , define the annulus @xmath94 and let @xmath95 be the haar measure on @xmath96 normalized so that @xmath97 . using that the transformation @xmath98 preserves @xmath96 we choose a real valued orthonormal basis @xmath99 of @xmath100 satisfying @xmath101 the band limited gaussian field @xmath102 is defined to be the random real valued functions @xmath2 on @xmath11 given by @xmath103 where @xmath104 and the @xmath105 s are identically distributed , independent , real valued , standard gaussian variables . we note that the field @xmath106 does not depend on the choice of the orthonormal basis @xmath107 . the distributional identity @xmath108 on @xmath96 together with lead to the explicit expression for the covariance function : @xmath109 from , or directly from , it follows that almost all @xmath2 s in @xmath102 are analytic in @xmath110 @xcite . for the monochromatic case @xmath111 we have @xmath112 where to ease notation we have set @xmath113 in this case there is also a natural choice of a basis for @xmath114 given by spherical harmonics . let @xmath115 be a real valued basis for the space of spherical harmonics @xmath116 of eigenvalue @xmath117 , where @xmath118 . we compute the fourier transforms for the elements of this basis . for every @xmath119 and @xmath120 , we have @xmath121 we give a proof using the theory of point pair invariants @xcite which places such calculations in a general and conceptual setting . the sphere @xmath122 with its round metric is a rank @xmath4 symmetric space and @xmath123 for @xmath124 is a point pair invariant ( here @xmath125 is the standard inner product on @xmath11 restricted to @xmath122 ) . hence , by the theory of these pairs we know that for every function @xmath126 we have @xmath127 where @xmath128 is any spherical harmonic of degree @xmath129 and @xmath130 is the spherical transform . the latter can be computed explicitly using the zonal spherical function of degree @xmath129 . fix any @xmath131 and let @xmath132 be the unique spherical harmonic of degree @xmath129 which is rotationally invariant by motions of @xmath122 fixing @xmath133 and so that @xmath134 . then , @xmath135 the function @xmath136 may be expressed in terms of the gegenbauer polynomials ( * ? ? ? * ( 8.930 ) ) as @xmath137 now , for @xmath138 , @xmath139 where we have set @xmath140 . hence , by we have @xmath141 with @xmath142 the last term in can be computed using ( * ? ? ? * ( 7.321 ) ) . this gives @xmath143 as desired . the monochromatic gaussian ensemble @xmath144 is given by random @xmath2 s of the form @xmath145 where the @xmath146 s are i.i.d standard gaussian variables . the functions @xmath147 , @xmath148 with @xmath149 , and those in for which the series converges rapidly ( eg . for almost all @xmath2 in @xmath144 ) , all satisfy , that is @xmath19 . in addition , consider the subspaces @xmath150 and @xmath151 of @xmath1 defined by @xmath152 @xmath153 [ e : approximation ] let @xmath19 and let @xmath154 be a compact set . then , for any @xmath155 and @xmath156 there are @xmath157 and @xmath158 such that @xmath159 that is , we can approximate @xmath2 on compact subsets in the @xmath160-topology by elements of @xmath150 and @xmath151 respectively . let @xmath161 . since @xmath2 is analytic we can expand it in a rapidly convergent series in the @xmath162 s . that is , @xmath163 moreover , for @xmath164 , @xmath165 in polar coordinates , @xmath166 , the laplace operator in @xmath11 is given by @xmath167 and hence for each @xmath168 we have that @xmath169 where @xmath129 is some positive integer . there are two linearly independent solutions to . one is @xmath170 and the other blows up as @xmath171 . since the left hand side of is finite as @xmath172 , it follows that the @xmath173 s can not pick up any component of the blowing up solution . that is , for @xmath174 @xmath175 for some @xmath176 . hence , @xmath177 furthermore , this series converges absolutely and uniformly on compact subsets , as also do its derivatives . thus , @xmath2 can be approximated by members of @xmath150 as claimed , by simply truncating the series in . to deduce the same for @xmath151 it suffices to approximate each fixed @xmath178 . to this end let @xmath179 be a sequence of points in @xmath122 which become equidistributed with respect to @xmath180 as @xmath181 . then , as @xmath182 , @xmath183 the proof follows since @xmath184 . indeed , the convergence in is uniform over compact subsets in @xmath110 . for @xmath185 open , let @xmath186 denote the eigenfunctions on @xmath187 satisfying @xmath188 for @xmath189 . any function @xmath29 on @xmath187 which is a limit ( uniform over compact subsets of @xmath187 ) of members of @xmath190 must be in @xmath186 . while the converse is not true in general , note that if @xmath191 is a ball in @xmath11 , then the proof of proposition [ e : approximation ] shows that the uniform limits of members of @xmath190 ( or @xmath150 , or @xmath151 ) on compact subsets in @xmath192 is precisely @xmath193 . with these equivalent means of approximating functions by suitable members of @xmath144 , and particularly @xmath1 , we are ready to prove theorems [ main theorem r a ] and [ main theorem r b ] . indeed , as shown in @xcite the extension of condition @xmath194 of ( ? * theorem 1 ) suffices . namely , for @xmath195 it is enough to find an @xmath196 with @xmath197 containing @xmath12 as one of its components for theorem [ main theorem r a ] , and for @xmath22 it suffices to find an @xmath196 such that @xmath23 for some component @xmath12 of @xmath197 . in this section we prove theorem [ main theorem r a ] . by the discussion above it follows that given a representative @xmath12 of a class @xmath46 , it suffices to find @xmath19 for which @xmath7 contains a diffeomorphic copy of @xmath12 . to begin the proof we claim that we may assume that @xmath12 is real analytic . indeed , if we start with @xmath198 smooth , of the desired topological type , we may construct a tubular neighbourhood @xmath199 of @xmath200 and a smooth function @xmath201 note that without loss of generality we may assume that @xmath202 . fix any @xmath203 . we apply thom s isotopy theorem ( * ? ? ? * thm 20.2 ) to obtain the existence of a constant @xmath204 so that for any function @xmath205 with @xmath206 there exists @xmath207 diffeomorphism with @xmath208 to construct a suitable @xmath205 we use whitney s approximation theorem ( * ? ? ? * lemma 6 ) which yields the existence of a real analytic approximation @xmath209 of @xmath210 that satisfies @xmath211 . it follows that @xmath212 is diffeomorphic to @xmath213 and @xmath12 is real analytic as desired . by the jordan - brouwer separation theorem @xcite , the hypersurface @xmath12 separates @xmath11 into two connected components . we write @xmath214 for the corresponding bounded component of @xmath215 . let @xmath216 be the first dirichlet eigenvalue for the domain @xmath214 and let @xmath217 be the corresponding eigenfunction : @xmath218 consider the rescaled function @xmath219 defined on the rescaled domain @xmath220 . since @xmath221 in @xmath222 , and @xmath223 is real analytic , @xmath224 may be extended to some open set @xmath225 with @xmath226 so that @xmath227 where @xmath228 is the rescaled hypersurface @xmath229 . note that since @xmath217 is the first dirichlet eigenfunction , then we know that there exists a tubular neighbourhood @xmath230 of @xmath228 on which @xmath231 ( see lemma 3.1 in @xcite ) . without loss of generality assume that @xmath232 . we apply thom s isotopy theorem ( * ? ? ? * thm 20.2 ) to obtain the existence of a constant @xmath233 so that for any function @xmath2 with @xmath234 there exists @xmath235 diffeomorphism so that @xmath236 since @xmath237 has no compact components , lax - malgrange s theorem @xcite yields the existence of a global solution @xmath238 to the elliptic equation @xmath239 in @xmath11 with @xmath240 we have then constructed a solution to @xmath239 in @xmath11 , i.e. @xmath161 , for which @xmath197 contains a diffeomorphic copy of @xmath12 ( namely , @xmath241 . this concludes the proof of the theorem . we note that the problem of finding a solution to @xmath242 for which @xmath243 contains a diffeomorphic copy of @xmath12 is related to the work @xcite of a. enciso and d. peralta - salas . in @xcite the authors seek to find solutions to the problem @xmath244 in @xmath11 so that @xmath243 contains a diffeomorphic copy of @xmath12 , where @xmath245 is a nonnegative , real analytic , potential and @xmath12 is a ( possibly infinite ) collection of compact or unbounded tentacled " hypersurfaces . the construction of the solution @xmath2 that we presented is shares ideas with @xcite . since our setting and goals are simpler than theirs , the construction of @xmath2 is much shorter and straightforward . the proof of theorem [ main theorem r b ] consists in perturbing the zero set of the eigenfunction @xmath246 so that the zero set of the perturbed function will have the desired nesting . the nodal domains of @xmath247 build a n - dimensional chess board made out of unit cubes . by adding a small perturbation to @xmath247 the changes of topology in @xmath248 can only occur along the singularities of @xmath248 . therefore , we will build an eigenfunction @xmath2 , satisfying @xmath249 , by prescribing it along the singularities @xmath250 of the zero set of @xmath247 . we then construct a new eigenfunction @xmath251 which will have the desired nesting among a subset of its nodal domains . the idea is to prescribe @xmath2 on the singularities of the zero set of @xmath247 in such a way that two adjacent cubes of the same sign will either glue or disconnect along the singularity . the following theorem shows that one can always find a solution @xmath2 to @xmath249 with prescribed values on a set of measure zero ( such as @xmath252 ) . we prove this result following the first step of carleson s proof @xcite of mergelyan s classical theorem about analytic functions . in the statement of the theorem the function @xmath254 can be replaced by @xmath256 , where @xmath185 is any open set with @xmath257 . this is because @xmath258 is dense in @xmath259 in the @xmath260-topology . consider the sets @xmath261 and write @xmath262 for the restrictions of @xmath263 to @xmath264 . both @xmath265 and @xmath266 are subsets of the banach space @xmath267 , and clearly @xmath268 . it follows that the claim in the theorem is equivalent to proving that @xmath269 to prove , note that a distribution @xmath270 in the dual space @xmath271 can be identified with an @xmath272-tuple of measures @xmath273 with @xmath274 for each @xmath275 . that is , for each @xmath276 , @xmath277 since @xmath278 , proving is equivalent to showing that for each @xmath279 satisfying @xmath280 for all @xmath281 one has that @xmath280 for all @xmath282 using that each @xmath283 is supported in @xmath264 , we have reduced our problem to showing that @xmath284 we proceed to prove the claim in . fix @xmath283 satisfying the assumption in . given @xmath285 we need to prove that @xmath286 . consider the fundamental solution @xmath287 where @xmath288 is the volume of the unit ball in @xmath11 . note that there exists @xmath289 so that @xmath290 for all @xmath291 therefore , for @xmath292 fixed , @xmath293 and @xmath294 are locally integrable in @xmath11 . in particular , @xmath295 and @xmath296 are integrable on the product @xmath297 , where the @xmath298 s are as in . also , note that @xmath299 by these observations , and since @xmath264 has measure zero , we may apply fubini to get @xmath300 where @xmath301 the claim that @xmath286 follows from the fact that @xmath302 for @xmath303 . to see this , let @xmath304 be large enough so that @xmath305 . then , for @xmath306 , the map @xmath307 is in @xmath308 . applying proposition [ e : approximation ] we know that there exists a sequence @xmath309 for which @xmath310 hence , by the assumption in , for each @xmath306 @xmath311 now , the integral defining @xmath312 converges absolutely for @xmath313 and defines an analytic function of @xmath110 in this set . since @xmath312 vanishes for @xmath306 , and @xmath314 is connected , it follows that @xmath315 as claimed . we will give a detailed proof theorem [ main theorem r b ] in @xmath316 since in this setting it is easier to visualize how the argument works . in section [ higher dimensions ] we explain the modifications one needs to carry in order for the same argument to hold in @xmath11 . + let @xmath317 be defined as @xmath318 its nodal domains consist of a collection of cubes whose vertices lie on the grid @xmath319 . throughout this note the cubes are considered to be closed sets , so faces and vertices are included . we say that a cube is positive ( resp . negative ) if @xmath247 is positive ( resp . negative ) when restricted to it . we define the collection @xmath320 of all sets @xmath187 that are built as a finite union of cubes with the following two properties : + * engulf operation . * let @xmath324 . we proceed to define the engulf " operation as follows . we define @xmath325 to be the set obtained by adding to @xmath326 all the negative cubes that touch @xmath326 , even if they share only one point with @xmath326 . by construction @xmath327 . if @xmath328 , the set @xmath325 is defined in the same form only that one adds positive cubes to @xmath326 . in this case @xmath329 . + * join operation . * given @xmath330 we distinguish two vertices using the lexicographic order . namely , for any set of vertices @xmath331 , for @xmath332 we set @xmath333 in the same way we define @xmath334 replacing the minimum function above by the maximum one . for @xmath330 , let @xmath335 be the set of vertices of cubes in @xmath326 . we then set @xmath336 given the vertex @xmath337 we define the edge @xmath338 to be the edge in @xmath339 that has vertex @xmath337 and is parallel to the @xmath110-axis . the edge @xmath340 is defined in the same way . we may now define the join " operation . given @xmath341 and @xmath342 we define @xmath343 as follows . let @xmath344 be the translated copy of @xmath345 for which @xmath346 coincides with @xmath347 . we join " @xmath348 and @xmath345 as @xmath349 + _ definition of the rough nested domains . _ let @xmath353 . a rooted tree is characterized as a finite set of nodes @xmath354 satisfying that @xmath355 to shorten notation , if @xmath356 is a node with @xmath357 children , we denote the children by @xmath358 . given a tree @xmath47 we associate to each node @xmath359 a structure @xmath360 defined as follows . if the node @xmath356 is a leaf , then @xmath361 is a cube of the adequate sign . for the rest of the nodes we set @xmath362 where @xmath357 is the number of children of the node @xmath363 . it is convenient to identify the original structures @xmath364 with the translated ones @xmath365 that are used to build @xmath361 . after this identification , @xmath366 * we let @xmath370 be the set of edges in @xmath371 through which the structures @xmath372 are joined . we will take these edges to be open . that is , the edges in @xmath370 do not include their vertices . + * we let @xmath373 be the set of edges in @xmath374 that are not in @xmath370 . here @xmath375 is the surface @xmath376 if @xmath363 is a leaf , we set @xmath377 . all the edges in @xmath373 are taken to be closed ( so they include the vertices ) . + * we let @xmath378 be the set of edges that connect @xmath379 with @xmath380 for some @xmath381 . if @xmath363 is a leaf , then we set @xmath382 . note that if @xmath356 , and @xmath383 , then @xmath384 is the set of positive cubes that are in the bounded component of @xmath379 and touch @xmath385 . also , if a negative cube in @xmath386 is touching @xmath369 , then it does so through an edge in @xmath387 . + given a node @xmath363 with children @xmath388 , let @xmath389 be the set of edges in @xmath390 . it is clear that for each @xmath391 the set @xmath389 is connected . also , @xmath392 . since the edges in @xmath370 are open , the structures @xmath373 are connected . we proceed to define a perturbation @xmath393 , where @xmath394 we note that by construction @xmath264 is formed by all the edges in @xmath395 . also , it is important to note that if two adjacent cubes have the same sign , then they share an edge in @xmath264 . the function @xmath224 is defined by the rules @xmath396 , @xmath192 and @xmath326 below . + * _ perturbation on @xmath373 . _ let @xmath356 and assume @xmath397 . we define @xmath224 on every edge of @xmath373 to be @xmath4 . if @xmath398 , we define @xmath224 on every edge of @xmath373 to be @xmath399 . + rule @xmath396 is meant to separate @xmath369 from all the exterior cubes of the same sign that surround it . note that for all @xmath356 we have @xmath400 , where @xmath401 is any of the children of @xmath363 , so rule a is well defined . + * _ perturbation on @xmath402 . _ let @xmath403 be an edge in @xmath378 . then , we already know that @xmath224 is @xmath4 on one vertex and @xmath399 on the other vertex . we extend @xmath224 smoothly to the entire edge @xmath403 so that it has a unique zero at the midpoint of @xmath403 , and so that the absolute value of the derivative of @xmath224 is @xmath404 . we also ask for the derivative of @xmath224 to be @xmath73 at the vertices . for example , if the edge is @xmath405\}$ ] where @xmath406 , we could take @xmath407 . + rule @xmath192 is enforced to ensure that no holes are added between edges that join a structure @xmath369 with any of its children structures @xmath408 . + next , assume @xmath409 . note that for any edge @xmath403 in @xmath370 we have that the function @xmath224 takes the value @xmath4 at their vertices , since those vertices belong to edges in @xmath410 and the function @xmath224 is defined to be @xmath4 on @xmath410 . we have the same picture if @xmath411 , only that @xmath224 takes the value @xmath399 on the vertices of all the joining edges . we therefore extend @xmath224 to be defined on @xmath403 as follows . + * _ perturbation on @xmath412 . _ let @xmath356 and assume @xmath397 . given an edge in @xmath370 we already know that @xmath224 takes the value @xmath4 at the vertices of the edge . we extend @xmath224 smoothly to the entire edge so that it takes the value @xmath399 at the midpoint of the edge , and so that it only has two roots at which the absolute value of the derivative of @xmath224 is @xmath404 . we further ask @xmath224 to have zero derivative at the endpoints of the edge . for example , if the edge is @xmath413\}$ ] where @xmath406 , we could take @xmath414 . in the case in which @xmath411 we need @xmath224 to take the value @xmath415 at the midpoint of the edge . + [ extension ] by construction the function @xmath224 is smooth in the interior of each edge . furthermore , since we ask the derivative of @xmath224 to vanish at the vertices in @xmath264 , the function @xmath224 can be extended to a function @xmath417 where @xmath418 is an open neighborhood of @xmath264 . [ perturbation ] given a tree @xmath47 , let @xmath417 be defined following rules a , b and c and remark [ extension ] , where @xmath418 is an open neighborhood of @xmath264 . since @xmath264 is compact and @xmath419 is connected , theorem [ t : perturbation ] gives the existence of @xmath420 that satisfies @xmath421 for @xmath156 small set @xmath422 we will show in lemma [ limit ] that the perturbation was built so that the nodal domain of @xmath423 corresponding to @xmath356 is constituted by the deformed cubes in @xmath424 after the perturbation is performed . we illustrate how rules a , b , and c work in the following examples . in what follows we shall use repeatedly that the singularities of the zero set of @xmath247 are on the edges and vertices of the cubes . therefore , the changes of topology in the zero set can only occur after perturbing the function @xmath247 along the edges and vertices of the cubes . + * example 1 . * as an example of how rules a and b work , we explain how to create a domain that contains another nodal domain inside of it . the tree corresponding to this picture is given by two nodes , @xmath4 and @xmath425 , that are joined by an edge . we start with a positive cube @xmath426 and work with its engulfment @xmath427 . all the edges of @xmath428 belong to @xmath429 . therefore , the function @xmath423 takes the value @xmath430 on @xmath429 . also , all the positive cubes that touch @xmath428 do so through an edge in @xmath429 . it follows that all the positive cubes surrounding @xmath428 are disconnected from @xmath428 after the perturbation is performed . the cube @xmath428 then becomes a positive nodal domain @xmath431 of @xmath423 that is contractible to a point . next , note that all the negative cubes that touch @xmath428 ( i.e. , cubes in @xmath432 ) do so through a face whose edges are in @xmath429 , or through a vertex that also belongs to one of the edges in @xmath429 . therefore , all the negative cubes are glued together after the perturbation is performed , and belong to a nodal domain @xmath433 that contains the connected set @xmath429 . so far we have seen that @xmath433 contains the perturbation of the cubes in @xmath432 . we claim that no other cubes are added to @xmath433 . indeed , all the negative cubes that touch the boundary of @xmath434 do so through edges in @xmath435 . then , since @xmath423 takes the value @xmath436 on @xmath435 , all the surrounding negative cubes are disconnected from @xmath437 after we apply the perturbation . since along the edges connecting @xmath438 with @xmath439 the function @xmath423 has only one sign change ( it goes from @xmath430 to @xmath436 ) it is clear that @xmath433 can be retracted to @xmath440 . + * example 2 . * here we explain how rule c works . suppose we want to create a nodal domain that contains two disjoint nodal domains inside of it . the tree corresponding to this picture is given by three nodes , @xmath4 , @xmath425 , and @xmath441 . the node @xmath4 is joined by an edge to @xmath425 and by another edge to @xmath441 . assume that @xmath428 and @xmath442 belong to @xmath320 . then , @xmath443 . when each of the structures @xmath444 or @xmath444 are perturbed , we get a copy of the negative nodal domain in example 1 . since in @xmath348 the structures @xmath444 and @xmath444 are joined by an edge , the two copies of @xmath433 will also be glued . the reason for this is that the function @xmath423 takes the value @xmath430 in the middle point of the edge joining @xmath444 and @xmath444 . therefore , a small negative tube connects both structures . in this section we explain what our perturbation does to the zero set of @xmath247 at a local level . given a tree @xmath47 , and @xmath156 , let @xmath445 be defined as in definition [ perturbation ] . using that @xmath2 is a continuous function , and that we are working on a compact region of @xmath11 ( we call it @xmath270 ) , it is easy to see that there exists a @xmath446 , so that if @xmath447 is the @xmath448-tubular neighborhood of @xmath264 , then @xmath423 has no zeros in @xmath449 as long as @xmath450 and @xmath451 where @xmath452 is some positive constant that depends only on @xmath453 . this follows after noticing that @xmath454 takes the value @xmath4 at the center of each cube and decreases radially until it takes the value @xmath73 on the boundary of the cube . the construction of the tubular neighborhood @xmath447 yields that in order to understand the behavior of the zero set of @xmath423 we may restrict ourselves to study it inside @xmath447 for @xmath450 . we proceed to study the zero set of @xmath423 in a @xmath448-tubular neighborhood of each edge in @xmath264 . assume , without loss of generality , that the edge is the set of points @xmath455\}$ ] . + * vertices . * at the vertex @xmath456 the function @xmath224 takes the value @xmath4 or @xmath399 . assume @xmath457 ( the study when the value is @xmath4 is identical ) . in this case , we claim that the zero set of @xmath458 near the vertex is diffeomorphic to that of the function @xmath459 provided @xmath448 ( and hence @xmath460 ) is small enough . to see this , for @xmath461 set @xmath462 to be one of the connected components of @xmath463 intersected with @xmath447 . we apply the version of thom s isotopy theorem given in [ ep , theorem 3.1 ] which asserts that for every smooth function @xmath129 satisfying @xmath464 there exists a diffeomorphism @xmath465 making @xmath466 we observe that the statement of [ ep , theorem 3.1 ] gives the existence of an @xmath467 so that the diffeomorphism can be built provided @xmath468 . however , it can be tracked from the proof that @xmath469 can be chosen to be as in the rhs of . applying [ ep , theorem 3.1 ] to the function @xmath470 we obtain what we claim provided we can verify . first , note that @xmath471 . it is then easy to check that @xmath472 for some @xmath473 depending only on @xmath474 . next , we find a lower bound for the gradient of @xmath423 when restricted to the zero set @xmath475 . note that for @xmath476 we have @xmath477 on the other hand , since @xmath478 for all @xmath479 , we conclude @xmath480 whenever @xmath448 is small enough . this shows that at each vertex where @xmath224 takes the value @xmath399 the negative cubes that touch the vertex are glued together while the positive ones are disconnected . + * edges . * having dealt with the vertices we move to describe the zero set of the perturbation near a point inside the edge . there are three cases . in the first case ( case a ) the perturbation @xmath224 is strictly positive ( approx . @xmath436 ) or strictly negative ( approx @xmath430 ) along the edge . in the second case ( case b ) the perturbation @xmath2 is strictly positive ( approx . @xmath436 ) at one vertex and strictly negative ( approx . @xmath430 ) at the other vertex . in the third case ( case c ) , the edge is joining two adjacent structures so the perturbation @xmath2 takes the same sign at the vertices ( it is approx . @xmath436 ) and the opposite sign ( it is approx . @xmath436 ) at the midpoint of the edge having only two zeros along the edge . + in * case a * the zero set of @xmath458 near the edge is diffeomorphic to the zero set of the map @xmath484 . the proof of this claim is the same as the one given near the vertices , so we omit it . in the picture below the first figure shows the zero set of @xmath247 near the edge while the second figure shows the zero set of @xmath470 . this shows that two cubes of the same sign , say negative , that are connected through an edge are going to be either glued if the perturbation takes the value @xmath399 along the edge , or disconnected if the perturbation takes the value @xmath415 along the edge . + in * case b * , it is clear that the only interesting new behavior will occur near the points on the edge at which the function @xmath2 vanishes . since @xmath485 and @xmath486 , there is only one point at which @xmath2 vanishes ; say the point is @xmath487 . note that @xmath2 was built so that @xmath487 is the only zero of @xmath2 along the edge . we claim that the zero set of @xmath423 near @xmath487 is diffeomorphic to the zero set of the map @xmath488 . the proof of this claim is similar to the one given near the vertices , so we omit it . the only relevant difference is that in order to bound @xmath489 from below , one uses that @xmath490 , and that @xmath491 in a ball of radius @xmath492 centered at @xmath487 while @xmath493 . of course , if one is away from the value @xmath494 , then the analysis is the same as that of case a. the first figure in the picture below shows the zero set of @xmath247 along the edge while the second figure shows the zero set of @xmath470 when @xmath495 . this shows that two consecutive cubes sharing an edge along which the perturbation changes sign will be glued on one half of the edge and disconnected along the other half . + in * case c * , the zero set of @xmath423 is diffeomorphic to that of @xmath496 where @xmath2 satisfies @xmath485 and @xmath497 and @xmath498 . the zero set of @xmath470 when @xmath499 is plotted in the figure below . given a tree @xmath47 and @xmath156 we continue to work with @xmath500 as defined in definition [ perturbation ] . fix @xmath356 , and suppose it has @xmath357 children . assume without loss of generality that @xmath411 . for every @xmath501 the perturbed function @xmath423 takes the value @xmath436 on @xmath502 , and @xmath502 is connected . it follows that for each @xmath503 there exists a positive nodal domain @xmath504 of @xmath423 that contains @xmath502 . we define the set @xmath505 as @xmath506 throughout this section we use the description of the local behavior of @xmath475 that we gave in section [ local ] . in the following lemma we prove that @xmath507 is a nodal domain of @xmath423 . [ nodal d ] let @xmath47 be a tree and for each @xmath156 let @xmath423 be the perturbation defined in . then , for each @xmath156 and @xmath356 , the set @xmath508 defined in is a nodal domain of @xmath423 . let @xmath356 and suppose @xmath363 has @xmath357 children . assume without loss of generality that @xmath397 . by definition , @xmath509 where @xmath504 is the nodal domain of @xmath423 that contains @xmath502 . to prove that @xmath507 is itself a nodal domain , we shall show that @xmath510 for all @xmath511 . the edge @xmath514 is shared by a cube @xmath517 and a cube @xmath518 . note that every cube in @xmath512 has at least one vertex that belongs to an edge in @xmath502 ( same with @xmath513 ) . let @xmath519 be a vertex of @xmath520 that belongs to an edge in @xmath502 . in the same way we choose @xmath521 to be a vertex in @xmath522 that belongs to an edge in @xmath523 . in particular , by rule a we have that @xmath524 and @xmath525 . first , we show that all the cubes in @xmath535 glue to form part of @xmath507 after the perturbation is performed . assume , without loss of generality , that @xmath411 . then , @xmath536 for every child @xmath401 of @xmath363 . all the cubes in @xmath535 have an edge in @xmath502 . since such cubes are positive , and @xmath423 takes the value @xmath436 on @xmath502 , it follows that the cubes become part of the nodal domain that contains @xmath502 . that is , all the cubes in @xmath535 become part of @xmath507 after the perturbation is added to @xmath247 . second , we show that no cubes , other than those in @xmath537 , will glue to form part of @xmath507 . indeed , any other positive cube in @xmath538 that touches @xmath539 does so through an edge in @xmath540 . since the function @xmath423 takes the value @xmath430 on @xmath540 , those cubes will disconnect from @xmath541 after we perturb . on the other hand , any positive cube in @xmath542 is touching @xmath543 through edges in @xmath544 where @xmath545 is the number of children of @xmath401 . since @xmath2 takes the value @xmath430 on @xmath544 , the cubes in @xmath546 will also disconnect from @xmath537 . it is convenient to define the partial collections of nested domains . given a tree @xmath47 , a perturbation @xmath423 , and @xmath356 , we define the collection @xmath547 of all nodal domains that are descendants of @xmath507 as follows . if @xmath363 is a leaf then @xmath548 . if @xmath363 is not a leaf and has @xmath357 children , we set @xmath549 we will use throughout this section that we know how the zero set behaves at a local scale ( as described in section [ local ] ) . let @xmath47 be a tree and for each @xmath156 let @xmath423 be the perturbation defined in . we shall prove that there is a subset of the nodal domains of @xmath423 that are nested as prescribed by @xmath47 . since for every @xmath356 the set @xmath507 is a nodal domain of @xmath423 , the theorem would follow if we had that for all @xmath356 statements ( i ) , ( ii ) and ( iii ) imply that @xmath555 has @xmath556 components . one component is unbounded , and each of the other @xmath357 components is filled by @xmath557 for some @xmath558 . we prove statements ( i ) , ( ii ) and ( iii ) by induction . the statements are obvious for the leaves of the tree . + the proof of claim ( iii ) actually shows that @xmath507 can be retracted to the arc connected set @xmath559 where @xmath560 is the curve introduced in lemma [ nodal d ] connecting @xmath561 with @xmath523 that passes through the midpoint of the edge joining @xmath512 with @xmath513 . assume without loss of generality that @xmath411 . then , for every child @xmath401 , all the faces in @xmath565 belong to cubes in @xmath408 that are negative . also , all the other negative cubes in @xmath566 that touch @xmath565 do so through an edge in @xmath502 . since the function @xmath423 takes the value @xmath436 on @xmath502 , all the negative cubes in @xmath567 are disconnected from those in @xmath566 after the perturbation is performed . while all the negative cubes touching @xmath567 are disconnected , an open positive layer @xmath568 that surrounds @xmath569 is created . the layer @xmath568 contains the grid @xmath502 and so it is contained inside @xmath507 . the result follows from setting @xmath570 . + _ proof of claim ( ii)_. this is a consequence of how we proved the statement ( i ) since both @xmath557 and @xmath571 are surrounded by a positive layer inside @xmath572 . + _ proof of claim ( iii)_. note that @xmath573 and that by the induction assuption @xmath574 has no bounded components . on the other hand , we also have that @xmath575 this shows that , in order to prove that @xmath554 has no bounded components , we should show that the cubes in @xmath576 glue to those in @xmath577 leaving no holes . note that all the cubes in @xmath576 are attached to the mesh @xmath578 through some faces or vertices . assume without loss of generality that @xmath411 . for each @xmath381 the layer @xmath568 is contained in @xmath507 and all the cubes in @xmath384 are glued to the layer thorugh an entire face or vertex . the topology of @xmath507 will depend exclusively on how the cubes in @xmath579 will join or disconnect each other along the edges that start at @xmath502 and end at a distance @xmath4 from @xmath502 . the function @xmath423 takes the value @xmath436 on @xmath502 . also , note that if a pair of positive cubes in the unbounded component of @xmath580 share an edge @xmath403 that starts at @xmath561 and ends at a distance @xmath4 from it , then the end vertex belongs to @xmath373 , and the function @xmath423 takes the value @xmath430 at this point . since the function @xmath423 has only one root on @xmath403 , we have that no holes are added to @xmath507 when applying the perturbation to those two cubes . for cubes in the bounded component that share an edge one argues similarly and uses the value of @xmath423 on @xmath581 where @xmath545 is the number of children of @xmath401 . to finish , we note that two consecutive structures @xmath512 and @xmath513 are joined through an edge separating two cubes as shown in figure [ figure join ] . the function @xmath423 is negative ( approximately equal to @xmath430 ) at the vertices of the edge , and is positive at the middle point ( approximately equal to @xmath582 ) . since along the edge @xmath423 was prescribed to have only two roots , no holes are introduced when joining the structures . the argument in higher dimensions is analogue to the one in dimension 3 . we briefly discuss the modifications that need to be carried in this setting . let @xmath583 we will work with cubes in @xmath11 that we identify with a point @xmath584 . that is , the cube corresponding to @xmath585 is given by @xmath586\}$ ] . as before , we say that a cube is positive ( resp . negative ) if @xmath247 is positive ( resp . negative ) when restricted to it . the collection of faces of the cube @xmath12 is @xmath587\;\ ; \forall k \neq i\}$ ] . the collection of edges is @xmath588 where each edge is described as the set @xmath589 \;\ ; \forall k \neq i , j\}.\ ] ] we note that if two cubes of the same sign are adjacent , then they are connected through an edge or a subset of it . in analogy with the @xmath316 case , we define the collection @xmath320 of all sets @xmath187 that are built as a finite union of cubes with the following two properties : + * engulf operation . * let @xmath324 . we define @xmath325 to be the set obtained by adding to @xmath326 all the negative cubes that touch @xmath326 , even if they share only one point with @xmath326 . by construction @xmath327 . if @xmath328 , the set @xmath325 is defined in the same form only that one adds positive cubes to @xmath326 . in this case @xmath329 . + * join operation . * given @xmath330 we distinguish two vertices using the lexicographic order . for @xmath330 , let @xmath593 be the set of its vertices . we let @xmath337 be the largest vertex in @xmath594 and @xmath595 be the smallest vertex in @xmath594 . given the vertex @xmath337 we define the edge @xmath338 to be the edge in @xmath339 that contains the vertex @xmath337 and is parallel to the hyperplane defined by the @xmath596 coordinates . the edge @xmath340 is defined in the same way . given @xmath341 and @xmath342 we define @xmath343 as follows . let @xmath344 be the translated copy of @xmath345 for which @xmath346 coincides with @xmath347 . we join " @xmath348 and @xmath345 as @xmath597 + * definition of the rough nested domains . * given a tree @xmath47 we associate to each node @xmath359 a structure @xmath599 defined as follows . if the node @xmath356 is a leaf , then @xmath361 is a cube of the adequate sign . for the rest of the nodes we set @xmath600 where @xmath357 is the number of children of the node @xmath363 . we continue to identify the original structures @xmath364 with the translated ones @xmath365 that are used to build @xmath361 . after this identification , @xmath601 + * building the perturbation . * let @xmath356 be a node with @xmath357 children . we define the sets of edges @xmath410 , @xmath378 and @xmath370 in exactly the same way as we did in @xmath316 ( see section [ s : perturbation ] ) . we proceed to define a perturbation @xmath393 , where @xmath394 the function @xmath224 is defined by the rules @xmath396 , @xmath192 and @xmath326 below . + let @xmath602 \to [ -1,1]$ ] be a smooth increasing function satisfying @xmath603 we also demand @xmath604 * _ perturbation on @xmath373 . _ let @xmath356 and assume @xmath397 . we define @xmath224 on every edge of @xmath373 to be @xmath4 . if @xmath398 , we define @xmath224 on every edge of @xmath373 to be @xmath399 . + * _ perturbation on @xmath402 . _ let @xmath605 be an edge that touches both @xmath373 and @xmath606 for some of the child structures @xmath607 of @xmath369 . assume @xmath608 . then we know that we must have @xmath609 and @xmath610 . let @xmath611 be the set of directions in @xmath605 that connect @xmath373 and @xmath606 . we let @xmath612\ ] ] be defined as @xmath613 with this definition , since whenever @xmath614 we have @xmath615 for all @xmath616 , we get @xmath617 . also , whenever @xmath618 we have that there exists a coordinate @xmath619 for which @xmath620 . then , @xmath621 and so @xmath622 . note that @xmath224 vanishes on the sphere @xmath623 and that @xmath624 on @xmath625 because of . if @xmath398 , simply multiply @xmath626 by @xmath399 . + * _ perturbation on @xmath412 . _ let @xmath356 and assume @xmath397 . we set @xmath627 where @xmath628 ranges over the indices @xmath629 . with this definition , whenever @xmath110 is at the center of the edge @xmath630 we have @xmath631 . also , if @xmath632 we have @xmath633 for some @xmath634 , and so @xmath622 . also note that @xmath224 vanishes on a sphere of radius @xmath635 centered at the midpoint of @xmath630 and that the gradient of @xmath224 does not vanish on the sphere because of . if @xmath398 , simply multiply @xmath626 by @xmath399 [ extension2 ] by construction the function @xmath224 is smooth in the interior of each edge . furthermore , since according to we have @xmath636 and @xmath637 , the gradient of @xmath224 vanishes on the boundaries of the edges in @xmath264 . therefore , the function @xmath224 can be extended to a function @xmath417 where @xmath185 is an open neighborhood of @xmath264 . given a tree @xmath47 , let @xmath417 be defined following rules a , b and c and remark [ extension2 ] , where @xmath185 is an open neighborhood of @xmath264 . since @xmath264 is compact and @xmath253 is connected , theorem [ t : perturbation ] gives the existence of @xmath638 that satisfies @xmath421 for @xmath156 small set @xmath422 the definitions in rules a , b and c are the analogues to those in dimension @xmath639 . for example , when working in dimension @xmath639 on the edge @xmath640\}$ ] , we could have set @xmath641 and @xmath642 if two adjacent cubes are connected through a subset of @xmath643 , then the cubes will be either glued or separated along that subset . this is because the function @xmath2 is built to be strictly positive ( approx . @xmath436 ) or strictly negative ( approx . @xmath430 ) along the entire edge . if two adjacent cubes share an edge through which two structures are being joined , then they will be glued to each other near the midpoint of the edge . this is because @xmath2 is built so that it has the same sign as the cubes in an open neighborhood of the midpoint of the joining edge . if two adjacent cubes in @xmath369 of the same sign share a subset of an edge in @xmath644 , then with the same notation as in rule b , there exists a subset of directions @xmath645 so that the set @xmath646 \ ; \forall t=1 , \dots , s \}$ ] is shared by the cubes . by construction , the cubes will be glued through the portion @xmath647 of @xmath648 that joins @xmath649 with the point @xmath650 near the midpoint @xmath651 , while being disconnected through the portion @xmath652 of @xmath648 that joins the point @xmath650 with @xmath653 . this is because @xmath2 is prescribed to have the same sign as the cubes along @xmath647 , while taking the opposite sign of the cubes along @xmath652 . let @xmath654 , with @xmath655 . running a similar argument to the one given in @xmath316 one obtains that all the cubes in @xmath656 will glue to form a negative nodal domain @xmath507 of @xmath423 . we sketch the argument in what follows . all the negative cubes in @xmath657 that touch @xmath369 do so through an edge in @xmath373 since they will be at distance @xmath4 from the children structures @xmath658 . since the perturbation @xmath2 takes a strictly positive value ( approx . @xmath582 ) along any edge in @xmath373 , the negative cubes in @xmath657 will be separated from those in in @xmath659 . simultaneously , for each @xmath129 , all the cubes in @xmath660 are glued to each other since they are negative cubes that touch @xmath661 and @xmath661 is a connected set on which the perturbation @xmath2 takes a strictly negative value ( approx . @xmath430 ) . this gives that @xmath661 belongs to a negative nodal domain of @xmath423 , and that the negative cubes in @xmath660 are glued to the nodal domain after the perturbation is performed . furthermore , two consecutive structures @xmath662 and @xmath663 are joined through an edge in @xmath402 . this edge , which joins a negative cube in @xmath662 and a negative cube in @xmath663 has its boundary inside @xmath661 . since @xmath2 is strictly positive ( approx . @xmath582 ) on @xmath661 , we know that the parts of the two cubes that are close to the boundary will be disconnected . however , since the perturbation was built so that @xmath2 is strictly negative ( approx . @xmath430 ) at the midpoint of the edge , both negative cubes are glued to each other . in fact , one can build a curve @xmath664 contained inside the nodal domain that joins @xmath665 with @xmath666 . it then follows that all the cubes in @xmath667 are glued to each other after the perturbation is performed and they will form the nodal domain @xmath507 of @xmath423 containing @xmath668 . one can carry the same stability arguments we presented in section [ local ] to obtain that at a local level there are no unexpected new nodal domains . for this to hold , as in the @xmath316 case , the argument hinges on the fact that in the places where both @xmath247 and @xmath2 vanish , the gradient of @xmath2 is not zero ( as explained at the end of each rule ) . finally , rule b is there to ensure that the topology of each nodal domain is controlled in the sense that when the cubes in @xmath656 glue to each other they do so without creating unexpected handles . indeed , the cubes in @xmath656 can be retracted to the set @xmath669 where @xmath670 and @xmath671 are the children of @xmath672 . the argument we just sketched also shows that the nodal domains @xmath507 with @xmath356 are nested as prescribed by the tree @xmath47 . indeed , claims ( i ) , ( ii ) and ( iii ) in the proof of theorem [ main theorem r b ] are proved in @xmath11 in exactly the same way we carried the argument in @xmath316 .